Felix Candela Concrete Shell Structures

Finite-Element Analysis of Felix Candela’s Chapel of Lomas
de Cuernavaca
Powell Draper1; Maria E. Moreyra Garlock2; and David P. Billington, Hon.M.ASCE3
Abstract: The structures of Felix Candela stand as elegant examples of the art of thin shell concrete structural design. However, few thin
shell structures are built today and many of the most enduring structures, such as those of Candela, have not been analyzed. This project
Downloaded from ascelibrary.org by UNIVERSITE LAVAL on 03/04/15. Copyright ASCE. For personal use only; all rights reserved.
performed a finite-element analysis of Candela’s Cuernavaca Chapel in order to better understand its structural efficiency and gain insight
into its design. Analysis results show that Candela’s striking structure has very low stresses resulting from its own self-weight.
DOI: 10.1061/共ASCE兲1076-0431共2008兲14:2共47兲
CE Database subject headings: Finite element method; Thin shell structures; Structural analysis; Hyperbolic parabolic shells.
Introduction tures in order to appreciate and understand his works.

In this project, we performed a finite-element analysis of the
Felix Candela designed and built a number of thin shell concrete Chapel of Lomas de Cuernavaca. By examining the results of our
structures in and around Mexico City during the 1950s and 1960s. analysis of the structure under its own dead load, we have at-
Nearly all of these structures still stand in good condition after tempted to model, analyze, and understand the structural behavior
50 years of continuous use often under harsh seismic loads. Can- of the shell as its designer did. We performed a study that evalu-
dela employed the form of a hyperbolic paraboloid for many of ates the stresses and deflections of the Cuernavaca shell using
these structures, the strength of which allowed for a remarkable three different models of differing complexity and precision: 共1兲 a
thinness. shell with uniform thickness and no beams on the free edge; 共2兲 a
Despite the low cost, low maintenance, and attractiveness of shell with uniform thickness and a beam on the free edge of the
Candela’s structures, few thin shell concrete structures are built low-rise side; and 共3兲 a shell with varying thicknesses and a beam
today. One of the reasons suggested for their absence is a percep- on the free edge of the low-rise side. The third model is the
tion of difficulty of their design and analysis 共Meyer and Sheer closest approximation to the built structure. Ultimately we seek to
2005兲. However, a few detailed structural analyses have been per- confront the issue of the difficulty of design and analysis of thin
formed on the successful designs of the past. For example, an shell concrete structures in order to determine methods to better
evaluation of Candela’s Chapel of Lomas de Cuernavaca 共shown understand them.
in Fig. 1兲 was published in the Mexican architectural journal Bi-
tacora Arquitectura in 2001 共Otero 2001兲. The article included
images of the results of an analysis done using the finite-element Chapel of Lomas de Cuernavaca
program ANSYS. However, neither the images nor the article
mentions units or specific load cases used in the analysis and it The Chapel of Lomas de Cuernavaca, or Cuernavaca Chapel 共Fig.
did not include a detailed description of its behavior or design. 1兲 is located on a plateau overlooking the town of Cuernavaca,
Alternately, engineers could study Candela’s own calculations Morelos Mexico. For its design, Candela, collaborating with the
in an attempt to understand his methods and their mechanics, as architects Guillermo Rosell and Manuel Larrosa, employed a dra-
has been done with the work of Anton Tedesko, another master matically modified saddle shape hyperbolic paraboloid. The taller
builder in the medium of thin shell structures 共Saliklis et al. free edge at the front soars to a height of 21.9 m 共72 ft兲. At its
2007兲. Unfortunately, though, few of Candela’s calculations sur- widest point, the shell spans 31 m 共102 ft兲共Fig. 2兲. To account for
vive. We therefore instead modeled and analyzed one of his struc- the asymmetry, Candela added an edge beam to the opposite back
side of the shell. The shell’s thickness is primarily 4 cm 共1.6 in.兲
1
Graduate Student, Dept. of Civil and Environmental Engineering, throughout, but it is thickened along the sides. Specific details of
Princeton Univ., Princeton, NJ 08544 共corresponding author兲. E-mail: these thickened regions have not been found, other than first-hand
pdraper@princeton.edu observations of the structure by the first writer. For example, at
2
Assistant Professor, Dept. of Civil and Environmental Engineering, the base of the taller opening, the shell was measured to be 52 cm
Princeton Univ., Princeton, NJ 08544.
3
wide 共20 in.兲.
Gordon Y.S. Wu Professor, Dept. of Civil and Environmental Engi- There has been some reference made to the partial failure of
neering, Princeton Univ., Princeton, NJ 08544. the Cuernavaca Chapel during construction. The 2001 article in
Note. Discussion open until November 1, 2008. Separate discussions
Bitacora Arquitectura refers to “Los análisis demostraron que el
must be submitted for individual papers. To extend the closing date by
one month, a written request must be filed with the ASCE Managing fallo al descimbrar se debió a un problema constructivo y no de
Editor. The manuscript for this paper was submitted for review and pos- cálculo” 共Otero 2001兲, or “The analyses demonstrate that the col-
sible publication on March 2, 2007; approved on August 17, 2007. This lapse during decentering was due to a problem related to a con-
paper is part of the Journal of Architectural Engineering, Vol. 14, No. struction problem, not design.” This somewhat cryptic reference
2, June 1, 2008. ©ASCE, ISSN 1076-0431/2008/2-47–52/$25.00. apparently refers to a partial collapse of the taller free edge of the
JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008 / 47
J. Archit. Eng. 2008.14:47-52.

Fig. 3. Hyperbolic paraboloid
his engineering knowledge and expertise as a builder to select

Fig. 1. Chapel of Lomas de Cuernavaca, by Felix Candela, 1958– those forms that were rational structural solutions.
1959 共with permission from Princeton Maillart Archives兲 For Candela, the doubly curved hyperbolic paraboloid 共Fig. 3兲
had two primary attributes that made it an ideal shape for a thin
shell concrete structure. The first was its ease of construction.
shell during construction. Whether Candela was experimenting Designed properly, thin shell concrete structures have proven du-
with the form during construction is unclear. However, the current rable, lasting, and economical over long periods of time. How-
structure is still in good condition and displays no threat of fail- ever, their initial construction costs can appear prohibitive if they
ure, so this partial collapse during construction does not affect our require extensive specialized formwork. Candela realized that one
assessment of the structure. of the unique properties of the geometry of the doubly curved
hyperbolic paraboloid is that it can be generated from straight
lines 共as shown in Fig. 3兲 and hence constructed from straight
Hyperbolic Paraboloid formwork, eliminating extra expense from customized, curving
formwork 共Fig. 4兲.
As a designer of shell structures, Candela did not arbitrarily pick The other benefit of the hyperbolic paraboloid is that in Can-
forms that seemed attractive or adventurous. Rather, Candela used dela’s parlance it is a proper shell form 共Candela 1958兲. By this
Fig. 2. Dimensions of Cuernavaca Chapel 共in ms兲; letters represent reference locations in Table 2
48 / JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008
J. Archit. Eng. 2008.14:47-52.

Fig. 4. Straight formwork of Cuernavaca Chapel 共with permission
from Princeton Maillart archives兲
he meant the form could distribute stresses without excessive

bending arising in the shell. Bending is anathema to thin shells.
When large moments arise in a shell structure, greater thickness is
required to assume and distribute the load without excessive de-
formation, cracking, or failure. Other shell designers have often
employed large ridge or edge beams in their shell structures to
counteract bending moments that arise. But this increase in thick-
ness has been seen by some designers as contrary to the very
nature of a thin shell structure. Also, ribs or edge beams obscure
a structure’s display of its thinness, which Candela valued. Can-
dela therefore instead designed forms that through their very
shape would minimize the bending in the shell. These he referred
to as proper shells.
The general equation of the hyperbolic paraboloid 共Fig. 5兲 is Fig. 6. Finite-element model of Cuernavaca Chapel
given by
y 2 x2
z= − 共1兲 values of h1 and h2, and therefore the equation and surface of the
h2 h1 shell 共Draper 2006兲. For this project, the modeling program Rhi-
where noceros 共Robert McNeel & Associates 1993兲 was used to ascribe
the hyperbolic paraboloid surface to the model of the Cuernavaca
a2 Chapel based on its bounding parabolas.
h1 = 共2兲
c1
and
Structural Model
b2
h2 = 共3兲
c2 The finite-element model of the Cuernavaca Chapel is shown in
Fig. 6. This project employed the computer program SAP2000
If we have a set of points 共x , y , z兲 describing a hyperbolic parabo- 共Computers and Structures 2004兲 for finite-element analysis.
loid, we can apply multiple linear regression to determine the Element size on the finite-element model was reduced until
displacements converged. The model uses primarily quadrilateral
thin shell elements 共which neglect transverse shear deformations兲
with a minimum of triangular elements in transition zones or
regions of unique geometry. An analysis of the same model with
thick shell elements 共which include the effects of transverse shear
deformations兲 produces nearly identical results.
The model was analyzed under the effects of dead load only.
The model did not account for the steel reinforcement in the shell,
in actuality, a “doble parrilla,” or double grillage, throughout. It is
assumed that any representation of the reinforcement by, for ex-
ample, a transformed section, would call for adding more thick-
ness to the model. Thus, not including the reinforcement in the
model is considered more conservative for this analysis. Further-
more, the master builders of thin shell structures, such as Candela,
Anton Tedesko, Pier Luigi Nervi, and Heinz Isler, have tradition-
ally designed their reinforced concrete shells by assuming that the
load is resisted by the concrete alone. Thus, analyzing the Cuer-
navaca Chapel without its reinforcing steel adheres more closely
Fig. 5. Hyperbolic paraboloid bounded by parabolas to the original design methodology.
J. Archit. Eng. 2008.14:47-52.

shown in Fig. 8. The forces are in tension at the top of the shell
and compression towards the sides, as statics would predict. From
these forces we could calculate the bending moment at the section
cut.
The results from the finite-element analysis show a moment
along the section cut of 713 kN m 共526 K ft兲. This corresponds to
a 1% difference from the hand calculation of 707 kN m
共521 K ft兲, thus verifying the finite-element model. Calculating
the entire area of the structure 共839.5 m2, or 9,036 ft2兲 and its
corresponding weight, we arrive at a total dead load of 792 kN
共178 K兲, which equals the sum of the base reactions from the
finite-element analysis. We therefore consider the results of this
finite-element analysis of the Cuernavaca Chapel valid.
Fig. 7. Section cut on model of Cuernavaca Chapel
Finite-Element Analysis Results

Validation of Finite-Element Model
The accuracy of the finite-element model of this structure was The force and stress results of the finite-element analysis rein-
first verified through a simplified analysis. The structure was ini- force Candela’s selection of a structurally efficient form. Using
tially modeled with a uniform 4 cm 共1.6 in.兲 thickness. A vertical the uniform shell thickness model, principal shell stresses reach a
section cut was taken from the fixed supports at the base nearest maximum of 4,488 kPa 共651 psi兲 in compression and 1,131 kPa
the taller edge 共Fig. 7兲. The area of the unsupported section 共in 共164 psi兲 in tension. Compressive stress values are well under
front of the supports兲 was determined from the Rhinoceros model what we can assume to be a minimum compressive strength of
of the geometry. Rhinoceros also computed the centroid of this concrete used. Tension is kept low enough so that it can suitably
section. With this information and the thickness of the structure be assumed by the reinforcement in the shell. Maximum defor-
共in the model兲, it is possible to compute the bending moment mation in the model was 0.5 cm 共0.2 in.兲. No significant cracking
created by the self-weight of the unsupported section 共or “over- was observed in the actual structure so we may conclude defor-
hang”兲 at the vertical section cut. mations are kept within reasonable limits.
The area of the shell extending in front of the supports was The highest stresses are found in the structure along the lines
calculated 共from Rhinoceros兲 as 267.7 m2 共2,881.5 ft2兲. The cen- of supports and the back edge of the structure. First hand obser-
troid of this section was calculated at 2.8 m 共9.2 ft兲 from the vation revealed that at these regions, Candela thickened the struc-
section cut. Assuming a concrete density of 2,400 kg/ m3 ture to accommodate the higher stresses 共Figs. 9 and 10兲. The
共150 lb/ ft3兲, the bending moment at the section cut was calcu- model thus serves to provide us with a clearer understanding of
lated to be 707 kN m 共521 K ft兲. The results of the finite-element Candela’s design by demonstrating its efficient distribution of
analysis therefore had to be in agreement with this value. forces and resulting stresses.
The finite-element model of the structure was analyzed under To further investigate the structural response of the Cuerna-
the effect of its own dead load. The results of the analysis of the vaca Chapel, we next altered the finite-element model by first
finite-element model gave joint forces acting at the nodes of each adding the thick edge beam at the shorter lip of the structure and
element. The distribution of joint forces at the section cut is then varying the thickness of the shell. Again, no specific dimen-
Fig. 8. Distribution of joint forces along section cut, plan view 共uniform thickness兲
J. Archit. Eng. 2008.14:47-52.

loid would distribute stresses effectively throughout the shell and
would allow for remarkable thinness, even at the free edge.
Conclusion
Felix Candela’s Chapel of Lomas de Cuernavaca proves an excel-

lent example with which to study the formulation, design, and
construction of thin shell concrete structures. By looking at vari-
ous aspects of Candela and his structure, we can demonstrate the
benefits of thin shell concrete structures.
A structural analysis of Candela’s Cuernavaca Chapel reveals
Fig. 9. High minimum principal stresses 共compression兲 in front edge the genius behind this form. The shape of the hyperbolic parabo-
of Cuernavaca Chapel model and corresponding actual increase in loid minimizes bending in the shell and precludes the need for
thickness ribs or a masking of the thinness. Despite the large size of this
structure, stresses are kept low enough to allow for a remarkable
thinness and free edge.
sions of these details have been found, so approximations were Comparison of three finite-element models of varying com-
made based on measurements observed on the actual structure. A plexity and precision shows that the simplest model, which rep-
comparison of the results may be seen in Tables 1 and 2. The resents the shell with uniform thickness and no edge beams,
model with varying thickness, as the actual structure has, shows a yields conservative stress results. Yet these results indicate that
general reduction of stress from the models with uniform thick- the stresses are well within the strength limits.
ness and a more gradual distribution of stress throughout the The structural analysis also allows for an interpretation of
shell. Candela’s design methodology. The basic form of the hyperbolic
For the model with uniform thickness and an edge beam, prin- paraboloid need not bind the designer into a predetermined ge-
cipal stresses reach a maximum in compression of 4,695 kPa ometry. Within the discipline of the hyperbolic paraboloid, Can-
共681 psi兲 and a maximum of 1,586 kPa 共230 psi兲 in tension. For dela found room for creative design, by maximizing the effect of
the model with varying thickness, principal stresses reach a the free edge with a strikingly tall opening. Candela also under-
maxima of 2,523 kPa 共366 psi兲 in compression and 1,503 kPa stood that the function follows the form, and by stretching the
共218 psi兲 in tension. Thus, in each case, we see that stresses are height he also opened the structure to an increase in dead load and
kept low, verifying the efficiency of Candela’s choice of form. wind load. This he countered by adeptly thickening the edges and
Computational analysis of structures such as the Cuernavaca back of the structure. The result is a structure that adheres to an
Chapel serves to verify what Candela seemed to know at the advantageous geometrical form while allowing the designer to
outset of design: that employed properly, the hyperbolic parabo- manipulate it into a singular work of structural art.
The current dearth of thin shell concrete structures continues
to confound those who are aware of how effectively they provide
an attractive, efficient structural solution. This study has sought to
demonstrate how the medium of thin shell structures was mas-
tered by one structural artist through a thorough investigation of
one of the master works of his oeuvre. Through careful attention
to the design and construction of structurally efficient forms,
Felix Candela was able to demonstrate the strength and durability
of a remarkably thin structure.
Acknowledgments
The writers gratefully acknowledge Mrs. Dorothy Candela’s do-

nation of photographs of Felix Candela’s works to the Princeton
Fig. 10. High maximum principal stresses 共tension兲 in back edge of Maillart Archives. Some of those photographs were included as
Cuernavaca Chapel model and corresponding actual increase in thick- figures in this paper.
ness
Table 1. Comparison of Finite-Element Model Results 共Uniform Thickness, Uniform Thickness with Edge Beam, and Variable Thickness with Edge
Beam兲
Uniform Uniform thickness with Varying thickness with
thickness edge beam edge beam
Max displ. 关cm 共in.兲兴 −0.5 共−0.2兲 −0.5 共−0.2兲 −0.3 共−0.1兲
Min principal stress 关kPa 共psi兲兴 −4,492 共−651兲 −4693 共−681兲 −2525 共−366兲
Max principal stress 关kPa 共psi兲兴 1,134 共164兲 1,586 共230兲 1,505 共218兲
Note: Negative values in compression; positive values intension.
J. Archit. Eng. 2008.14:47-52.

Table 2. Comparison of Finite-Element Model Results 共Uniform Thickness, Uniform Thickness with Edge Beam, and Variable Thickness with Edge
Beam兲 at Identical Locations
Uniform 4 cm Uniform 4 cm thickness with Varying thickness with
thickness edge beam edge beam
共a兲 Location A
SMAX 关kPa 共psi兲兴 21 共3兲 21 共3兲 21 共3兲
SMIN 关kPa 共psi兲兴 −234 共−34兲 −228 共−33兲 −228 共−33兲
共b兲 Location B
SMAX 关kPa 共psi兲兴 −76 共−11兲 −69 共−10兲 234 共34兲
共c兲 Location C

共d兲 Location D
共e兲 Location E
SMAX 关kPa 共psi兲兴 97 共14兲 359 共52兲 −14 共−2兲
Note: SMAX= maximum principal stress; SMIN= minimum principal stress; negative values in compression; positive values in tension. Locations as
shown in Fig. 2.
References Meyer, C., and Sheer, M. H. 共2005兲. “Do concrete shells deserve another
look?” Concr. Int., 27共10兲, 43–50.
Candela, F. 共1958兲. “Understanding the hyperbolic paraboloid.” Architec- Otero, A. B. 共2001兲. “Félix Candela y el borde libre.” Bitacora Arquitec-
tural Record, 123共7兲, 191–195. tura, 5, 38–47, 71.
Computers and Structures, Inc. 共2004兲. CSI analysis reference manual for Robert McNeel & Associates. 共1993兲. Rhinoceros user’s guide, Seattle.
SAP2000, Berkeley, Calif. Saliklis, E. P., Billington, D. P., and Carmalt, A. W. 共2007兲. “Tedesko’s
Draper, P. 共2006兲. “Felix Candela and the chapel of Lomas de Cuerna- Philadelphia skating club: Refinement of an idea.” J. Archit. Eng.,
vaca.” Master of Science in Engineering thesis, Princeton Univ., 13共2兲, 72–77.
Princeton, N.J.
J. Archit. Eng. 2008.14:47-52.

Nexus Network Journal
https://doi.org/10.1007/s00004-018-00424-1
RESEARCH
Geometry and Construction Optimization: An example

using Felix Candela’s Church of St. Joseph the Craftsman
in Mexico
Zlata Tošić1 · Sonja Krasić1 · Naomi Ando2 · Milos Milić1
© Kim Williams Books, Turin 2019
Abstract
This paper presents a contemporary architectural working method that encompasses
digitalization and parameterization of existing buildings and optimization of new
buildings designed with ruled surfaces. The method uses parametric modeling
and computational structural analysis in order to simplify contemporary building
processes. As an example of the application of these techniques, in this paper they
are applied to Felix Candela’s Church of St. Joseph the Craftsman, a design which
features hyperbolic paraboloids that are considered difficult to design, calculate and
build. The optimization method introduced in this paper seeks to explore different
possibilities for designing and modifying buildings designed using non-standard
geometry allowing them to be built out of simplified elements but also keep
construction and visual properties of their shape. This method is also useful for
students and young engineers to expand their skills in structural analysis, parametric
modeling and optimization methods with contemporary tools.
Keywords Construction optimization · Geometry · Ruled surfaces · Digital

fabrication · Felix Candela
* Zlata Tošić
zlata10@live.com
Sonja Krasić
sonjak@gaf.ni.ac.rs
Naomi Ando
n‑ando@hosei.ac.jp
Milos Milić
milos.cicevac@gmail.com
1
Faculty of Civil Engineering and Architecture, Niš, Serbia
2
Faculty of Engineering and Design, Hosei University, Tokyo, Japan
Vol.:(0123456789)
Z. Tošić et al.
Introduction
To talk about architecture without using the word “form” may now seem
inconceivable (Forty 2013). Much of the architecture of the twentieth century is
characterized by its experimentation with the use of geometric shapes in building
design. Reinforced concrete is one of the most used materials for building these
designs. Not only because it has the ability to cover large areas, but also because
it enables physical continuity of the shape. However, in shell structures with non-
standard geometry, the complexity of the geometry, the large continuous surface
area and material use, demands certain phases in the building process that are
costly in order to get the best final results. These include: the cost of wooden
formwork, the time for concrete to stabilize and how to get the right shell shape
(Jovanović et al. 2014).
Topology optimization can be described as a family of computational methods
aimed at finding optimal structural layouts and configurations (Aage et al. 2014).
There are a range of methods for this optimization including solid isotropic
microstructures with penalty (SIMP) methods (Mariano et al. 2009), heuristic
methods of topology design and the homogenization method (Bendsøe and Kikuchi
1988). The method also depends on the type of object. For example, for optimization
of higher buildings there are manufacturing type constraints, in particular pattern
gradation and repetition, in the context of building layout optimization (Stromberg
et al. 2011). For optimal reduction of material the Evolutionary Structural
Optimization (ESO) method for developing conceptual forms of complex
structures can be used also (Xie et al. 2005). Although, it is powerful tool, the
important steps must be chosen in order to have the best overview of the effects
of optimization. Developments in technology have brought new programs which
enable us to optimize, design parametrically and rationalize shapes and elements
of construction, therefore making them more efficient, while still taking care of the
aesthetic component of the final product (Stavrć and Wiltsche 2015). Furthemore,
the structural behavior of shells is developed essentially due to their form. Some
researchers deal with small modifications in the geometry of the form without
modifying their initial aesthetic configuration too much (Cavieres et al. 2011). In
contrast, this paper uses discretization of the shape, for construction simplification,
while keeping the visual effect of the hyperbolic paraboloid (HP).
The aim of this paper is to demonstrate a method for using geometry properties
for optimization and to place constructive elements in the optimal positions. The
method supports appropriate material choice, structural system selection and
the development of additional elements of the façade. The method supports the
examination of these properties of both new and existing construction, including
a consideration of mass, stress in elements, necessary process of building,
architectural properties. It allows for defining analyzed parameters of both models in
the form of percentages for comparison and the analysis of geometrical properties of
elements for potential digital prefabrication (Tomas and Marti 2010a).
The case study used to demonstrate the method is the Church of St. Joseph
the Craftsman in Mexico, designed by architect Felix Candela and built in
Geometry and Construction Optimization: An example using…
1959 (Wortman and Tuncer 2017). The method is undertaken using Rhinoceros
with plug-ins Grasshopper, Karamba, Silvereye and FEMAP with NX Nastran.
Using the method we present the geometry and construction optimization
of the new contemporary variation or model of the case study building. The
main characteristics of this building, which are changed to create the new
contemporary model, are the execution cost and performance process. Although
it will be defined in detail in the next section, the principle of shortened workflow
can be seen in this scheme (Fig. 1). In the workflow the operations are lined up
by order. The first three are dealing with definitions of geometry, parametrization,
construction elements, material, support and loads. Operation 4 is a calculation
of defined construction (using plug-in Karamba) and operation number 5 is
minimizing deformation in function of number “N” (using plug-in Sivereye).
Operation 6 is checking the other geometrical and construction characteristics.
If they are in an acceptable range of numbers (length and number of elements,
mass) and satisfy visual criteria (shape) for the process of prefabrication, then the
number “N” from step 5 is the final solution. Otherwise, the next number “N” is
taken and the operation 6 is repeated. The loop from 6 to 6′ will be repeated until
we satisfy conditions from the operation 6, in which case the number “N” will be
defined in the final step 7.
Geometry and Construction Optimization
A level of complexity arising from the application of topology optimization

procedures is the geometric rationalization of results required for translating
optimized rods into buildable geometry (Aage et al. 2014). Past research suggests
that massive constructive systems, depending on the stress placement, can be
turned into a spatial grid, arch or beam system (Radivojević and Kostić 2011).
In order to define the precise dimensions of Candela’s building, we were able to
use Radivojević and Kostić’s (2011) drawings (Fig. 2) while geometrical methods
helped us determine approximate dimensions of the starting HP.
The geometry of the Church of St. Joseph the Craftsman is formed out of two
identical HPs (Moreyra Garlock and Billington 2010), obtained by using a spatial
Fig. 1 Workflow for designing and optimization of a new model

Z. Tošić et al.
Fig. 2 Plan and front view of the Church of St. Joseph the Craftsman in Mexico (Source: Radivojević
and Kostić 2011)
quadrangle. The space between two HPs is formed for daylight illumination (Fig. 3).
The other four remaining sides are used as glass walls with thin columns.
Typically, the three observation levels micro, meso and macro are distinguished
in reinforced concrete construction. A new idea is to transfer this principle to shell
construction. It is not absolutely necessary to construct the entire load-bearing
structure on the macro level through one shell, but rather to understand the shell as
part of an overall structure on the meso scale. For example, ceiling structures can be
composed of individual shell elements, which in turn act as a beam (Faber 1965). To
accomplish this it is necessary to think about the elements of the construction and
their connections, which are the main properties. Geometry optimization of the shell
can be done through shell tessellation with many algorithms (Hegger et al. 2018).
Using geometry properties this type of optimization is possible on the double curved
surface of the HP form between two straight lines (generatrices of both systems);
thus, Candela achieved economy of construction by avoiding use of curved panels
for his formwork in construction (Moreyra Garlock and Billington 2010).
To digitize the building’s shape, it is necessary to analyze its dimensions
and geometry first. This building is an example of ruled surface where we can
use its characteristics for optimization of its geometry. In this case, geometry
optimization is defined by using discretization of the geometry by planarization
of the continuous, double curved form of the HP, with triangular panels. The
goal is to keep a similar visual effect of the HP form, achieve good construction
Fig. 3 Rhinoceros drawings with approximate dimensions which demonstrate the geometry of Candela’s
design
properties and at the same time facilitate the process of building with
prefabricated elements.
The whole geometry of the building is then defined in Grasshopper. The idea is
then to modify a new HP surface using its straight generatrices. Parameterization
is defined by number “N” which represents the number of divisions of HP
directrices (guidelines) on equal parts in order to get straight lines (Krasić 2012).
These creates the basis of the grid, for both systems of generatrices are the main
rods in the construction (Fig. 4). The second step is to find the average point
coordinates of every special quadrilateral and connect them with all four points.
These are the second type of rods called diagonals. Triangle modules represent
the planar panels, which are not susceptible to bending during construction (grid)
deformation in order to form stable structural geometry. The third type of rods
which connect two HPs are ties.
Then pre-calculation of the new construction design is necessary, because
even though the geometry optimization represents discretization geometrical
features of the shape in order to facilitate the assembling of building elements,
they are also part of the construction system and all construction properties
have to be satisfied. The pre-calculation process is using Karamba where we
are able to define: cross-section of elements, loads, thickness of the panels,
material and support. For material, steel type S355 is chosen, not only because
of its mechanical properties, but also because of the shape of the elements and
their connections. These loads are taken into account in the decision: mass of
the construction and weight of triangular panels. For determining support three
degrees of freedom (rotation around all three axis) were defined, in marked blue
points (Fig. 4).
We then calculated mass and deformation of the new design and using
Grasshopper all properties of construction are directly related to the number of
division “N”. Changing that value all characteristics (except cross sections) are
simultaneously changing as well. For every selected value “N” the Karamba
calculations are represented in Table 1.
Fig. 4 Grasshopper drawing of the position of all elements of the new model

Z. Tošić et al.
Table 1 Geometry and construction properties of the new HP in function of number “N”

N (number of division) 4 6 8 10
Mass 19,449 kg 27,810 kg 36,173 kg 44,538 kg

Deformation (m) 0.263 0.185 0.158 0.154
Min–maximum length of rods (m) 3.30–13.80 2.20–9.20 1.64–6.9 1.31–6.22
Number of elements 183 396 688 1060
Shape of geometry Polygonal Polygonal Continuous Continuous
The next step is to define the number “N”, which is directly related to geometrical
properties of construction elements including number and length of elements
(geometrical features) and mass and deformation (construction features). These
are all intended to be minimized for the assembling process and level of stress in
elements. In order to see relations between construction properties and number
“N”, the process of optimization is done in Silvereye. The goal is to find the number
“N” (input parameter) for minimal values of deformation (fitness value) of the
construction system. As mentioned, geometry and construction optimization are
co-dependant so we compare the two most important properties of geometry and
construction optimization (Fig. 5).
From Fig. 5 we can see that the value of deformation varies from N = 1–4 because
of the placement of the supports, length and cross-section of the elements. After
stabilization maximal deformation dramatically decreases when the value of number
N > 4 and then slowly rises after N = 11. Nevertheless, divisions more then 10 aren’t
analyzed further because of the large number of elements and their connections.
If changes in the original shape are made in order to place planar elements, the
lower boundary of the discretization must be defined in order to preserve the fluid
continuity of the geometry as much as possible, as it presents the whole architectural
beauty of the shape. The similarity between curve and polygonal line depends of
scale and individual perception, so it is up to the designer to select the satisfactory
discretization. Authors decided that options 4 and 6 are eliminated as the line from
Fig. 5 Interdependence of number of division and maximum deformation in the new HP model

Fig. 6 Grasshopper drawing of the shape of discretized geometry with planar panels with different “N”
value
the “discretization of the original HP” shape which is achieved with division points
of generatrices (blue full line) does not follow/overlap the “curvature of the original
HP” (yellow dashed line). Therefore, it does not create the same effect and satisfy
the visual criteria (Fig. 6).
In the end, we focus on the number and length of the elements. Even though they
are inversely proportional, it can be noticed that for N = 10, length of the elements
does not decrease much when compared to N = 8, but the number of elements is
drastically higher. The last characteristic is mass, which is increasing with every
number „N”. After overviewing all parameters it can be concluded that chosen
number for division is 8.
Calculation of Structural Properties of Both Models Using Femap

with NX Nastran
Structural Properties of the Existing Building
Candela produced structures of astonishing elegance and efficiency thanks to his

intuition and experience (Moreyra Garlock and Billington 2010). Even though it
was an advanced approach at the time, its execution was not easy and demanded a
series of precise engineering phases. As such, complications may appear in these
structures as in the example of the Chapel of Lomas de Cuernavaca at Palmira in
Mexico where, during the building process, the part of the structure suddenly
collapsed while scaffolding was being removed. Such accident are due to problems
of construction (too young concrete, scaffolding removed defectively, existence of
local defects) (Tomas and Marti 2010b). That is one of the reasons that motivated
us to explore a different solution. Structural system of the existing object is defined
with HP shells which rely on four beams (3.0 × 0.5 m) and are connected with
number of ties and supported on the ground by two large eccentric foundations.
On other four sides the reinforced console beams are also supported by the same
foundations. Due to all the experiments he achieved to make shell that is only 0.04 m
thick (Moreyra Garlock and Billington 2010) (Fig. 7). This construction is already
Z. Tošić et al.
Fig. 7 Rhinoceros drawing of the existing Church of St. Joseph the Craftsman with main constructive
elements annotated
effectively optimized and is one of the most advanced buildings, not only in terms of
construction system, but also building technology in that time.
The chosen geometry and material require physical continuity, which makes
the process of building complicated and lenghty. It also requires a lot of precision
especially during placing the formwork (Fig. 8). Structural calculations for the
existing building are done with all the stated elements using FEMAP with NX
Nastran. It is important to define mass and maximum deformation in the system
with reinforced concrete to be able to compare the results with the calculations of
the new designed building.
When analyzing stress distribution we found that maximum intensity is located in
the spikes of the shell as well as connections of shells and main beams, as expected
for this system. Maximum stress is 12.64 MPa as it can be seen in Fig. 9.
Figure 10 shows total translation which is located at the spikes of both shells; the
maximum amount is 0.244 m. The mass of the whole construction of the existing
object is 565.805 t. The structure is analyzed with geometric nonlinear theory and
linear behavior of the concrete. Conditions are formed for the basic load case, which
includes only self-weight of the structure. Deformation on Fig. 10 is enlarged three
times.
Fig. 8 Rhinoceros drawing of
the formwork that was used
for the building process of
the Church of St. Joseph the
Craftsman (adapted from: Smith
et al. 1967)
Fig. 9 Stress distribution in the existing object—max: 12.64 MPa (FEMAP with NX Nastran)
Fig. 10 Total translation in the existing object—max: 0.244 m (FEMAP with NX Nastran)
Structural Properties of the New Model
The authors wanted to analyze results of the calculation for the new model in the
same program as they used for the built Church. All of the inputs (position of
elements, material, load and support) are defined to be the same as in previous
Karamba calculation (Fig. 11).
Points marked “123” in Fig. 11 present supports with three degrees of freedom
(rotation around all three axes). Cross sections are tubular and their diameter is
different for every type of element: main rods (168.3 mm), diagonals (133 mm)
and ties (88.9 mm). Moreover, because of the buckling in the longest ties at the
bottom, they are replaced with simple beams, which will serve only for glass
panel’s position.
After the definition of all elements, connections, materials, loads and supports,
the results of the stress distribution are shown in Fig. 12. Maximum stress is
located on the connections of rods and it is max = 62.72 MPa.
Figure 13 presents the model of total translation, where the location of
the maximum level of deformation is in the same place as it is in the existing
Z. Tošić et al.
Fig. 11 Elements and loads of the new model (FEMAP with NX Nastran)
Fig. 12 Stress distribution of the new model—max: 62.72 MPa (FEMAP with NX Nastran)
Fig. 13 Total translation of the new model—max: 0.164 m (FEMAP with NX Nastran)
building. This value is shown in the contour view and it is max = 0.164 m. Total
mass of the object is 36.039 t. This value is less than stated previously, because
of the reduction of cross sections during the design, but the value of deformation
has not varied a lot.
Comparison of Construction Properties of the New and the Existing

Building
The comparison of the construction properties of the new and existing or referenced
building is shown in the Fig. 14 according to the previous analysis in the paper.
There are two charts with mass and deformation characteristics presented in
percentages.
The first chart defines the percentage of mass of the new and referenced
construction, where it can be seen that the choice of different material and
construction elements can reduce mass by about 16 times. The second chart presents
analysis of maximum deformations, where the new construction has around 66% of
maximum deformation of the existing construction. This comparison demonstrates
the level of control of the methods used throughout the previous process. Moreover
it shows that wide range of possibilities of different analyses in contemporary
technology can be effective on these shapes.
The use of optimization techniques in the design process of structures widens the
field of use of computers and allows the user to obtain optimum designs for stated
design conditions. There are other researches that are achieving optimization due
to transformation of the entire geometry (dimensions, thickness, curvature), while
preserving aesthetic appearance similar to that initially planned by the designer
(Tomas and Marti 2010a). In future research the combination of these two methods
could be explored.
Digital Fabrication of Elements
After analysis of the construction and geometry, optimizing and finding element’s
number and positions, elements are placed in the construction system (Fig. 15).
Optimal design of rods offers, by comparison, a path for closer integration
between optimization and fabrication, and it is therefore arguable that they should
take a more prominent role in the investigations of the field (Aage et al. 2014).
Fig. 14 Percentage of mass and

maximum deformation
Z. Tošić et al.
Fig. 15 Scheme of all elements of the new model construction as well as facade expression (Rhinoceros)
Using technology the next step is to digitalize elements for potential fabrication.
This is an indicator of the geometrical properties of the elements.
Firstly, we can define types of rods in the construction. Rods are divided in five
groups regarding their length which is from 1.4 to 6.9 m. Figure 16 demonstrates
how every group of elements is distributed through the model. Table 2 shows
specifications of every group length and the number of elements as well as the
percentage of elements that the group covers in the whole structure. It can be
concluded that most of the elements are from 1.64 to 4.65 m long and only around
6% are from 5.08 to 6.90 m. All of the 5th group rods are ties. The total number
of rod elements is 688. This geometrical configuration is beneficial as the main
rods, which have the biggest percentage of stress, are the shortest and the longest
ties have only tension.
The second phase is prefabrication and division of panel’s area. The groups are
formed depending on the surface area of the panels. This analysis can show the
curvature of the surface, as the zone with the smaller panels has bigger curvature
Fig. 16 Grasshopper drawing of five groups of elements divided by length

Table 2 Group’s specification of rod elements of the new model

1st Group 2nd Group 3rd Group 4th Group 5th Group
Length of the rods (m) L = 1.64 − 2.15 L = 2.36 − 3.57 L = 3.74 − 4.65 L = 5.08 − 6.22 L = 6. 32− 6.90

Number of the rods in n = 198 n = 314 n = 136 n = 26 n = 14
group
Percentage of rods in ≈ 29 ≈ 46 ≈ 19 ≈4 ≈2
every group (%)
than the area of the bigger and identical panels. When adding all numbers from
Table 3 the total number of panels is 512.
The most curved part of the surface is the middle of the HP where the first
group is. The difference in the surface areas in every group varies. In the first
group it is 0.19 m2, then it rises to 0.38 m2 in the second, third and fourth group
and finally the fifth group has four identical panels (Fig. 17). Most of the panels
are from the first three groups (90.61%). According to the surface area of the
panels it can be concluded that the distribution during the process of building will
not be complicated, as the panels are from 1.54 to 3.26 m2.
Every panel is attached to three intersection points of the rods. Weight is also
not the issue during assembling, as its amount is around 10 kg/m2. Composite
aluminum panels are fireproof, water resistant, with excellent heat and accoustic
insulation. Some of the aesthetic values are (1) surface flatness and smoothness,
(2) superior weather, corrosion and pollutant resistance, (3) even coating, (4)
super peeling strength, (5) impact resistance, (6) lightweight and easy to process
and maintain. As they can be custom made, and all dimensions are specified, it is
possible to make prefabrication precise (Lee et al. 2018).
Finally, it can be noted that in many topology optimization procedures, certain
regularization is necessary in order to control the physical size of structural
members appearing in the new design (Aage et al. 2014). Although these
dimension values apply only to this example, this process is not only defining
overall properties about construction, but helping in practical digitalization
of the elements as well. It facilitates organization in the whole structure and
gives better input when it comes to construction process. Figure 18 shows the
final result of the new building’s design. Given the change in the construction
system it is possible to cover the spatial grid with lightweight aluminum panels
with openings which can provide more daylight in the Church. There is a lot of
freedom in design options. The planarization of the panels enables placing glass
in place of the panel’s openings.
One deficiency of the building process concerns the precise spatial placement of
the rods and their connections. It is planned to start from the support placement,
and then to be tied with matching rods. It will be necessary to finish the middle
of the object first in the shape of cantilever going to the ends. In future work such
issues have to be thought through in more detail. For example, when it comes to the
production process one of the priorities is to create a safe waterproof shield for the
building, therefore, the shell element’s connections must be designed as continuous.

Table 3 Group’s specification of the panels of the new model

1st Group 2nd Group 3rd Group 4th Group 5th Group
Area of the panels (m2) A = 1.54 − 1.73 A = 1.80 − 2.18 A = 2.20 − 2.57 A = 2.63 − 3.01 A = 3.26

Number of the panels in group n = 104 n = 204 n = 156 n = 44 n = 4
Percentage of the panels in every group (%) ≈ 20 ≈ 40 ≈ 30 ≈9 ≈1
Z. Tošić et al.
Fig. 17 Grasshopper drawing of five groups of panels divided by area
Fig. 18 Final model of the new building
Conclusions
New developments in technology have spread across all the areas of expertise in
architecture and civil engineering, making new technology and processes available
in the building process. Complex geometric shapes can now be achieved using
prefabricated elements and the visual identity of the final object can also be
preserved at the same time.
The geometry optimization method described in this paper helps in defining
the element’s shape, position and dimensions in order to analyze its construction
properties. Both are co-dependant, as geometric properties have influence on some
of the construction characteristics and vice versa. Together they form elements
which can be precisely analyzed for the process of fabrication. This process is one of
the main benefits of efficient building. Now, not only will the process be quicker and
less expensive, but also less risky when it comes to the process of building.
This paper presents an analysis of a case study building, taking Candela’s built
object as a good example of HP shape design, and offers a contemporary solution
with different materials and changes in the constructive system which can make the
Z. Tošić et al.
building process much easier. Candela’s significant structures were all of HP forms
(Moreyra Garlock and Billington 2010). This research shows methods which can
modify this approach to contemporary designing, optimizing the construction and
geometry in ruled surfaces geometry with strictly defined dimensions. However, it
is not possible to use this method for more free-form and complex geometries as
the starting point. Construction properties, such as mass and deformation, of the
new and referenced HP shape are compared in order to verify the right methods.
In addition it is important to see in which ways are contemporary programs and
technology upgrading the process of modelling, digital fabrication (Tomas and Marti
2010b) and optimization. The process is not only important for existing objects such
as Candela’s Chapel, but also for new design works.
References
Aage, N., O. Amir, A. Clausen, L. Hadar, D. Maier and A. Søndergaard 2014. Advanced topology
optimization methods for conceptual architectural design. In: Advances in Architectural Geometry,
eds. P. Block, J. Knippers, N. Mitra and W. Wang. 159-179. Cham : Springer International
Publishing : Springer.
Bendsøe, M. P., and K. Kikuchi. 1988. Generating optimal topologies in structural design using a
homogenization method. Comput Methods Appl Mech Eng 71 (2): 197-224.
Cavieres, A., R. Gentry and T. Al-Haddad. 2011. Knowledge-based parametric tools for concrete
masonry walls: Conceptual design and preliminary structural analysis. Automation in Construction
20 (6): 716-728.
Faber, C. 1965. Candela and his bowls (Candela und seine schalen). Munich: Verlag Georg D.W.
Callwey.
Forty, A. 2013. Words and buildings: a vocabulary of modern architecture. London : Thames & Hudson.
Hegger J., M. Curbach, A. Stark, S. Wilhelm and K. Farwig. 2018. Innovative design concepts:
Application of textile reinforced concrete to shell structures. Structural Concrete 19 (3): 637-646.
Jovanović N., P. Pejić and S. Krasić. 2014. The use of voronoi diagram in the design of an object
containing a hyperbolic paraboloid, 18th Geometrical Seminar, Vrnjačka Banja, Serbia, Book of
Abstracts. ISBN 987-86-6275-027-3, 55. http://tesla.pmf.ni.ac.rs/people/geometrijskiseminarxviii/
Book%20Abstracts-VrnjackaBanja2014.pdf. Accessed 4 March 2018.
Krasić, S. 2012. Geometrical surfaces in Architecture. Nis, Serbia: Faculty of Civil Engineering and
Architecture, University of Nis.
Lee, A. D., P. Shepherd, M. C. Evernden and D. Metcalfe. 2018. Optimizing the Architectural Layouts
and Technical Specifications of Curtain Walls to Minimize Use of Aluminium. Structures. 13: 8-25.
Mariano, V., M. Pascual and M. Q. Osvaldo. 2009. Topology design of two‐dimensional continuum
structures using isolines. Computers & structures. 87 (1-2): 101-109.
Moreyra Garlock, M. E. and D. P. Billington. 2010. Structural Art and the Example of Félix Candela.
Journal of structural engineering. 136 (4) : 339-342.
Radivojević G., and D. Kostić. 2011. Construction systems in architecture. Nis: Faculty of Civil
Engineering and Architecture, University of Niš, Serbia.
Smith C. B., M. Pani, M. Goeritz, F. Candela, L. Barragán and J. O’Gorman. 1967. Builders in the sun :
five Mexican architects, New York : Architectural Book.
Stavrić M. and A. Wiltsche. 2015. Ornamental plate shell structures. https://www.researchgate.net/publi
cation/268368071_Ornamental_plate_shell_structures. (accessed 5 January 2018).
Stromberg L. L., A. Beghini, W. F. Baker and G. H. Paulino. 2011. Application of layout and topology
optimization using pattern gradation for the conceptual design of buildings. Structural and
Multidisciplinary Optimization. 43 (2) : 165-180.
Tomas A., and P. Marti. 2010. Shape and size optimization of concrete shells. Engineering Structures.
32(6): 1650-1658.
Tomas A., and P. Marti. 2010. Optimality of Candela’s concrete shells: A study of his posthumous design,
Journal of the International Association for Shell and Spatial Structures, 51:.67.
Wortman T. and B. Tuncer 2017. Differentiating parametric design: Digital workflows in contemporary
architecture and construction, In Parametric Design Thinking. Design Studies 52: 173-197.
Xie Y., P. Felicetti, J. Tang and M. Burry. 2005. Form finding for complex structures using evolutionary
structural optimization method. Design Studies 26 (1): 55-72.
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
Zlata Tošić is PhD Candidate in the Faculty of Civil Engineering and Architecture in Nis, Serbia at
the Department of Architecture. Her field of research implies the geometry, topology and structural
optimization of shell constructions. She is also interested in exploring genetic algorithms and using this
method for optimization of grid shells.
Sonja Krasić is Associate Professor in the Department of Visual Communication, Faculty of Civil
Engineering and Architecture, Niš, Serbia. Dr. Krasić undertakes research in the field of descriptive
geometry and into applications of photogrammetry in architecture.
Naomi Ando is Fulltime Professor, Faculty of Engineering and Design, Hosei University, Tokyo, Japan.
Dr. Ando’s speciatly is Architectural, urban planning and geometry. Various forms of architecture and
cities (e.g., plazas and streets) and the qualities and characteristics of the spaces that those forms produce.
Milos Milić is PhD Candidate in the Faculty of Civil Engineering and Architecture in Nis, Serbia at
the Department of Civil Engineering. His field of research implies steel constructions, materials and
structural engineering. Resent research is regarding clinging timer slabs with steel beams and their
structural analysis.
JOURNAL OF THE INTERNATIONAL ASSOCIATION FOR SHELL AND SPATIAL STRUCTURES: IASS
FELIX CANDELA, ELEGANCE AND ENDURANCE:

AN EXAMINATION OF THE XOCHIMILCO SHELL
NOAH BURGER1 AND DAVID P. BILLINGTON2

1
Former Undergraduate Student, noah.burger@gmail.com
2
Gordon Y.S. Wu Professor of Engineering, billington@princeton.edu
Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ, USA
Editor’s Note: Manuscript submitted 9 September 2005; revisions received 20 September and 17 November 2006;
accepted for publication 18 November 2006. This paper is open for written discussion, which should be submitted to the
IASS Secretariat no later than August 2007.
SUMMARY
Studying landmark works of structural engineering is essential to the advancement of the field and to the
continuing education of the engineer. This report presents a study of such a structure, Felix Candela’s Los
Manantiales shell in Xochimilco, Mexico City. The shell’s structural action is discussed, and then examined
through finite element analysis using SAP software. Analysis results are supplemented by field observations of
the shell’s current condition, and examinations of archival photos of the shell taken during its construction.
Keywords: Felix Candela, concrete shell, hyperbolic paraboloid, free edge, creep, shrinkage
Figure 1. Restaurant Los Manantiales [5]
1. INTRODUCTION Today, the romantic setting has faded. The canals

have sunk and the floating gardens have
In 1958, Felix Candela completed his most disappeared. The enduring shell, however, has not
significant work, the Los Manantiales Restaurant lost its majesty.
shell, in Xochimilco, Mexico City. It sat at the
edge of a wide waterway, surrounded by floating Los Manantiales is an eight-sided groined vault
gardens (Fig. 1), and could be reached directly by composed of four intersecting hyperbolic
embarcaderos, long, colorful gondolas that still paraboloid saddles. An elevation reveals canted
ferry passengers along the canals of Xochimilco. parabolic edges, which display its striking thinness
VOL. 47 (2006) No. 3 December n. 152
of 1-5/8” (40mm). Candela was taking a risk when

he built Los Manantiales. The form was original,
unexplored, and impossible to analyze precisely.
Candela’s career, however, habitually flew in the
face of precise analysis. In his first acclaimed shell,
the Cosmic Ray Pavilion of the University of
Mexico City, he also designed an unprecedented
form, using almost no calculation; a University
committee requesting equations for review was
Figure 2.1. Sketch of groin deflections
instead presented with a written explanation [5, p.
14]. Candela’s subsequent designs relied
increasingly on structural understanding and
practical experience. As a designer-contractor, he
had the unique responsibility of building his own
solutions. By closely observing his buildings, and
using smaller projects to test new ideas, he
developed an acute sense of concrete shell behavior.
When he began his work on free shell edges, an
exact analytical solution was nearly impossible to
derive. But Candela sensed that free edges were
possible, even without complex equations.
Figure 2.2. Normal forces
This report examines the structural action of Los
Manatiales, using SAP2000 software to conduct a
finite element analysis. A trip was also made to
Mexico City to see the shell firsthand, and to
document its current condition. The findings of this
trip ultimately enhanced the significance of the
analysis, but more importantly, it brought this study
out of the laboratory and into the field.
2. CANDELA’S CONCEPT
Prior to Candela’s free edge shells, most shells used

edge ribs. Hypar shells generate large forces
normal to their edges, but their uniquely thin cross
Figure 2.3. V-beam (groin) cross-section, 1”=25.4mm
sections, allowed by double curvature, have little
stiffness in the normal direction. Left unchecked, a
shell’s normal forces would crack it from the edges
inward. To prevent this, most engineers of
Candela’s time secured their shells’ edges with ribs.
These effectively reined in the normal forces,
carrying them down to the supports tangentially
along the edges. Unfortunately, such ribs made a
shell appear heavy by obscuring its true thinness. A
form existed that generated forces only tangential to
the edges, with no normal forces, but it was nearly
impossible to calculate. Ribs appeared the only
viable solution.
To Candela, the edge rib seemed cumbersome; the

shell, while sturdy, was not true to form, not pure,
and could not visually express its thinness. In his Figure 2.4. Umbrella footing, 1”=25.4mm
technical writings, Candela never boasted of his Table 1. Finite element model reactions, 1k = 4.45kN
structures as ‘works of art,’ but in many ways, he
considered himself an artist. He enjoyed integrating Model Reaction (moment reactions ≈ 0) Value (k)
Vertical, V_model 57.21
aesthetics and sound structure, using his imagin- Horizontal, H_model 35.92
ation to find new solutions. He saw the free, thin,
crisp edge as the ultimate shell builder’s challenge,
software was used to tabulate the shell’s surface
and his solution, found after less than a year of
area, as shown in Fig. 3.1. The surface area was
research, is simultaneously an elegant and efficient
then used to calculate the shell’s total dead load.
innovation. Colin Faber describes how, in 1955,
This method deviated from ideal hand-made
Candela began working on the free-edge problem
calculations, but as the software accurately
for the roof of the Mexico City Stock Exchange [5].
produced surface areas for simpler shapes, its use
The project ended with a ribbed shell, but Candela
was decided an acceptable break with intended
continued exploring the problem on his own. His
procedure. The results are presented in Table 2
next large project, the San Antonio de Las Huertas
below.
Church, built in mid-1956, was completed with free
edges.
Normally, a shell’s dead load causes bending in its

groins, bowing them outward (Fig. 2.1). The
resulting deflection pushes against the saddles,
generating forces normal to the shell’s surface. At
the edges, the normal forces cause the shell to
balloon out, forming cracks that propagate from the
edges inward (Fig. 2.2). Edge ribs can be used to
Figure 3.1. 3D surface model
mitigate these cracks, but Candela, recognizing the
behavior described above, addressed the groins Table 2. Dead load calculations
rather than the edges. Each of Los Manantiales’ 1ft = 0.305m & 1k = 4.45kN
eight groins is thickened into a V-shaped beam, as
seen in Fig. 2.3. These beams enhance the groins’ Shell Surface Thickness Concrete Dead
stiffness, reducing bending and deflection, and Portion Area (ft) weight Load
thereby eliminating normal edge forces. Instead of (ft^2) (k/ft^3) (k)
V-beam 2,758 0.3937 0.15 162.7
membrane stresses reaching the shell’s edges and Saddle 14, 975 0.1312 0.15 294.7
generating tension, as in Fig. 2.2, they are Total Dead Load = 457.6
essentially “pulled” to the groins, making edge ribs
unnecessary. At the supports, the V-beams anchor Vert. reaction = 457.6/8 = 57.2 k (255 kN) = Vhand
into inverted umbrella footings, seen in Fig. 2.4.
These unique footings cup the earth, preventing the Vhand exactly matched Vmodel.
shell from sinking in the wet Mexican soil. To
resist lateral thrusts, adjacent footings are linked A second check was conducted against the
with ∅1” steel bars. Candela’s design produced horizontal reaction, H. While not required to
startling thinness; between V-beams, the shell’s validate the model, as an exercise it provided a
saddle spans are only 1-5/8” thick. useful opportunity to compare the arch behavior
assumed for a 2D calculation and the more complex
3. FINITE ELEMENT ANALYSIS shell behavior modeled by SAP.
Prior to a full stress analysis, SAP2000 was first From Fig. 3.2, tanα = 4d/L, where α is the angle
used to calculate the shell’s support reactions. between N and the horizontal, d is the rise, and L is
These are presented in Table 1. Moments about the the span.
supports were negligible.
tanα = 4d/L = (4 x 5.8m)/(32m) = 0.725
A global equilibrium check was conducted against Hhand = V/(tanα) = 57.2/0.725 = 79 k (351 kN)
the vertical reaction, V, to test the finite element
model’s accuracy. First, Rhinoceros 3D drafting Hmodel = 35.92k (160 kN), less than half of Hhand.
VOL. 47 (2006) No. 3 December n. 152
3.1 Groin (V beam)
The groin was examined first, using membrane

stresses S11, S22, Smin, and Smax. S11 stresses
(Fig. 3.1.1) are circumferential, perpendicular to the
groin and crown lines. S11 compression in the
shell is greatest at the groins, where the full weight
of the saddles descends to their corresponding
groins. As the groins descend from the center of
the shell, these stresses begin to taper, due to the
slope of the shell’s saddle surfaces increasing as
Figure 3.2. Reaction diagram for hand calculation they near the supports. As the slope perpendicular
to the groin increases, the component of the
Two aspects of the shell’s structural action explain saddle’s weight perpendicular to the groin (the
this large discrepancy in the two calculated values component producing the S11 stresses) decreases.
of H. First, the total dead load of the shell does not The maximum S11 compression stress therefore
reach its supports exclusively through its groins. occurs at the shell’s crown, where the slope is
As the shell surface between the groins approaches smallest and the danger of buckling the greatest.
the supports, an increasing portion of the dead load However, at –101 psi (–696 kN/m2), the S11
descends directly through the surface itself, rather maximum is low, presenting little danger in terms
than through transfer to the groins. Reducing the of buckling. S22 stresses (Fig. 3.1.2) are radial,
load on the groins reduces their outward thrust, parallel to the groin and crown lines. Maximum
thereby reducing H. Second, the shell’s edges S22 compression also occurs at the shell’s center,
extend past its supports. The vertical load but at –71 psi (–489 kN/m2), it is even smaller than
generated by these outer sections must therefore in S11. As the groin descends, S22 compressive
travel to the supports against the direction of H, stresses shrink to approximately 0 psi at the center
reducing H even further. The finite element model of the groin, and then rise to –78 psi (–537 kN/m2)
implicitly takes these two considerations into near the support. These stresses are extremely
account, and therefore gives a lower horizontal small, but they do hint at the ‘bowing’ trend in the
reaction. groin suppressed by the groin’s increased thickness,
previously discussed in connection with Fig. 2.1.
The two most stress-critical areas of the shell are its
groins and the edges. These were examined using Smin stresses (Fig. 3.1.3) represent the greatest
two lines of data points along the model’s surface: compression resultants of the S11 and S22 stresses.
one along a groin, from the center of the shell down Maximum compression occurs at the shell’s crown
to the support, and one just before the edge, from and the supports. The crown, the area most
just behind the support up to just behind the apex of vulnerable to buckling, experiences a total
an edge parabola (due to the limitations of the finite compression of –124 psi (–854 kN/m2), while the
element model, in it was impossible to obtain support experiences –142 psi (–978 kN/m2).
stresses exactly at the edge, and points slightly Between these two points, compression lessens
behind the edge were deemed sufficient). Since slightly due to the ‘bowing’ trend. Smax stresses
snow loads are not a concern, only a dead load (Fig. 3.1.4) represent the greatest tension resultants
analysis was conducted for this study. Stress values of S11 and S22. About the crown, SAP produces
given in the following sections were calculated by no Smax stresses above zero, indicating the area is
averaging SAP top-surface and bottom-surface in pure compression. Again due to bowing, tension
stresses, which are shown in the accompanying occurs one-third down the groin, quickly peaking at
figures. Positive stresses indicate tension, negative 20psi (138 kN/m2) and gradually decreasing to zero
stresses compression. The difference between the at the support. Actual bending induced in the shell,
top and bottom-surface stresses gives an indication however, is negligible. All Smin and Smax stresses
of the bending. For all graphs in this Section 3, 1ft are small, proving that Candela’s choice of form
= 0.305m & 1k/ft2 = 6.94psi = 6.89 kN/m2 (For a more renders the stresses in the groin insignificant, and
comprehensive description of the analysis presented provides superbly high safety against buckling in
here, see [2]). the thin shell.
V-beam S11 Stresses [(+) Tension]

3.2 Edge parabola
0
The line just behind an edge parabola, the area of
-2 0 20 40 60 the shell most vulnerable to tension cracking, was
Stess in local x direction
-4 analyzed next. The maximum S11 compression

-6 (Fig. 3.2.1) is –30psi (–207 kN/m2), just above the
(k/ft^2)
Top Surface
-8
-10
Bottom surface support. Compression then decreases steadily to
-12
zero at the apex. S22 compression (Fig. 3.2.2)
-14 begins at zero at the apex, and steadily increases to
-16 a peak of –68 psi (–469 kN/m2) two-thirds down the
Distance from center of shell (ft)
edge. Compression then decreases again, but
returns to a final peak of –118 psi (–813 kN/m2), in
Figure 3.1.1. S11 stresses along groin the V-beam at the support. These stresses are
small, even for the shell edge. However, their rise
and fall suggests that as the shell leans outward past
V-beam S22 Stresses [(+) Tension]
its supports, its cantilevered weight bends the edge
2 slightly inward, as if pulling it into one wide inward
Stress in local y direction
0
curve. Compression rises to a peak two-thirds down
-2 0 20 40 60
-4
the edge, the deepest point of the curve, and then
(k/ft^2)
-6
Top Surface steadily decreases as the edge moves away from
Bottom Surface
-8 that point. At the groin, the full weight of the
-10 saddle bears down on the support to create a final
-12
compression peak. Bending in the thin shell is
-14
Distance from center of shell (ft)
negligible, but the stress pattern itself is notable and
will be discussed later, with respect to evidence of
slight creep in the concrete.
Figure 3.1.2. S22 stresses along groin
The Smin and Smax stresses (Fig. 3.2.3 & Fig.
V-beam Minimum Stresses [(+) Tension]
3.2.4) follow the same trend as those of S11 and
S22, respectively. Of special interest are the Smin
0 stresses, which begin at –122psi (–841 kN/m2) at
0 20 40 60
Minimum Stress in
-5 the support and steadily decrease to negligible

element (k/ft^2)
-10 Top Surface

values at the apex. This trend expresses the
-15 Bottom Surface function of the groins as stress attractors, effectively
-20
pulling stresses away from the edges. Deformations
due to edge stresses are on the order of 0.001in
-25
Distance from Center of Shell (ft)
(0.025mm), impossible to observe with the naked
eye and completely negligible in terms of the
structure’s behavior. Normal edge stresses output
Figure 3.1.3. Smin stresses along groin by SAP, S23 and S13, are on the order of 0.1psi
(0.69 kN/m2), practically nonexistent, and most
V-beam Maximum Stresses [(+) Tension] likely attributable to finite element approximations.
The miniscule deformations, and the absence of
6
tension and normal stresses in the edge, affirm that
Maximum Stress in element
4
2
Candela’s design successfully eliminated tensile
0 edge stresses. (The full analysis appears in [2].)
(k/ft^2)
Top Surface
-2 0 20 40 60
Bottom Surface
-4 4. PRESENT-DAY LOS MANANTIALES
-6
-8 Figure 4.1 shows Los Manantiales in its present
-10 condition. It is currently under renovation, but
Distance from Center of Shell (ft) many sealed cracks can still be observed on the
shell’s inner and outer surfaces. The most probable
Figure 3.1.4. Smax stresses along groin
causes of these cracks are shrinkage and creep.
VOL. 47 (2006) No. 3 December n. 152
Edge S11 Stresses [(+) Tension]
2
Top Surface
0
0 10 20 30 40
Bottom Surface
-2
-4
-6
Vertical distance, from support to apex
along perimeter arch (ft)
Figure 3.2.1. S11 stresses (k/ft2) along edge parabola
Edge S22 Stresses [(+) Tension] Figure 4.1. Present-day Los Manantiales
5
Figure 4.2, a photograph of the shell taken shortly
0
0 10 20 30 40
after concrete had finished being poured, shows
-5 faint, damp cracks on the shell’s surface. Figure
-10
Top Surface 4.3, a photo taken after construction was completed,
Bottom Surface
-15
shows a network of cracks spreading in hard lines
over the shell’s surface. That the cracks became
-20
more prominent from the initial pouring to the
-25 finishing of construction indicates they developed
Vertical distance, from support to
apex along perimeter arch (ft)
as the concrete dried. Los Manantiales was poured
by hand, one section at a time, and as workmen
Figure 3.2.2. S22 stresses (k/ft2) along edge parabola moved over the formwork, finished areas were left
to dry uncovered. The shell’s exterior surfaces thus
Edge Minimum Stresses [(+) Tension]
dried faster than its interior, creating “mud-cracks”
5 as shrinkage pulled the outer surface into tension.
0 Over time, moisture and pollution deepened some
0 10 20 30 40
-5 of the cracks, and today, a number of drip stains can
-10
Top Surface be observed where cracks once broke completely
Bottom Surface
-15
through the shell, as show by the patched leaks in
-20
Fig. 4.4. A number of these cracks are still in the
process of being sealed, but none has done any
-25
Vertical Distance, from support to apex
significant damage to the shell.
A second crack pattern can be observed on the
Figure 3.2.3. Smin stresses (k/ft2) along edge parabola
underside of the shell. In Fig. 4.5, a now-sealed
Edge Maximum Stresses [(+) Tension]
crack propagates in from an edge along the
underside of each saddle. These cracks appear on
8 each side of every edge, about 10ft (~3m) up from
6
each support and symmetric about the saddle’s
4
2 Top Surface
center.
0 Bottom Surface
-2 0 10 20 30 40 While the cracks occur systematically, they are not
-4 the result of form. Their correspondence to the
-6 edge stress trends of the finite element analysis
Vertical Distance, from support to apex
instead suggests they have occurred due to creep.
Xochimilco is hot all year round, threaded with
Figure 3.2.4. Smax stresses (k/ft2) along edge parabola canals, and suffers the air pollution abundant
Figure 4.2. Photograph showing shrinkage cracks during Figure 4.3. Photograph showing more diffuse shrinkage
construction. The cracks are most visible on the left side of cracks after the structure has been completed
the image.
Figure 4.4. Patched leaks and drip-stains on the underside Figure 4.5. Patched creep crack and crack pattern layout
of the shell, and their location in plan
VOL. 47 (2006) No. 3 December n. 152
throughout Mexico City. Heat, moisture, and elegant structure that avoided what he called
pollution all accelerate creep, and for decades, Los “extravagant architectural dreams”[4] that can lead
Manantiales’ small mud cracks increased its to great expense and questionable structural
susceptibility to these three elements. As with any performance. It is critical to the future of structural
concrete structure in permanent compression, Los engineering to recognize the high quality of such
Manantiales most likely experienced some small work, significant for its technical virtuosity and
degree of creep independent of its environment. sensitivity to aesthetics.
This natural creep, however, would then have been
magnified by its unfavorable surroundings to REFERENCES
produce the cracks. It is important to note that the
cracks have in no way compromised the structural [1] Billington, David P., “Felix Candela and
integrity or safety of the shell, and that any Structural Art,” Bulletin of the International
deformations in the shell’s present state are so small Association of Shell Structures, no. 88,
as to be invisible to the eye, even in the cracked January 1986, pp. 5-10.
areas. After 46 years, the concrete is still in good
condition, and with vigilance and occasional [2] Burger, Noah, The Xochimilco Shell, Senior
maintenance, will most likely last through another Thesis, Department of Civil and
46. Environmental Engineering, Princeton
University, 2004.
5. CONCLUSION [3] Candela, Felix, personal letter to Anton
Tedesko, December 5th, 1963, Princeton
Because the finite-element analysis in this report is University Tedesko Archive.
relatively straightforward, it has substantial
educational value, especially when connected to [4] Candela, Felix, New Architecture, Maillart
field experiences and simple calculation to explain Papers, Ed. Billington, D. P., Mark, R., &
structural behavior. However, the analysis Abel, J. F., Dept. of Civil Engineering,
ultimately demonstrates not that such analysis was Princeton University, 1973, pp. 119-126.
necessary for the design, but rather that the design
renders the stresses essentially insignificant. It [5] Faber, Colin, Candela: The Shell Builder,
confirms the correctness of the design while Reinhold Publishing Corporation, 1963.
reinforcing Candela’s insistence on the PHOTO CREDITS
unimportance of such analysis when the designer
chooses a structurally appropriate form. As Figure 1 – Erwin Lang. Photo taken from [5, p.
Candela was fond of saying, “the quality of a 159].
structure is in inverse proportion with the amount of
calculations…”[3] (For further insight into Figure 2.3, 2.4 – Cubiertas Ala. Drawings obtained
Candela’s ideas, see [1] and [4].) from Avery Library Archive of Architectural
Drawings at Columbia University.
Even more important is the first-hand inspection of Figures 4.1, 4.4, 4.5 – Noah Burger
the 46-year-old shell, and its evaluation as still in
fine condition. This is a dazzling accomplishment Figures 4.2, 4.3 - Antonio Candela. Photos
for such a thin shell, and is further testimony to obtained from Avery Library Archive of
Candela’s high merit as a structural artist. With Architectural Drawings at Columbia
Los Manantiales, Candela succeeded in creating an University.
DEVELOPMENT AND PERFORMANCE OF A PRECASTING
SYSTEM, FOR CONCRETE HYPERBOLIC
PARABOLOID SHELL CONSTRUCTION
By
Ronald Tacie Noyes
\\
Bachelor of Science
Oklahoma State University
Stillwater, Oklahoma
1961
Submitted to the faculty of the Graduate School of

the Oklahoma State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
May, 1964
OKLAHOMA
ll'ATE UNIVERSITY
LI BRARY
JAN 8 19.,j ~
t
•i
•
'
DEVELOPMENT AND PERFORMANCE OF A PRECASTI NG

SYSTEM FOR CONCRETE HYPERBOLIC
PARABOLOID SHELL CONSTRUCTION
Thesis Approved:
Dean of the Graduate School
570282
ii
ACKNOWLEDGEMENT
The writer wishes to express his appreciation to those

who aided him in this investigationo The writer is indebted
to the Portland Cement Association, the Oklahoma Experiment
Station, and the Oklahoma State University for their finan-
cial support which made this study possible. The valuable
guidance and encouragement of Doctor G. L. Nelson through~
out this study is great:Ly appreciated.

Appreciation is expressed to Professor G:. W. A. Mahoney,
Mro Jim McTaggert, and Mr. Rolla King for their helpful sug=
gestions during the course of the study.
The writer wishes to thank Mr. J. L Fryrear and Mr. D.
R. McCracken for their help in preparing tables and graphs,
and for their assistance and advice during the construction
and testing phases of the research project.
The cooperation from the Civil Engineering Department
and the Mechanical Engineering Department in the loaning of
strain gage equipment is gratefully acknowledged.
The helpful suggestions of Doctor J. W. Gillespie and
Professor R. L. Peurifoy in analyzing the test results are
greatly appreciated.
The writer also wishes to thank his wife, Zona, for her
patience and helpful suggestions throughout the past year.
iii
TABLE OF CONTENTS
Chapter Page
INTRODUCTION • • • • • • * • • • • • • .. fll l
The Problem • • • • • • • • • • • • • • • • • 2
Objectives • • • • • • • • • • • • • • • • • 4
Limitations • • • • • • • • • .. • • • • • • • 4
IL LITERATURE REVIEW e • • • • • • • • • • • • s • • 6
Introduction • • • • • ,, • • • • • • • • • • 6
Precasting. ~ • • • • • • • • • • • . • • . • 7
Hyperbolic Paraboloid Shell Structures • • • 13
Major Problem Areas in H-P Shell
Development • • • • • • • • • • • • • • • • 17
Results and Conclusions of Testing • • • • • 23
III. THE EXPERIMENTAL STUDY • • • • • • • • • • • . . .. JO
Research Outline • • • • • • • • • ·• • • • &> 31
IV. DESIGN OF STRUCTURAL ELEMENTS .... • • 33
Shell Design • • • • • • • • • •_ • • • • • * 3;
Support System Analysis • • • --· • • • • • • • 4o
Tie Bar Design • • • • • • • • • • • • • • • 53
Footing Design • • • • • ~ • • •- • • • • • • 54
v. DEVELOPMENT OF ASSEMBLY COMPONENTS • • • • • • • • 59
Introduction • • • • • • • • • • • • • • • • 59
Formwork and Precasting • • • • • • • • • • • 59
Curing Precast Elements • • • • • • • • • • • 67
Footing Steel • • • • • • • • • • • • • • • • 68
Assembly Supports • • .. • • • • • • • • • • • 68
Lifting Frame • • • • • • ~ • • • • • -• • • • 76
Column Supports for Foundation Casting • • • 79
iv
Chapter Page
VI. ERECTION PROCEDURE • • 0 • • • 0 • • 0 0 0 0 0 80
Site Layout • • • • • • . • • • • • • 81
Column Erection • • • • • • • • • • • • • 84
Tie Bar Connection~ • • • • • • • • • 87
Support System~ ~ ~ • • • • • • • • • • • 87
Shell Assembly. • • • • • • • • • • • • • 89
Welded Shear Connections. • • • • • • • • 90
VII. TESTING APPARATUS AND PROCEDURES • • 0 0 • 0 • 95
Tie Bar Calibration Tests • • • • • • • • 95
Concrete Test Samples • • • • • • • • • • 96
Structural Testing; • ~ • • • • • , • • • 97
Testing Procedure • • • • • • • • • • • • 102
VIII. PRESENTATION AND ANALYSIS OF DATA. • • • • 0 0 105
Analysis of Construction Costs. • • • 10 5
Analysis of Load Test Data. • • • • • • • 112
IX. DISCUSSION OF RESULTS. • • • • • • • • 0 0 • • 135
Assembly Components and Techniques • • • o 135
Discussion of Cost Analysis • • • • • • • 145
Discus~ion of Load Test Results • • • • • 164
X. SUMMARY AND CONCLUSIONS. • • · • • • • • • • • • 169

Results and Conclusions • • • • . • . • • 170
Suggestions for Further Study • • • • • • 173
SELECTED BIBLIOGRAPHY • • • 0 0 ~ 0 0 • • . . . . . . 175
APPENDIX:
A. Tie Bar Calibration Test Data. • 0 0 0 • • 0 • 179
B. Shell Erection Strain Gage Data. .. 0 • • 181
c. Deflection Data from Load Tests I-IV • • • • • 183
D. Strain Gage Data from Load Tests II-IV • • • • 187
v
LIST OF TABLES
Table Page
I. Labor Costs • • • 0 0 • 0 0 0 • 0 0 0 0 0 0 • 106
II. Equipment Costs • • • • 0 • 0 • • 0 0 0 0 0 0 0 0 109
III. Material Costs • • 0
• • • 0 0 ... • 0 0 • • 0 111
IV. Structural Tests Applied to Shell •• • • • 0 0 0 113
v. Properties of Column Sections •• • 0 •
• • • • 0 115
VI. Test IV Strain and Stress Data • • 0 0 0 0 0 0 0 116
VII. Axial Strain Under Uniformly Distributed Load •• 121
VIII.· Stress Conditions Under Uniformly Distributed
Gravity Load from Test IV Data • • • • • • • • 122
IX. Test III Strain and Stress Data. 0 0 0 0 • 0 • • 126
X. Summary of Maximum Stress Values from Test III • 128
XT. Material Savings by Concentric Load Design • 0 0 146
XII. Adjusted Material Costs for Twenty Foot
Square H-P Shell , • , • • • • • • • • • • . • • • 0 148
XIII. Labor Adjustments by Skill Factors • • • 0 0 • • 149

XIV. Labor Reductions Due to Concentric Load Design • 155
XV. Adjusted Equipment Costs for Second Shell Unit • 156
XVI. Projected Labor Estimates for Forty Foot
Square H-P Shell • • • • • • • • • • • 0 0 • 0 160
XVII. Variable Costs for Forty Foot Square Shell • 0 0 163
vi
LIST OF FIGURES
Figure Page
1. Four Basic Hyperbolic.Paraboloid Shell
Configurations • • • • • • • • • • • 0 0 • 0
• • 5
2. The Hyperbolic Paraboloid Shell Configuration
Used in This Study • • • • • • • • • • • • • • 0 5
3. A Typical Parabolic Arch with Horizontal Load
Distribution • • • • • • • q • • • • • • • • • 0 35
Quadrants Showing Dimensions and Stress··
Distribution in the Parabolic Arches • • • • • • 35
5. Shear Distribution in the Shell 0 0 0 0 0 • 0 0 0 37
6. Cross-sectional Views of Exterior Edge Beam
Sections • • • • e o o o o o o o o o o .. • • 0 38
Shell and Edge Beam Reinforcing Steel •• • • 0 0 0 42
Tie Connection Welded to Bearing Plate on Top
of the Column. • • • • • • • • • • • • • • • 0 • 42
Precast Quadrapt Showing Lift Rings •• 0 0 0 0 0 0 45
Lift Ring Detail • • 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 45
11. Side View of Structure Showing Lift, Drag, and
Moment Reactions • • • • • • • • • • • • • • • • 48
Column Steel Arrangement • 0 0 0 0 0 0 0 • 0 0 0 0 48
13. Haunch Detail • • • • • • • 0 0 0 0 0 0 0 0 0 0 0 0 52:
14. Column and Haunch Steel Cage • • 0 0 0 • 0 0 • 0 0 52'
15. Fboting Details. • • • • • • • •• 0 0 0 0 0 0 0 • 55
16. Soil Reactions to Footitig atid Wing Walls Against
Overturning Moments. • • • • • • • • • • • • • • 55
17. Shell Form Ready for Precasting. • • • • 0 0 0 0 • 65
vii
LIST OF FIGURES (Continued)
Figure Page
18 o As:sembly Jig • • • • • • • • • • • • • • • • .. o .. 70
19. Temporary Center Support • 0 • 0 • 0 0 0 0 75
zo. Lifting Frame •• 0 0 0 0 • 0 0 0 0 0 0 0 0 0 0 78
21. Erected Column Showing Cribbing. • 0 0 0 0 0 0 0 0 78
Worksite Layout Showing Construction Control
Points o o • o o c o o o o o o o o o o o o 0 0 0 82
23 0 Material Layout for Assembly • • 0 0 0 0 0 0 0 0 0 82
24. Method of Lifting ?recast Column During Erection • 85

Second Quadrant Being Lowered During Assembly. 0 0 85
Support System Holding Shell Quadrants for
Weldingo o o o o o o o t) o o o o o o o o 0 0 0 0 91
Roof Center Showing Method of Pulling Quadrants
Together • • • • • • • • • • • • • • • • • • 0 0 91
28. Completed Structure • • • • 0 0 0 0 0 0 0 0 0 0 0 0 93
29. Positions of Strain Gages on Columns and
Haunches • • • • • • • • • • • • • • • 0 0 0 0 0 100
30. ManDmeter Deflection Apparatus Showing Reservoir
and Moveable Section • • • • • • • • • • • • o • 100
31. Columns Showing Directions of.Bending~.
Dimensions, and Reactions • • • • • • • • 0 0 0 0 117
32. Axial and Bending Strain Distribution During

Sustained Loading in Test IV o •••• 0
0 • 0 0 0 119
33. Free Body Diagrams of Theoretical Stress
Conditions Acting on Column During
Uniformly Distributed Roof Loading • • • • • • • 120
34. Axial and Bending Strains in Direction of
Overturning During Test III • • • • • • • • 0 0 0 127
35 0 Definition Sketch for Heavy Bending Strain • 0 0 0 12?
36. Neutral Axis Locations in Columns Under
Cantilever Loads • • • • • • • • • • • 0 0 0 0 0 131
viii
Figure Page
37. Roof and Column Deflection Curves for Test II •• 132
38. Roof and.Column Deflection·cu.rves for Test.III.
and IV. • • '.~ • 0 0 0 0 ~ 0 0 0 0 0 ~ 0 0 0 0 133
39. Stress Versus Load and Time Curves, ~est IV • • • 134
40:. Labor Requirements.Adjusted by Three Methods
for Comparison. • • • • • • • • • • • • • • • 0
41. Variation of Labor and Erection Apparatus Costs

with Increased Number of Shells Constructed •• 162
ix
CHAPTER I
INTRODUCTION
The uses of reinforced concrete have been greatly

expanded as a competitive structural material in the
building industry during the last decade. American archi-
tects and design engineers have recently begun to utilize
reinforced concrete in designing such configurations as
the folded plate, dome, barrel shell~ and the hyperbolic
paraboloid, henceforth discussed as the h-p. Because of
its inherent qualities of variable strength~ plasticity~
11
built-in 11 fire protection and low maintenance costs, this
iv versatile product Vi has become popular with designers as
a structural material.
Until the late 1950us the principal precast concrete
construction consisted of the use of concrete blocks~ which
limited construction mostly to modular·units. Some exper=
imental work in tilt-up construction was also conducted.
However, during the past five years the use of concrete
shells utilizing the four basic geometric shapes previously
listed, has broadened the outlook of the concrete industry
considerably. A major advantage which the use of concrete
shell roofs has introduced is the large savings in material
and reinforcement while covering large floor areas. Due to
1
2
the basic shapes of the shell structures~ large vertical

loads are readily transmitted into axial stresses in the
plane of the roof where they are transferred into edge
memberso Thus~ the tensile and compressive properties of
reinforced concrete are utilized effectivelyo
If these advantages of reinforced concrete shell con-
struction can be further enhanced by the development of
suitable methods to precast and erect these structures~
costly f ormwork and large crews of skilled labor can be

eliminatedo Only then can reinforced concrete compete
favorably with other producers of pre-packaged buildings
on the consumer market.
The Problem
Trends in light building construction in the United

States during the last decade indicate the need for a prac-
tical~ economical~ one story structure for use as a farm
or light industrial buildingo Several types of buildings
of various building materials have been developed and are
now available on the market~ however~ at the present time
few concrete structures are available that are ready for
assembly on a selected siteo
A limited amount of research has been done on pre~
casting light concrete structures. Most of the work done

previously has been on prefabricated steel or wooden frame
structureso Some work has been done on the design and in-
sitµ construction of the basic thin shell concrete
3
structures using the principles of the folded plate, dome,

barrel shell, and the h-p configurations. To compete
favorably with steel and wooden prefabricated structures
in satisfying the demands of the light construction indus-
try, the concrete producer needs structures which can be
precast and assembled rapidly, have versatility, and have
close tolerances.
The general requirements of a structure which will
satisfy the demands of agricultural or light industrial
use are defined as follows:
It should (1) be structurally sound,(2) be attractive
(3) be erected by small contractors, (4) provide good fire
protection, (5) provide low maintenance costs and depreci-
ate over a long time period, and (6) be functionalo
The major requirements which must be overcome in the
design of a concrete shell precast structure are: (1)
suitable joints and connections, (2) a simple system of
precasting, (3) capability to resist moving or transporta=
tion stresses in the precast elements intransit to the
construction site, (4) efficient use of formwork, (5)
adequate lifting and placing equipment, (6) a temporary
support system and sequence for erecting and assembling
the precast elements, (7) adequate lifting devices and
attachments on all precast elements, and (8) an adequate
footing system to support the building during adverse
atmospheric and soil conditions.
4
Objectives
The objectives of this study are:

l~ To design a precast shell which can be incorp-
orated with a precast column system to form a
h-p shell structure.
2~ To develop a readily transportable support
framework which will rigidly stabilize the
structure during erection and assembly.
3; To develop a step-by-step procedure for the
assembly of a precast h-p shell.
4* To load test the structure to verify the
structural design.
5. To evaluate construction costs of the system.
Limitations
Of the four basic shell structures which are most

frequently used today, the h-p, Figure 1, has proven to be
one of the most efficient and easily analyzed. Because of
the financial limitations of this project, the study was
limited to one structure, a four quadrant h-p with two sup-
porting c6lumns, Figure 2. This structure seems to compare
favorably with the other three basic h-p structures, Fig-
ure 1, as far as the rural customer is concerned, because
it is suited to a number of agricultural uses.
The prototype used for this research was limited to
a 20 by 20 foot shell as this size was adequate for test-
ing the procedures developed in this study.
5
(8)
(C)
Figure 1. Four Basic Hyperbolic Paraboloid

Shell Configurations.
Figure 2. The Hyperbolic Paraboloid Shell Con-

figuration Used in This Study.
CHAPTER II
LITERATURE REVIEW
Introduction
The precasting of reinforced concrete shells is a

relatively young field in modern construction. This meth-
od of erecting concrete structures lacks only the develop-
ment of assembly line procedures in factory or job site
casting beds and standardized erection methods on the site
before it can be a competitor with other prefabricated
structures. Precast concrete elements are becoming in-
creasingly popular in the building industry due to the
savings which are realized by simplified or eliminated
formwork and by the multiple use of forms. Less skilled
labor is required during erection, quality control is
greatly improved, and structural elements may be factory
cast year 'round, which reduces lost time on the job for
curing the concrete. Curing processes can be more closely
controlled by steam or hot water curing in plant precastin.g.
Precast units can be stockpiled in erection sequence, thus
eliminating excessive handling Q.nd storage on the worksite.
6
7
Precasting
Beauchemin (1) discussed the advantages and disadvan-

tages of precast concrete. He concluded that the advan-
tages outweigh the disadvantages by a large margin. Listed
below are the specific advantages and disadvantages which
are usually encountered in precasting with a brief discus-
sion of each.
1. The principle disadvantages of precasting are:
a. Shrinkage
To completely hydrate, one bag of cement
requires approximately 2 U.S. gallons of
water. Extra water added to give plas-
ticity to the mix, not only weakens the
paste but also causes the concrete to
shrink when portions of it dry out.
Added steel reinforcement will aid in
resisting stresses due to shrinkage.
b. Weight
The average concrete mix weighs in the
range of 140 to 150 pounds per cubic foot.
This means that in precast concrete work,
handling becomes more difficult and, there~
fore, more expensive. Hauling and shipping

costs.are higher, placing requires special
lifting and handling equipment; and the
over-all size of the precast unit is
limited.
8
These weight difficulties can be reduced

by use of (1) hollow cores, (2) high
strength concrete, (3) lightweight aggre-
gates, (4) prestressing, or (5) various
combinations of these items.
c. Assembly and Continuity
Practically every precast concrete product
must eventually be connected to another
concrete product or to some other con-
struction material. Large pieces, such as
columns, roof slabs, wall panels, and
beams, present assembly and joint difficul-
ties. Reinforcing bars and dowels which
protrude from precast elements are some-
times welded or bolted together or to
other members at construction joints to
develop continuity. Post-stressing is
also used to join several components into
a composite unito
2. The principle advantages of precasting are:
a. Economy
Precasting economy is incurred by labor
and forming cost reductions.
b. Quality
Quality is closely controlled by the use
of right mixes and by maintaining optimum
humidity and temperature during the curing
9
process. These control measures can best

be achieved in a concrete products plant.
Concrete placement is facilitated because
of low level formwork and mechanical
placement aids, such as vibrators.
c. Speed of Construction
The erection of precast concrete takes far
less time than in-situ concrete, because
the former requires little or no -formwork
or curing period, unless grouting is re-
quired; even then curing time is reduced.
d. Flexibility
When concrete is poured, it is a plastic
material which can be molded into any shape
or form desired. This quality, peculiar to
concrete, is one reason why concrete has
always interested designers. A.l though some
shapes are difficult to achieve with in-
situ concrete, precasting supplies these
shapes with greater ease and economy.
e. Availability
Wherever the construction may take place,
the designer can usually locate a good
source of precast concrete products
within a reasonable distance.
Several precast structures built within the past four
years indicate the potential of precast construction.
10
Amirikian (2) designed and erected a multipurpose building

of precast thin shell panels. This structure was a frame
type building with bolted joints and connectors using a
modular system of panel assembly. For this frame design,
Amirikian used the statically determinate three hinged
bent principle. He stated:
As yet no standard procedure for the assembly and
erection of a panel building has been devised.
This is something which must be developed by
builders and fabrication geared for mass produc-
tion and erection of these structures.
In 1957, the gymnasium roof of the Westmore High
School of Daly City, California, was constructed of pre-
cast reinforced concrete barrel arches supported by precast
bents. The arch shaped units, each 61 feet long, 15 feet
wide, 31/2 inches thick, with a 31/2 feet rise, weighed 20
tons. The bents were three-hinged and spanned 91 feet.
The barrel arches were made of 3,000 psi lightweight aggre-
gate concrete which weighed 102 pcf. The contractor set
up a casting line using six sets of forms and cast six
shells in less than four hours. No camber was introduced
for the 52 foot span as the maximum deflection at midspan
was only one-half inch. The erection of the 20 ton units
was handled by two 50 ton capacity cranes.
Early in September, 1960, Hurricane Donna cut across
the island of Puerto Rico leaving hundreds of people home=
less. Within hours after the tragedy, IBEC Housing Corpor=
ation proposed a program of commercial housing to the
Puerto Rican Government at a cost of 1,000 dollars per
11
living unit. IBEC and the contributing suppliers delivered

a house shell, approximately 630 square feet in area on a
foundation slab measuring 21 by 31 feet with front and back
doors and eight jalousie-type windows installed. The
entire package included 12 elements; six flat wall sec~
tions, two curved wall sections~ two roof beams and two
roof sections. Components were precast in IBEC 1 s casting
yard near San Juan and trucked 45 miles to the site. A
test house was erected in one hour using a seven man crew.
Faerber (3) describes the construction of a precast
folded plate roof for a residence in Naples, Florida. The
building was designed in the shape of an octagon, incorpo-
rating eight separate gables, each designed as a folded
plate. The roof slabs were cast one on top of the other
as in the lift slab method of construction. Sheets of
polyethylene film were used to facilitate separation. Each
of the 400 square foot sections were four inches thick and
weighed 11 tons.
A slightly different approach toward precast concrete
was conducted by Riley (4). He constructed a barrel
shaped roof by shaping the earth into the desired form and
precasting his roof in place. The columns were placed
prior to casting the roof in order to allow sleeves cast
in the roof to utilize the columns for stability during
the erection sequence. The roof was cured on the casting
bed, then raised by means of hydraulic jacks. After the
roof was raised, the mound was leveled off and the floor
12
slab was cast in the conventional mannero Riley used a

polyethylene film between the earth and the concrete on
the casting bed which gave a smooth undersurf ace to the
roof~ making finishing the underside of the roof unneces=
saryo This particular method of construction has limited
uses but the method of forming the casting bed from an
earth mound has good future possibilities. This type of
form could be re-used many times by jacking or lifting the
forms off the casting bed or by the lift slab principle
where several slabs can be cast one on top of the other.
The second method could be used to store a limited number
of slabs in place for use on a specific project.
One of the largest precast concrete construction
projects yet undertaken is described by Thompson (5). This
project consisted of roofing the new terminal building and
ticket building of the Oakland International Airport in
Calif orniao Two types of precast shells were usedo The
roof of the t-erm.inal building consisted of inverted 1l.llll.=·
brella shaped surfaces; each precast element had a

h~p
minimum shell thickness of 21/4 inches and weighed 16 tons o

The conoidal shaped barrel vault was used to span the
ticket buildingo Both roofs used a mix design of 2~750
psio The h=p shells were designed to support a full load

of water in case of a plugged draino
Many more examples of precast shell construction sim=·
ilar to those mentioned above are in evidence at the

present timeo Extensive preparations and plans have been
13
made by several organizations to promote precast concrete

design and constructiono One of the major promoters of
precasting~ the Portland Cement Association~ publishes
several types of literature promoting the use of precast
shell structures utilizing new space frame techniqueso The
American Concrete Institute has placed additional emphasis
on 'the need for uniform practices in precast concrete
design and construction by additions and major revisions
of the ACI Building Codeo
Although there are various reasons for the design and
construction of each individual shell structure~ the pre=
dominate motive is economyo Because of the substantial
savings of time~ material~ and manpower which can be ob=
tained by precasting 9 the demand for precast reinforced
concrete elements in building construction will continue to
grow.
Hyperbolic Paraboloid Shell Structures
Felix Candela (6) ~ internationally recognized in t:he

architectural world for his extensive work with thin she
h=p surfaces~ stated~
Hyperbolic paraboloidal surf aces are extremely

interesting from a structural and constructive
point of view. Their use in reinforced concrete
shell roofs offers the same advantages inherent
to all shells of this material~ i.eo ~ lightness~
incombustibility~ economy of material~ security
against explosions~ bombardments and earthquakes~
and little sensitiveness to foundation settle-
ments" These last properties are consequences
of 'their structural action; not restricted to one
plane~ but working as space~frames.
14
The theory of the h-p is an old one but only in the

past decade have the basic principles of membrane stresses
been put to work as economical space enclosures. Although
they are doubly curved surfaces~ the surfaces are formed
by two systems of straight generatriceso This fact greatly
simplifies the basic f ormwork for casting the shell by
allowing the f ormwork to be composed only of straight
lumber~ provided the shell is rectangular in the horizontal
planeo The principle stresses which exist in the h=p
shell surface are tensile and compressive stresses which
form angles of 45 degrees with the direction of the
generatrix. These stresses accumulate and are transferred
by shear from the shell edge into the edge beams which are
parallel to the generatriceso The shears accumulate along

each side of the warped parallelogram~ resulting in either
tangential tensile or compressive forces which are redis=
tributed in the shell or act as compressive thrusts a-t the
column. .By taking one h=p quadrant and combining it with
three other quadrants of the same dimensions? a variety of
st;ructural shapes can be obtainedo
Candella (7)~ in his discussion of warped surfaces at
the Conference on Thin Concrete Shells in 1954 at the
Massachusetts Institute of Technology~ pointed out the
fact tha.t the h=p stress analysis does not involve higher
mathematics~ and is even elementary when surfaces with
small slopes are consideredo He stated that 00
on account
of their double curvature~ it suffices to investigate the
15
membrane state of stresses, without considering bending or

deformationo 00
Candella was one of the first designers in the

western hemisphere to experiment with h-p 0 s. His first
h=p shell was the Cosmic Rays Pavilion at the University
of Mexico. This structure has received much acclaim due
to its extremely thin surfaceo Because of a functional
requirement that the top part of the shell have no more
mass than eight pounds per square foot~ the shell thick-
ness was only five-eighths inch in the upper part of the
structure. The success of this structure prompted him to
design other structures as h~p shells; one of the most
notable was Rio's Warehouse in Mexico City. The basic
program requirements were to economically cover 55~000
square feet of floor space and at the same time to provide

a small amount of roof light~ a clearance height of 15
feet~ and 50 foot bays. The solution was found in a
reinf orced=concrete structure containing 36 umbrellas
which were approximately 30 feet by 50 feet. Standard
weight 2~000 psi concrete was used throughout and was
vibrated evenly in the thin shell roof. As a result 9 the
good compaction eliminated the need for waterproofing the
roof shells. By tilting each umbrella slightly~ Candella
obtained a north light effect in a very economical maru;,er.
In a warehouse built more recently from h=p shells~ he
used glass blocks cast in the roof slabs for additional
top lighting. To solve the problem of footings in one of
16
the world 9 s worst subsoils (150 feet of clay which varies

from 75 to 90 per cent water content by total weight)~
Candella designed an umbrella shaped footing which he cast

over a shaped earth moldo
Parme (8) presents a good mathematical analysis of
the h=p shell theory. His discussion shows that there are
no forces normal to the edges of an h=p shell subject to a
uniform load. Parme stated that~
For most hyperbolic paraboloid shells of moderate

rise~ it is deemed satisfactory to consider the
load as uniform. However~ when the rise is great~
the dead load can no longer be considered as uni=
form on the projected area.
One of the largest single h-p shell units in the
United States is the entrance to a new department store
which is part of the Denver Court House Square Development.
The shell~ designed by Tedesko (9) was opened to the public
in Augusts 1958 as an exhibit pavilion. The roof consists
of four h=:p surf aces and is supported through steel hinges
on buttresses at the four corners of a rectangle. The
three inch shell which rises to a height of 28 feet
spanned a floor area of 14~800 square feet.
An example of the economical large scale production
using h=p roofs is illustrated by Madsen and Biggs (10).
An h=p shell roof was designed for a shopping center in
Minneapolis~ Minnesota. The structure consisted of 44 h=p
shells? each 46 feet 4 inches by 48 feet 6 inches~ covering
a floor area of 100 ~000 square feet. The structural design
follows the classic formula developed by Felix Candela •..
17
This construction program was designed around the reuse of

movable forms. They scheduled all construction operations
on an assembly line basis. The steel was pre-tied where
possible so that placing could be accomplished in a mini-
mum time. The concrete placing was scheduled so that eight
h-p 0 s could be cast per week, allowing 44 hours curing
time before forms were stripped. Curing procedures were
started within five to ten minutes of final smoothing. A
check on deflections after the 28 day curing period indi-
cated that the corners of the h-p 0 s had deflected tbree-
fourths inch, while at the midpoint of the edge, the
deflection was also three-fourths inch.
Many applications of h~p shells are being used,
ranging from airport structures, hospitals, libraries, and
industrial buildings, to modern farm structures. In addi-
tion to its application as a roof surface, the h=p has
been well adapted to use as a foundation structure for
soils of low bearing pressures. New ideas are continuously
being develope~ to utilize its full potential as a struc=
tural shape. Many problems continue to exist in h~p con-
struction leaving opportunities for future development.
Major Problem Areas in H-P Shell Development
Although many problems exist in designing and con-

structing h-p shell structures, some are more predominant
than others. Some of the major problem areas are listed
below. Each problem will be discussed specifically and
18
analyzed in terms of this project.

(1) Footings Because of the various shapes and
sizes of h-p shells and the conditions under
which they are erected, footing problems will
vary from one location to another. Footing
systems are usually broken down into two groups~
the in-situ footing and the precast footing.

Each method of placement is dependent upon the
local soil conditions and~ therefore~ must be
designed under the same criteria. Footings of
both types can be standardized to some extent .
for a specific structure but must be checked
for each individual building site to determine
the design adequacy.
(2) Lifting Equipment - Building sizes 9 materials
to be handled, and location of construction
sites determine the types and sizes of lifting
equipment required for specific construction
projects. For in-situ construction~ the li..fting
requirements are usually limited to fairly small
loads such as steel members and concrete buckets.
However~ in precast construction~ the sizes and
weights of precast elements may be quite large
and are the major factors which determine the
crane sizes. Peurifoy (11) gives a good analy-
sis of the.safe lifting capacity and radius of
operation of several sizes of cranes which could
19
be used to lift bulky precast elements.

(3) Formwork - The costs of concrete f ormwork may
be excessive if multiple reuse of forms is not
made possible. Minor structural failures in
forming systems are relatively common. A study
of the cause of structural forming failures was
made by ACI Committee 622; the most common
deficiencies leading to form failures in build=
ing construction are listed in their report (12)o
(4) Cg!'ing - Proper curing of concrete elements is
one of the most difficult operations in construe=
tiono Optimum curing is usually desired on con-
struction projects to obtain maximum concrete
strengths~ but is usually difficult to attaino
The variables which control curing are (1)
temperature~ (2) moisture content~ (3) time~ and
(4) freedom from physical disturbanceso Pro~
viding the temperature is acceptable~ only the

moisture content need be controlled if the mass
is free of physical disturbances during the
curing period. ACI Committee 612 (13) makes the
following recommendations for optimum curing~
(a) Horizontal Units

1. Initial curing - As soon as finishing
operations are completed~ cover with
two thicknesses of an approved woven
fabric or quilted fiber mat which is
20
saturated when placedo Cover is kept

saturated with water and is kept in
place until the heat of hydration has
been dissipated.
2. Final curing - (a) Same cover left in
place throughout the curing period~ (b)
two inches of moist earth or sand con=
tinually saturated~ (c) three inches of
wet hay~ grass~ or clean straw uniformly
distributed and saturated continuously 9
(d) approved impervious light colored
paper or plastic covering placed in
constant contact with the concrete sur=
face~ or (e) approved impervious com=
pound or coating sprayed on the surface
in liquid form.
Coatings should be light in color when
concrete is exposed to the direct sun~,
lighto When the temperature is above

0
40 F~ the final curing agent should
remain in place at least 72 hours or
more as strength requires. When the
air temperature is less than 400 F~ con=
crete should be so protected to main=
tain 50 to 70°F.
(b) Precast Units
1. Initial curing - Immediately after the
21
casting operations~ enclose each member

by two layers of an approved water-
saturated fabric until placed in posi-
tion for final curing.
2. Final curing - Members may be cured
under the original saturated fabric~ or
moved to a special chamber where they may
be uncovered in a completely saturated
atmosphere of mist~ water~ or steam.
The temperature for a curing room
should be uniformly maintained between
50 to 180°F. Final curing may be per-
formed under a pressure between 100 psi
and 150 psi in saturated steam at 335
to 366°F.
In many cases~ precasting on the job site
will not permit use of factory controlled
final curing procedures. In this case~ the
final curing procedures listed under
Horizontal Units should be applied.
(5) Joints and Connections - Of the many problems
encountered in h-p shell development~ whether
precast or in-situ construction~ designing ade-
quate joints and connections is one of the most
difficult problems to overcome. Cazaly (14)
points out the fact that joints must: (a) with-
stand bending moments without breaking down~ (b)
22
absorb concentrations of stress and strain, and

(c) occupy minimum space and present a neat ap-
pearance when exposed. For economic reasons,
they must be: (d) safely formed by normal labor,
(e) cheap to fabricate without expensive or
excessive formwork~ (f) capable of erection in
all kinds of weather, (g) fast to erect without
cranes and other trades~ and (h) able to take a
considerable amount of tolerance. Mr. K. C.
Naslund (15) summed the problem of joints and
connections neatly when he stated that:
The engineer must determine his scheme
of erection, then design his members and
joints for the stresses that occur during
fabrication~ delivery, and erection, as
well as with final conditions. He must·
visualize how the members will be erected
to assure that the erection is safe,
feasible, and that it is economical.
(6) Safety = Safety is a continuous problem in prac=
tically all types of construction, yet it is too
often overlooked on the job as well as in the
designo Design safety factors and features
should be one of the first considerations given
to a structural design.
(7) Waterproofing = Waterproofing of shell surfaces
can usually be accomplished by three basic meth-
ods: (a) use of a built=up roofing surface such
as a bituminous coating~ (b) use of a sealing
compound such as a neoprene roofing material or
23
a light colored polyester-based paint to seal

the pores in the concrete, and (c) by designing
the concrete to obtain a dense, impermeable
masso Other methods of waterproofing are
available~ but are less frequently used. Of
the three methods mentioned~ the newest method
is the use of a neoprene roofing compound which
can be placed on the surface with paint rollerso
For example~ one commercial product,
Armstrong F/A Roofing is applied in three basic
steps: (a) joints are sealed with a deck sealer
and flashing tape is applied as a reinforcing
membrane where needed~ (b) two layers of the F/A
400 Base Course are applied in two colors to aid
in visual gaging of the film thickness during
the second application~ and (c) two applications
of F/A 600 complete the installation and provide
a final waterproof coating and a variety of
roofing colors for modern structures. The
favorable characteristics of the neoprene com-
pounds are its flexibility for expansion or
contraction of roofs, versatility for conforming
to any roof surface or slope, and ease of repair
or maintenance.
Results and Conclusions of Testing
1. H-P Models and Prototypes

24
Several tests have been conducted on h-p shell struc-

tures to determine the capacity of shells under a variety
of testso One recent test was conducted by the Structural
Development Section of the Research and Development
Division of Portland Cement Association (16). The shell
used for testing was an inverted umbrella with a 24 by 24
feet outside dimension~ a i1;2 inch shell thickness and a
2 feet 10 inch rise. The reinforcing in the shells con-
sisted of Noo 3 bars at 12 inch centers.
The loads which were applied were: (a) a uniform
vertical load of 50 psf ~ (b) four equal concentrated loads
applied symmetrically on the shell, one at the center of
each quadrant~ using a 2 by 3 inch washer as a contact
area~ and (c) an unsymmetrical loading of 75 psf was ap-
plied to two adjacent quadrants. During the uniform load,
the sum of dead and live load produced a calculated thrust
in the perimeter beam reinforcing of 26 9 300 psi. No exces=
sive stress was noted under this load. The concentrated
loads produced some minor radial and circumferential
cracking at the load points when the loads reached 5,000
pounds. This load also produced a local bending moment of
103 kip=ft/ft at the point of application of load with a
punching shear of 500 psi. No distress was observed dur-
ing the unsymmetrical load over the major portion of the
shell even near ultimate capacityo
The tests on this shell demonstrated that h-p shells
with a thickness of only i1;2inches can resist large
25
concentrated loads as well as unsymmetrical loads.
A similar umbrella shell was constructed and tested
at Oklahoma State University in 1962 (17). The size of
this shell was 20 by 20 feet with a minimum shell thick-
ness of 2 incheso Testing of this shell was accomplished
by closing the drain and filling the shell with water.
Deflection readings were taken at each corner after each
load increment was addedo A total load of approximately
14~000 pounds~ or 35 psf horizontal loading was applied.
The maximum deflection noted under the total design load
was 0.004 feet. No other effects of strain were noted.
Harrenstien (18) discusses tests conducted on two
reinforced concrete h-p shell prototypes which were con-
structed as a class project at Iowa State University. The
shells were 10 feet square in plan~ 1 inch thick, and had
a maximum rise of 1 foot 8 inches in 5 feet. The average
28 day compressive strength of the concrete was 7~500 psi,

6
and the average modulus of elasticity was 4o75 x 10 psi.
The inverted umbrellas were mounted inversely on an 8 inch
steel column and were loaded simultaneously by a single
hydraulic cylinder jack as a concentrated load on each
shell. Point loads were individually applied at 60 loca-
tions on each shell~ with a maximum applied load of 548
pounds at each point. A system of strain gages located
the principle stress contours for each individual load.
The test results were used to set up a prediction equation
to determine the final stresses in the shell ~ue to applied
loads.
26
Waling and Greszczuk (19) conducted experiments on

thin-shell h-p models at Purdue University~ using styro-
foam stretched on high strength wires as a formingmaterial,
Results of their studies indicated that styrofoam would
make a good forming material for field constructiono
Although other testing has been done on reinforced
concrete h-p shells, these examples indicate the types of
experimental work that have been conducted,
2o Lightweight Aggregate Concrete
Lightweight aggregate concretes are now generally ac-

cepted for conventional constructiono They are especially
useful when they will produce a structural strength equal
to that produced by normal weight concrete, and at a lower
cost. A savings can be obtained due to lower total weights
of structures~ which require smaller or lower strength sup-
port memberso Lightweight concrete also has up to 5.5
times the insulation quality of standard weight concretes.
To determine their structural qualities, a considera-
ble amount of testing has been done on lightweight aggre-
gate concretes. Several of these tests have been conducted
to compare lightweight aggregate with conventional aggre-
gate concretes.
Hanson and Kleiger (20) made extensive tests of the
freezing and thawing performances of lightweight aggregate
concrete as compared to normal weight concrete. They
tested nine lightweight aggregates and one sand and gravel
27
aggregate. Each aggregate was used in both an air-dried
and a saturated conditiono The ten samples were each
designed at two different strength levels~ 3~000 psi and
4~500 psio Various percentages of air entrainment were
introduced in the test sampleso The conclusions derived
from the freezing and thawing tests were~
(a) Entrained air increases resistance by
freezing and thawingo
(b) The amount of entrained air required
for both conventional and lightweight
concrete is approximately the sameo
(c) The' initial moisture condition of the
lightweight aggregate has a significant
influence on the resistance to freezing
and thawing compared to only minor in-
fluences for the standard concreteo
(d) The variation in durability among the
concretes made with the different light-
weight aggregates appears no greater than
might be encountered with normal weight
aggregateso
(e) Aggregate properties are obviously of
importance in determining the level of
durability~ even in air-entrained concreteso

)
Hanson (21) tested seven commercially available light=
weight aggregates in reinforced concrete beamso These
tests were part of an investigation by the Portland Cement

28
Association Laboratories to augment available technical
information necessary for the design of structural con-
crete using lightweight aggregateso In these tests, the
beams were loaded at third pointso The results indicated
that at comparable strengths the sand and gravel concretes
showed a nominal shear strength no greater than that of
the lightweight aggregate concretes. It was evident
throughout the tests that the lightweight beams failed
more suddenly than the sand and gravel concretes~ espe-
cially at higher concrete strengths.
Shideler (22) conducted tests on eight lightweight
concrete samples and one standard sample. Results from
his testing indicated that:
(1) Structural grade concrete was obtained with
each of the lightweight aggregates.
(2) The unit weights of the various lightweight
aggregate concretes in the lower strength
series (3~000 to 4,500 psi) ranged from 90
to 110 psf compared to 146 psf for standard
concrete a
(3) The various lightweight aggregates require
a wide range of cement content to produce
similar strengths.
(4) The modulus of elasticity of the lightweight
aggregate concretes in the 3~000 to 4,500 psi
series varied from 53 to 82 per cent of the
modulus of sand and gravel concrete at 28 dayso

29
(5) Flexural strengths of the lightweight and

sand-and-gravelconcrete were approximately
equal at early ages, but after 28 days the
standard concrete showed greater strength
gain with continuous moist curing.
(6) Bond strengths of some of the lightweight
concretes were approximately equal to those
of sand and gravel concretes.
Shideler (22) concluded that within the group of
lightweight aggregates studied, rather wide variations
were obtained in the structural properties of the concretes.
He felt that it was important that the individual producers
of lightweight aggregates for structural concrete conduct
investigations to provide reliable design data on the per-
formance of their product.
In recent tests, Hanson (23) has determined tensile
strength and diagonal tension resistances of structural
lightweight concrete. Comparisons of the unit shear
strengths at diagonal cracking with the ACI Building Code
working stresses reveal that inadequate factors of safety
existed for the lightweight concrete beams with long spans
and low steel percentages.
Due to the high moisture absorption characteristics
of lightweight aggregates, the American Concrete Institute
has published a new standard, "Recommended Practice for
Proportioning Lightweight Aggregate Structural Concrete
(ACI 613A-59)."
CHAPTER III
THE EXPERIMENTAL STUDY
The procedure used in this research program was to
design~ erect, and test a precast concrete shell. The
plan of research that was followed is discussed in the
following section. However, before a procedure was set up~
certain problem areas were outlined on which the study was
made. Problem areas which were investigated are~
(1) Use of lightweight aggregate concrete for the
shell surface.
(2) Joints and shear connections between precast
roof elements and edge beams.
(3) Support system to stabilize roof elements dur=
ing erection.
(4) Lifting device to lift shell elements into
position.
(5) Precasting system for shell.
(6) Suitable erection and assembly techniques.
(7) Foundations.
(8) Shell to column anchorage.
Due to the limitations of this program~ this study was
conducted on one specific configuration of the h-p shell.
30
31
This structure consisted of a four-quadrant shell with two

supporting columnso This type of structure appeared to be
more practical to precast than a structure with four sup-
ports when used as modular units for large roof areas such
as a hay storage shed, an equipment shelter, or a dairy
barno It seems that the two-support structure would be
more appealing to farmers for small structures as it ap-
pears more stable and yields a larger clear floor area
than the inverted umbrella structure, when used in widths
of only one modular unit.
Research Outline
This program was outlined to meet the objectives

stated in Chapter I by analyzing the problems listed in
Chapter II in a logical ordero Although separate stages
of research were carried on concurrently, the steps which
were f ollowe·d are~
(1) Design the shell and supporting structure for

precasting.
(2) Design and construct a simple support system
to stabilize precast elements during assembly
and erection •.
(3) Prepare a building site and layout plan for
precast units and construction material and
equipmento
(4) Cast shell components and take samples of
material at time of casting.
32
(5) Test samples at various time intervals to
determine strength of precast elementso
(6) Erect the columns and tieo
(7) Erect the roof and record time required for
a9sembly.
(8) Load test the structure to obtain load de-
flection data.
(a) Design live load was 40 pounds per
square foot~ uniformly distributed.
(b) Test to approximately 1 1/2 to 2 times
design live load.
(c) Test to design load on one-half the
roof surface.
(d) Measure deflections during sustained
loading.
(9) Analyze the test datao
(10) Prepare a detailed cost analysis on labor~
equipment~ and material requirements for
precasting and erection.
(11) Assemble pertinent data and combine results
of research on erection procedure~ testing~
and cost analysis.

CHAPTER IV
DESIGN OF STRUCTURAL ELE:MENTS
Shell Design
The basic design of the prototype shell was analyzed

using the basic equations listed in the Portland Cement
Association publication, Elementar_y Analysis of Hyperbolic
Paraboloid Shells (24).
Several basic decisions were made initially concern-
ing the desired parameters of the prototype. These design
factors were~
1. The over-all dimension in plan would be 20 feet

by 20 feet.
2.. The minimum shell thickness would be two and
one-half in~hes.
J~ The vertical rise, h, would be three feet.

4. Design static loading would be 40 psf in the
horizontal plane.
5. Lightweight aggregate concrete with an est-
imated density of 110 pcf would be used
for shell material.
JJ
34
Horizontal Thrust in Parabolic Arches
Because of the doubly curved surf ace of the h-p

shell, the load, W, is supported by two arch-like elements
so that each element will support one-half of the load
intensity, ~' Figure 3o The internal moment developed in
w ]}
this two-hinged arch is 2 ° 13- H (-hxy) = o~ or
w L2 W ]}
H (-hxy) = 2 13 o Thus, H = - 2 S-~ = - 16 h
WI}
0
xy xy
where
H = the horizontal thrust at the end of each arch
per foot of shell widtho
hxy = vertical rise of each archo
L = horizontal length of each archo
Further simplification of the analysis for horizontal
. ld s H
thrus t , H , y1e W•aob F" 4 h
2 h , 1gure , w ere
=
w = the total unit load in pounds per square foot.

a = the length of a horizontal side of one
quadranto
b = the length of the adjacent horizontal
side of the quadranto
h = the vertical rise of the shello
The approximate dead load per square foot = 2o5 ino x
1 110 lbo
12 in/ft. x f~ = 22.9 lb./ft~ or 23o0 lb./ft~
The design static load = 40 lb/ft!
Total design load = 40o0 + 23o0 = 63o0 14/fto
H = ± wab
2h -
- ± 63o0 x 10 x 10
2 x 3
= ±
35
1------L.--..L-,--L.-.-....1..-.,.......l--4--1-.......-'---'-~..........,_---l......,....---t-----.
fw
~--A---L--L-.L-4---L----J-~~~~
H
I
hxy H
L/2 . .I
L
Figure 3. A Typical Parabolic Arch With
Horizontal Load Distribution.
Figure 4. Quadrants Showing Dimensions and Stress

Distribution in the Parabolic Arches.
36
Tensile Shell Reinforcement
The area of steel required per foot of shell width in

the direction of the parabolic arches, Figure 4, is A~ = H =
fs
1,050 lbs. = $0525 sq. in~
20,000 psi
Area of Steel Reguired Perpendicular to Edge Beams
To simplify steel placement during forming, the shell

steel area perpendicular to the edge beam was computed;
thus, As = A~ x sec 45° = .0525 x 1~414 = .075 sq~ in~/ft.
No. 2 bars at eight inch centers were used to provide
~08 sq. in. of steel per foot~
Although theoretically no steel was required for the

parabolic arches in compression, the same amount of shell
steel was placed in both directions because of requirements
for temperature and shrinkage reinforcement.
Compressive St~ in the Shell Concrete
The maximum compressive stress in the shell concrete

under the design load was f c = 1 2 050 lbs~/ft. .. . = 35 psi.
2~3 in. x 12 in.;ft.
Horizontal ~rior .Ed~e Beam
The total force in any edge beam is equal to the sum

of the shear forces acting along its length. In the hori-
zontal interior edge beam, Figure 5, the shearing forces,
transmitted from the shell, build up to a maximum tensile
37
i..... ....._ ~ - """ 1 - __. _.... ~ ~

~~~~
tt
~
'
....... -..~~~
.cL
t
t~_..~ _..~ i
~
-,
I. b b J
Se ct ion PP
p
Pion
Figure 5. Shear Distribution in the Shell.

24"
r 't .......~ a"
tll~~==~#~6'~-~-;;:::;:::--;-;-··_/_#_2;_'~~~~~ij·
Section A A
24"
B A
Quadrant
Section BB
2
B A
c c
r
24"
·1
Quadrant Slope= 0. 72." I Ft. ~·
I
r
3.5" Slope= 1.62"
I Ft.
_J.
L Section CC
I. 4• .1 24"
--T 18"
yColumn S Hounch
I
I
D
I
I
I
D Quadrant
4
Quadrant
3
r
3.5"
__ JI l_
Section DD
1. 4"
.1
Figure 6. Cross-sectional Views of Exterior

Edge Beam Sections. CJJ
00
39
force in the edge beam at the center of the roof. There-

fore, the maximum force generated by the accumulated
shearing forces are the forces acting in the edge beam
over half of the roof span. Shearing forces on both sides
of an interior edge beam contribute to the total direct
stress in t~e edge beam.
Thus, the total tensi:le force, HT = 2 x H x a =
2 x 1,050 x 10 = 21,000 lb. The area of steel to resist
the t ensile
· · force was As -- 21,000 lbs. = 1 •05 sq. in.
20,000 psi 0
A welded shear connection was selected to connect the

precast quadrants; therefore, steel angles were used for
all interior edge beams. Two 2 in. x.2 in. x 3/8 in.
angles, As ~ 2.72 sq. in., provide adequate width and
thickness for the attachment of dowels and flat bars by
welding. The composite interior edge beam was made up of
two angles connected by a 1 ~.2 in. x. 3;s in. flat bar welded on
top of the angles to form a "T" shaped section.
Sloped Interior Edge Beam

1 2 l 2
From Figure 3, AB = [(A' B)Z + (A 1 A) 2 ] 1 = (100 + 9) 1
= 10044- fto
The total compressive force was He = 2H(AB) ~ 2(1050)

(10.44) = 21,960 lb. The compressive stresses are trans-
mitted by the edge beam steel to the column. The steel
21,960
area was As = 16 , 000 = 1.37 sq. in. This area of steel
was furnished by two 2 in. x 2 in. x 3/8 in. angles, As =
2. 72 sq. in.
40
Horizontal Exterior Edge Beams
The total compressive force in the horizontal exteri-

or edge beam was H0 = H•a = 1,050 x 10 = 10 ,500 lbs/sq. in~
The area of steel required to transmit the compressive
stresses was As = 102200
;I..6,000 = .656
sq. in.
Two No. 6 deformed steel bars~ As = 0.88 sqo in., were
used. The. shape and dimensions of typical sections of the
exterior edge beams are illustrated in Figure 6.
Sloped Exterior Edge Beams
From Figure 5, the slope length of the exterior edge

1 1
beam was DA = [(DA' )2 + (AA' )2] 12 = (100 + 9) 12 = 10.44
ft. The total compressive force was H0 = H (DA) = 1~050 x
10944 = 10,980 lb. The required area of steel to transmit
the compressive stress was As = ~ 80 6 6
16 , 000 = • 9 sq. in.
T wo
No. 6 bars were used. The compressive steel was also re-
quired to resist bending in the edge beam during eccentric
roof loads.
Tension Tie Plate at Center, of. Roof
An area of steel of 1.05 sq. in. was required to take

the calculated maximum tension at the center of the hori-
zontal interior edge beam. However, as this is a critical
structural point, an additional factor of safety was in-
troduced by using a 2.0 in. x 3/4 in. x 24 in. flat bar
which had an As = 1.5 sq. in. This was also used because
. .,
1+.l.
of dead loads greater than the design load which i1vere in-
duced during the testing procedure.
The welding of this plate to the horizontal interior
edge beams was also noted as a possible weak point; there-
fore, a weld leg width of 3/S in& was made on both sides
of the 24 inch bar. This gave a. calculated allowable load
capability bf q = 4S in~ x 3/S in. x .707 x 5,000 = 63,630
lbs. under dynamic loading, or q = 48 in. x 3/8 in. x .707

x 14,000 psi = 178,160 lbs. under static loading. This
weld connection itself allowed a factor of safety of 5 for
static loading and would limit any failure of the horizon-
tal interior edge beam to the steel.
Connection f3ei;.ween Edge Beam Angles and Shell
To provide adequate bond between the edge beam angles

and the shell concrete, 10 in. dowels of No. 6 bars were
used; each dowel was bent in the shape of an "L" to allow
approximately 2 inches of welding surface against the angle~
This allowed approximately 8 inches of length for the

dowels to develop bond with the concrete, Figure 7. The
dowels were spaced at 8 inch centers, to line up with the
shell steel. The calculated bond stress at design load
was u = (V) = _(l,050) = l,050 , or u = 125 psi, which
E · 7/Sd l.2(7/8)8 S.4
was less than the 350 psi allowable stress (25).
To counteract localized radial shearing and bending

42
Figure 7. Shell and Edgebeam

Reinforc ing Steel.
Figure 8. Tie Connection Welded to Bearing

Plate on Top of the Column.
43
stresses at the corner of the shell supported by the col-
umn~ a reinforcing mat was constructed of Noo 5 bars at 4
inch centers, Figure 7o This mat was approximately 18
inches square in plan with the bars bent to conform to the
slope of the shell surface.
The mat bars perpendicular to the interior edge beam
replaced the 8 inch dowels and were welded in the same
manner. The bars parallel to the interior edge beam were
welded to a 23 inch angle cast into the exterior sloped
edge beamo
The 23 inch angle was cast into the exterior sloped
edge beam at the lower corner? Figure 7~ to connect the
shell to the column. This angle was welded to the angle
on top of the column haunch to develop the tensile strength
of the haunch against overturning moments.

11
A special wide flange T 11 section was also utilized
to connect the shell to the column. This section was made
by cutting the web of a 10 ino x 5:j4in. wide flange on a
3 by 10 slope to match the slope of the interior edge beam.

This piece was then welded to a metal plate cast into the
top of the column~ Figure 8. When the two quadrants were
lowered onto the column? the top slope of the inverted 11

T 0u
section was even with the top slope of the edge beams for
welding.
Four lift rings were cast into each quadrant

44
approximately at quarter points in plan, Figure 9. The

lift rings were No. 5 bars, 18 in. long plus 6 in. pieces
of No. 5 bars, as illustrated in Figure 10.
The allowable shearing strength of each ring was cal-
culated assuming the total quadrant weight was carried by
one lift ring. Thus, S = pA = 2 3,260
x •
31
= 5 , 260 psi,
. which
is considerably less than the 13,000 psi allowable working

unit shear stress for structural steel (26). The rings
were placed diagonally beneath the shell steel and spot-
welded in order to transfer the lifting stresses into the
entire shell.
Shell Dimensions
The nominal thickness of the shell was 21;2inches.

This depth would not give adequate coverage of the edge
beam steel, therefore~ a depth of 31/2 inches was used for
the thickness of the exterior edge beams. A horizontal
surf ace 4 inches wide was specified for the bottom side of
the exterior edge beams to provide a uniform surf ace for
juncture with future wall construction, Figure 6~ section
D-D.
Shell Concrete
.
To reduce the over-all weight which was lifted during
assembly of the shell on the column, the shell quadrants
were p·recast of lightweight aggregate concrete which had a
21 day density of 117 lbs./ft~ Lightweight concrete,
45
Figure 9. Precast ' Quadrant Showing Lift Rings.
5" 3" .l. 3"

.I
Figure 10. Lift Ring Detail.
46
having an ultimate strength of 3,750 psi, was requested

from the concrete plant. The average test cylinder strength
from three samples was 4,480 psi.
The volume of concrete required for the shell was de-
termined by the 21/2 in. depth over 256 sq. ft. (horizontal
projection) and 3 in. average depth over 144 sq. ft.·
(horizontal projection). °
Thus, V0 = 2i 25 x 256 + 3;2 x 144
= 53.32 + 36 = 89.32 cu. ft. = 3.31 cu. yd. An additional
5% of volume was added to the total volume~ therefore,
VT = V
0
x 1.05 = 3.31 x 1.05 = 3.47 or 3.5 cu. yd. was
required.
Support System Analysis
The support system for the structure consists of the

column, haunch, tie bar, and footing. The design of each
item is considered individually in the following paragraphs.
Column Design
One of the first considerations given to column

design was the possibility of failure due to overturning
moment. This moment could be composed of static loads on
one-half of the roof (adjacent quadrants parallel to the
centerline) plus wind forces. Conventional wind data do
not apply to a roof configuration of this type9 also~ very
little has been published concerning the behavior of wind
forces on open sided h-p shells.
Mannschreck (27) conducted wind tunnel experiments on
47
h-p shell models of square configuration in 1963. His

tests were conducted for values of Reynolds numbers less
than 8.0 x 105. Mannschreck used Reynolds number, NR =
VLnN
r· e , where L was the length of one side of the roof.
-f".....--
For wind speeds of 60 feet per second on the 20 foot
square prototype in this study, a value of NR = 6.54 x 106
would be obtained.
Mannschreck developed three dimensionless coefficients
for the force components of lift, drag, and moment on the
model. He tested models which had rise to span ratios of
1/6, 1/8, 1/10, and 1/12 at various ratios of column height
versus roof span over a range of wind speeds up to approx-

imately 60 feet per second. Each of the coefficients was
plotted against NR. The resultant lift force was applied
at the center of the roof in Mannschreck's analysis,
Figure 11, and does not cause eccentricity, therefore,
only the coefficients for moment and drag will be
considered.
The coefficients which Mannschreck developed are~
Rx
1. Drag coefficient, C - ':::!r
x - Ne · H • w "/' • v-
R
2. Lift coefficient, Cz - z
- Ne • H • w •f ·v2
Moment coefficient, M0 = Ne • H • W2 "f' • v2
where:
1 lb.f Secz
. . , NewtoJl' s Second Law Coefficient =-~--·--
32.2 lbm. ft.
P = Air Density = ~070 lbm/ft!
48
Wind -
1·
....,::::-----'----flR;\--,.[R_y
.._R_:.;.~---~-::-- --"T"""f
·1
H
l
D
Figure 11. Side View of Structure Showing Lift,

Drag, and Moment Reactions.
2.5"
# 2 ~ Ties, 10 o. c.
11 I 7''
12
11
_J
11
2.5" 2. 5" 2.5
11
I0
Figure 12. Column Steel Arrangement.

49
V = Wind Speed ~!c)

By rearranging the values for the coefficients, equa-
tions can be set up to estimate the resultant reactions
acting on the structure, or
4o Drag, R = Cx 0
Ne °H•w•p 0
V2
x
5. Lift, Rz = e z 0
Ne °H 0
w f1°V2
0
6. Moment, M = Mo 0
Ne °H 0
w2 o;• V2o
For the plot of ex versus Ne for a model which was

dimensionally similar to the prototype in this study~ ex
decreased as NR increased. A conservative value of e x
would be the lowest value which was obtained during the
model testing, ex for NR = 7.91 x 10 6 would be approxi~
mately 0.180. For purposes of computing the moment due to
Drag, a value of ex = 0.250 will be assumed at NR = 7.91 x
106 .
As NR increased, M0 increased at a constant rate. It
is not likely that the slope of the Mx verus NR plot would
remain constant for NR up to 7.91 x 10 6 as this would give
a value of M of approximately 1.20. As NR increased from
0
8.0 x 10 5 , the slope probably decreases as separation of
streamlines occurred at the edge of the structure. The
highest values of Mo obtained on the model was approxi-
mately 0.245 at NR = 8.5 x 10 5 . A value of 1'1 0 = 0.70 will
be assumed to compute the wind load moment acting on the
shell. A val~e of wind speed = 66 fps will be assumed to
50
be one-half of the maximum wind speed that would act on

the structure at any time in the direction of overturning.
The maximum moment for design will consist of one-
half roof load plus a wind moment computed from the half
. ' 1
wind speed. Assuming!'= .070, V = 66 fps, and Ne = 32 • 2 ,
the Drag Force ' Rx = Cx • Ne · H • w •I ,a· V2 = O 25 • _
32.L2 · 0
3.0 • 20 • .070(66) 2 = 142.5 lb. Thus, M1 =Rx •(H+D) = 142.5

• 13.0 = 1,850 ft.-lbs·. The wind force moment, M2 =
Mo. Ne. H. w2 •[J. V2 = (o.70)(3.0§~~~) (0.070)(66) = 7 , 210 . 0
2 2
ft~lbs. The total moment due to one-half of the maximum

wind velocity would be Mw = M1 +M 2 =7,210.0 + 1,850 = 9,060
ft.-lbs.
The moment due to design roof load on one-half of the
~•~ p~w2 (40)(20)( 2 ~)
2
roof would be MDL= p • b · = = =
40, 000 ft.-lbs.~, or MDL = 20, 000 ft.-lb.s./c.olumn. The ref ore~
the total moment acting on each column will be MT = Mw +
MDL = 9,g 60 + 4 o,goo = 24,530 ft.~lbs.
By the method of ultimate strength design from
Reinforced Concrete 38 (28) a design factor of 2 is used.
The design axial load, P = 14.52 kips, and the ultimate
design load, Pu = 2(14.52) = 29.04 kips. The design
moment, MT = 24.58 kip-ft., thus, the ultimate moment,
Mu= 2(24.58).= 49.16 kip-ft.
From Table 6, "Eccentrically Loaded Tied Columns uu ( 28)
for f c = 3,000 psi, f y = 40,000 psi~ and column size = 10
in. by 12 in., p = 30 kips, and l'1u = 56 kip-ft ~ 4 No. 10
u 0
bars are recommended. However, for p u = 30 kips, and

M
1v1U ::::: L.v>2- kl' 1·1·J-+'1-
T ..:.. e
1.. 'll ' four No~ 9 bars are required, which have
an area of steel of 4.0 sq. in. It was considered desir-
able to substitute six No. $bars, A8 = 4~74 sq. in.,
Figure 12, as this bar combination gave a better steel
distribution in the column. The ties consisted of No., 2
bars spaced at 10 inch centers with two ties per set.
~ Design
The general dimensions of the haunch were selected

and were then checked by analytical methods for the re-
quired lengths and depths of section, Figure 13* The
top of the haunch was given a slope value of 17° which
was approximately the slope of the exterior edge of the
shell. The bottom of the haunch was assigned a slope of
30~ The horizontal length of the haunch from the face of
the column vras 18 inches. The vertical depth of the end
of the haunch was 8 inches. For the haunch section at the
face of' the column, MDL= W(~.3) = 40 x 500 = 20,000 ft.-lbs.,
and MJ.vlAX =MDL+ Mw = 20,000 + 4,530 = 24,530 ft.-lbs~
2
One-half the MMAX will be resisted by each haunch arm, thus,
M "" 12, 265 ft. -lbs. From Table 1, Reinforced Cort_srete
Handbook (29), K = 236 psi. Solving for the distance
from the center of the reinforcing steel to the extreme
.j."'b d = ( M)l/2 = (12,850 x l~.)1/2 = ( 65 • 4 )1/ 2 = Uc~.()>1:
~1 er, ~o \ 236 x 10 -V in*,
which was less than the 9.0 in. actual distance. The steel
~1
area rea.,,uired at the column 1ace was As - -----·
f -id
-- !") -
-- Sv
122:0.9 x 6.__ = 61L700 == "' 98 sq. in ~ Two No • 7 bars,
20,000 x 7/8 x 9 63,000
52
11
20
6-# 5 • 110 Bent 90~

11
11
6 o. c. Eoch Side, Buttwelded
To Angle 11
18
Welded
Colu mn Steel
#8<P
Figure 13. Haunch Detail.
Figure 14. Column and Haunch Steel Cage.

53
As = 1.20 sq. in. were required for the bottom steel in

each haunch. The top steel consisted of the three No. 8
bars which were continued from the column and bent to the
desired shape. The required amqunts of bond perimeter and
shear capacity are E0 = uj d = y; • uj d = 2 x 300 x2 7 8 x 9 =

V M 1 · 12 26~
V M 12,26~ _
2.62 in. < 5.5 in.' and v = bjd = Lbjd = 2xl0x7 8x9. -
5,i4 o = 81 •5 psi. < 90 psi,
. b oth of whi c h were a d equat e.
63
To develop the moment capacity of both haunch arms
simultaneously, a tension connection between the haunch
and shell was developed by precasting steel angles into
each member, Figure 14. These two angles were placed to-
gether during the erection process and were welded together
to form a positive load transferring connection.
For a moment of 12,265 ft.-lbs. and a moment arm of 2
ft., the equivalent force acting upward at the end of the
haunch will be 6,132 lb; therefore, two No. 5 dowels were
used near the end of the haunch, and the required bond
A • f
length was a = 8
·· ~ = .306 x 20,000/2
300 x 1.96 = 5· 2 in.
Three
u • E0
sets of No. 5 bars were spaced at 6 inch centers to anchor
the 3 in. x 5 in. angle on each haunch~ Figure 14.
An 8 in. x 10 in. x 3/8 in. bearing plate was cast
into the top of the column to act as a base for anchoring
the column tie. Four No. 8 bars, 6 inches long were used
for dowels on the base plate, Figure 14.
Tie Bar Design
The thrust in each sloped interior edge beam was

54
I
2 • H • aa
cos =2 · 10,500 · 10.43 = 21,950 lbs. The horizontal
17
component of this thrust was H0 = 2 · H · a = 21,000 lbs.
"He
As = 1'S = 21,000 The As was_ furnished by
20 , 000 = 1.05 sq. in.
one No.• 1.0 bar which had an area of l. 47 sq. in.
The length of weld which was necessary to develop the
p
full strength of the tie bar was L = • 707 h •S. =
9
21,000 The tie bar was welded
.707 (3/8)(14,000) = 5.66 inches.
to the inverted "T" sections, which were welded to the
bearing plate on each column, after the column footings
were cast.
Footing Design
The basic design of the footing was taken from a

study on pole type buildings (30). The depth of set for-
mula specified that the equation was used to determine the
required depth of embedment where no constraint was pro-
vided at the ground surface. This empirical equation~
1
l
d = ~[ 1 + ( 1 + 40 6 h) 12 ], specified the following
parameters:
p = Applied horizontal force or equivalent in kips.
s1 = Average soil resistance above the point of

rotation in ksf.
82 = Average soil resistance below the point of

rotation in :ksf.
bl = Diameter of round post or the diagonal dimension

of a square post, in ft.
I
t
I
s,
I
'
Pion
-I I
L 3
/
I
I
I
J,.. j
03 !'/ I
I .34 0
I
01 I
I
0=4.5
I
\ I
I
I .560
..,
-...
I
'
- I
I
I
S2
- I
I
.32 0
I
I
I
I
I I
-- ~
~ 0'2
L - --
·~
\1(2 '\'
~
---..:! >0
:------__
Section AA I. 20
11
.I
Figure 16 . Soil Reactions to Footing and Wing
Figure 15. Footing Details. .Walls Against Overturning Moments.
56
h = Distance, in feet, from ground surface to the

point of application of P.
d = Depth of embedment of post.
For the clay soils in this area, s1 was assumed to be
3,500 psf, M was 24.530 kip-ft.~ ~ ~ 10 ft., P = 2~45 kips,
and b 1 = 1.31 ft. 1
Thus, d = 2 • 2. "34p[1
.b
81 1
+ h
\1 + 2. 34P ;2
4·. 36h)
1
J
s1 ·b1
= 12 (1.310) [. 1 + (1 + 43.6
1 • 310 )
1
;2 J = (0.656)(6.86), or d =
4.50 ft. The depth of the foundation holes was approxi-

mately 4.5 ft.
The minimum recommended diameter for foundation holes,
Figure 15, is b 1 + 4 in. = 15.6 in. + 4 in. = 19.6 in., or
'-'\.
20 in. The footing excavation was dug with a 16 inch

rotary drill, then hand finished to a diameter of 20
inches.
To support the structure in bearing, the area of the
footing base was determined and checked for adequacy. The
total design weight at the bottom of the column was Wt./
column = WQ,ua d . +W01
o • +WF oo t+WD.
esign =6,400+2,265+
1,000 + 8,000 = 17,665 lbs. Assuming the soil bearing
capacity, P = 5 ,000 lhs./sq .. ft., the required bearing area
WT lbs.~ 3 53 q ft
was AB = p = 17,665
5 ,000 psf - · · s • •'
which left an
additional bearing area required of 3.53 - 2.18 = 1.35 sq.ft.,

where the area of the 20 inch diameter hole was 2.18 sq. ft.
57
The additional bearing area was supplied by two cantilever

wings 8 inches wide~ 12 inches long, and 18 inches deep,
which were placed parallel to the column tie bar along the
centerline of the structureo These two wings furnished a
bearing surface of ~. in. 4
12
sq. x in. I in.
2
sq. 2f t • = 1. 33 sri.
o
':1
ft ..'
which was sufficient.
The equation used earlier to determine the required
depth of the footing is an empirical expression which
accounts for the overturning moment. However, the wing
walls will also resist overturning by developing the pas=
sive earth pressure of the soil. Assuming the top 6 inches
of soil was disturbed and, therefore, not effective, the
overturning resistance developed by the wing walls was
·
CF igure 16) Q
3 = s x A = 3 ' 500 lb • /ft •2 • ( 12 in. x 12 in. x g2
144 in.2 /ft.2
= 7,000 lbs. Assuming that Q1 acted at 2 d and the over-
3
turning moment, M caused pivoting at .68 D, which was
approximately 2 t,
the force to be resisted by the wings
was F -- ~
h
= 24 2530 = 12,265 lbs. The maximum force, Q,
3,0-1.0
which had to be developed by the footing was 12,265 lbs.=
7,000 lbs. = 5,265 lbs. or a moment of 10,530 ft.-lbs.
The addition of the wing walls to the circular footing
gives a conservative value of resistance of overturning
moment of MT + MF + Mw = 24,530 + 14,000 = 38,530 ft.-lbs.
The wing wall must withstand moments in two directions;
when loaded, the wall must act as a cantilever beam in the
vertical direction, and when acted upon by overturning
58
moments, it must act as a cantilever beam in the horizontal

plane.
Checking the moment in the horizontal or overturning
plane, F1 = ~ = ?,~OO = 3,500 lbs. (maximum horizontal
F1 x L1 1
for?e), and M1 = ~ = 3,500 lbs. x -2 ft. = 1,750 ft.-
2
lbs. The maximum moment which the reinforced wall was
capable of developing was Mc = Kbd.2 = 236 x 18 2
12 x (5) =
1
8,850 ft.-lbs., which was greater than M1 = 1,750 lbs.
A _ M1 _ 1,750 _
Thus, the required area of steel, sl - ad - l.44 x 5 -
.243 sq. in. Checking the moment in the vertical plane
for bearing, <&' 2 ,= P x A= 5,000 x 8 1~4 12 = 3,330 lbs.
L2 1
M2 = F 2 x 2 = 3,330 x 2 = 1,665 .ft.-lbs., Mc = Kbd2 =
2
236\~l x (15) = 35,400 ft.-lbs. > 1,665 ft.-lbs., and
2
1 .$i
6
As = ~ 5 14 = .076 sq. in. Two No. 4 bars top and
2
bottom, As= .40 sq~ in., were used to satisfy As •
1
The As was the governing value of steel area, so
1
this area was checked for bond. For f'c = 3,000 psi~ and
Vmax 3,~00
vmax = s1 = 3 ' 500 1b s • ' u = E •jd = 3.1 x 7 8 x 6 =
0
215 psi < 300 psi. Four No. 4 bars were used ;for the wing
wall steel, Figure 15.
CHAPTER V
DEVELOPMENT OF ASSEMBLY COMPONENTS
Introduction
The development of several assembly components and
techniques~ in addition to the structural elements them-
selves, was necessary for final assembly of the precast
elementso This involved (1) the design and construction
of shell and column forms, (2) the footing reinforcing
cage~ (3) the assembly support system~ (4) the lifting

frame~ and (5) a means for supporting and stabilizing the
column while casting the footingso Also involved were (6)
the casting and curing operations for the shell quadrants
and columnso
The design procedures involved in developing these
components were mainly investigations of maximum stresses
to assure safe construction conditions during the erection
of the structure and were not intended to be a complete
and detailed designo
Formwork and Precasting
Column Forming
The column forms were constructed from 2 in. by 12in.
59
60
Douglas fir lumber. This type of lumber was rigid enough

to maintain form dimensions with a minimum number of
braces. Disassembly time of the forms was reduced by the
use of double headed forming nails which allowed workmen
to pull all nails with nail bars.
Steel fabrication was complicated by the special
haunch at the top of the column. This was formed by bend-
ing the column steel from one side of the column to form
the top steel in the opposite haunch, Figure 14. To keep
the bending process as simple as possible, the column
steel was offset one-half the width of the bars, 1;2inch
on each side of the column to allow the cross-over bars to
pass without special bending.
The bottom haunch steel was formed by No. 7 bars four
feet long with a 60° bend approximately 18 inches from the
end. Three bars were placed on each side. The 18 inch leg
of each bar was spotwelded to the three column steel bars
on the opposite side of the column. This placed each lower
haunch bar directly below the top bar.
By welding the lower haunch bars to the column steel
and spotwelding the ties, the steel was formed into a cage
which could be handled and moved about as a unit. This
was helpful when moving the steel from its construction
location to the casting site. Also, a minimum of supports
and attachments were needed to keep the steel properly
spaced when placing concrete.
An outdoor concrete floor slab was used for a casting
61
bed. The column forms were blocked up to facilitate lift-

ing with a fork truck. Each form was coated with a bond
breaker compound prior to placing the reinforcing steel.
The column steel cage, which weighed approximately 360
pounds, was moved to the casting site by a small hoist on
a farm tractor and lowered into the forms. The steel
reinforcing cage was supported on the haunch end of the
form by the 3 in. x 5 in. steel angle cast into the haunch
to serve as the shell connecter, Figure 14.
The 3,000 psi concrete was delivered to the casting
site by a ready-mix truck to simulate prototype casting
procedures. A three-man crew cast the concrete using an
electric vibrator for uniform placement. The entire cast=
ing operation including the finish troweling of the sur-
face lasted approximately 1 hour and 30 minutes.
Shell Forming
The forms used to precast the shell quadrants were

previously used to cast an inverted umbrella shell 20 feet
square in plan. The forms were made up of four feet square
modular units on a metal framework. Four of these units
were combined to form an 8 ft. x 8 ft. form unit. A 2 ft.
section was added between each 8 ft. quadrant unit to form
the interior edgebeam section, and a one ft. extension
section was added to the outer edge to form the horizontal
edge.
This set of forms was modified by deleting the two
62
form extensions and adding a new 2 foot extension to the 8

foot base to form the outer edge beams. The slope of the
form extension was tapered to a 3'l2 inch edge depth to pro-
vide a thickened edge beam section without altering the
shell's top slope, Figure 6.
Side forms were made from 1 in. x 8 in. yellow pine
lumber. The edge height was adjusted vertically to 3'l2
inches and holes were drilled through the boards to match
pre-located holes in the steel angles9 then, one-fourth
inch bolts were inserted and tightened. Braces were
bolted to the forms at the corners to increase form
stiffness.
The 2 in. x 2'l2 in. x 3;s in. angles, which made up
the interior edge beams, were cut to length. The corners
were cut at 45° angles to form a 90° corner angle between
the horizontal and sloped edge beam; both angles were bent
down slightly along the 45° cut for welding. Ten inch
long No. 6 dowels were welded to the edge beam angle to
bond the shell and edge beam. A 90° bend was made 2 inches
from one end of each dowel to provide a welding edge. The
two end dowels were tackwelded in place parallel to the
form slope. When the edge beam angle was removed from the
form~ a straight steel bar was clamped to the two end
dowels to act as a welding guide for the rest of the
dowels.
The two No. 6 bars which comprised the exterior edge
beam steel were heated and bent around the corners. The
63
bar ends were buttwelded to the interior edge beam angles

at each end. Thus, continuity, with rigidity and effective-
ness of stress transfer was obtained. The two No. 6 bars
were spaced at 2 inch centers with the outside bar centered
2 inches from the edge of the concrete. The bars were
placed at the center of the edge beam depth for efficient
stress transfer.
The reinforcing mat in the corner of the quadrant
above the column was constructed from 18 inch long No. 5
bars, which replaced the 8 inch No. 6 bars in the lower
corner of each quadrant. They were spaced at 4 inch cen-
ters along the edge beam and the 23 inch angle cast into
the exterior edge beams. The bars welded to the short
angle were bent upward 4 inches from the angle to conform
to the slope of the shell. The short angle was notched to
fit against the end of the edge beam angle for welding.
Preparation of the shell steel consisted of cutting
the No. 2 bars into 10 foot lengths and making a 180° bend
in the bars 4 to 5 inches from one end. The other end of
the bars were left straight to overlap the 8 inch dowels
of the interior edge beams.
Prior to placing the steel in the forms, all cracks
between sections of the form plywood were covered by
strips of plastic stretched tight and stapled to the form.
Then, the form surfaces were given a heavy coat of form oil
to keep the concrete from bonding to the forms.
After spraying the forms with the bond breaker, the
,.
64
steel angles were placed in position and clamped to the

side forms. Then, the four quadrants were placed together
and leveled. The edge beam angles between quadrants were
checked and all corners found to be within 1° of 90°
angles. The edge beams were aligned with a maximum allow-
able clearance gap of one-half inch, which was the design
tolerance.
The steel was placed by a five-man crew, consisting of
a foreman, two welders, and two laborers. The shell steel
spacings were marked off by one workman on the form side-
walls for rapid alignment of the steel. Two workmen
placed the steel on the forms while one workman spaced and
tied the steel~ Figure 17.
One welder welded the corners of the interior edge
beams and the short angles at the low corner. The second
welder buttwelded the exterior edge beam bars to the inte-
rior edge beam, then tightened the shell steel and spot-
welded the shell steel to the edge beam dowels to keep the
steel network rigid. The shell steel was tied at alter-
nate junctions in both directions. Four lift rings were
placed under the shell steel at quarter points from the
edges and tackwelded. Three-fourth inch thick wooden
blocks were placed at various points under the steel junc-
tions so that the steel would not be over one-fourth inch
from the center of the shell thickness at any time. A
length of wire was attached to each block so that they
could be removed as the concrete was poured with the steel
Figure 17. Shell Form Ready for Precasting .
66
supported by the fresh concreteo

Final steel forming was completed prior to the arrival
of the concrete ready mix trucko A five-man crew~ ex-
cluding the truck operator, was used to place and work the
concreteo Two men began moving the concrete on the forms
while the other three men worked the concrete around the
steelo The concrete had to be rodded and vibrated under
the angles and shell steel to reduce voidso
After the first quadrant was cast, one man placed the
concrete as it came from the chute, one man worked the
concrete under the angles and vibrated the forms~ two men
worked and screeded the concrete on the main part of the
shell, and one man finished with a wooden trowelo
When approximately one-'third of the shell was covered
androughly smoothed by rake and shovel to the approximate
depth, two workers began screeding the concrete with a 14
foot screedo The third man continued working the concrete
under the edge beam angles and the shell steelo Just
before the first quadrant screeding was completed~ one
worker moved to the next quadrant form and began placing
the concrete on ito The finish man began wood troweling
the first quadrant when the screeding was past the mid-
point in the shell.
The concrete began to dry and became stiff by the
time the third quadrant was cast and had to be tempered.
This was due to the length of time required to cast all
four shellso The concrete would have been more consistent
67
and workable throughout the casting period if the load had

been ordered on two trucks spaced at one hour intervalso
The total time required for casting the four quadrants was
two and one-half hours.
Curing Precast Elements
Column Curing
The curing operation on the columns was begun after

the concrete had hardened for approximately two hourso Two
layers of burlap material were placed over the top surf ace
of the columns. A perforated sprinkler hose was laid down
the center of each column. The sprinkler hose pressure
was adjusted to keep the burlap continuously soaked. After
the columns had cured under moist conditions for eight
days~ the sprinkler hoses were removed. The burlap mate-
rial was left in place until the end of 14 dayso Then the
covering was removed and the columns cured in the forms
with no covering.
Shell Concrete Curing
The shell curing process was initiated approximately

one hour after the fourth quadrant was cast. Each quadrant
was covered with two layers of burlap material and a four
milli-inch thickness of clear plastic. The plastic was
weighted down securely so that wind gusts would not blow
it off. The quadrants were watered twice daily during the
68
first four days, and in the mornings only during the next
four days. No water was added after the eighth day. The
sun shining through the plastic covering during the day
raised the curing temperature and vaporized the moisture.
This produced a curing condition similar to factory con-
trolled curing. After 14 days~ the plastic covering~ bur-
lap material and side forms were removed. Figure 9 shows
a typical quadrant ready to be removed from the form.
Footing Steel
The footing steel was formed into a rectangular cage

with the inside dimensions ,approximately 11 in. x 13 in.
The cage was designed to allow a clearance of approxi-
mately one-half inch on all sides of the precast column as
it was lowered into the footing excavation. This configu-
ration of steel reinforcement was designed to give maximum
anchorage and bond to the wing wall steel in order that
\
the wall could develop its full potential in bending.

Additional anchorage of the steel was also provided by
hooking the ends of the footing steel.
Assembly Supports
The construction procedure selected for this study

specified that a system of supports be developed to hold
the precast roof quadrants in place during the erection
process. The procedure also required that the quadrants
be held rigidly~ without uncontrolled movement or
69
deformation of supports until the erection was completeqo

These requirements were met by the development of a rigid
support frame or assembly jig~ which was supplemented by
three wooden supports"
Assembl;y Frame
The assembly jig, Figure 18, was developed to fulfill

four basic requirements prior to and during the erection
of the structureo These were:
(1) To rigidly support the corners of the shell
quadrants during the assembly and final
erection steps.
(2) To provide vertical adjustment of corner
towers for precise control of the shell
corner elevations.
(3) To provide a means of elongating or shortening
the distance between the tower caps for ease
of horizontal spacingo
(4) To provide a method of clamping the quadrants
together for welding.
To satisfy these requirements~ the rigid support
framework illustrated in Figure 18 was designed to provide~
( 1) A tower cap which could be adjusted to

various slopes of exterior edge beams.
(2) A set of top and bottom horizontal braces
to stabilize the corner towers.
+¢=0
2.l'x 21\ ..L' Angle +¢=goo

2 2 4
-
1
12 -0
11
-- '"
1~"x 12
11
r:
Turnbuckle
Lap Joint For

Height Adjustment 11 II I"
4 -6
1 11
:
All Angle Iron 2 x 2 x 4
Except As Nate d
3 -0
~
1 11
..,
Screw Jack
Adjustment Joint
f
2{ x 2f x Angle
1 11
3 -6
l Figure 18.
1
20 -0
Assembly Jig.
11
1-
:I
-.J
0
?l
(3) A turnbuckle mounted on the top horizontal
brace which could be extended or retracted
over a range of approximately 12 incheso
(4) Three screwjack legs in the base of each
tower to provide fine adjustment vertically
and to plumb the towerso
(5) Large increments of vertical adjustments in
the corner towers by overlapped tower leg
sections and braces with spaced bolt holeso
The design of the assembly frame was based on a
weight per quadrant of approximately 3,200 poundso By
using a design factor of 2.0, the working load per quad-
rant was 6~400 poundso During the assembly~ each quadrant
was supported by the assembly jig tower, the concrete
column, and two wooden supportso The maximum stress con-
dition for the tower would probably occur with the quad=
rant supported by the tower and the wooden support at the
centero Then, the tower would support one=half of the
working load, or 3~200 poundso
The corner leg of the tower was designed to carry the
full load of 3~200 poundso The slenderness or l/r ratio
governed the design; thus, for steel columns (26),
18/000 =
(84_o 39) 2
18,000
P = f ·A = 5~030 x 094 = 4~730 pounds was the safe load

which the support could carry concentrically~ compared to
72
the applied load of 3,200 pounds.

The frame was welded at the corners where the legs
joined the base and top angles of the towers. Each end
of the diagonal brace was bolted to the top and base
angles to allow rotation when the tower height was changed.
The screwjack legs were fitted into one and three-fourths
inch inside diameter pipe sleeves which were welded to the
tower base. The tower cap was bolted to the top of the
tower by eight 1 1/2 in. x ~8 in. counter-sunk headed bolts.
Slope adjustment of the tower cap was made by shims or
washers placed between the cap and the top of the tower.
Slight adjustments in the slope of the cap could be made
by adjusting the screwjack adjacent to the concrete column.
The horizontal frame braces were bolted to the corner and
outside leg of the tower and at the two center adjustment
slots by ~8 inch bolts. The turnbuckles were connected to
one section of the top brace by a welded ring; the opposite
end of the turnbuckle was bolted to a ring on the over-
lapping brace section.
Temporary Wooden Supports
Three wooden supports were necessary to stabilize the

roof quadrants vertically. Adequate horizontal stabiliza=
tion was provided by the assembly frame and the two columns.
The wooden supports were constructed principally for this
project, therefore, they were designed to give small
vertical adjustments.
From the topographic survey of the construction si.te~
the ground elevations at the support points were deter-

mined. The distances from the ground to the lower side of
the roof at these three points were then determined so that
the towers could be constructed to the approximate hei.ghts
necessary to maintain the proper horizontal interior edge
beam elevation.
Center Support
The center support legs were constructed from two

pieces of 4 in. x 4 in. x 14 ft. lumber spaced 12 inches
apart, with 2 in. x 4 in. members for diagonal braces. Two
2 in. x 6 in. x 20 in. members were nailed horizontally on
both sides of the.support legs at the base.
The total calculated load carried by the center sup-
port assuming that one-third of the weight of each quadrant
was supported by the center support, was 3, 200 x 1;3 x 4 =
4,2?0 pounds. For a design factor of 2, the design load
was 8,540 pounds. The cross-sectional area of the support
legs was 26.28 in2 which gave the support a load capacity
t
of P = A•·f ::26.28 • c(l - '85d) ....2..6
= 26 .. 28 • 1,200(1 - 8ox3766)
= 26.28 • 1,050 = 27,590 pounds.

Vertical movement of the support was supplied by two
screwjacks on metal brackets bolted to the base of each
leg. A 3 inch length of 13/4 inch inside diameter pipe was
welded on the bracket to act as a sleeve for the screwjacks.
The top of the support was made up of a 4 in. x 4 in.
?4
cap and a 2 in. x 12 in. scabbing plate on each side,
Figure 19. A 4 in. x 4 in. member was bolted to each
scabbing plate. The top surface of the support was planed
down at a 17 ° angle to conform to the slope at the interior
edge beam.
The placement and removal of the center support was
complicated by the tie bar which it straddled. This prob~
lem was solved by the removal of an 8 inch section of one
leg while the support was being placed over the bar or
being removed. The short section was braced by 6 inch
metal plates which were bolted in place.
End Supports
The two end supports were designed to support a maxi=
mum load of one-half the working load on the center sup-

8
port or P = 25~0 lbs.= 4,270 lbs. The construction of a
satisfactory supporting surf ace was completed by using a
24 inch column cap held in place by a 2 in. x 12 in.
scabbing plate on both sides of the cap. This gave the
support top dimensions of approximately 7 inches by 24-
inches. The legs were spaced 8 inches apart and braced at
24 inch intervals by a 2 in. x 6 in. member on each side.
The base of the support was constructed so that small
height adjustments could be made by wooden wedges. Larger
adjustments were made by placing shims beneath the base of
the support.
75
11
36
11 11
12 12" 12
f l T :1
c
_i;i_ - -
c
_si _ _ - _c_
D c
-'1- - I 12
11
_J 12
11
11 11
4 x4
11
32
11 11
12'-6
11
2 x4
[J
11"
2
x 16 11
Screw Jock
Figure 19. Temporary Center Support.

76
Lifting Frame
A special frame was developed to provide a vertical

lift on all lift rings during removal of the quadrant from
the forms and during erection, Figure 20. The lift frame
also worked quite well when the quadrants were to be
lifted with the surface sloped at various angles.
The frame had a square configuration with a diagonal
brace. The frame sides and brace were constructed from
2 in. x 2 in. x 3/8 in. steel angles. Braces were placed
across each corner at 45° angles. The main diagonal brace
was welded to two of the corner braces. A 4 inch length
of 3 inch pipe was welded flush with the top of the frame
in each corner; one-fourth inch holes were drilled 3 inches
from the top of the pipe section so that a one-fourth inch
bolt could be inserted. These bolts were placed through
one link of a three-eighths inch diameter chain to main=
tain the angle between the chain and lifting frame.
The lifting mechanism was completed by two lengths of
three-eighths inch chain with hooks on each end. The
chain ends were placed through the pipe sections in adja-
cent corners of the frame so that lifting stresses would
be evenly distributed into both chains. The vertical
chain angle was adjusted to approximately 45 °, then the
chains were bolted at the corners. Each chain hook was
passed through a lift ring on the shell, then hooked back
to the chain to give the quadrant surface the desired
slope.
77
The maximum compressive stress in the frame occurred

in the diagonal brace. The calculated load in each of the
four chain legs was ~ x sin k5 _ = 4~~?~o~~·

0 = 1~130 lbs.
tensions under a static load (31). The dynamic load which
would be developed by a dynamic load design factor of 3.0
was Pw = 3,390 lb. per chain section. The critical load
condition would occur when the entire load was supported
by a chain connected to diagonal corners of the frame
parallel to the diagonal brace. This situation would pro=
duce a calculated dynamic load, P w = 6,780 lbs. and. a com-
pression of Pw cos 45° = (6,780)(.707) = 4,800 lbs. in the
diagonal brace. Checking for buckling gives Fallow. =
A. f _ ( 94 )(18t000j _ 16,920 _ 4 560 lbs. For Fallow=
- • i .e/d 2 - 3. 71 - '
1 +181,000
4,560 lbs.• , the calculated factor of safety under the

. Pall ow
critical condition for static loading, F = p =
s static
h560
1 , 600 = 2.859 therefore, care was exercised in connecting
the chains to the frame through adjacent corners and in
lifting the quadrants.
The safe working chain loa.d was T = 8D2 = 8( 3/8) 2 =
1.125 tons, or 2,250 lbs., tll.~re
., D = 9-iameter of one side
of the chain link in inches (31). The equation employs a
Fs of 4.0; a factor of 3.0 would give an allowable stress
of 3,000 lbs. This is close to Pw=3,390lbs. which was also
computed with a design factor of 3.0.
78
Figure 20 . Lifting Frame.
F igure 21 . Erected Column Showing

Cribbing Clamps.
79
Column Supports for Foundation Casting
A special arrangement of cribbing was used to main-

tain the vertical position of the column during the time
that the footings were cast and cured. This cribbing,
Figure 21, consisted of 4 in. by 4 in. members, 36 in.
long, clamped to the column in both directions to support
the column weight. Additional vertical support was pro=
vided by 2 in. by 12 in. members clamped vertically to the
sides of the column and butted against the bottom of the
haunch at the column face.
The maximum stress in the cribbing was produced by a
cantilever moment when each end of the cribbing in one
direction supported the entire weight of the column. One=
fourth of the column weight, 2,200 lbs., supported by each
member, or 550 lb. per member, produced a calculated load
on the end applied over a distance of 12 inches. The mo-
ment, M = P x L = 550 x 12 = 6,600 in.=lbs. and the shear
force, V = 550 lbs. The allowable extreme fiber stress for
Douglas fir (framing and joint grade) ·was 1,200 psi~ the
allowable shear stress perpendicular to the grain was
325 psi (26). Thus, f = r = 1 ~?~4 = 41.85 psi which is
,
less than the 325 psi allowable. The section modulus re-
. d was S = M
qui.re f = 6,600 in.-lb. = 5 •5 in.
. , wh'i.c h was l ess
12200 .psi
than the value of 7.94 in.' for a 4 in. by 4 in. member.
CHAPTER VI
ERECTION PROCEDURE
The erection of the h-p shell from prefabricated ele-

ments required that five separate construction phases be
integrated into a continuous operation. These phases con~
sisted of site layout, column and tie erection, assembly

of the support system, shell assembly, and final shear
connections.
To make the study as realistic as possible, the con-
struction of the quadrants and columns, the column and tie
erection, and the assembly of the structure were carried
out by an untrained crew with one of the departmental
staff members acting as general contractor or foremanc The
author was available for coordination with the foreman on
construction procedures and plans, but did not actively
supervise. In general, the entire construction phase was
carried out as if this was a general contractor's crew~
unfamiliar with the construction of an h-p shell"

The following paragraphs discuss the methods used to
carry out each phase of the erection procedure. It should
be noted that some of these steps were carried out concur-
rently as would be done on a prototype construction projectc
80
81
Site Layout
The building site was on the Animal Husbandry farm,

one and one-half miles west of the Agricultural Engineer-
ing Laboratory. The site was cleared and leveled by per-
sonnel from the Animal Science Department. After completion
of this study, the structure was to be used as a machinery
storage shelter.
The project foreman coordinated with the farm super-
intendent on the approximate location and the general
orientation of the structure, and discussed the movement
of the fences. Then, a two-man team surveyed the topogra-
phy of the site and staked the principle building points.
Figure 22 indicates the general layout of the structure~
elevation points, and column locations. The site layout

and construction staking required 31/2 hours for the two-
man crew.
The foreman contracted a rotary drilling truck and
operator to dig the foundation holes; 20 inch diameter
holes were required, but the maximum bit size on the drill
rig was 16 inches in diameter. The two holes were drilled
to approximately 54 inches depth in one hour. The neces-
sary reaming from 16 to 20 inches plus the excavations for
the wing walls required two additional hours for a three-
man crew. After the footing excavations were completed~
the holes were covered to keep out moisture until the

columns were erected.
82
Roof Outline
Outline Of Footing
,........l(f-+-t'-E~xcovo~~ _ J ____
1
19 -2"
9 7
I. 20'-o"
.I
Figure 22. Worksite Layout Showing Construction
Control Points.
[I](
~
,
®
D Support
Column
Storage CD System
Storage
~
®
Assembly Sequence
CD -Precost Columns Erected
® - Support System Assembled
@ - Roof Assembled
Figure 23. Material Layout for Assembly.

83
The columns were cured and erected in the forms to
prevent damage to the concrete during lifting and movingo
Both of the columns were loaded on a three-ton equipment
trailer with a fork truck for transportation to the con-
struction site. A three-man crew moved the columns to the
site in one hour. The columns were stored at the site as
indicated in Figure 23.
Because of the sequence for precasting the columns
and shells, and constructing the assembly supports, all of
the materials were not completely laid out at the worksite
at the same time; however, this plan of material location
was followed as closely as possibleo The storage area in-
dicated in Figure 23 for the rigid frame and temporary
wooden supports was not ut.ilized due to space limitations
on the east side of the construction siteo The north,
south, and west sides were relatively unrestricted for
locating and moving construction equipment and materialso
The shell quadrants were moved to the site after the
columns and tie had been erectedo Each quadrant was lo-
cated as indicated in Figure 230 The assembly jig and
wooden supports were moved to the site and assembled in
their approximate locations.
No provision was made at the worksite layout for the
parking or storing of major items of construction equip-
ment. For this study, only one modular unit was erected,
thus heavy equipment was required for short periods of
time.
84
Column Erection
The column forms were used in the erection procedure

in this study. Ordinarily, these forms would be removed
by the second or third day for re-use in casting more
column units. Two clamps, each made from two 21 inch
double-threaded bolts and two 2 in. by 4 in. members were
used to hold the side members against the column, Figure 8.
The upper ends of these two members were placed against the
lower surf ace of the haunches at the column face and were
utilized as vertical supports during erection.
When the footing steel had been placed in the wing
wall excavations, the airport crane lifted the column by a
chain around the haunch arms, Figure 24. The crane lowered
the column tbrough the footing steel cage into the footing
excavation until it was at the desired elevation. A
target elevation had been marked on the column face five
feet below the column top. By using the transit height of
instrument reading from a temporary bench mark (used for
the initial topographic survey), the exact elevation was
determined.
The vertical support members on the sides of the
column had been cut off to rest on the cribbing for column
support. The outside face of the south column was aligned
with the corner stakes and centered between them. The
column was then plumbed in both directions when final ele-
vation changes were made. Brace boards were placed in the
85
Figure 24. Method of Lifting Precast Column

During Erection.
Figure 25. Second Quadrant Being Lowered

During Assembly.
86
four principle directions and fastened securely before the

lift chain was released.
The second column was lowered into the footing exca-
vation and the cribbing was fastened in place. The dis-
tance between the outside column faces and the relative
elevation of the second column was checked by transito The
column was then aligned between the corner stakes, set at
the proper elevation and plumbed. Because of the dimen-
sions of the precast elements, the distances between the
outside column faces at the top of the column was 19 feet
11 inches.
When the second column was correctly aligned and
plumbed, it was braced rigidly, Figure 21~ and the crane
support was released. The time required by a four-man
crew plus crane operator to erect the columns, from the
time the first column was ready for lifting until the
column footings were ready for casting, was 2 hours and 50
minutes.
The footing concrete was delivered to the site by a
ready-mix truck. The concrete was placed by a two-man
crew and required approximately 30 minutes of working time.
The footings were difficult to rod because of the small
amount of clearance between column reinforcing steel and
side of the excavation. A 22 or 24 inch diameter footing
would have been easier to place, especially with an elec=
tric vibrator. No special curing procedures were used
because of the small amount of surface area. The braces~
87
cribbing, and vertical supports were removed after two

days.
Tie Bar Connection
The tie bar was cut to a length of 18 feet 8 inches

to allow a 4 inch overlap on each 11
T" sectiono The bar
was laid in place and clamped to the 11
T 11 section at each
end. The welder spotwelded the bar securely to one tie
plate, then the bar was raised slightly at midspan and was
spotwelded to the tie plate on the opposite columno The
bar was then welded securely to the column tie sections on
each column.
Support System
The assembly jig was assembled in its approximate

location when it was moved to the building siteo The cor-
ner towers were tilted up into position by the three-man
assembly crew after the screw jack footings were inserted
into the pipe sleeves. The towers needed no adjustment at
the vertical lap joints, because the ground elevations did
not vary by more than 8 inches. The horizontal braces had
been previously marked according to their tower connection
and frame position; these members were bolted to the towers
and connected at the slotted overlap joint at midspans.
The tower elevations were adjusted roughly for ease in
connecting the horizontal braceso The tower assembly, not
including final alignment, was completed by hand by a
88
three-man untrained crew in three hours.

During the final tower alignment, the corner supports
were centered on the columns, then the four tower caps
were adjusted to 20 feet 2 inch horizontal spacings for
assembling the quadrants. The lower horizontal frame
members on the north and south sides were clamped to the
concrete columns with 12 inch "C" clamps to prevent move-
ment of the frame under an unsymmetrical load. Final
adjustments were made in the corner elevations by checking
the tower cap height with a survey rod and raising or
lowering the towers by the screw jack legs to the correct
heights.
The wooden end supports were placed in position at
points 2 and 8 in Figure 22. Each end support was con-
nected to the top and bottom horizontal frame members.
This stabilized the support until the roof quadrant was
lowered onto it. Small elevation changes were made by the
use of shims beneath the base and by wooden wedges.
The center support was positioned over the tie bar 9
then the metal braces were fitted onto the removable 8inch
leg section. The vertical height was set by adjusting the
two screwjacks at the base and checking the elevation with
the transit and survey rod. Then, the support was centered
horizontally and diagonal braces were set in tbe four
principle directions, Figure 25. The time required for
the alignment of the support system by a four-man crew was
2 hours and 45 minutes.
89
Shell Assembly
The procedure used during the shell assembly was to

erect the quadrants in a pattern that would keep the sup-
port system stable during all phases of the assembly
process. This requirement was met by erecting the quad-
rants in the sequence illustrated in Figure 23. The first
and second quadrants were placed in positions adjacent to
the column tie so that the low corner of each quadrant
rested against the inverted "T" section which formed the
column tie connecter; these column elements formed an ef-
fective guide during the assembly. The two adjacent quad-
rants rested against each other along the horizontal
interior edge beam. The same procedure was carried out
with the third and fourth quadrants so that the entire
erection took place by rotating the crane's position in a
clockwise direction around the structure to minimize crane
movement.
The assembly crew consisted of the construction fore-
man~ the crane operator, and three workmen. When lifting
the quadrants, two workmen used tag lines to guide the
quadrant into position on the frame. The movements of the
crane were supervised by the foreman.
When the first quadrant was test lifted into position,
the lower corner of the quadrant did not fit well on top
of the column; close observation revealed that the quad-
rant was resting on the point of the shell corner. It was
90
also noted that the tie rod would keep the third and fourth
quadrants from fitting against the vertical web of the tie
connection. The first quadrant was lowered and a portion
of the corner was removed. The lower corners of the re-
maining tl:).ree quadrants were also corrected in the same
manner.
The first q11adrant was , lifted into position and t~e
interior edge ~eam was visually aligned along the tie bar
as it was lowered. The second quadrant was lowered into
position in the same manner. Approximately 3;4 inch sepa-
rated the two quadrants along the horizontal interior edge
beam after they were initially set in place. The thirdand
fourth quadrants were lowered onto the support system with
their sloping edge beams against the first and second
quadrants. The total assembly time required to connect
the lifting frame to all quadrants and set them in posi-
tion was one hour and twenty minutes. Figure 26 shows the
shell immediately after assembly and prior to welding.
Welded Shear Connections
The quadrants had to be adjusted vertically and

pulled together before the edge beam plates could be
welded. After the quadrants were adjusted at the column
top, a heavy weld was made connecting the sloped edge of
the inverted "T" section to the edge beam angles, which
fit directly against it. This weld on each end served to
tie the shells solidly to the column top.
91
Figure 26 . Support System Holding Shell

Quadrants for Welding.
Figure 27. Roof Center Showing Method of

Pulling Quadrants Together.
92
Initial closing along the horizontal interior edge

beam was done by lowering the center and end supports.
Instead of using the turnbuckles on the assembly jig to
pull the quadrants together, a chain was connected between
the two lifting rings, parallel to the horizontal edge
beam and lo~d binders were used to pull the sloped edge
beams together, Figure 27. Because of the lack of com-
plete uniformity in casting, the quadrants did not match
at the center of the roof. The maximum desired allowance
for misalignment and spacing between edge beams was one-
half inch. This was the maximum that actually occurred
due to warping of the edge beam angles when the dowels
were welded to the edge beam, and due to the spacing in-
duced by the web of the inverted "T" section on the column.
As soon as the quadrants were bound together, the angle on
the haunch was welded to the precast 23 inch angle in the
lower corner of the shell. Two of the angles in the shell
quadrants did not fit up against the haunch angle so a
one-fourth inch steel bar was used as a filler and welded
to the two angles.
Next, the 24 inch tension bar was centered over the
intersection of the four quadrants and was welded in place.
The 1's in.
x 1 )12 in. bars were then centered over the edge
beam angles on the four edge beams and welded. A 1/4 in.
to 3;s in. leg fillet weld was used throughout the edge
beam welding except on the center tension bar. The total
time required by the welder to complete the shear and
Figure 28. Completed Structure.
94
moment connections was 15 hours. As soon as the final

welds were finished, the shell was structurally complete
and the support system was removed, Figure 280
CHAPTER VII
TESTING APPARATUS AND PROCEDURES
This phase of the study included experiments to ana-

lyze the properties of the structural elements, and tests
of the structure under two types of static loadso First~
the testing of the tie bar material and the concrete sam-
ples will be discussed. Then, the procedures and equip-
ment used for load tests on the structure will be
explained.
Tie Bar Calibration Tests
To determine the modulus of elasticity of the tie

bar, two bar samples were tested in tension in the Riehle
100,000 Pound Testing Machine located in the Agricultural
Engineering Laboratory.
A section at the center of each bar was ground down
and smoothed on a belt sander. Two gage locations were
marked 180 degrees apart near the center of the baro By
using an accelerator with the cement, the gages were
bonded and ready for testing in approximately three
minutes.
A Baldwin strain indicator and 10 channel Baldwin
96
switching and balancing unit were used to indicate the

strain. A temperature compensating gage placed on a bar
sample was used to complete the external portion of the
Wheatstone bridge~
Loa4 was applied in increments of approximately 2~000
pounds. At each load increment, the scale was balanced

and the load and strain readings were recorded simultane-
ously. The maximum loads placed on the two samples were
33,690 pounds and 35,230 pounds.
The recorded loads were converted to stress values by
dividing each load by the cross section area of the bar.
A plot of stress versus strain was made by regression anal-
ysis to determine the slope of the curve, which was the
modulus of elasticity. An average value of the modulus of
elasticity of Es = 30.48 x 106 psi was obtained. The ob~

served data for,the tie bar samples testing is in Appendix A.
Concrete Test Samples
Samples of the standard concrete in the precast col-

umns and the lightweight aggregate concrete in the shell
were taken during casting. These were cured under the
burlap material with the columns and shells.
Three column test samples were cast in 3 inch diame-
ter molds, 6 inches deep. These samples were tested in
the Riehle 100 ,00 Pound Testing Machine after 14 days of
curing time to determine the strength of the column
concrete.
97
The average 14 day ultimate strength of the test

cylinders was 3,086 psi, From the strength of these
samples, the modulus of elasticity of the standard con-
crete was determined. The ACI recommendation for the
modulus of elasticity for concrete was Ec = 1~000 f~,
providing the concrete was moist cured for 28 days (25)o

To adjust the results of the 14 day test, Figure 7 and
Figure 9 from Design and Control of Concrete Mixes (32)
were used. The adjusted 28 day ultimate strength of the
column concrete, moist cured for 10 days, then air cured 9
z.
was f Ic = _2 086
90
)Si
x .95 = 3,260 psi.
= 3.26 x 10 6 psi was used.
Three lightweight aggregate concrete samples were
taken during the casting of the first, second, and fourth
quadrants. The samples were removed from the mold after
the first day and continued to cure under the plastic
shell covering until the moist curing was completed at the
end of 14 days. The three cylinders were tested at 21
days to check the strength of the quadrants for removal
from the forms and movement to the site. The average 21
day strength of the samples was f 'c = 4 ,480 psi.
Structural Testing
Tie Bar Testing
The tie bars were tested by mounting two sets of foil

strain gages approximately 3 feet from each end of the bar.
98
The bar was ground down to approximately 1 1/4 inches diame-

ter and smoothed to present a uniform gage mounting sur-
face. The gages were mounted 180 degrees apart longitudi-
nally in the same manner that was used on the test samples.
The tie bar was welded between the columns so that the
gages were vertically opposite. A temperature compensating
gage was mounted on 1 1/4 inch diameter steel bars which
were located on the tie bar near each set of gages.
Tension Bar Testing
The steel bar connecting the four quadrants at the

center of the roof was tested for tensile stresses by a
strain gage centered on the bar over the edge beam gage.
A temperature compensating gage was placed on a 6 inch
length of the same material. During load testing 1 the
gages were covered by a galvanized steel box formed to fit
the roof slope and bar protrusion. It was bolted down to
the concrete by 3;8 inch diameter nail-set bolts. The lead
wires were protected by a flexible conduit connected to the
side of the box and extended to the edge of the roof.
Column and Haunch Testigg
To determine the magnitudes of strains induced into

the column and haunches under varied loading conditions~
two gages were placed on each haunch arm along the center=
line of the bottom side, one gage was centered 4 inches
directly beneath the haunch on each side of the column~
99
and one gage was centered approximately 6 inches from the

base on each of the four sides, Figure 29. Six inch paper
backed gages were used. Two compensating gages were
mounted on a 10 inch concrete cube, cast to simulate the
column dimensions; each compensating gage served the
active gages on one column.
Strain Gage Equipment
The strain gage testing equipment consisted of a

Baldwin strain indicator, one 20 channel, and one 10
channel Baldwin switching and balancing unit. The twenty
gages mounted on concrete were connected to the 20-channel
unit and the five gages mounted on steel were connected to
the 10-channel unit.
The strain gage equipment was placed in a small
wooden building approximately 12 feet from the south
column. This building protected the instruments and did
not interfere with the loading of test material onto the
structure. The maximum lead wire distance was limited to
approximately 50 feet while the shortest lead wires were
approximately 25 feet long.
Deflection Apparatus
A manometer type deflection device was constructed to

measure the vertical roof deflection at seven points and
the relative vertical movement of the two columns. The
reservoir of the manometer was a large coffee urn~
100
West
+~=O
4 13 14
No. 7 On Bockside No. 17 On Bockside

Of Column 19 .' Of Column
10
20 ~ 18
North Column South Column
Figure 29. Positions of Strain Gages

on Columns and Haunches.
Figure 30. Manometer Deflection Apparatus Showing

Reservoir and Moveable Section.
101
Figure 30. The urn was clamped to the inner face of the
south column so that all movement of the structure could
be related to one point.
A datum line was scribed around the glass water level
tube on the urn. A 10 inch section from a scale with 50
divisions per inch was mounted vertically against the tube.
· A 3;s inch outside diameter plastic tube was attached to
the spigot directly below the water level tube to connect
the manometer reservoir to the movable end of the manome-
ter. The movable section of the manometer consisted of a
glass tube clamped to a 1 in. x 2 in. x 18 in. board, a
50th scale which was attached to the board behind the
glass tube, the 3;s inch plastic connector tube~ and a 1/4
inch steel rod approximately 7 feet long, Figure 30. The
bottom end of the steel rod was rigidly attached to the
top end of the board. The top end of the rod was formed
into a ring to use in suspending the manometer board from
hooks which were clamped to the edge of the shell.
The manometer was open to the atmosphere on both ends
so that no pressure differences were developed. The manom-
eter reservoir was filled with approximately 31/2 gallons
of a water and alcohol mixture to prevent freezing during
cold weather. During the testing period, water level
readings were taken each day from the datum line on the
water level tube to correct for evaporation. Readings
were taken on the movable end of the manometer at all 8
points before and after each roof load change.
102
Testing Procedure
The structural load-testing consisted of three uni=

form load tests and one eccentric load testo Each test
included one or more load increments.
The procedure used for each load increment was~
1. Zero the strain gages at 1~000 on the

indicator.
2. Record the initial datum reading on the
water level tube of the manometero
3. Record the zero reading on the north
column and at each of the four corners of
the roof, at midspans of each horizontal
edge~ and the center of the roof.
4. Place the roof load increment on the shell
by loading alternate quadrants on each
side of the column tie~ and spreading the
load material uniformly.
5. Take depth measurements at 13 points on
each quadrant to obtain an average depth.
Average the four quadrant depths to obtain
the average roof depth.
6. Take density samples during the loading in
12 inch square pans which were 2 inches~ L-i.
inches~ and 6 inches deep. Fill each pan

in the same manner that the rest of the
roof was loaded.
103
7. Weigh the density samples and average the

densities to obtain an average density
for the roof load.
8. Record the strain readings for all 25
gages after the instrument has warme d up
for approximately 5 minutes .
9. Record vertical roof deflection data .
This sequence was repeated throughout the testing
period with the exceptions of steps 6 and 7. The densi ty
of the gravel was checked periodically during each test
phase, especially after a change of weather conditions .
To load the structure, a three- point hookup tractor
slip mounted on the lift arms of a fork lift truck was
used. The operator filled the slip by driving it into a
gravel pile, raising the slip above the roof height, and
dumping the load. The load material, consisting of ~8
inch unwashed chat, was held on the roof by 8 inch depth
wooden forms.
The initial loading phase consisted of a sustai ned
uniform roof load of 21.6 lbs . /ft. 2 of horizontal proj ec-
tion. This load was placed on in one increment an d served
as a preliminary load to settle the structure. Deflection
and strain readings were taken during a 72 hour load
period, then the shell was unloaded and the final zero
readings were taken.
The second load condition was a uniform r oof l oad
with increments of 25.0, 22 . 0, and 14.7 lbs . /ft. 2 to gi ve
104
a total load of 61.7lbs./ft. 2 • The maximum load which was

approximately 1.5 times the design load~ remained on the
shell 114 hours.
The third load consisted of an eccentric roof load of
41.3 lbs./ft. 2 on one-half of the roof surface which caused
a cantilever load centered 5.0 feet from the column tie.
This load, which was approximately design load~ was placed
on the roof in increments of 25.0 and 16.3 lbs./ft. 2 and
remained on the structure for approximately 45 minutes
while the readings were taken.
The fourth load consisted of a uniform total load of
57.0 lbs./ft. 2 placed on in progressive increments of 19.3~
16.1, 13.6~ and 8 lbs./ft. 2 ; the duration of load for this

test was 46 hours.
CHAPTER VIII
PRESENTATION AND ANALYSIS OF DATA
The data from this construction engineering s t udy

will be analyzed in two categories: (1) Analysis of con-
struction costs, and (2) Analysis of load test data . The
results of these analyses will be discussed in Chapt er IX .
Analysis of Construction Costs
The analysis of construction costs will be divided

into three sections: (1) labor costs, (2) material costs ,
and (3) equipment costs. The observed data from this
study are valid only for this project. A set of skill or
experience factors will be discussed in Chapter IX . These
may be used to estimate actual construction labor c ost s by
adjusting the observed data.
The labor wage scales were estimates from , Estimating
Construction Costs ( 33), Table 1-2, "Union Wage Scale I n
The United States, In Dollars. " This table list s an e sti-
mated average rate, and a range ~n rates. The average
rate for building laborers, $2.18 per hour, will be used
for unskilled labor costs, and a rate of $3 . 13 per hour ,
which is the average rate for carpenters , will be used f or
skilled labor and supervision.
105
106
Other costs for material or equipment were either

actual costs incurred or estimates obtained from local
sources.
Labor Costs
The labor costs will be tabulated for each phase of

construction for unskilled labor and skilled labor (or
supervision) on a man-hour basis. The final cost for
labor will be computed from the total man-hours.
TABIE I
LABOR COSTS
----
Item Skilled Unskilled
1. Column Construction
(a) Forms 2 40
(b) Steel Forming 5 54
(c) Casting and Curing 4 12
SUBTOTAL (Man-hours) 11 106
2. Shell Construction
(a) Forms 16 96
(b) Shell Steel Forming 8 78
(c) Form Preparation and
Casting 4 16
(d) Curing -2 10
107
TABLE I (Continued)
3. Support System Construction

(a) Cutting Out Parts 1 7
(b) Welding Tower Frames 36 0
(c) Assembly of Bolted
Components 0 3
(d) Wooden Supports 2 16
(e) Adjustments on Steel Frame _l ~
4. Lift Frame Construction 6 0
5. Site Preparation
(a) Leveling and Smoothing 0 2
(b) Survey and Layout 2 2
(c) Foundation Excavation 2 --2
SUBTOTAL (Man- hours) 4 13
6. Site Layout
(a) Hauling Columns and
Placing 2 2
(b) Construction of Shell
Supports 2 4
(c) Removing Forms, Loading ,
and Transporting Shells
to Site 4 7
(d) Moving Support System 1 4
7. Column Erection
(a) Development of Column
Support System for
Stabilizing Column 4 25
(b) Cutting and Bending
Footing Steel 0 4
(c) Column Erection and
Plumbing 4 14
(d) Casting Column Footings 0 1
(e) Removal of Braces and Site
Cleanup 0 4
108
TABLE I ( Continued)
7. (Continued)
(f) Welding Tie 2 2
8. Support System Erection
(a) Initial Erection of
Corner Towers 3 9
(b) Final Alignment of
Tower s f or Shell
Erection 4 12
9. Erection of Structure
(a) Initial Assembly 4 12
(b) Prepar ation for Welding 2 4
(c) Welding Edge Beams and
Column to Shell
Connections 15 0
(d) Support Removal and Site
Cleanup 0 6
(e) Grouting Top of Columns 0 4
(f) ·Waterproofing Interior
Edge Beams 1 5
(g) Final Cleanup 0 4
FINAL TOTAL (Man- hours) 140 472
The total cost for labor was Labor Cost = 140($3.13)

+ 472($2.18) = 438 .20 + 1,028.96 = $1,467 . 16.
Equipment Costs
Equipment charges were made for all equipment used,

109
whether rented or obtained from the Agricultural Engineer-

ing Labor ator y. Labor was included with rental equipment
charges on the crane, tractor dozer, and rotary drill rig;
all other equipment was laboratory property or operated by
departmental personnel. Labor charges were shown in Labor
Costs . Table II shows the types of equipment used and the
number of equipment- hours for the specific jobs. The
local electric welder rates varied from $2.00 in the shop
to $3.00 for portable welders.
TABLE II
EQUIPMENT COSTS
Item Hours Cost
1. Acetylene Welder (Labor and

material separate)
(a) Heating and Bending
Column Steel 13
(b) Heating and Bending
Shell Steel 6
(c) Tower Support Frame 7
(d) Footing Steel Cage 1
(e) Lifting Frame 1
SUBTOTAL ( At $3.00/hr.) 28 $84.00
2. Electric Welder
(a) Column Steel Forming, $2.00/hr. 3 6 . 00
(b) Tie Bar and Column Tie Plate,
$3.00/hr. 2 6 . 00
(c) Shell Steel Forming
(1) Welding dowels to edge
beam, $2.00/hr. 10 20 . 00
(2) Corner reinforcing mat,
$2.00/hr. 4 8 . 00
110
TABLE II (Continued)
Item Hours Cos t
(3) Final welding on shell

and edge beams, $3 . 00/hr. 5 $15.00
(d) Rigid Frame Supports,
$2.00/hr. 33 66·.oo
(e) Footing Steel Cage, $2.00/hr . 1 2.00
(f) Lifting Frame, $2 . 00/hr . 2 4-.00
(g) Portable Welding on Edge
Beams During Shell
Erection, $3.00/hr. 15 4-5... 00
SUBTOTAL $ 172.00
3. Tractor and Equipment Trailer,
$2.50/hr . 9 22.50
4- . Tractor With Drawbar Hoist, $2.00/hr. 2 4-.00
5. Fork Truck, 10 Ton Capacity, $3.00/hr. 2 6.00
6. Crane , 10 Ton Capacity (With operator),
$6.00/hr . 8 4-8.00
7. Tractor Dozer for Site Leveling (With
Operator), $6 . 00/hr . 2 12.00
8. Ro t ary Drill Truck (With Operator),
$12 . 50/hr . 1 12.50
9. Power Hacksaw , 20 cuts per hour ,
$0 . 10/ct, or $2 . 00/hr . 9 18.00
FINAL TOTAL $379 .00
Materi al Costs
The cost of materials was separated from labor and

equipment to provide a clear outline of the expenditures
charged to each part of the project . The materi al costs
are listed in Table III .
111
TABLE III
MATERIAL COSTS
Item Quantity Cost

1. Welding Materials
(a) Welding Rod 75 lbs., $ . 20/lb. $15.00
(b) Acetylene 1- 100 cu. ft. bottle 5.70
(c) Oxygen 1 - 224 cu. ft. bottle 5.65
SUBTOTAL $26.35
2. Concrete
(a) Standard Weight,
3,000 psi 3 cu. yd. , $14.75 44.25
(b) Lightweight Aggre-
gate, 3,750 psi 3! cu. yd., $18.25 63.87
SUBTOTAL $108.12
3. Steel Material
(a) Assembly Support
SysteJJl
(1) Steel 1,540 lbs., $.097/lb. 149.69
(2) Jacking Screws 12, $9.62 each 115. 44
(3) Turnbuckles 4, $3.20 each 12.80
(4) Bolts and Pipe 14.10
(b) Lifting Frame
(1) Steel 99 lbs., $.097/lb . 9.62
(2) Pipe 16 in., $.25/ft. . . 33
(3) Bolts 4 - i in. x 4 in., $.10 .40
(c) Shell and Column
Steel 2,056 lbs., $ . 097/ lb. 199.84
(d) Jacking Screws,
Center Support 19.74
SUBTOTAL $521.96
4. Lumber and Miscellaneous
(a) She 11 Forms 170.39
(b) Column Forms 24 . 15
(c) Assembly Supports 30.62
SUBTOTAL $225.16
FINAL MATERIAL COST TOTAL $881.59
112
From Tables I, II, and III, the total combined costs

for labor, equipment, and material was determined. The
total initial cost of precasting the 20 foot square h-p
shell was $2,727.75. This would be an initial cost of
$6.82 per square foot of horizontal projection, for one
use of forms and erection apparatus.
Of the total cost, 53 . 8 per cent was for labor, 32.3
per cent for material, and 13.9 per cent of the total was
charged to equipment. The multiple use of forms and erec-
tion equipment, and a discussion of tbe cost of construct-
ing a 40 foot square prototype will be discussed in
Chapter IX.
Analysis of Load Test Data
The analysis of the data from the load tests compares

theoretical computations with the observed data from the
structural tests . The data c onsists of observed strain
and deflection readings recorded during uniform and eccen-
tric roof load tests. The strain data were readings from
strain gages on the steel tie bar and the horizontal in-
terior edge beam at the roof center, and strain gages
mounted on the columns . The deflection data were differ-
ential elevation readings taken at seven roof points and
on the north column.
Although three uniform load tests were run, Table IV,
only the strain data from Test IV are analyzed . The eccen-
tric load data from Test III are also analyzed . The
lJ_J
strain data are analyzed first, then the deflection data

are presented.
TABLE IV
STRUCTURAL TESTS APPLIED TO SHELL
Test No~ Type of Loading l'•Iaximum Load Time Duration
I Uniformly Distributed
Gravity Load, 1 Load
Increment 21~6 psf 73 1/2 hrs~
II Uniformly Distributed
Increments 61@7 psf 117 1/2 hrs.
III Half-roof Eccentric Load,
Uniformly Distributed, 2
Load Increments 41.3 psf 3 hrs.
IV Uniformly Distributed
Increments 57~0 psf 74 1/2 hrs*
·-----..-------. ----·
The strain values of gages 1-20 for Test III were not
adjusted due to the residual strain which remained in the
structure after the test was completed. Strain values for
gages 21-25 for Test III and gages 1-25 for Test IV were
adjusted by using a ratio of the time of reading against
the total time of the test and adjusting the final zero
load values.
114
Properties of the Column Section
The properties of the column in both directions, ~ =

0 and ~ = 90 °, Figure 4, were determined for the analysis .
The values that were determined for each direction were
(1) the column width, b, (2) the distance from the center
of tension steel to the extreme compression face of the
column, d, (3) the depth of column section, t, (4) the
location of the neutral axis, N.A., which is the distance,
kd, from the extreme compression fiber, (5) the distance
from the extreme compression fiber to the center of the
resultant compressive force, z, (6) the distance between
the resultant tensile and compression forces in bending,
jd, and (7) the moment of inertia of the transformed
section, It.
The column section properties were analyzed for both
the cracked and the uncracked sections, and the values are
listed in tabular form in 'Table V.
The modulus of elasticity of the concrete was adjusted
from t he 28 day value, Ee' because of the influence of
creep strain. The sustained modulus of elasticity , Ect =
1 psi where:
0 t + 6 t'
ot =Axial creep strain (specific creep), the time-

dependent unit creep strain of concrete per psi
of sustained axial stress, in millionths .
6 = Axial elastic strain = 1 psi
t Ee
115
From formula (10-1 ), (34), ot = c (t) l /r, where:

1
cl = A coefficient determined by tests, expressed

in millionths, the first days creep strain
under a stress of unity_ O. 500
- (a)o . 4o ·
r = A root deduced from tests;

t = Time, the duration of the loading, in days.
a = Age when loaded, in days.
TABLE V
PROPERTIES OF COLUMN SECTIONS
Property Values
Uncracked· Section · Cracked Section
~ =0 ~ = 90° ~ =0 ~ = 90°
(1) b 10.0 :in. 12.0 in. 10.0 in . 12~0 in.
(2) d (or d ) 9.5 ."in. 6.~5 · in. 9.5 in. . &..25 .:i.n,.
a:ve
(3) t 12. 0 "in. 10.0 in. 12.0 in. 10.0 in.
(4) kd 6 . o :i:n. 5.0 in. 4.15 in. 3 . 45 in.
(5) z 2.14 in. 1.81 in. 1.69 in. 1.35 in.
(6) jd 7.36 in. 4.4o in. 7.81 in. 4.90 in.
(7) 4 4 4 4
It 1957.0 in. 1176.o i n. 921.6 in. 410.2 in.
For Test III and Test IV, which were loaded on successive
TABL;E VI
Test IV Strain and Str e ss Data
Load 19.3 psf 3 5.4 psf 49 .0 psf 57 . 0 psf

Time l ! hrs. 2 } hrs. 3} hrs . 4 ! hrs . 20 hrs . 50 hrs .
!
Gage No. (
I er ( I (T c I (r E I er € I () ( '' (}
1 +12 +33.4 +13 +36.2 +24 +66.7 +42 +116.8 +18 +48.1 +15 +3 9.1
2 +2 + 5.6 -6 -16.7 +5 +13 . 9 +28 +77 . 9 +4 +10.7 0 0
3 +10 +27.8 +20 + 55.6 +41 +114 .0 +81 +225.0 +84 +224.l +50 +130.5
4 +27 +75 . 0 +24 +66 .7 +45 +125.1 +28 +77.9 +2 +5.3 +10 +26.1
5 +7 +19.5 +3 +8.4 +14 +38.0 +27 +75.0 -2 - 5.3 -5 -13 . 1
6 +12 +33 . 4 +3 +8.4 +14 +38.0 +26 +72.3 +9 +24 . 0 D 0
7 +25 +69.5 +27 +75.0 +40 +111.0 +55 +153 .0 +37 +98.9 +60 +156.8
8 +32 +89 . 0 +24 +66 .7 +45 +125.1 +28 +77 . 9 +22 +58.7 0 0
9 +8 +22.2 +14 +38.0 +25 '+69.5 +34 + 94 . 5 +16 +42.7 + 50 +130 . 5
10 - 6 -16.7 +1 4 +38.0 +8 +22.2 +57 +158.5 +52 +138.8 +120 +312.5
11 +11 +30.6 +12 +33.4 +13 +36.1 +24 +66.7 -12 - 32.0 -15 -39.1
12 +11 +30.6 +7 +19.5 +13 +36.1 +24 +66 . 7 -22 -58 . 7 - 25 -65 . 3
13 +3 +8.3 +18 +50 . 0 +17 +47.4 +45 +125.1 +5 +13.1 -30 -78.4
14 +22 +61.1 +23 +64.0 +45 +125.1 +37 +103 . 0 0 0+10 +26 . 1
15 +12 +33 . 4 +3 +8 . 4 +14 +38.0 +17 +47.2 -12 -32.0 -5 -13 . 1
16 +7 +19.5 +3 +8.4 +14 +38. 0 +17 +47 . 2 -2 -5.3 +5 +13.1
17 0 0 0 0 -10 -27.8 0 0 -20 -53.5 -40 -104.5
18 +16 +44 . 5 +9 +25 .0 +22 +61 . 1 +3 +8.4 +118 +315.0 0 0
19 +22 +61 . 1 +53 +147 . 2 +54 +150.0 +67 +186.1 +133 +355 . 5+120 +313.0
20 +4 +11 .1 +37 +103.0 +27 +75 . 0 +56 +156.0 -4 -10.7 0 0
21 -35 -1067 -50 -1524 -65 -1982 -75 -2285 -62 -1890 -55 -1677
22 -160 -4880 - 260 -7930 - 360 -1Q980 -422 -1Z860 -445 -14560 -405 -1ZJ50
23 -110 -3350 -190 -5790 -270 ~ 8240 -320 -9750 -320 -9750 -310 -9450
24 -1 30 -3960 -230 -7090 -310 -9450 -374 -1~400 -391 -1~920 -370 -1~280
25 -120 -3660 -210 -6440 -280 -8540 -340 -lD,370 -358 -1Q920 -340 -1Q370
6
Remarks: (1) Ee = 1 x lo- 6 in. / in . (2 ) u = psi. (3) Ec t(l) = 2.78 x 10 ps i at
t ime, 0 through 4 ! h rs : (4) Ect ( 2 ) 2.67 x 10 6 p s i at time , 20 hr s.
6
( 5 ) Ect ( 3) = 2.61 x 10 p s i at time, 50 hr s . (6) Es = 30 . 48 x 106 p s i .
117
11
h =113
M
M~
..Gage-4:- _- -
ELEVATION ELEVATION
I
b=IO
0
II
0
2
I
0
L...___0_10
Kd=4.15'~9.5J I
i:=, =12''~
PLAN PLAN
(a) Bending in ~ = 90° Direction (b) Bending in ~ = 0 Direction
Figure 31. Columns Showing Directions of Bending,
Dim~nsions, and Reactions.
118
days, c 1 remained constant; c 1 = 0.0533 x 10-6 • For Test

~
III, ut =_ 0.0533 x 10-6 , 6t = 0.307 x 10 -6 , and Ect = 2.78
x 106 psi. For ~est IV, the values of Ect( 2 ) = 2.67 x 10 6
,psi, and Ect( ) :;: 2.61 x 106 psi.
3
Analysis of Load Strain Data
The axial load imposed upon each column was N = 57.0

x 200 = 11,400 lbs. According to elastic theory, both the
concrete and reinforcing steel would deform equally due to
bond. Thus, 6t = es·h = ec·h, where h =the height of the
column above the base gage centerline, Figure 3l(a). From
p -·h PT
this, 6t = A:•Es' and Ps = The maximum axial
Ac
1+ -
n•A s
load per column which was supported by the steel was P 5 =
3,180 lbs. Thus, At = 2.48 x 10-3 in. The axial deforma-
tion for each column was the average of the four strain
readings at the base of each column. These values are
tabulated in Table VI.
The axial strains for both columns are shown in
Table VII for the entire load period during Test IV. The
values for the bottom gages of each column are compared
with the values of the other column. By comparing the
values of axial strain for the north and south colum~s in
Table VII, it is evident that the columns did not deform
ideally. The differences in values between column (2) and
column (3) in the table may be due to unequal settlement
of the column footings. Thus, one of the .columns resists
119
{a} Time - 4~ Hours
0 0 0 0
l I l l l Jec=33.5 l J 111 Jec= 29.5
- - - -
(b) Time - 50 Hours
0 0 0 0 0 0
OJIIJec=55 mec=60 [ 1111 J ec =40
ebt =-80
eot11=-40
~~ ...............
-
+¢= 90°
Figure 32.
+¢ = oo
- e =Ix 16 6
in./in.
Axial and Bending Strain Distribution

-
+¢ :QO
During Sustained Loading in Text IV.

120
q q
11 I I I I JA H ~ I I I II
I. 11
36
I
-T -T
11
113
h~
114
11
~
l R
v
(a) Condition 1 • , Table VIII (b) Condi tioo 2. , Table VIII
q q
~A
A
-H H
fR
R
(c) Condition 3., Table VIII (d) Free Body at Base of, Tie Connection
Figure 33. Free Body Diagrams of Theoretical Stress
Conditions Acting on Column During Unifor'mly
Distributed Roof Loading.
121
more of the load. The bending strains in both directions

are shown in Figure 32 as an illustration of the change in
loads resisted by each column. These plots show the vari-
ation in bending due to unequal settlement of the footings.
TABLE VII
AXIAL STRAIN UNDER UNIFORMLY DISTRIBUTED LOAD
Time Strain (Micro-in./in.)

(Load) North Column South Column Average, All· Theoretical
Base Gages Base Gages Column Gages Strain
(2) (3) (4) (5)
17'2 hours 14.75 10.50 13.58 8.96

(19.3 psf)
1
2/4hours 16.40
19.75 24.75 21.91
(35.4 psf)
1
3/4hours 29.50 23.25 29.92 22.70
(49.0 psf)
1
4/2hours 43.50 31.50 49.17 26.40
(57.0 psf)
20 hours 31.75 56.50 37.00 27.50
(57.0 psf)
50 hours 57.50 20.00 29.17 28.20
(57.0 psf)
The values of strain in column (4) do not compare

favorably with the theoretical strain values, column (5).
This can be attributed to the fact that the footing on the
122
prototype does not hold the column base rigid, thus, the
footings deformed and the column could not resist the
entire force in bending.
Table VIII summarizes the stress conditions of the
columns under the uniformly distributed load during Test
IV for several structural conditions, Figure 33. Condi-
tion 3 was the nearest to the actual conditions at the
site. Condition 2 assumes pinned connections at the
haunch and ridge. All of the values in Table VIII were
determined by using the maximum values from the Test IV
data.
TABLE VIII
STRESS CONDITIONS UNDER UNIFORMLY DISTRIBUTED
GRAVITY LOAD FROM TEST IV DATA
Condition Tie Bar Shear at Bending at Bending at

Load Column Base Column Base Haunch
(lbs.) (lbs.) (in.=lbs.) (in.=lbs.)
1. Ideal Situation, Tie 19,000 0 0 0
Bar Carries All (Cale.)
Thrust, No Bending
in Ridge, Haunch, or
Column
2. No Bending in Haunch 16,990 2,010 227,000 0
or Ridge, Bending in (Meas'd) (Cale.) (Cale.)
Column
3. No Bending in Ridge, 16,990 4~760 21,900 520~000
Bending in Haunch (Meas'd) (Cale.) (Meas'd) (Cale.)
and Column
4. Bending in Ridge, 16,990 Indeter- 21,900 Indeter~
Haunch, and Column (Meas'd) minate (Meas 9 d) minate
123
According to Portland Cement Association design pro-

cedure (24), the horizontal th.rust acting at the top of
the column was Ph = 2•H•a = 2 ·w2~~·b = 57.0 ~l,OOOl =
19,000 lbs., if no bending stress exists in the edge
beams. This force is resisted by the columns in bending
and by the tie bar. From Test IV, the average maximum
stress in the tie bar under the 57.0 lb./ft. 2 roof load
was 11,540 psi, which produced a tensile force in the tie,
T = fs·As = 11,540 x 1.47 = 16,990 lbs. The shear resisted
by the column at the top was P v = 19,000 - 16,990 = 2,010
lbs. The calculated bending moment at the base gages 0
centerline due to shear at the top of the column was 2,010
x 113 = 227,130 in.-lbs. The moment der:Lved from the ob-
served strain data at the base gages' centerline was
fc•It
Mb = c = 21,900 in.-lbs. for an uncracked section and
11,080 in.-lbs. for a cracked section.
Checking the north column base strain data for the
second day of sustained loading, eb = ez ; ee. = 10 5 micro=
0
in./in., which was the same reading obtained the previous

day. For the south column, the observed bending strain~
6
eb = el9- - 17- = 76. 5 micro-in./in. The theoretical value
2
was eb = 48.1 micro-in./in. for an uncracked section. By
assuming the section was cracked, this value was eb = 33.2
micro-in./in.
The values of bending strain during the third day
e - e _ el9 = el2
were eb = 7 9 = 5 micro-in./in., and eb - - 2 - =
2
80 micro-in./in. for the north and south columns,
124
respectively. The theoretical strain assuming first an

uncracked section, then a cracked section, were eb = 45.2
micro-in./in. and eb = 31.2 micro-in./in.
A noticeable trend was developing during the sustained
load period; this was indicated by the decrease in bending
strain in the north column from 10.5 to 5.0 micro-in./in.
and from 33.5 to 80 micro-in./in. in. the south column.
During this same time, the deflection data indicated a
settlement of the south column of 0.04 in. between the
first and second day readings.
Throughout the three-day period, gages 7 and 9
(Figure 29) on the north column indicated an increasing
bending moment toward the+~= 0 direction, while gages 17
and 19 on the south column indicated bending toward +~ =0
on the first day but shifted to -~ = 0 on the second day
and back to zero on the third day.
The average compressive strain values of the base
gages on the north column during the two-day sustained
load period were ec = 43.5, 31.75, and 57.5 micro-in./in.,
while for the south column, ec = 31.5, 56.75, and 20.0
micro-in./in. These values indicate a shift of the
structural stresses.
The strain gage on the tension bar at the center of
the horizontal interior edge beam did not develop the
stresses for which the bar was designed. For the 57.0
lbs./ft. 2 uniform load, the calculated tensile stress in
the tension bar was 19,000 lbs. The maximum stress
125
measured by the gage was 3,430 lbs. The remainder of the

load, 15,570 lbs. was resisted in tension by the welded
plates in the sloped edge beams on both sides of the
horizontal edge beam.
Analysis of Eccentric Load Strain Data
The strain and stress data for Test III are shown in
Table IX. The values of the moments calculated from the
base gage strain data are tabulated in Table X. The cal-
culated maximum overturning moment due to the eccentric
roof load, Figure 3l(b), assuming idealized conditions
was M0 = 248,000 in.-lbs. for each column. Assuming an
uncracked section, Mb = 360,000 in.-lbs. for the north
column and Mb = 318,000 in.-lbs. for the south column.
Assuming the section was cracked, Mb = 249,000 in.-lbs.
for the north column and Mb = 220,000 in.-lbs. for the
I
south column. A check of the ~ ratio indicates that the
columns should be investigated for the cracked section
condition. The comparison of the strain values against
the theoretical value shows that the cracked section
values check very closely with the idealized moment.
To compare the observed strain to the calculated
values, the maximum bending strain in the direction of
overturning for both columns was derived from the data.
The values of actual bending strain were derived from the
observed values, Figure. 34, by the following relationships.~
Figure 35: (1) ebc = t~~d • ebt' (2) e 0 t = ebt + eac' and
126
TABLE IX
TEST III STRAIN AND STRESS DATA
Load 25.0 psf 41.3 psf Remarks
Gage No. € er I E I er
1 +30 +83.4 +20 +55.6 1. E.c = 1 x lo-6
2 +60 +167.0 +65 +180.5 in./in.
4 +150 +417.0 +280 +778.0
4 -70 -194.5 -420 -1168.0 2. CT= psi.
5 +10 +27.8 -10 -27.8 3. Ect = 2.78 x
6 0 0 -30 -83.4 106 psi.
7 -25 -69.5 -185 -514.0 4. Es = 30.48 x
8 -250 -695.0 -550 -1529.0 106 psi.
9 0 0 +40 +111.1
10 +200 +556.0 +430 +1195.0
11 -20 -55.6 +5 +13.9
12 -10 -27.8 +5 +13.9
13 +90 +250.0 +230 +639.0
14 -120 -334.0 -350 -972.0
15 -50 -139.0 -50 -139.0
16 -40 -111.1 -50 -139.0
17 -100 -278.0 -240 -666.0
18 -230 -639.0 -470 -1308.0
19 -40 -111.1 -90 -250.0
20 -150 -417.0 +365 +1015.0
21 -26 -792 -10 -304
22 -78 -2378 -134 -4090
23 -60 -1830 -110 -3358
24 -64 -1950 -117 -3570
25 -64 -1950 -117 -3570
-"'=
+ 0
ebt =-515
+"' =0
----
et= -499.I
1~.-1--J......L.LIO + 01~11-+~~
ebc=401
eA: 15.9
Theoretical Values
ec= 416.9
It Kd By Similar Triangles
eot =- 550 ebt

ebt
-- = i>c
t -Kd Kd
0
Kd
or ( I) ebe = - - x ebt
t-Kd
e0 c = 365 ebc
+
=
01 f f f *
eAC o, ! t ! i t t ,o e ( 2)
eAC 14.C eot = ebt + eAc
+
eot
( 3) e = e +e
oc be AC
0
e
Observed Values oc
e = 1 x lo- 6 in./in.
Figure 34. Axial and Bending Strains Figure 35. Definition Sketch for
in Direction of Gverturn- Heavy Bending Strain.
ing During Test IIIo
128
where:
ebc = compressive bending strain (unknown)
ebt = tensile bending strain (unknown)
eac = compressive axial strain (unknown)
e0 c = observed compressive strain
e 0 t = observed tensile strain.
TABLE X
SUMMARY OF MAXIMUM STRESS VALUES FROM TEST III
Location Type of Stress Idealized Analysis Experimental Results

(cp = 0) North Colo South Colo
Top of Shear (lbs.) 0 0 0

Column Compression (lbs.) 4,130 27~400 21~100
Moment (in.-lbs.) 248~000 142s200 122~200
Base of Shear (lbs.) 0 1~520 19520

Column Compression (lbs.) 4~130 5~850 1~955
Note (2) Moment (in.-lbso) 248~000 / 360~000 318~000
Note (3) 248~000 249~000 220~000
Tie Bar Tension (lbs.) 6~880 5~360 5~360
Notes~
(1) Maximum load = 41.3 psf.
(2) Assume an uncracked section.
(3) Assume a cracked sectiono
(4) See Figure 36 for location of N.A. for eccentric loading.
The maximum axial load per column during the eccentric

129
loading was N = 4,130 lbs. This load was accompanied by a

calculated horizontal thrust at the top of the column of
41
p = w2~~·b = ·3 2 (~0~0) = 6,880 lbs., which was resisted
by the tie bar, and by the column and haunch joint in
bending. From the test data, the average stress in the
tie bar was 5,360 lbs., which left 1,520 lbs. to be re=
sisted by the column in shear and bending.
The orientation of the N.A., the axis along which
strain is zero, Figure 36, shows the influence of the
bending moments in the direction, ~ = O. If the tie bar
resisted all of the horizontal thrust, the orientation of
the N.A. would probably be in the~ = 90° direction.
Figure 36(a) shows that the N.A. has shifted far enough
over from the column center to cause tension in gage 7.
The south column, Figure 36(b) shows that tension existed
in gages 17 and 19, which are on opposite faces of the
column.
The complete strain relationship at the maximum
eccentric load condition is illustrated by the three dimen~
sional sketches in Figure 36. Because bending moments

occurred in two directions, the resultant N.A. was located
by plotting the known values of strain, which showed that
the N.A. was skewed in the same direction for both
columns.
The deflection data are presented in Figure 37 and
Figure 38. The data for Test II were used to illustrate
the uniform deformation of the roof. The deflections of
130
the roof during Test IV were influenced by the residual

strain remaining in the structure after Test III had been
completedo The presence of strain is clearly apparent in
the final unloaded condition at the conclusion of Test
III, Figure 38(a)o This may indicate that the column had
exceeded the elastic limit during heavy bending under the
eccentric loading and could not return to its normal
state. Although residual strain was recorded by the
column gages after unloading~ part of the roof deformation
may have been due to a slight yielding of the soil around
the footings and wingwalls and tilting of the structure in
the + cp = 0 directiono
The values of stress versus load and time for Test IV

are shown in Figure 390 Values for Test III were not
shown as there were only two load increments and no sus-
tained loading.
e11 :-240
e'11 :-\25
Ad\. Value :
(al soutb Co1""" (bl North Co1umD
Figure 36. Neutral Axis Locations in Columns

Under Cantilever Loads.
o S. E. Corner
.6 ll E. ,Mid - Span
.5 o N. E. Corner
• N. W. Corner
.4 • W. Mid-Span
• S. W. Corner
en
Q)
.3
.s=
u
c: .2
c: .I
-
0
-
u
Q) 0
Q)
c -.I
-.2
-.3
0 10 20 30 40 50 60 0 20 40 60 80 100 120
Uniformly Distributed Load - Lb./Sq. Ft. Time - (Sustained Max. Load) Hrs.
Figure 37. Roof and Column Deflection Curves
For Test II.
0
S. E. Corner
t:. E. Mid-Span
o N. E. Corner
• N. W. Corner
"" W. Mid-Span
• S. W. Corner
0
S. E. Corner
.6 t:. E. Mid-Span
a N. E. Corner
.5 • N. W. Corner
... W. Mid- Span
.4
• S. W. Corn er
A----~
A-----,
.3
"'
Q)
..t::.
u 0 .2
-= -I .I
c:
0 "'
Q)
..t::.
t; -2 u 0
-
Q) c:
Q)
Cl
-3 c: -.I
.9
u -.2
-5 -
.!!
Q)
Cl -:3
-4
. 0
0 10 20 30 40 50 10 20 30 40 50 60 0 20 40 60
Eccentric Load - Lb. I Sq. Ft. Uniformly Distributed Load - Lb./ Sq. Ft. Time -(Sustained Mox. Load) Hrs.
(a) Test III Data (b) Test IV Data
Figure 38. Roof and Column Deflection Curves for

Test III and IV.
Gage No. (See Fig. 291
I a Gage No. (See Fig. 291
2 0 13 A
3 5 D 14 o.
6 • 4 17 a\
c 2 18 ...
c- I
c: 3 19 •
"'
....... & 20
.d
.....I 0 "'
....... 2
0 .d
!:2 -I .....I I
0
.; ~ 0
"'~ -2
en -3 "'"' -I
~
0 20 40 60 en -2 0 10 20 30 40· 50 0 20 40 60
0 10 20 30 40 50
Uniformly Distributed Load, Lb./ Sq. Ft. Time ( Sustoined Max. Lood= Uniformly Distributed Load, Lb./Sq. Ft. Time (Sustained Max. Loo d =
5 7 Lb.I Sq. Ft.I Hrs. 57 Lb./ Sq. Ft.I Hrs.
Gage No. (See Fig. 291

3 A
4 0
c: 3 7 a -14
~ 2 8 ... -13
:ci 9 •
.....I 10 • -12
~ o~~~~~~ -11
Gage No. (See Fig. 29 I
21 A
"'
"'~ -IL-~_l..~~..L-~-'-~~--'-~--'~~....._~----''--~~~~~ 22
-10
ii) 0 10 20 30 40 50 0 20 40 60 23
Time (Sustained Mox. Load= -9 24 a
Uniformly Distributed Load, Lb./ Sq. Ft.
57 Lb./Sq. Ft.I Hrs. -8
25 •
c -7
.,;.
-6
Gage No. (See Fig. 29 I "'
.......
II A "".,; -5
12 "' -4
~
15 .a ;;;
16 -3
-2
-I
60 0 20 40 60
Time (Sustained Max. Load= Uniformly Distributed Load, Lb. I Sq. Ft. Time (Sustained Max. Load=
Uniformly Distributed Load, Lb./Sq. Ft.
57 Lb. /Sq. Ft. I Hrs. 57 Lb. I Sq. Ft.) Hrs.
Figure 39. Stress Versus Load and Time Curves, Test IV.
CHAPTER IX
DISCUSSION OF RESULTS
The analysis of test results and a set of construe~
tion skill factors will be discussed in this chapter. The

method of erection of the prototype structure used in this
study will be examined and an erection procedure based on
the research experience from this study will be recommended
for use in the construction of h-p shells.
Assembly Components and Techniques
The discussion of assembly techniques includes the

initial construction of forms and apparatus necessary for
the assembly of the structure as well as the actual erec-
tion of the structural elements.
Column Forms
From the observations made during this study~ the

Douglas fir material used for the column forms would not
be satisfactory for multiple reuse if extensive reuse was
planned. After a period of approximately two weeks, the
first 10 days of which the forms were constantly soaked,
the side f orm.s were warped to the extent that bracing or
clamps would be necessary for reuse. It should be noted
i'35·
136
that under ordinary conditions, the forms would probably

be removed after two to three days of curing; even then,
this material, due to its non-homogeneous nature~ would
tend to warp unless well braced. A material which would
provide adequate stiffness and strength for continuous
reuse would be a 5 ply exterior grade or marine plywood.
By using non-corroding hinges between the base and the
sides of the form, and braces or stiffeners across the top
face of the form, the column could be easily removed and
the form could be quickly_prepared for casting the next
columns. By using a form which could be removed and pre-
pared quickly, labor cost for forming could be reduced.
The haunches cast at the top of the column were devel-
oped for h-p structures consisting of only one unit with
no walls or supports. This feature could be eliminated for
structures composed of two or more shell units in which
overturning moments were not acting upon the structure.
However, the haunches provided a greater surface area for
ease during the roof assembly. Eliminating the haunches
would affect a savings in labor and equipment due to the
large amount of special forming required by heating and
bending the steel.
Even though the column may not be subjected to loads
and moments as great as those applied in this study~ the
designer should consider transporting, lifting~ and assem-
bly loads which the column may be subjected to before it
has been erected, as well as loads which it may receive
137
during the. roof assembly or before the individual struc-

ture is completed.
Shell Forming
The shell forms for this study were revised from a

set used to cast an inverted umbrella shell at the Agri-
cultural Engineering Laboratory in 1962. These form· sur-
faces, which were used the second time during this study
and had been stored out of doors, were showing signs of
weathering. Covering the surface with a plastic coating,
after the original surface was not usable, would permit
additional uses to be obtained inexpensively. The metal
base of these forms provided a rigid framework to keep the
shell surface in its original shape and would stand the
abuse of being transported to worksites for on-site cast-
ing. Due to the symmetry of the h-p shell, a minimum of
two forms could be used if casting was done year-round and
production demands were not excessiveo Thus, material and
labor costs of construction would be reduced.
Forming the shell steel for precasting required
greater precision and more material than would be required
in a cast-in-place shell. Lower design loads would allow
the shell steel to be spaced wider. This would require
fewer interior edge beam dowels to overlap with the shell
steel. The reduction in the number of dowels used plus
reducing the length of weld on the dowel base to approxi-
mately one inch per side of the dowel would reduce the
138
warping problem which was encountered in this study. The

shell steel forming could be further simplified by leaving
out the reinforcing mat at the corner of the shell, used to
resist local bending and radial stresses during eccentric
loading. This mat would not be necessary for a shell
loaded uniformly or connected to another shell or wall.
The shell casting operation would have been simpli-
fied by using steel chairs to support the shell steel in-
stead of the three-fourth inch wood blocks. Several of
these blocks were not removed during the casting opera-
tions. This could be critical in a building where several
shell units were connected, and waterproofing and drainage
were necessary. The concrete screeding operation was dif-
ficult in the area around the lift rings, but handworking
around them was satisfactory. The lifting rings were
easily installed and both the structural and functional
design seemed to work satisfactorily.
At first, the low corner of each quadrant did not fit
properly where it was seated on top of each column. This
was detected during the test lift of the first quadrant
onto the supports G The bottom tip of each corner had to be
removed from each quadrant, which caused a delay of the
assembly of approximately two man-hours. This method of
connecting the quadrant to the column should be considered
when designing the form surface in this area of the shell
forms. A flat area could be formed easily by placing a
wooden wedge in the low corner of each quadrant form.
139
Footing Methods
The footing used in this study performed satisfactor-
ily, however, by not knowing the exact soil shear strength
and bearing capacity, the footing may have been overdesign-
ed both in regard to size of wing walls and the depth of
the footing. The footing reinforcement cage used in this
study functioned we1'1 in the assembly of the column but
required special bending during construction. This prob-
ably could be eliminated by another type of footing. For
a smaller design load and no overturning moments, a cylin-
drical footing with sufficient bearing area would be
adequate.
Temporary Support System
The rigid assembly frame performed satisfactorily dur-

ing the shell assembly process. For a larger shell, such as
a 40 foot square structure, the corner towers would have to
be braced so that the l/r ratio was less than the ratio used
in this study and the possibility of buckling was reduced.
The horizontal braces would have to be supported between the
two towers. A metal or wooden post could serve the purpose
of supporting the horizontal braces and also support the
corners of the two roof quadrants; thus, the horizontal
braces would reduce the unsupported length of the midspan
supports.
The initial positions of the corner towers were not
marked during the assembly of the rigid frame. If these
positions had been established and marked and the towers
140
positioned over the marks accordingly, time would have been

saved during the initial frame assembly and much of the
final frame adjustment.
The wooden center support was effective. However, im-
provements could be made in the bracing method used to keep
the structure centered. Some methods which could be used
. are: ( 1). A metal. stake driven into the ground with a metal
brace from the tower to the stake. Bolt connections on
each end of the brace with slotted adjustment holes for the
lower end would provide the necessary adjustment. (2) A
wooden member with a metal bracket bolted to the lower end

to resist the wear of making repetitious connections. (3)
A metal or wooden brace with a steel loop connected to the
lower end to receive a steel stake. (4) A steel or wooden
brace with a heavy-duty turnbuckle fixed rigidly to the
lower end of the brace, adjusted to position or plumb the
support. The particular method used would depend upon the
amount of intended use and the relative cost of the alter~
native methods of bracing.
Lift Frame
The lift frame configuration used in this study would

operate effectively on larger shells providing the unsup-
ported length of the diagonal brace was not excessive. The
addition of a second diagonal brace connected to the orig-
inal brace at the center would make the frame more rigid
and allow the use of materials of approximately the same
dimensions that were used in this frame.
141
By using a three point arrangement of lift rings, a

triangular shaped frame would work satisfactorily providing
a standardized system of cables or chains was devised to
complete the system. This configuration would lend itself
well to a bolted frame assembly which could be assembled
and dismounted ~apidly.
A second alternative frame would be a simple "I 11 beam

with a clevis on each end for the sling attachment. Bal-
ance of the quadrant could be maintained by two clamps on
each edge of the shell parallel with the beam, connected to
the center lift ring of the sling by small cables.
Column Erection
The column erection was costly in labor requirements

as the complete erection required nine man~hours per column
and was completed by a four man crew plus a crane operator.
For structures with no heavy bending, a system of leveling
bolts mounted on the base of the precast column would pro-
vide a satisfactory means of erecting and plumbing the
column. This would require that the footings be located
precisely before casting. After the column was plumbed by
the bolts, the bolts would be welded, making the reinforce-
ment continuous, and the joint would be completed by an
expanding grout pack.
Other methods of providing rapid erection of precast
columns by construction joints are presented by Rensaa (35),
who used a precast footing socket; Naslund (15)~ suggests
(1) a baseplate connection, (2) reinforcing bars from the
142
column which extend into holes in the footing which are pre-
viously filled with grout, and (3) a slotted bar, cast into
the footing and column, which is welded and the joint groutedo
Cogan (36) developed an effective pipe connection by pre-
casting into the footing a 4 inch pipe sleeve which was cut
off at the correct elevation; the 4 inch pipe sleeve fit
over a 3 inch pipe insert cast into the column. The insert
had a steel ring or shoulder welded around it to give the
exact elevation. The reinforcing bars were overlapped be-
tween the footing and column, and welded. The joint was
completed by grouting with an expanding grout mix which pre~
stressed the column reinforcement, thus giving a highly ef-

ficient joint.
Weld,ed Connections
Four weld connections were made during the erection of

the shell; these were: (f) the tie bar connection to column,
(2) column and haunch to shell connection, (3) tension bar
welded at center of roof, and (4) the interior edge beamso
The haunch to shell connection would be eliminated in
a shell which was not designed for overturning moments or
for a column with a different method of resisting momentso
The connection on this shell worked adequately during ec-
centric loading. The column to shell connection was made
quickly and efficiently by welding the interior edge beam
angles and the web of the tie connector together. No
change would be recommended for this erection step.
143'
The tie connections performed satisfactorily in the

structural sense but was not satisfactory functionally.
During the erection of the roof quadrants, the tie bar pro-
truded too far back along the web of the inverted "T 11 sec=
tion, thus keeping the quadrants from seating properly;
this caused approximately 3 man-hours delay while the quad=
rant corners were adjusted. This situation could be avoided
by shaping the quadrant corners to compensate for the tie
bar, or by a different method of attaching the tie baro One
method of adjusting the tie bar connection used in this
study would be to use two 1 inch wide by 3/4 inch thick bars~
welded on both sides of the web to the flange of the tie

connection, with the 1 inch side placed horizontally. The
tie bar could be placed between the two bars and fillet
welded. A second method would be to notch the web of the
11
T" section from the flange up to a height equal to the
tie bar diameter, and approximately 3 to 4 inches back
from the end of the flange; the tie bar could be inserted
and welded to the web on the top of the bar and to the
flange on both sides of the bar at the base. Either method
would have performed better functionally than the method
which was used· in this study.
0
Both the tension bar at the roof center and the edge
beams should have been welded in sections or strips. This
part of the shell design should have been examined more
critically. The factor of safety of the weld on the ten-
sion bar was 5.0; this indicates that the welding should
have been reduced to half of the amount used.
144
The edge beams could have been welded securely at the

horizontal ends and at spaced intervals along the length.
This would have reduced the welding time on the shell by at
. '
least one-half and saved on welding.
Recommended Shell Erection Procedure
The procedure for the erection of a two-column h-p

shell of the c.onfiguration in Figure 2 could be carried
out in the step-by-step procedure outlined below:
(1) Prepare site by clearing, leveling, construction
staking, and excavating footings.
(2) Precast columns.
(3) Precast shell quadrants.
(4) Cast footings, if constructed separately.
(5) Move structural elements and construction ap-
paratus to site and place in prescribed posi-
tions, Figure 23, (This should be accomplished
while precast footings are curing for the first
or second day.)
(6) Erect column and complete tie connection.
(7) Assemble and align the support system while
columns are being erected, or while the footing
or column construction joints are curing.
(8) Assemble the shell quadrants on the support
system, adjust roof elevation, and pull quad~
rants together for welding.
(9) Connect quadrants to top of column by welded
connections and weld tension bar at center of
roof.
(10) Remove support system after tension bar at
center of roof is welded and shell to column
connection has been completed.
(11) Complete welding of edge beams.
(12) Waterproof steel edge beams.
145
This list of steps constitutes a procedure which can
be utilized on one shell, or can be modified for use in
erecting multiple shell structures; however, the erection
of a number of shells to form a continuous structure is
beyond the scope of this study.
DISCUSSION OF COST ANALYSIS
The observed cost data from this study have small sig=
nificance in its present form; however, if this data can be
adjusted by appropriate estimates based on the experience
gained from this study, the adjusted data may serve as a
useful guide for construction estimates on this type of shell.
The next four sub-sections will consider the data by
(1) adjusting the material cost where appropriate, (2) ad-
justing the labor cost data by an appropriate skill factor
based on the experience that a crew would have after becom-
ing familiar with the construction routine, (3) adjusting
the equipment costs that are related to the labor and mater-
ial reductions, and (4) converting the values to a cost per
square foot for the shell used in this study.
The final sub-section will be used to adjust the pro-
totype data to a 40 foot square shell. Only the variable
cost factors will be considered, Interest rates will not
be considered in this study; the cost data will be consid-
ered as capital costs.
Material Cost Adjustments
By designing the columns by elastic analysis for a

concentric load or small bending loads, the amount of st~el
146
reinforcement and the column size could be reduced. Table XI

illustrates the reduction in material for a 20 foot square
shell, which would result by designing for axial loads onlyo
TABLE XI
MATERIAL,.SAVINGS BY CONCENTRIC COLUMN DESIGN
Item of Material Qµantity Unit Cost Cost
No. 10 bar (Reduction due to

No. 9 bar used for tie) 29.6 lbs. $.097/lb. $2.88
No. 8 bar (Column steel reduc-

tion to 4-No. 7 bars) '26'r.7 lbs. .097/lb. 25.78
No. 6 bar (Haunch dowels) 24.o lbs. e097/lbo 2.33

No. 5 bar (Reinforcing mats
and footing steel) 83.5 lbs. .097/lb. 8.12
3 in. :im: 5 in. x 3/8 in. angle

(Haunch-shell connector) 78.4 lbs. e097/lbo 7.62
2 in. x 2 in. x 3/8 in. angle 37.6 lbs. .097/lb. 3.66

Welding rod 15.0 lbs. .20/lb. 3.00
Acetylene (Heating and bending) 1/4 bottle 5.70/bottle lo42
Oxygen (Heating and bending) 1/4 bottle 5.65/bottle lo4l
TOTAL $56.22
From Table XI, the savings that could be realized be=

tween a structure which was subjected to overturning moments
and one which was concentrically supported was evident. The
147
savings of $50.39 on steel comprises approximately 25 per-

cent of the shell steel costs. The concrete which would be
saved on the haunch arms would probably be used to enlarge
the footing diameter from 20 inches to 26 inches in order
to provide adequate bearing area.
The reduction in equiyment costs could be r~lated to
both the efficiency developed on each job and to the re-
duction in materials by a change in the support system de-
sign.
Another aspect of material costs is the cost for all
shells after the construction of the first structure. By
assuming that forms, the support system, lifting frame, and
column cribbing will be reused indefinitely, the direct
material cost for each future shell can be estimated from
Table III. Table XII lists the costs of materials required
for a 20 foot square h-p shell for (1) eccentric loading,
and (2) for concentric loading. The adjustments which
were made in the eccentric load costs are for deletion of
costs for the erection apparatus. The adjustments in the
concentric load cost include the deletion of costs of erec-
tion apparatus plus the items listed in Table XI. Thus, a
savings in material costs of $56.18 would be made per 20
foot square shell by designing for concentric loading.
Labor .Q..Q..§1 Adjustment
The labor costs which were tabulated in Table I are

not usable except for estimating the time requirements
of construction on a 20 foot square h-p shell with an
TABLE XII
ADJUSTED MATERIAL COSTS FOR TWENTY FOOT SQUARE H-P SHELL
Item Quantity Eccentric Concentric

Eccentric Concentric Cost Cost
1. Welding Material
(a) Welding rod 40 lbs., $.20/lb. 25 lbs., $.20/lb. $8.oo $5.00
(b) Acetylene 7/8 bottle, $5.70 5/8 bottle, $5.70
per bottle per bottle 4.98 3.56
(c) Oxygen 7/8 bottle, $5.65 5/8 bottle, $5.65
per bottle per bottle 4.94 3.53
2. Concrete
(a) Standard weight, 3 cu. yd., $14.75 3 cu. yd., $14.75
3,000 psi per cu. yd. per cu. yd. 44.25 44.25
(b) Lightweight Ag- 3 1/2 cu. yd.' 3 1/2 cu. yd.,
regate, 3,750 $18.25/cu. yd. $18.25/cu. yd. 63.87 63.87
psi
3. Steel Material 2056 lbs., $.097/lb. 1538 lbs., $.097/lb. 199.84 149.49
4. Form Oil 5 gal., So80/gal. 5 gal., $.80/gal. 4.oo 4.oo
TOTAL $329.88 $273-70
149
unfamiliar crew. Table I was examined to arrive at appropri-

ate skill factors for each type of operation which could be
repeated.
The quality of labor used in the study was excellent,
considering the attitude, education, and previous work ex~
perience. The productive working time per hour was esti-

mated at 45 minutes per hour over all project operations.
According to Dallavia (37) this would represent a working
efficie~cy of 75 percent. Table XIII lists the opergtions
with estimated skill factors based on expected construction
efficiency during construction of the second h-p shell of
this type.
TABLE XIII
LABOR ADJUSTMENTS BY SKILL FACTORS
Operation $J;c.;illed Unskilled Adjusted Adjusted

Labor Labor Skilled Unskilled
Factor Factor Labor Labor
i; Column Construction
(a) Frame Assembly .50 .85 1.0 30.0

(b) Steel Forming .50 .80 2.5 43.2
(a:) Casting and
Curing .50 .75 2.0 2.0
SUBTOTAL (Man-hours) 5.5 82.2
,,
2; Shell Construction
(a) Form Assembly .25 .80 4.o 76.8
(b) Shell Steel
Forming .60 .80 4.8 62.4
150
TABLE XIII (Continued)
Operation Skilled Unskilled Adjusted Adjusted

2. (Continued)
(c) Form Preparation
and Shell
Casting .60 .Bo 1.2 1208
(d) Curing .4o .75 1.2 7o5
SUBTOTAL (Man-hours) 12.4 159·5
3. Support System Construction
(a) Fabricating Parts .50 .Bo Oo5 5.6
(b) Welding Tower
Frames .Bo 2B.B
(c) Assembly of Bolted
Components .B5 2.5
(d) Wooden Support
Fabrication .40 .75 o.B 12.0
(e) Final Adjustments
on Steel Supports .50 .50 0.5 2.0
4. Lift Frame Construction .70 4.2
5. Site Preparation
(a) Leveling and
Smoothing .90 1.8
(b) Survey and Layout .75 .75 1.5 1.5
(c) Foundation
Excavation .50 .50 1.0 4•.z
6. Site Layout
(av Hauling and Plac-
ing Columns .90 .90 l.B 1.8
(b) Wooden Support for
Shell Transport .50 .75 1.0 3.0
(c) Removing Shell Forms
Transporting, and
Placing Shells .40 .50 1.6 3.5
151
TABLE XIII (Continued)
Operation Skilled Unskilled Adjusted Adjusted

6. (Continued)
(d) Transporting and

Placing Supports .50 .75 0.2 2.0
7o Column Erection
(a) Construction of
Cribbing for
Column .50 .75 2.0 18.8
(b) Cutting and Bending
Footing'Steel .75 3.0
(c) Column Erection .60 .75 2.4 10.5
(''.~~(:;). (d) Casting f:ooting .85 o.8
(e) Removal of Braces
and Site
Cleanup 085 3.4
(f) Tie Er~yction .80 .80 1.6 1.6
SUBTOTAL ( Man-hours) 6.o ·38.1
8. Support System Erection
(a) Initial Assembly .50 .50 1.5 4.5
(b) Final Alignment .50 .50 2.0 6.o
9. Erection of Shell
(a) Initial Assembly .40 .40 1.6 4.8
(b) Adjustment for
Welding .50 .50 1.0 2.0
(c) Welding .50 7°5
(d) Support Removal .70 4 .. 2
(e) Grouting Haunches .85 3 .. 4
(f) Waterproofing In-
terior Edge
Beams .50 .,75 0.5 3.8
(g) Final Clean-up .. 75 3 .. 0
SUBTOTAL (Man-hours) 10.6 ?l.2
FINAL TOTAL (Man::..hours) 86::.·ra 352~7
-~-------- -=
152
By adjusting the observed time data for the first con-

struction cycle, the supervision and skilled labor cost was
reduced to 57.1 percent of the initial supervision labor
cost. The adjusted· man.;..hours-for the. unakfi'.'Il:T'etl·lab•or catego-
ries was74.8 percent of the initial unskilled labor cost.
The total adjusted man-hours for both labor categories was
70.7 percent of the combined man-hours of the initial con-
struction phase. However, the final totals listed in Table
XIII reflect datC3.,. which would be reproduced only periodical-
;.';';~ '
ly; several operations which are shown would not be repeated

in construc·t:ing the second shell. The items which would be
deleted from Table XIII would be items l(a), 2(a), 3, 4, 6(b),
and 7(a). ·Thus, the revised total man-hours for supervision
was 37.4 man-hours, while the total for unskilled labor was
202.0 man-hours. Comparing these values to the first data
with the same sections dele.ted, the second combined man-
hour total of 239.4 was 67.9 percent of the initial adjust-
ed total. This total was only 39.1 percent of the initial
phase combined total of 612 man~hours. Figure 40 shows the
projected trends of the percentage of first unit man-hours
under three conditions. For reuse of equipment, the repet~
itive operations are considered in Curve C. This is the

estimated true situation; the trend of the percent of first
unit labor would probably level out at around 25 percent of
the FIRST UNIT values due to the high man-hour totals re-
quired to build the forms, lift frame, column cribbing,
and support framework.
153
The estimated cost of labor for the second shell con-

structed would be $120~19 for supervision and $446.90 for
unskilled labor, which would give a labor total cost of
$567~09~ This would be 3g.7 percent of the total cost. If
the direct labor cost on this size of shell could be reduc-
ed to 25 percent of the original first unit cost, the pro-
jected cost would be $367~00 which would be $.92/ft. 2 for
labor. The labor cost for erection equipment which were
constructed for multiple use, items l(a), 2(a), J, 4, 6(b),
and 7(a) in Table I, was $629e00~ This cost is reduced for
each reuse of the equipment by a proportional amount~ For
example, if a 200 foot by 80 foot warehouse were built using
20 foot sqt1are h-p shells, 40 h-p units would be required;
thus, assuming all equipment was used 40 times, the equiv-
alent cost per use would be $15.72, or $80393/ft~ 2
By considering a concentric load design, a further
reduction in labor cost could be made by simplifying the
column forming and footing, removing the shell angle used
for the haunch connecter and removing the reinforcing mat
in the corner of the shell. The labor reductions which
this would create would be 5 man-hours for supervision
and 5$ man-hours for unskilled labor, Table XIV. These
reduced items were taken from Table I. The labor savings
obtained by the concentric design would be $142.09. This
would reduce the estimated second unit labor cost from
$567~09 to $425.00, which is 28.9 percent of the original
labor cost, $1467.16.
LEGEND
Symbol Item
100
----e A. First Unit Total Labor Requirements
Adjusted by Skill Factors
0 0 B. First Unit Shell Construction Labor
Requirements Adjusted by Skill Factors
80
C. Percentage Item B of First Unit Total
en
.... Labor Requirements Unadjusted
::I
0 68.9
..c
I
c:
0
60
:E
- c:
- -- -- . -----·-- ----
-....
::::>
en 40
--
0
c:
Q.>
u.... 20
Q.>
a..
Number Units Constructed

Figure 40. Labor Requirements Adjusted by
Three Methods for Comparison.
155
TAJB:LE XIV
LABOR REDUCTIONS DUE TO CONCENTRIC LOAD DESIGN
Items Reduced Supervision Labor Unskilled Labor
lo Column Construction
(a} Forms 1 10
(b) Steel forming 2 JlO
(b) Steel forming 2 10
7o Column Erection
( b) Cutting and bend-
ing footing
steel 0 .4
9o. Erection of Structure
~e) Welding
d) Grouting haunches
0 0
0 --1±_
TOTA:L· (Man-hburs) 5 58
Equipment Cost Adjustments
The adjustment in equipment costs was dependent upon

the adjusted labor and material costs., The equipment usage
costs were adjusted on the basis of (1) eccentric design
with labor adjusted for skill factors and reuse of erection
apparatus, and (2) concentric design with labo~ adjusted
and reuse of erection apparatuso The results of the equip=
ment cost: adjustments.are tabulated in Table XV for the
I
second shell unito

156
T'.ABLE XV
ADJUSTED EQUIPMENT COSTS FOR SECOND SHELL UNIT
ITEM HOURS COST'

Eccentric Concentric Eccentl...ic Concentric
Design Design Design Design
lo Acetylene Welder,
$3.00/hr.
(a) Forming column
·steel ,,,,.10.4 0 $31•20 0
(b) Forming shell
steel "~ 4.8 4.o 14.40 $12.00
2o Electric Welder
(a) Forming column
steel, $2.00/hr. r~:". 2.4 " 1.0 0 ,.4 ;80 2.00
(b) Tie bar erection,
$3.00/hr. ~ 1.6 ' 1.6 0 4.80 4.80
(c) Forming shell steel,
(1) Shop welding,
$2.00/hr. ,,11.2 .. 8.o .22.40 16.oo
(2) Field welding,
$3.00/hr. 3.2 2.7 9.60 8.10
(d) Footing cage, .. e
$2.00/hr. o.8 0 1.60 0

(e) Portable welding on
erected shell,
$3.00/hr. 7.5 7.5 22.50 22.50
3o Tractor and Equipment
Trailer, $2.50/hr. 6.o 6.o 15.00 15.00
4. Tractor and Lift Arm,
$2.50/hr .. 2.0 2.0 5.00 5.,00
5 .. Fork Truck, $3.00/hr. 2.0 2.0 6.oo 6.oo
6. Crane, with Operator,
$6.00/hr. 4.6 4.6 27.60 27.60
7. Tractor Dozer for'Site ,. . "
Leveling; $6.00/hn. 1.8 1.8 10.80 10.80
8. Rotary Drill Rig,
$12.50/hr. 1.0 leO 12.50 12.50
9. Power~HacRsaw, $2.00/hr. 3;5 2;5 7;00 ~~~00
TOTAL $191;20 $137.30
157
From Table XV, an expect<?d savings of $187.80 could be

expected between the first and second units prov{ding the
equipment costs of the erection apparatus was included in
the first unit cost for this comparison. A total equip-
ment cost of $214.70 would be saved by constructing the
second unit by concentric design compared to the eccentric
design of the first shell. The difference of $53.90 be-
tween the costs in Table XV is attributed to the change in
column and shell reinforcing with labor adjusted.
Shell Costs Per Sguare Foot of Horizontal Projection
By combining the data for labor, equipment, and mat-

erials for constructing a shell designed for eccentric
loading with the costs of the erection apparatus adjusted
for the number of uses, a cost of construction per square
foot of horizontal roof surf ace can be obtained.
Assuming that the shell and column forms were used
at least 10 times without repair or replacement of sur-
faces, the cost for the tenth shell could be estimated.
From the original data, the costs of labor, materials,
and equipment required to construct the support system,
lifting frame, column cribbing, and forms was $644.20 for
labor, $101 .. 00 for equipment, and $556.07 for material;
thus, the erection equipment total cost was $1,301.27.
Pro-rating the total cost over 10 uses would give $130.13
per unit.
Figure 40 shows expected trends in labor man-hours.
158
requirements of future units based on increasing skills

and job efficiency during the first few units constructed.
The total labor requirement was described as First Unit
Total Labor Requirements. This labor total included all
labor used in fabricating the assembly components such as
forms, steel and wooden supports, column support cribbing,
and lifting frame, plus .the precasting and erection of the
structural elements. These labor requirements were
adjusted by construction skill factors in Curve A.
The First Unit Total Labor Reguirements were divided
into two categories; these were labor for: (1) fabrica-
tion of casting and erection equipment (forms, column and
shell erection supports, and lifting frame), (2) Column
and shell casting, on-site assembly of supporting systems,
and column and shell erection.
Curve B shows the labor requirements of item (2), in
the previous paragraph, adjusted for increasing skillo
Curve C shows item (2) adjusted for skill as a percentage
of the First ~ Total Labor Requirements, unadjusted.
From Figure 40, the labor value could be estimated
from Curve B to approach 50 per cent or slightly below
after the 7th or 8th unit constructed. Subtracting the
labor cost for forms and erection equipment, $644.20, from
$1,467.00, the original cost, and multiplying by 0.50
gives a projected labor cost of $411.40. The equipment
costs would be proportional to the adjusted labor, except
for the rotary drill truck; the adjusted equipment costs
159
were $153.25. The equipment costs due to the construction

of the shell would be the original equipment cost, $379.00,
minus the cost due to forms and erection equipment,
$101.00 = $278.00 for the construction of the first shell.
The material cost was $881.59 - $556.07 = $325.52.
The total cost for the tenth shell would be $411.40
for labor, $153.25 for equipment, $325.52 for material,
plus $130.13 for each use of forms and equipment =
$1,020.30. This gives a cost of $2.55 per square foot of
horizontal projection, which is 37 per cent of the orig-
inal cost for the first shell. Further uses of the erec-
tion apparatus could be readily projected as the data for
the time~ equipment, and material will not be expected to
change appreciably.
Estimated Variable Costs for Forty Foot Square Shell
The material costs of a forty foot square shell could

be considered to be directly proportional to the cost for
material of a prototype of the same characteristics. The
labor, material, and equipment costs, however, would vary
from one size of shell to another. Table XVI lists esti-
mated labor costs for a 40 foot square shell with a six
foot rise and a column height of 10 feet.
The adjusted supervision labor averaged 1.21 more
for the 40 square shell than the prototype, while the
adjusted unskilled labor was 1.298 greater. The total
adjusted labor cost for shell production was $242.58 plus
160
$712. 00 = $954. 58. These cbsts do not include cost of

construction of erection appar~tus. The adjusted labor
values for a 40 foot square shell were .. projected over sev-
eral units constructed, Figure 41 . The 100 per cent value
represents the total of Table '.XVI .
The values for steel forming w~re estimated ,, to take
approximately the same time .for a 40 : ;foot square i. Shell as
a 29 foot square shell. The steel would be ,p·l aced in .•: ..:,~
longer lengths. However, the forming and tieing .,W01fld re-
quire more time due to the larger number of :. junctions in
\
the shell steel of the 40 foot square shell.
TABLE XVI
PROJECTED LABOR ESTIMATES FOR FORTY FOOT SQUARE H- P SHELL
Operation Labor for 20 ft. Adjustment Adjusted Labor

x 20 'ft. ··shell Factor for 40 ft. x
40 ft. Shell
Super. Unsk. Super . Unaj<. ,· . Super. Unsk.
1. Column Construction
(a) Steel Forming 5 54 1.0 1.0 5 54
(b) Casting and :
Curing 4 12 1.2 1.4 4.8 16.8
(a) Steel Form-
ing and 8 78 1.1 1.3 8.8 101.4
Tieing
(b) Form Prepa-
ration and
Shell
Casting 4 16 1.0 2.5 4.o 40.0
(c) Curing 3 10 1.0 1.5 3.0 15.0
161
TABLE XVI (Continued)
Operation Labor for 20 ft. Adjustment Adjusted Labor

x 20 ft. Shell Factor for 40 ft. x
40 ft. Shell
Super. Unsk. Super. Unsk. Super. Unsk.
3. Site Preparation
(a) Leveling and
Smoothing 0 2 1.5 0 3.0
(b) Survey and
Layout 2 2 1.3 1.3 2.6 2.6
(c) Foundation
Excavation 2 9 2.0 2.0 4.o 18.o
4. Site Layout
(a) Hauling and
Placing
Columns 2 2 1.0 1.0 2.0 2.0
(b) Removing
Forms, Loading,
and Transport-
ing Shells to
Site 4 7 1.2 lo2 4.8 804
(c) Moving Support
System 1 4 1.1 1.1 1.1 4.4
5. E+ecti9n of·S~pports 7 21 1.1 1.1 7.7 23.1
('.;
Erection of Structure ..
6.
(a) Initial
Assembly 4 :12 1.0 1.0 4.o 12.0
(b) Preparation for
Welding 2 '.4 1.1 1.1 2.2 4.4
(c) Welding Time 15 0 1.5 22.5
(d) Support Removal
and Site
Cleanup 10 1.1 11.0
(e) Grouting
Haunches 4 1.0 4.o
(f) Waterproofing
Edge Beams 1 5 1.0 1.3 1.0 6.5
TOTAL (Man-hours) 64 252 775 3~6.6
The shell form costs were adjusted by determining the

LEGEND
Symbol Item
oo~~~o A. Labor Costs Adjusted for Job Experience
100
A .6. B. Shell and Column Form Costs Adjusted Over
Number of Shells Constructed
t:s------6 C. Combined Labor and Form Costs (Labor ad-
justed for experience, forms adjusted
80 for use)
- --- D. Erection Apparatus Cost Adjusted for
Number of Units Erected
c:
60
-,,,
:::>
....
0 40
Q)
-
C"
0
c:
Q)
u
~
a... 20
---·-- ----·------·- -------

3 4 5 6 7 8 9 10 II 12 13 14 15 16 17 18
Number Of Shell Units Constructed
Figure 41. Variation of Labor and Erection Apparatus Costs
With Increased Number of Shells Constructed.
163
amount of increase in material costs and adjusting the

labor costs of the 20 foot square shell forms. The column
forms were not adjusted as the same column size and shape
could be used for the larger shell. Table XV lists the
pertinent variable costs for the 40 ft. x 40 ft. shell.
These values are used in Figure 41 to show the cost varia-
tion adjusted for experience and use. By adjusting the
material costs in direct proportion to the prototype and
increasing the equipment costs, the costs of constructing
a 40 foot square shell can be readily estimated.
TABLE XVII
VARIABLE COSTS FOR FORTY FOOT SQUARE SHELL
Item Cost
1. Total Form Costs

(a) Form Material $528.00
(b) Form Labor 501.00
2. Labor Cost for First Shell Constructed 954.58
3. Total Cost of Refinishing Column and
Shell Forms 531.50
4. Labor Costs for Constructing Erection
Apparatus 291.50
5. Material Costs for Erection Apparatus 388.00
6. Equipment Costs for Erection Apparatus 108.00
164
Discussion of Load Test Results
The analysis of the test data in Chapter VIII was

made to determine whether the shell and columns reacted
according to the design. A discussion of factors which
were not covered in Chapter VIII will be presented in the
following paragraphs.
The maximum shell stress at the center of the inte-
rior horizontal edge beam was only 18 per cent of the
design load, 19,000 lbs., therefore, it was assumed that
the sloped interior edge beams in the local area surround-
ing the tension bar at the roof center actually took 82
per cent of the tension.
From the strain readings in the tie bar, an account
was maintained of the horizontal shear forces acting at
the intersection of the column and roof. The construction
joint at the haunch and column absorbed much greater bend-
ing stresses than the design indicated. This was pointed
out by the fact that the maximum tensile force in the tie
bar was 16 ,990 lbs . for a live roof load of 57.0 lbs./ft. 2
compared to the design load in the tie of 19,000 lbs. The
maximum total force that the tie measured was composed of
the load of the shell, approximately 32 lbs./ft . 2 , plus a
uniformly distributed gravity load of 61.7 lbs . /ft. 2 , or
93.7 lbs./ft. 2 ; this gave an observed tensile force which
was slightly less than the allowable load for the tie bar.
This load was computed from the original strain readings
165
on the tie bar just after the shells had been welded
together and the supports lowered, which averaged 21805
micro-ino/in., plus a maximum strain of 405 micro-in. /ino
measured during Test II.
By recalculating the value of tensile force which
should have been taken by the tie according to design,
using the actual dead load of 32.0 lbs./fto 2 , the tensile
force should have been 32,540 lbs. Assuming that the
strain of 218.5 micro-in./ino had not changed during the
four months period between the shell completion and the
testing period, the difference between 32,540 lbso and
27~955 lbso, 4,585 lbs., was the calculated maximum
shearing force resisted by each column. The calculated
unit shearing stress in the concrete by the method for
flexural members was 43.8 psi, which was approximately
one-half of the allowable shear stress.
An i nteresting correlation was noticed between the
observed strains in the tie bar after the quadrants were

pulled together before welding , and after welding was com-
pleted and the supports were lowered, compared to the max-
imum strains during Test II. The first two average
readings were 165 micro-in./in. before welding of the edge
beams and 218.5 after lowering the supports; both of these
readings were taken after the column to shell connection
had been made so the difference in strain was considered
to be induced by lowering the supports. The static load
of 6lo7 lbso/ft. 2 , which was nearly double the dead load~
166.
was placed on the roof after the column haunches had been
grouted. If the load- strain trend had continued linearly,
and the columns had not been grouted, an expected strain
value of 422 micro- in . /in. would have been obtained . The
difference between 422 and the obser ved str ain value of
405 micro-in./in., or 17 micro- in./in. was considered to
be a measure of increased stiffness of the joint due to
grouting . The bending resistance was increased by approx-
imately 4 per cent by gr outing the haunch.
It was believed that the rigidity exhibited by the
construction joint was due to the wide haunch and shell
connection. This connection including the edge beam to
column weld spanned a horizontal distance of 4 feet and
was "V 11 shaped with a side slope of 3 to 10. Thus, the
joint could not react as a pinned or simple connection,
which was the design aSSUlJlption.
The tie bar area could be reduced by 14.0 per cent
because of the difference between the design tensile
stress and the measured stress. However, changes in soil
conditions due to ground water and moisture infiltration
should be considered before changing the design.
The shell quadrants were assumed to transfer all roof
lo ads, including eccentric loads, as shearing forces
through the parabolic arches into the e.dge beams of t he
quadrant . However, from visual observation during the
cantilever load of 41 . 3 lbs . /ft. 2 , the shell was subjected
to bending stresses. Cracks perpendicular to the edge
167
beam centerline extended to the midpoint of the edge beam

on the loaded side of the structure . This indicated heavy
bending stresses in the beam, causing the section to crack.
Three cracks were observed approximately 12 to 15 inches
apart, starting about 12 inches past the end of the haunch
on the south column. A similar pattern was observed on
the north edge of the shell.
During the maximum cantilever load, the shell raised
off both haunches on the unloaded side by approximately
1/16 to 1/8 inch . This occurred due to the slack in the
welded connection. Two hairline cracks appeared in a
radial direction around the outside of the reinforcing
steel mat location on the north unloaded quadrant. These
cracks had radii of approximately 2 feet and 2 1;2 feet
from the column center. No cracks or other visible stress
were noticed around the column in the quadrants which were
loaded . However, after the shell had been loaded for 45
minutes, the roof edges were still deflecting slowly.
The maximum observed deflections were +5.0 inches and
-5.0 inches at the west and east roof edge mid-spans . The
weld connection at the mid-span of the roof edges showed
greater deflections than the corners. The maximum ob-
served deflections of the corners during the half-roof
load were -4.16 inches at the southeast corner, -4. 36
inches at the northeast corner, +4 .18 inches at the
northwest corner, and +4 .34 inches at the southwest corner.
During the sustained uniformly distributed load test
of 61.7 lbs./ft. 2 , the average maximum deflection of
168
corresponding roof points was 0.30 inches at the east and

west roof edge mid-spans. Under the same load~ the average
deflections were 0.11 inches for the northeast and north-
west corners and 0.18 inches for the southeast and south-
west corners. On the basis of a deflection to span ratio,
the mid-span deflection would be 0.9 in./360 in.
CHAPTER:-X
SUMMARY AND CONCLUSIONS
A construction engineer ing study was conducted on a two-

~olumn h-p shell to (1) develop a step- by-step precasting pro-
cedure, (2) test the structural design values , and (3) to study
the costs of production of the shell and correlate the observed
cost data to usable data for use in construction cost estimat-
ing on this type of shell .
A construction procedure was developed for trial dur ing
the study . This procedure was carried out to simulate proto-
type conditions by using an untr ained crew with a staff mem-
ber from the Agricultural Engineer ing Department acting as
construction foreman . After the erection procedure was com-
pleted, the individual constr ucti on steps and methods were
analyzed critically and alternate methods wer e suggested.
These were discussed in Chapter IX.
The cost study was divided into three sections for analy-
sis of the data; these were (1) labor costs, (2) material costs,
and (3) equipment costs . After the raw data were analyzed in
Chapter VIII , the labor data were adjusted by skill factors
for each labor operation in Chapter IX. Estimates were made
from construction experience for adjusting the variable cost
data from a 20 foot square shell to a 40 foot square shell.
169
170
Curves were plotted showing the expected trend of labor, form,

and erection apparatus costs as a function of number of units
constructed .
The structural test data included strain data plus roof
a nd column deflection data . The strain data were obtained
from strain gages mounted on the columns, as illustrated in
Figure 29 , four strain gages mounted on the tie bar, and one
strain gage mounted at the center of the horiz ontal interior
edge beam. The deflection data were obtained from a differ-
ential water level manometer with the reservoir connected to
to a ~ moveable end by a plastic tube. Observations were taken
at six points' on the edge of the ' roof, at the roof center,
and on the column opposite the reservoir.
RESULTS AND CONCLUSIONS
The following conclusions were drawn from the investi-

gation:
(1) Precast h- p shells can be erected by rural build-
ers and general cohtract9rs by following the sim-
ple erection sequence outlined in Chapter IX.
(2) Pre cast h- p shells can function as well structur-
ally as conventional , cast-in- place shells.
(3) Prefabricated h-p shell elements can be stock-
piled , transported to a building site, and erected
efficiently.
(4) Precasting reduces the amount and complexity of
formwork as compared to conventional h-p shell
171
construction, thus , reducing forming costs due to

labor.
(5) The welded edge beams with reinforcing dowels func-
tioned well as shear transfer members.
(6) Welding the shell quadrants together is a fast, ef-
fective means of obtaining an efficient construe~
tion joint .
(7) The cribbing used for supporting the precast columns
during erection was an inefficient means of erection.
(8) The steel assembly jig effectively controlled the
vertical and horizontal positions of the shell
quadrant corners.
(9} The configuration of the haunch connection to the
shell produced a joint capable of resisting moment
stresses between the columns and edge beams. Grout -
ing the hauncheG increased the rigidity by 4 percent.
(10) The exterior sloped edge beams on the loaded quad-
rants were subjected to bending stresses when the
shell was eccentrically loaded. Bending cracks
occurred along the edge beam under the maximum cant-
ilever load of 41.3 lbs . /ft}
(11) The measured tensile force in the horizontal interior
edge beam was approximately 18 percent of the design
stress, 19 , 000 lbs . for a uniform roof load of 57.0
lbs./f~~ The remainder of the stress was resisted
by the splice plates joining the sloped interior
edge beams.
172
(12) The bending moments from observed column bending

strains analyzed by the cracked section method
were 249,000 and 220,000 in.-lbs. for the north
and south columns compared to the theoretical
value of 248,000 in.-lbs. during eccentric
loading in the direction of overturning.
(13) The shell was subjected to radial bending stresses
around the reinforcing mat, cast into the corner
of each shell, during eccentric loading; this was
indicated by hairline cracks in a radial direc-
tion around the column in the unloaded quadrants .
(14) The haunch arms were effective in resisting over-
turning moments. The welded angles connecting the
shell and haunch allowed 1/16 to 1/8 inch deflec-
tion of the shell above the haunch arms on the
unloaded side of the shell during eccentric
loading.
(15) The steel edge beams and column haunches plus the
wingwall footing incr~ased the shell material
costs by $56.18 above the cost for a similar shell
design for only axial loads. However, this add~
tional expense can be offset by reduced labor

costs.
(16) The calculated difference in total cost between
the prototype shell and a shell designed for
concentric loading was $252 . 17 or $0.63/ft. 2
(17) The maximum roof deflection under the cantilever
173
roof load was 5.0 inches at the mid-span of the hor~
izontal exte:rior edge beam.

(18) Under a sustained uniformly distributed roof load of
: .....·
61.7 lbs./ft!, the average -maximum deflection of cor-

responding roof points was .JO inches for the east
and west midespans. The same deflection was observ-
ed for a sustained load of 57.0 lbs./ft. 2 This gave
a deflection to span ratio of 0.9 in./360 in.
(19) After ten uses of forms and erection apparatus~ the
cost of erecting a 20 foot square precast h-p shell
Ran be reduced to $2.55/ft. 2 , or 37 percent of the
first cost total.
(20) The h-p shell roof ~uadrants were assembled on the
support system in 1 hour and 20 minutes.
(21) The 40 foot square shell could be erected with an
estimated 21 percent increase in supervision and
2~8 percent increase in unskilled labor costs
based on estimates from the construction exper-
ience of this study.
SUGGESTIONS FOR FURTHER STUDY
Investigate further the stresses under uriiform and ec-

centric loading by mounting strain gages on the reinf orcin~
steel in the-9olumns and haunches.
Fabricate an efficient deflection measuring apparatus
for future testing using the principle of the system used
in this study. Hook gages could be used for accuracy.
174
Exper.tm~ntally investigate the stresse:Srdhr1<thededgt:!RJ.i>.eam

steel for a prefabricated h-p shell with welded Bf~el~edge
beams.
Conduct a construction engineering study for a prefab-
ricated h-p shell utilizing new methods of column erection
which were suggested in Chapter IXo
Analyze the stresses in the edge beams, tie, and column
on an h-p shell to determine the stresses resisted by each
member.
SELECTED BIBLIOGRAPHY
1. Beauchemin, R. 0. "Both Sides of Precast Concrete,"

Concrete 65:6:22-3. June, 1957.
2. Amirikian, Arsham. "Multipurpose Building of Precast

Thin-Shell Panels," Journal of the American
Concrete Institute 35:12:124~5~June, 1960.
3. Faerber, Nelson A. "You Can Raise the Roof With

Concrete," Proceedings, Journal of the American
Concrete Institute 59:8:1047-53.~August, 1962.
4. Riley, Walter E. 11 Shell Construction - A New Approach,"

Journal of the American Concrete Institute
32:2:1361-7Y-:-- April, 1961.
5. Thompson, Isadore. "Mass Production of Shells for

Oakland International Airport," Proceedings,
Journal of the American Concrete Institute
59:7:949-57-.~July, 1962.
6. Candela, Felix, "Understanding the Hyperbolic

Paraboloid," Architectural Record 123:7:191-5.
July, 1958. 124:2:205-7, 215. August, 1958.
7. Candela, Felix. "Hyperbolic Paraboloids," Proceedings

of the Conference on Thin Concrete Shells,
Massachusetts Institute of Technology, pp.91-8,
1954.
8. Parme, A. L. "Hyperbolic Paraboloids and Other Shells

of Double Curvature," Proceedings, American
Society of Civil Engineering 82:N-ST5.
September, 1956.
9. Tedesko, Anton. "Shell at Denver - HP Structure of

Wide Span," Journal of the American Concrete
Institute 32:4:403-12:" October, 1960.
10. Madsen, Gordon, and Dutton Biggs. "Building for

Economy With Hyperbolic Paraboloids," Journal
of the American Concrete Institute 32:4:373-83.
October, 1960.
175
176
11. Peurifoy, R. L. Construction Planning, Equipment 1 and

Methods. New York: McGraw-Hill Book Company,--rllc.,
pp. 151-2. 1956.
12. "Formwork for Concrete," ACI Committee 622,

Journal of the American Concrete Institute 32:2:997-8.
March, 1961-.-
13. "Curing Concrete," ACI Committee 612, Journal

August, 1958.
14. Cazaly, Laurence. "Neat Joints are Good Business,"

Journal, Prestressed Concrete Institute 5:42:52-3,
56-9. December, 1960.
15. Naslund, K. C. "Precast Concrete Joinery: Basic Design
Principles," Architectural Review 129:6:166-9.
June, 1961.
16. Reinforced Concrete 35. Chicago: Portland

Cement Association. pp. 17-8. 1960.
17. Mahoney, G. W. A., and Earl R. Bell. "New Concrete Shell

Construction. Systems for Agriculture," Lubbock,
Texas: Southwest Section, Amerl.can Society of
Agricultural Engineers. 1963.
18. Harrenstein, Howard P. "Hyperbolic Paraboloidal Umbrella

Shells Under Vertical Loads," Journal of the American
Concrete Institute 32:1:385-401. October-;-1960.
19. Waling, J. L., and Longin B. Greszczuk. "Experiments

With Thin-Shell Structural Models, 11 Journal of the
American Concrete Institute 32:1:413-32. October,
1960.
20. Kleiger, Paul, and J. A. Hanson. "Freezing and Thawing
Tests of Lightweight Aggregate Concrete," Journal
January, 1961.
·-·
21. Hanson, J. A. "Shear Strength of Lightweight Reinforced
Concrete Beams," Journal of the American Concrete
Institute 30:3:387-403. September, 1958.
22. Shideler, J. J. "Lightweight-Aggregate Concrete for

Structural Use," Journal of the American Concrete
Institute 29:4:299-328. October, 1957.
177
23. Hanson, J. A. "Tensile Strength and Diagonal Tension

Resistance of Structural Lightweigpt," Proceedings,
Journal of the American Concrete Institute 58:1:
1-39. July-;-1961.
24. "Elementary Analysis of Hyperbolic Paraboloid
Shells," Bulletin No. ST85, Portland Cement
Association, pp. 1-20. 1960.
25. Urquhart, L. C., Charles E. O'Rourke, and George Winter.

Design of Concrete Structures. New York: McGraw-
Hill Book Company, Inc. pp.26-8. 1958.
26. Parker, Harry. Simplified Engineering for Architects

and Builders. New York: John Wiley and Sons, Inc.,
2nd Edition. 1959.
27. Mannshreck, Karlson E. "A Model Study of Wind Forces

on Hyperbolic Paraboloid Shells," Unpublished
Maste~s Thesis, Oklahoma State University. May, 1964.
28. Reinforced Concrete 38. Chicago: Port-

land Cement Association. P:-17. 1961.
29. Reinforced Concrete Design Handbook, Com-

mittee 317, American Concrete Institute. 2nd
Edition. 1955.
30. Patterson, Donald. How to Design Pole-Type Buildings.

Chicago: American-Wood Preservers Institute.
pp. 34-5. 1957.
31. Engineer Field Data, FM 5-34: Washington,
D. C.: Headquarters, Department of the Army.
pp. 332-54. June, 1962.
32. Design and Control of Concrete Mixes.
Chicago: Portland Cement-Xssociation. 10th
Edition. p. 7. 1952.
33. Peurifoy, R. L. Estimating Construction Costs. New York:

McGraw-Hill Book Company, Inc. pp. 6-8. 1958.
34. Large, George E. Basic Reinforced Concrete Design:
Elastic and Creep. New York: The Ronald Press
Company.----:lnd Edition. 1957.
35. Rensaa, E. M. "Joints in Precast Concrete Building

Frames," Engineering Journal (Montreal)
43:8:64-7. August, 1960.
178
36. Cogan, Myles H. R. "New, Fast, Economical Joint Makes

Precast Concrete Practical," The Oil and Gas
Journal 58: 19: 158·-60. May 9, 1960-.- - -
37. Dallavia, Louis. Estimating General Construction Costs.

New York: F. W. Dodge Corporation. 2nd Edition.
1957.
APPENDIX A
TIE BAR CALIBRATION TEST DATA
179
TABLE A-1
OBSERVED STRAIN DATA FROM TESTS
OF TIE BAR SAMPLES
-6
'Reading Load (lbs.) - Strain (in./in. x 10 )
I "
No. Bar #1 Bar #2 Bar #1 Bar #2
. (rage #1 Gage #2 Gage #3 Gage #4
1 2,100 1,830 . -190 +105 0 -70
2 3,950 3,990 .· -310 +145 ..;..10 -155
3 . 6,100 5,640 -300 +150 -15 -230
4 7,980 8' 070 • -480 +135 -30 -330
5 9,.680 9,820 -525 +100 -35 -400
6 11,540 13,000 -560 +50 -75 -510
7 17,640 15,180 -675 -1~0 -110 -580
8 18,860 17,880 ' -690 -145 -160 -650
9 22,800 19,500 -770 -260 -190 -700
I
10 23,840 22,520 ' -785 -280 -255 -775

' 11 25,070 25' 100. . -810 -310 -310 -840
12 26,090 27,230. ..;;8'!o .... 340 -360 -890
13 28,000 30,020 -870 -390 -420 -960
14 29, 210 32,200 -895 -425 -470 -1020
15 31,480 35,230 -920 -500 -535 -1095
16 33,690 -·--- -930 ..-550 ---- ----
\
APPENDIX B
SHELL ERECTION STRAIN GAGE DATA
181
182
TABLE B:-1
STRAIN GAGE DATA FROM TIE BAR
DURING SHELL ERECTION
Condition Observed Strain (in./in~ x io-6) ·

Gage #22 Gage #23 Gage #24 Gage #25
1st Quadrant Erected

(4: 00 p .m.) 0 -15 0 -25
2nd ·Quadrant Erected
(4:25 p.m.) -5 -10 -50 -35
3rd Quadrant Erected
(4: 45 p .m.) -30 +10 -20 -40
4th Quadrant Erected
(5: 20 p .m.) -50 0 -15 -80
Quadrants Drawn
Together -200 -120 -80 -260
Quadrant Edge Beams
Welded, Column Weld
ed to Shell Angles -135 -85 +30 -130
Support System
I Lowered -250 -225 -115 -275
Remarks:
(1) Shell erection was begun on October 8, 1963.
(2) Supports were lowered on October 25, 1963.
(3) Tension is indicated by minus (-) signs.
APPENDIX C
DEFLECTION DATA FROM LOAD TESTS I-IV
183
184
.REMARKS AND DEFINITION~
~efl. =Deflection
Acc .. Defl. =Accumulated Deflection
S.E.C. = Southeast Corner of Roof
E.M.S. = East Mid-span of Roof Edge
N.E.C. =Northeast Corner of Roof
.N. Col. =North Column
.N.W.C. = Northw~st Corner of Roof
W.M.S. = West Mid-span of Rooi Edge
S.W.C. =Southwest Corner of Roof
C. of' R .=<Center of Rbof-
Deflections ~were_measul'.led ...::.in' inches.
Positive (+) "qeflection was measured downward.
TABLE C-1
TEST I ROOF AND COLUMN DEFLECTION DATA
'•
Load 21.6 psf 21.6tpsf 21.6 psf 21.6.psf 0

Time 1 hr·. 24 hrs, 48 hrs. 72 hrs. 74,5.hrs.
Location Deflection (Inches)
Defl. Acc. Defl, Ace Defl . . Ace. Defl. Acc. Defl, Acc.
Defl, Defl, Defl, Defl. Defl.
S.E.C. +.08 +.08 0 +.08 0 +.08 0 +.08 =.10 ~.02
E.M.S. +.10 +.10 0 +.10 +.04 + 014 0 +.14 -.20 -.06
N.E.C •. + .10 +.10 =.02 +.08 ~.04 +.04 +.04 +.08 -,06 +.02
N.W.C. +,08 +.08 ..... 04 +.04 +.02 +,06 +.02 +.08 -,02 +.06
W.M.S. +.08 +.08 -.02 +.06 +.04 +.10 -.06 +.04 -,10 -,06
s.w.c. +.02 + d-02 +.06 +.08 -.02 +.06 -.10 -.04 +:·02 ... 62 \. 0 •. ,.;,
c. of R. +.06 +.06 +.06 +.12 -.18 -.06 +.02 -.04 -.08 -.12
,.;.,
TABLE C-2
TEST II ROOF AND COLUMN DEFLECTION DATA
Load 25.3 psf 46.9 psf 61.7 psf 61.7 psf 61.7 psf 61.-7 psf 61.7 nsf 61.7 nsf 0
'
·Time . . l .l
2 hrs. 3 hrs. 4 hrs. 24 hrs. 49 hrs. 74 hrs. 96.5 hrs. lla hrs. 117! hrs .
Loe a- .,
tion - Deflection (Inches)

l>ef l Acc. Defl. Acc._ Defl. Acc. Defl. Acc. Oefl. Ace. Defl. Ace. Defl. Ace, Defl. Acc. )efl. Acc.
Defl. l>efl. [Defl. Defl. Defl. Defl. Defr. Defl. Defl.
S .E .C. ft.06 +.06 .... 26 -.20 I- .06 ... .,26 +~iT4 -.22 +.20 -.02 ft-. 04 ft-. 02 -.04 ,... . 02 -r. 08 . II-. 06 lt-.12 +-.18
E.M.S. ft .12 +.12 .... 10 ft 02 ..... 06 .... 04 +.04
0 0 +.14 +-.14 0 f!-.14 +-.02 ft- .16 0 II- .16 ... 04 +-. 20
N.E.C. ft.04 +.04 .... 18 ..-.14 11-.02 .-.12 -.02 -.14 +.14 0 ,... . 06 .... 06 +-.02 -.04 0 .... 04 .... 20 -.16
N.Col. H-.06 +.06 .... 02 ~.04 I- 02 r+.02
0 0 +.02 +.02 +-. 04 i-.02 11-.02 -.06 .... o4 .... 02 06 ... 0 .... 06 0
N.W.C. r--.12 -.12 ft.24 ft .12 I+-. 06 ft .18 +.10 +.28 ..... 12 +-.16 -.04 II- .12 +-. 02 ft- .14 +-. ·02<
\
~<16 ~ ... 26 .... 10
·.)'.~
W.M.S. ""'". 08 -.08 ft.42 ft 34 I+- .18 fto52 +.06 +.58 ..... 16 +-.42 ft- .'"02 I+-. 41'.:l
0 0 It-. 44 -.02 11-.42 ... 46 .:. . 04
s.w.c. ft.02 +.02 ft-.20 H-.22 ft-. 08 +.30 +.10 +.40 .... 16 +.24 ft-. 02 i-. 26 +-.02 f\-.28 II-. 02 It-. 30 ... 30 0
..
C.ofR. ft .14 +.14 ft,04 ft .18 --.to +.08 +.06 ft- .14 0 if- .14 ft-.10 i- .24 -.02 ft- .22 .... 04 fl-.18 ... 14 ~.04
'
TABLE C-3 186
TEST III ROOF AND COLUMN DEFLECTION DATA
Load 25.0 psf 41.3 psf 0

Time l l hrs. 2 hrs. 3 hrs·•~
Location
cc.
efl .:
S.E.C. 2.5 3.06 1.10
E.M.S. 3.0 3.66 1.34
N.E.C. 2.5 3.16 1.20
~.Col. 0 -.02 -.06
N.W.C. 2.5 3.26 +.92
W.M.S. 3.1 3.90 1.10
s.w.c. 2.7 3.36 +.98
C.ofR. -.0 +.02 0
TABLE C-4
TEST IV ROOF AND COLUMN DEFLECTION DATA
L-oad 19. 3 psf 35.4 psf 49,0 psf 57.0 psf 57.0 psf 0
Time l~ hrs. 21 hrs. 3l hrs. 4!_ hrs. 20 hrs. 51 hrs.
'Loca-
tion ·- Deflection (Inches
Defl. Acc. pefl. Acc. Defl. Acc. Defl. Acc. Defl. Acc. Defl. Acc.
Defl. Defl. Defl. Defl. Defl. Defl.
S.E.C. +.10 +.10 +.12 +~-22 +.14 +.36 -.io +.26 f+.04 +.30 0 +.30
E~M.S. +.30 +.30 +.04 +.34 +.18 +.52 -.16 +.36 f+.02 +.38 -.04 +.34
_N .E .C, +.16 +.16 +.04 +.20 f+.22 +.42 -.28 +.14 1+.04 f+,18 +.14 +.32
N.Col. 0 0 -.02 -.02 0 -.02 1+.02 0 ""'. 04 ~.04 0 -.04
N .W.C. -.10 -.10 : ,,0 -.:.. .i..i:F· 1--'~ 1:4: I- .2,4 f+: 24 - ... ,0 ::....02~ + .02·' -·30 -.32
W.M.S. ..... 10 -.10 +.14 +.04 -.12 i-. OR.
l -
f+.32 +.24 t- .10 +.14 -.40 --.26
s.w.c. -.12 -.12 f+.06 -.06 -.12 -.18 f+.26 +.08 -.06 +.02 -.30 --.28 I
C.ofR. 0 0 f+.04 +.04 f+,10 +.14 1+.04 f+ .18 i-,06 +.12 I - .10 f+.02
APPENDIX D
STRAIN GAGE DATA FROM LOAD TESTS II=IV
187
TABLE D-1
TEST II OBSERVED STRAIN
(Inches x io-6)
~ ~---~ :~4 hr-s~124hr-;~~ ::fhrs. hrs. 115fthrs-1~7 h:J

·-·--41
:::. 96
Gage No.
1 0 -20 +20 +60 +50 -40 +10 +35 +20 +100
2 -35 +10 +45 +20 -65 -10 +20 0
3 -40 +25 +70 +40 -40 +10 +30 +10 I -10 ,.
4 -40 -5 +40 +10 -50 -10 +15 +15 +5
5 -30 +10 +40 +30 -50 0 +30 +120 +90
6 +250 +220 +280 +335 +220 +220 +185 +100 +115
7 -50 -20 +15 +10 +220 -25 -40 -50 -85
8 -40 -20 +5 0 -45 +10 -10 0. -5
9 +10 +10 +50 +35 +55 +25 +10 +20 +10
10 -5 +20 +60 +55 +50 +20 0 0 -40
11 -10 +10 +50 -10 -45 -10 -10 -30 0
12 -20 -10 +20 -10 -80 -20 -20 -25 0
13 +5 +60 -85 0 -30 -45 -50 -60 -60
14 +20 -80 -15 -90 -160 +30 +15 +30 +60
15 -20 -20 +10 -20 -50 -10 -15 -15 0
16 -10 0 +30 -10 -35 0 0 -15 0
17 -25 -20 +30 +60 -65 +10 -65 +5 +10
18 -30 -40 -5 +5 -50 0 -30 +15 +10
19 -20 -40 -20 -45 -40 -90 +20 -130 -110
20 -5 +30 +70 +80 +15 +20 +20 +15 -10
21 --- --- --- --- -315 -320 -310 -285 -190
22 0 ~320 -415 -415 -420 -430 -440 -440 0
23 -15 -265 -365 -340 -370 -370 -380 -370 -30
24 -5 -290 -395 -420 -420 ~415 -420 -425 -10
25 0 ~5 -270 ~360 -380 -380 -370 -380 -385 -10
Remarks: (1) Test dates were February 24-29, 1964. (2) Loading, uniformly distribut-
ed gravity load. (3) Steel gage factor, F=2.ll. (4) Concrete gage factor, F=2.13.
(5) Steel gage resistance, 120 ohms. (6) Concrete gage resistance, 300 ohms. (7)
Strain data is uncorrected. (8) Minus (-) sign -~-ndicates tens~__o_n__._______________
189
TABLE D-2
TEST III OBSERVED STRAIN

(Inches x l0-6)
Load 0 25.0 psfl 41.3 psf! Unloaded

Time 0 1 ! hr s . 2 hrs . I 3 hr s .
Gage No.
1 0 +30 +20 0
2 +60 +65 -10
3 +150 +280 +20
4 --10 -420 -30
5 +10. -10 0
6 0 -30 -10
7 -25 -185 -50
8 -250 -550 0
9 0 +40 +30
10 +200 +430 +60 l
11 -20 +5 +15 I
12 -10 +5 -30
13 +30 +230 0
14 +120 -350 -40
15 +50 -50 -25
16
17
18
+40
-100
+230
-50
-240
-470
-15
-85
-60
I
·I
19 I
-40 -90 -110
20 +150 +365 +85 II
21 -30 -15 -10
I
22 -70 -120 +20
23 -60 -110 0
24 -60 -110 +10
25 0 -60 -110 +10
I Remarks:
(1) Testdate, March 2, 1964.
(2) Load, uniformly distributed on half of
the roof.
($)Same as Remarks (3) through (8), Test II.
TABLE D-::S
TEST IV OBSERVED STRAIN

(Inches x io-6)
Load 0 19.3 psf 35.4 psf 49.0 psf 57.0 psf 61.7 psf 0
Time 0 l! hrs. 2t hrs. 3! hrs. 4! hrs. 20 hrs. 50 hrs. 51 hrs.
Gage No.
1 0 +10 +10 +20 +35 -10 -55 -70
2 0 -10 0 +20 -30 -85 -85
3 +10 +20 +40 +80 +80 +40 -90
4 +25 +20 +40 +20 -30 -70 -80
5 +5 0 +10 +20 -30 -75 -70
6 +10 0 +10 +20 -15 -60 -60
7 +20 +20 +30 +40 -25 -95 -155
8 +30 +20 +40 +20 -10 -80 -80
9 +5 +10 +20 +25 -20 -40 -90
10 -10 +10 0 +45 0 -10 -130
11 +10 +10 +10 +20 -30 ... so -45
12 +10 +5 +10 +20 -40 -70 -45
13 +5 +20 +20 +40 +25 +20 +50
14 +20 +20 +40 +30 -30 -65 -75
15 +10 0 +10 +10 -40 -75 -70
16 +5 0 +10 +10 -30 -65 -70
17 0 0 -10 0 -20 -40 0
18 +20 +15 +30 +15 +170 +130 +130
19 +20 +50 +50 +60 +105 +50 -70
20 +5 +40 +30 +60 +20 +40 +40
21 -35 -50 -65 -75 -60 -50 +5
22 -160 -260 -360 -420 -430 -380 +25
23 -110 -190 -270 -320 -320 -310 0
24 -130 -230 -310 -375 -390 -380 -10
25 0 -120 -210 -280 -340 -360 -345 -5
Remarks: (1) Test dates were 3-5 March, 1964. O ) Loading was by uniformly distributed
gravity load. (3) Same as Remarks (3) through (8)' Test II.
VI'F'A
Ronald Tacie Noyes
Candidate for the Degree of
Master of Science
Thesis: DEVELOPMENT AND PERFORMANCE OF A PRECASTING SYS'REM

FOR CONCRETE HYPERBOLIC PARABOLOID SHELL
CONSTRUCTION
Major Field: Agricultural Engineering
Biographical:
Personal Dat~: Born near Leedey, Oklahoma, Januaty 4 3
1937, the son of Johnny L. and. Aqna Mo Noyeso
Education: Graduated from high school at Leedey, Okla=
homa, in May, 1955. Received the Bachelor of Sci=
enc.e degree in Agricultural Engineering, January
21, 1961, from Oklahoma State Universityo Com=
pleted the requirements for the Master of Science
degree in May, 1964.
Professional Experience~ Worked t.hree summers in the·
Engineering Department, Kootenai National Forest 9
Libby, Montana, as an engineering aid. Worked
part time at the Stillwater Outdo.or Hydraulic ·
Laboratory from November, 1959, to January 9 1961 9
and· fu£l:iL:.time fr,am< January to April, 1961 9 as an
engineering aide Served as a platoon leader,
communications officer, and adjutant for a combat
engineer battalion for 21 monthso Served as a
graduate research assistant for one year and a
graduate teaching assistant one semester for the
Agricultural Engineering Department, Oklahoma
State University, Stillwater, Oklahoma.
Missouri University of Science and Technology
Scholars' Mine
Wei-Wen Yu Center for Cold-Formed Steel

Center for Cold-Formed Steel Structures Library Structures
01 Jan 1971
The analysis and behavior of thin-steel hyperbolic paraboloid

shells
Peter Gergely
P. V. Banavalkar
J. E. Parker
Follow this and additional works at: https://scholarsmine.mst.edu/ccfss-library
Part of the Structural Engineering Commons
Recommended Citation
Gergely, Peter; Banavalkar, P. V.; and Parker, J. E., "The analysis and behavior of thin-steel hyperbolic
paraboloid shells" (1971). Center for Cold-Formed Steel Structures Library. 160.
https://scholarsmine.mst.edu/ccfss-library/160
This Technical Report is brought to you for free and open access by Scholars' Mine. It has been accepted for
inclusion in Center for Cold-Formed Steel Structures Library by an authorized administrator of Scholars' Mine. This
work is protected by U. S. Copyright Law. Unauthorized use including reproduction for redistribution requires the
permission of the copyright holder. For more information, please contact scholarsmine@mst.edu.
CCFSS LIBRARY Gergely, Peter, Banavalkar, P.
22 3 * 178 V., Parker, J. E., THE ANALYSIS
1971 AND BEHAVIOR OF THIN-STEEL
HYPERBOLIC PARABOLOID SHELLS
CCFSS LIBRARY Gergely, Peter, Banavalkar, P.

22 3 * 178 V., Parker, J. E., THE ANALYSIS
1971 AND BEHAVIOR OF THIN-STEEL
Technical Library
Center for Cold-Formed Steel Structures
University of Missouri-Rolla
Rolla, MO 65401
THE ANALYSIS AND BEHAVIOR OF THIN-STEEL
by
Peter Gergely
P. V. Banavalkar
and
J. E. Parker
January 1971
Report No. 338
Department of Structural Engineering

Cornell University, Ithaca, N.Y.
ABSTRACT
An extensive experimental anri analytical investigation

of thin-steel hyperbolic paraboloid (hypar) sgt'Uctures
was carried out to provide design information. As a result
of this work, empirical data is provided regarding the
behavior of :·.such structures and computer programs are
presented for the analysis of thin steel hypar structures.
Hyperbolic paraboloid structures possess a unique

combination of structural and architectural properties; some
of then are the followi~ga l) Due to the double curvature of
the surf ace the internal stresses in the deck are generally
low and the def tections are small. 2) Since a hypar surface can
be generated by straight lines, thin-steel or light-gage
panels may be used to form the shells furthermore such panels
are well suited to carry the in-plane shear forces in hypar
shells. 3) Basic hypar units can be combined in a large
variety of ways to produce attractive roofs (Fig. 1-2, page 216).
4) The dead load to live toad ratio ia very low in the case
of thin-steel shell structures.
A hypar unit is a warped surf ace bounded by straight

lines (Fig. 1-1. page 215). The equation of the surface
is z ~ cxy/ab. According to the simple membrane theory, a
uniform load p produces pure shear forces N s abp/2c.
xy
This membrane shear transmits uniform eccentric axial forces
to the edge members.
- i -
The following are the major problems associated with
the design of thin-steel hypar structuresa 1) The deflections,
stresses and the stability of hypars depends greatly on
the shear rigidity of the thin-steel deck. This property
must be evaluated experimentally for each combination of
decking, connections to edge members, and seam connections.
Furthermore, in the case of hypars the deck is warped
and thus the shear rigidity may be different from that
of an equivalent flat diaphragm. 2) The deck may buckle
due to the shear a~resses1 and the bucklinl0 "'1st be
evaluated for highly orthotropic shells. 3) The design
of thiR•steel h}"P&r structures ia generally governed
by stiffness (deflections or buckling) requirements. The
evaluation of thP. deflections is a very complex matter
because it depends on the deck rigidity, the edge member
axial and bending stiffnesses, and on the eccentricity
of the deck-to-edge member connection. 4) lf the curvature
(or rise-to-span ratio) of a hypar is small, the deflections
may be very large and a conaidera~le portion of the load
is carried by bending oather than by membrane shear.
5) Partial or concentrated loads may cause large local
deflections, eepecially if single-layer decks are used.
The present investiaation studied all the above-
mentioned factors. The experimental and the analytical
studies are summarized briefly in the following paragraphs.
- ii -
The experimental investigation consisted of four
types of tests& a) Four medium-scale (12 ft by 12 ft in plan)
inveJrted umbrella tests to study the stresses, deflections,
&nd the deck bucklings b) Test on a small-scale (2 ft by
2 ft) inverted umbrella structure to •tudy scaling effects
and the overall buckling of hyparss c) Sixteen flat shear
tests to determine the shear rigidity of the decks used
in the hypar testes d) Twelve saddle-shaped hypar tests
(5 ft by 5 ft in plan) with various rise-to-span ratios to
evaluate the effect of rise or warping on the shear rigidity
and to study other factors such as partial loading and
single versus double layered decks. Photos of the various
types of tests are shown in Figs. 7.1 to 7.5. The experimental
program is described in detail in Chapter VII.
Prior to the main test program, several small-
scale (2 ft by 2 ft) four-quadrant tests and medium-scale
single-quadrant tests were also conducted. These tests
were however discontinued because of the severe .. ecaling
effects in the case of the small-scale models and the
violation of the synvnetry ronditions in the case of
single-quadrant experiments where the neighboring
quadrants were missing. Nevertheless, these teats produced
useful qualit•tive information and experience with
manuafacturing and testing thin-steel hypar structures.
The edge members of the umbrella-type specimens
were made of tubular members since this afforded easy
connection of the warped surface to the straight edges.
- 111 -
The decking consisted of single or double layers of
standard corrugated panels. One layer was connected
to the edge members with sheet metal screws at various
spactngs. The seam connections between the panels
was also by means of sheet metal screws. In the
case of shells with two layers, the top layer was
connected to the bottom layer in a similar manner.
The medium-scale umbrella models were loaded

nsing air bags under each of the four quadrants. The
saddle-shaped hypars were loaded with sand, whereas
the small-scale models were loaded through loading
pads and suspended weights.
The following are the principal conclusions

of the experimental part of this investigations
The effective shear stiffness of the cold-formed
deck and the rise (or curvature) of the structure are
the most important factors influencing the behavior
of hypara. For low shear stiffnesses and for small
rise-to-span ratio•~the deflections may be large,
the bending stresses tend to increase relative to
the membrane stresses, and the possibility of deck
buckling increases. As in the case of flat shear
diaphragms, the shear stiffness depends stron9ly on
the seam and edge connections.
- iv -
The incl!"eaae in shear stiffness due to the addition
of a second layer of deck was found to be only about
1/3 if the aecon4 layer was connected only to the
first layer and not directly to the edge members. Similarly,
the deflections of a double-layered shell are more than
half of those of a corresponding single shell. If the
two layers are interconnected with sheet metal screws
(on an 8 in. grid in the present saddle-shaped hypar
tests), the deflections are further reduced by about
10 to 20J., depending on the rise ratio.
A particular problem of certain types of hypar

structures is the deflection of unsupported outside
corners (see Fig. 1-2, page 216). The membrane shear
cannot carry the load over such flat corners and thus
considerable bending and def lectiona may develop. The
tests showed that the bending stiffness of the edge
members has a great effect on the corner deflections,
in fact, they indicate that the design of the edge
members in hypars with flat corners is probably
governed by deflection limiaations.
The measured bending strains in flat saddle

shells (rise-to-span ratio of 1/8) was much greater
than the bending in hypars with greater curvature (rise
ratio of 1/3). The membrane theory ia insufficient
- ,, -
for the design of flat hypar structures. However, the
design of the connections (seam or edge) may be based
on the shear forces obtained from the simple membrane
theory.
Several single and double layered saddle-shaped

models were tested under partial loading. Since
such loads must be carried mainly by bending of
beam strips along the deformations of single decks,
relatively large deflections were noted. The deflections
under the 8 in. by 8 in. loaded area were about three
times greater in the single decks than on the double-
layered stuctures.
Since the effective shear rigidity of the deck
is of paramount importance, the effect of curvature
(warping) on it is an important question. The effective
shear rigidity of various deck, edge member and connection
configurations are determined by tests on flat diaphragms.
The comparison of the measured deflections for saddle
hypars with various rise-span ratios and th~ evaluation
of the effective shear rigidities backwards from the
measured deflections indicated that the shear rigidity
is reduced by about 20% due to the warping effect.
The buckling of the deck is one of the design

factors. For small rise-span ratios and for low deck
- vi -
shear rigidities the deck may buckle. As an example, a
12 ft by 12 ft model having a single layer 24 gage
corrugated sheet deck buckled at a uniform load of 70 psf
(see Fig. ). Thia model had relatively stiff edge
members (3 in. dia. tubular sections). The corner
deflections remained linear with increasing load beyond
the buckling load.
The buckling load of double-layered structures

is much larger than that for single deck shells. A
model, similar to the above but with two layers of
28 gage standa~d corrugated decks, did not buckle up
to a load of 145 psf, when the test was discontinued.
The major part of the analytical investigation

consisted:.of two finite element approaches for th~
calculation of deflections, stresses, and instability.

In addition, two simple methods were developed for
estimating tthe deck buckling load and the buckling of
the compression edge members, which would suffice in
preliminary designs.
Two types of finite elements were used1 curved
shallow shell elements and flat elements. The details
of the analysis are described in Chapter III. Both
approaches were verified by comparisons with existing
experimental and analytical results.
- vii -
The stiffness of the eccentric edge members were
properly accounted for in the mathematical representation
of the structure. The connection of the decks ~6 the
edge members may allow rotation about the axis of
the edge members and movement normal to the edges due
to slip at the connections. These possibilities were
also considered in the analysis.
The instability of the decks was studied with

the help of the incremental stiffness rnatr~ approach.
~he effective stiffness of the system is reduced a~e to
the in-plane f orcea •n the deck. The in-plane forces
depend on the deflections of the shell and to obtain
the buckling load, the eigenvalues of a large· order
system needs to be evaluated. In the present study the
load incrementation method was used instead. The effect
of the in-plane forctes was evaluated iteratively at
succes•ive load increments. The buckling load is obtained
from the non-linear load deflection eurve1 (Fig. 6-6, Page 280).
The compar.ison of the results of the f lat-eleR\fl!nt

and the curved-element approaches reveals that both
give good results for shells supported around the
perimetry, However, the flat element method gave better
results in the neighborhood of unsupported flat corners.
The analysis of the structures tested in this and
in other studies confirmed the conclusions of the experimental
- viii -
part of the investigation. The stresses in most types
of hypars are low and th design is usually controlled
by deflection or buckling limitations.
The relative stiffness of the deck and the edge

members is an important factor. For stiff edge members
the deck.tends to bend between opposite edges, whereas
in the case of flexible cantilevered edge members the
shell partially supports the edge members. Analysis of
a structure including the weight of the edge members
indicated that this effect may have to be considered

in the design of hypar structures.
The analysis of buckling of hypar decks showed'~

load
that the buckling of double-layered shells is three to
four times greater than that of single decks, The predicted
buckling loads compared well with experimental or previous
analytical evidence, The buckling load does not depend
much on pre-buckling deflections, however it depends
on the axial stiffness of the edge members.
The finite element analysis was also used to

calculate the deflection of an unsymmetrically loaded
inverted umbrella structure, The results, which compared
well with experimental data, showed that these deflections
- ix -
are about four times greater than those due to symmetric
loading. This increase of deflections obviously depends
on the type of structures in this case much of the
flexibility was due to the bending of the central column
of the umbrella structure,
Since the instability analysis of hypars by the
finite element method involves considerable amount
of computer capacity and expense, approximate methods
were developed for the calculation of buckling loads.
The buckling of the compression edge members was studied
by isolating them from the structure, The instability
of columns loaded by tangential axial forces that
to the member
remain parallelAduring deflection was evaluated. The
results are tabulated in Fig. 6-13, page 287.
The buckling of hypar decks was also investigated

by the energy method (Section Vl-7). The resulting
equation has to be minimized to get the critical
loada this can easily be done with the help of a
computer. This approach is much simpler than the finite
element instability analysis and is preferable in
preliminary designs.
A few buckling analyses of cold-formed hypar

shells showed that the critical load for double-layer.s
is about three to four times greater than a shell with
a single deck.
- x -
-INDEX
Pase
NOMEl':CLATURE l
I. INTRODUCTION 7
1. HYPAR ROOFS 7
2. LITERATURE REVIEW 9
3. SCOPE AND OBJECTIVES OF THE

il PRESENT INVESTIGATION 16
• ~CktJO\JUU>Gt\eWT.5
II. DECK PROPERTIES 18
1. INTRODUCTION 18
2. ELASTIC PROPERTIES OF DECK 19
A. MEMBRANE PROPERTIES 20
B. BENDING PROPERTIES 22
c. ELASTIC PROPERTIES FOR ARBITRARILY
ORIENTED CO-ORDINATE AXES 24
3. SHEAR RIGIDITY OF ORTHOTROPIC DECKS 25
4. DETERMINATION OF THE BFFECTIVE SHEAR
RIGIDITY FOR ORTHOTROPIC HYPARS 28
III. FINITE ELEMENT METHOD FOR STIFFNESS AND

STRESS ANALYSIS 3:.1
1. INTRODUCTION 31
2. ELEMENT STIFFNESS 33
A. DISPLACEMENT FUNCTIONS 33
ME.rh~D
'( ,, )
B. CURVED ELEMENT ( ct - 3:6
c. FLAT ELEMENT (~e -r1-1oi) " t "J 40

o. EDGE MEMBER 43
E. ECCENTRICALLY CONNECTED NON·
COMPATIBLE SUPPORTING EDGE MEMBER 45
F. ELASTIC SPRINGS 47
Page
3. MASTER STIFFNESS MATRIX 47
4. LOADING 53
S. BOUNDARY CONDITIONS 54
6. SOLUTIONS OF EQUATIONS 56
7. STRESS ANALYSIS 57
A. DECK STRESSES 58
B. BEAM STRESSES 59
IV. GENERAL COMPARATIVE STUDY 61
1. INTRODUCTION 61
2. COMPARISON OF 'METHOD 'a' WITH OTHER
NUMERICAL SOLUTIONS 62
3. COMPARISON OF METHOD 'a' AND

METHOD 'b' 63
4. DISCUSSION OF METHOD 'a' 67
S. COMPARISON OF ANALYTICAL AND
EXPERIMENTAL WORK 71
A. EDGE-SUPPORTED HYPARS 72
B. CONCRETE UMBRELLA SHELLS 75
C. UMBRELLA SHELLS WITH STANDARD
CORRUGATED DECKS 79
1. INVERTED UMBRELLA SHELL WITH
FLEXIBLE EDGE MEMBERS 81
STIFF EDGE MEMBERS 86
V. QUALITATIVE EFFECTS OF PRINCIPAL VARIABLES
ON BEHAVIOR OF HYPARS 92
1. INTRODUCTION 92
Z. GEOMETRICAL SHAPE 92
Page
3. DECK PROPERTIES 94
A. SHEAR RIGIDITY 94
B. THICKNESS OF THE CORRUGATED DECK 95
C. NUMBER OF DECKS 96
4. BOUNDARY CONDITIONS 97
A. EDGE MEMBER PROPERTIES 98
B. EDGE MEMBER AND DECK CONNECTION 99

S. LOADING 100
A. EDGE MEMBER WEIGHT 100
VI.. INSTABILITY ANALYSIS OF HYPARS 104
1. INTRODUCTION 104
2. INCREMENTAL MATRIX FOR DECK AND
EDGE MBMBBR 107
3. CHECKING OF THE INCREMENTAL MATRICES 108
4. INSTABILITY OF HYPARS 111
S. DETERMINATION OF THE BUCKLING LOAD 112
6. NUMERICAL RESULTS AND THEIR DISCUSSION 114
BUCKLING ANALYSIS
BY THE ENERGY METHOD 11.0
\ · STAB IL I TY OF ISOLATED EDGE 21EMBERS 1&4
Vll. EXPE:{l:1ENTAL INVESTIBATION

1. INTRODUCTlON
~. FLAT SHEA~ T:!STS
3. SADDLE -SHAPED HYPAR TESTS

4. MEDllJM-SCALE INVERTSB UMBRELLAS WITH STIFF eDGE BEA.1'1S ( !(;
5. 1.... ---- - , __ .. INVERTED UMBR~LLAS WITH FLFJCIBLE F.DGE BEA.t..tS 13&
~
·-
.# ... ' .....
I, DISCUSSlON OF EXPE RlMENTAL RESULTS

·""-·SUMMARY, CONCLUSIONS AND RI.iCOMMUNflATTONS \4t
BIBLIOGRAPHY \ S'C.
APPENDIX A. DECK PROPERTIES \l.L
APPENDIX B. STIFFNESS MATRICES \,0
APPENDIX c. DECK STRESSES \ 1~

APPENDIX D. INCREMENTAL ,~TRIX FOR DECK \/°l
APPENDIX E. STIFFNESS AND INCREMENTAL MATRICES FOR
A FULLY COMPATIBLE ECCENTRIC STIFFENER li3
-vi·
CLflSSAPY OF TERt'S
x,y,z ~rthoeonal cartesian plobal co-ordinates.

x' ,y' ,z' Orthogonal axes other than the global cartesian
axes.
ox' ,oy' Generators of the hypar surface.
02.' Normal to the hypar surface at point o.
-
- -
ox,oy,oz Local orthogonal clement axes.
xo,yo,zo Cartesian co-ordinates of the center point o
of an element.
Length of hypar quadrant in x-direction, inch.
B Length of hypar quadrant in y-direction, inch.
c TUsc of hypar quadrant) inch.
A.,B.,C.
l l 1
Dimensions of the i-th quadrant.
x.l x-co-ordinate of the local oriein of a hypar
quadrant with respect to the global co-or-
dinate systerr:.
-
y. y-co-ordinate of t~e local origin of a ~ypar
1
quadrant with respect to the global co-or-
dinate systerr..
a Length of the eleMent in x-direction, inch.
b Length of the element in y-direction, inch.
""-t. s-.;.... t "'-Ce. i!(.... ' ~)
Derivatives of '}..with respect to x and y.
Displacement along x-axis, inch.
v Displacement along y-axis, inch.
w Displacement along z-axis, inch.
~otation about x-axis, radian.
Rotation about y-axis, radian.
Rate of change of rotation~ radians/inch.
-1-
-Z·
w ,w Difforcnciation of w-displaceMcnt with respect

'x ,y to x and y.
u. ,v. ,w. Displacements of the i-th node.
1 1 1
Generalized displacement at the i-th node.

Angle of twist of a beam at the i-th end, radian.
Rate of change of anple of twist at the i-th
end, radians/inch.
w-displacement at the point a.
0 Twist rotation of a beam, radian.
e, Rate of change of t~.;ist rotation.
radians/ inch.
Average axial displacement of a beam section
measured at its centroid, inch.
Displacement of the shear center measured along
y-axis, inch.
111
··s Displacement of the shear center measured along
z- axis, inch.
Increment in the displace~ents due to the
application of the incremental load 6p.
Strain in x-direction.
Strain in y-dircction.
Shearinp, strain in the x-y plane.
Strains measured along x' nnd y' axes
Bending curvature in x-direction, inch- 1 .
-1
Bending curvature in y-direction, inch •
. .
Tw1st1ng . h-1 .
curvature, inc
Stress in x-direction, lb/inch 2 .
ay Stress in y-direction, lb/inch 2 .
TXY Shearinp stress, lbs/inch 2 .

a x ''ay ' ,Tx 'y' Stresses measured along x' and y' axes.
Force per unit length alonr the x-direction,
lb/inch.
Force per unit lcnfth along the y-dircction,
lb/inch.
Shear force, lb/inch.
Moment about y-axis, inch· lb/inch.
.' y
Moment about x-axis, inch-lb/inch .
M
xy Twisting moment about x and y axis, inch-lb/inch.
E Young•s Podulus, lb/inch 2 .
G Shear modulus# lb/inch 2 .
-I
C:i 1·.:,.~er.v<" ~cv.'O..: "-\ cf ~~ \-::: G ""'1lt) / \\-1 / lAcl-..
\) Poisson's ratio.
r'x Modulus of elasticity in x-circction, lb/inch 2 •
f;
y
Modulus of elasticity in y-direction, lb/inch 2 •
D Modulus of elasticity due to Poisson's effect,
1 lb/ inch 2 •
Effective shear modulus, lb/inch2.
Poisson's ratios in x and y-directions re-
spectively for the equivalent orthotropic plate.
a natio of shear rigidity of a corrugated deck to
that of a flat deck with the same thickness.
Extensional rigidity of the deck along the x-
dircction, lb/inch.
Bxtensional rigidity of the deck along the y-
direction, lb/inch.
Inplane shear ripidity of the deck, lb/inch.
Inplane coupling rigidity of the deck accounting
for Poisson's effect~ lb/inch.
Bending rigidity of the deck in x-direction,
lb-inch 2 /inch.
Bending rigidity of the deck in y-direction,
lb-inch 2 /inch.
Torsional rigidities of the deck, lb-inch·/inch.
-4-
~irirlity due to Poisson's effect, lh-inc~ 2 /inch.

Bcndin~ of an isotropic deck,
lb· i11cn 2 /inch.
rigidity
t Thickness of the deck, inch.

Length of a basic repetitive unit of the deck,
inch.
s Developed length of a basic repetitive unit of
the deck, inch.
Thickness of the base plate in a cellular
deck inch.
p
Thickness of the hat section in a cellular

deck, inch.
. h2 .
Area of deck alonr x-direction, 1nc
. hz .
Area of deck alonr y-dirccticn, inc
.
Area enc 1 osec1 b y t h e ce 11u 1 ar portion, . hz .
inc
- tl
~ 1 oment of inertia of thr decl: section, = IT ,
inch .. I inch.
r··o:ment of inertia of a repetitive deck unit
ahout the x-axis, inch 4 /inch.
I I
Mowent of inertia of a repetitive deck unit
y
about the lire of connection parallel to the
x-axis, inc~~/inch.
I
ed Ef fectivc mo~ent of inertia of a corrugated
deck, inch .. /inch.
r \~!arpinr constant of the deck., inch6 /inch.
n
··red neduction factor for the torsional rigidity of
a cellular deck.
:!umber of in terconnccted decks.
.
Cross sectiona . h2.
1 area o f t h e b earn, inc
~omcnt of inertia of a beam about the y-axis,
inch ...
l Ioment of inertia oi a beam about the z-axis,
inch ....
J St. ''enant torsional constant., inch4.
Harpin~ constant of a beam, inch6.
Eccentricity of t~c centroid of the cross-
scction of the bea~ in the y-direction fro~
the shell surface, inch.
Eccentricity of the centroid of the cross-
scction of the beam in the z-direction from
Eccentricity of the shear center of the cross
section of the beam in the y-direction from
the shell surface. inch.
section of the beam in the z-direction from
the shell surface. inch.
T
F
Coefficient for torsional fixity between the
deck and the edge me~ber.
The coefficient for inplanc fixity bet11een the
deck and the edge member.
s.l Stiffness of the i-th sprinp.
qo Uniformly distributed load, lb/inch 2 •
c5p Load increment used in load incre~entation
method, lb/inch 2 •
Eigenvalue for the critical buckling load.
lJ Str.ain energy. i::lch-lb.
v r~tential energy of the applied loads, inch-1~.
Totcl potential energy of a system, inch-lb.

Strain energy due to bending of deck, inch-lb.
Strain energy d\!e to the membrane action of the
deck, inch-lb.
uwarp Strain energy due to the warpinr, of the deck, inct-lb
ubb Strain energy of a beam, inch-lb.

Vu Potential energy of the in-plane forces, inch-lb.
[k] Stiffness matrix.
k .. Element of the stiffness matrix in the i-th
1)
row and the j-th column.
-6-
[kl curved EleIDcnt stiffness matrix for the curved elewent.

[k]flat Element stiffness matrix for the flat clement.
[k]m Element membrane stiffness matrix.
[k]bm Flexural and membrane coupling element stiff-
ness matrix.
[k]bb Element stiffness l'latrix for plate bendinp.
[k]bbm Element stiffness ~atrix for u-displacemcnt.
[k]et Stiffness matrix of the plate element in the
global co-ordinate.
Effective stiffness matrix of the supportinp
edge member with respect to the eccentric axes.
[PJ Load vector.
[6] Generalized displacement vector.
[T]s Translational transformation matrix for a
beam element.
[T] Component transformation matrices for the
plate element.
[N] Incremental matrix.
[k]eff Effective s·;.iffness of the deck after inclusion
of the instability effects.
[N]c Incremental stiffness for the flexural dis-
placements (w).
[N]global Incremental ~atrix for deck in the global

c·ordinate system.
N.l j ; Element of the incremental matrix in the i-th
:row and j·th column.
notational transformation matrix for the beam
element.
tx,f.y,tz
"'x ' my ' "' z Direction cosines of the local element axes
nx,ny,nz ox, oy and oz.
CliAPTE r. I
Ii'JTRODUCTION
I. 1. HYPJl..R ROOFS
- ----
T!-ic hyperbolic paraboloid shell ro6f, like any other
form of shell is one of the types of construction that makes
efficient use of materials by depending primarily upon the
form or shape for strength rather than on ~ass. The doubly
curved surface of a hypar shell is composed of straight lines
in two directions (Fig. 1.1). From the construction point of
view, this property is very attractive. It facilitates the
use of straight members for formwork and reinforcing steel in
the case of concrete hypars. This very feature also allows
the use of light gage steel deck panels, which could be easily
warped to the required degree to form the hypar surface.
The hypar surface shown in Fig. 1.1, can be pener-
atcd in two ways 1 : (1) The surface can be defined by woving
a convex parabola one in a direction parallel to itself, over
a concave parabola BDA. The parabola ODC lies in the plane
perpendicular to that of BDA. (2) The surface can also be
defined as a warped parallelogram. The surface can be gener-
ated (Fir. 1.1) by movinr along y-axis, a strai~1t line that
remains parallel to the xz-plane at all tiwes but pivots while
sliding alonr. the straight line AC. Physically the surface
can be visualized as a warped parallelogram OBCA, obtained by
depressing the corner H through a distance CH. By means of
s i mi 1 a r tr i an g I cs {Fig . 1. 1) ~ it can b c ea s i 1 y sh own that the
surface equation is,
PQ =Zc Kxy where K • OA CHx OB
The surface is called hyperbolic paraboloid because

any plane parallel to the xy-plane, intersects it in a hyper-
bola; whereas a plane perpendicular to ODC intersects it in
a parabola.
For simplicity, tho structural action of a hypar
can be visualized as a net of intersecting arches and cables.
The convex parabolas (arches) parallel to ODC carry compres-
sive stresses, whereas the concave parabolas (cables) parallel
to ADB carry tensile stresses. This implies· that the element
I is in a state of biaxial stress, compression parallel to
the arches and tension parallel to the cables. On the other
hand, the element II is in a state of pure shear. In the in-
terior, the membrane shear is carried by the shell. Along the
free edges, stiffening edge members are usually provided to
sustain the membrane action. These edge members themselves
carry gradually increasing tensile or compressive forces de-

pcndin~ upon the geometry of the structure.
By the combination of a basic hypar unit, such as
sh.own in Fig. 1.1, different elegant hypar structures can be
built. Four such configurations are shown in Fig. 1.2. The
edge members are provided to build up the membrane action and
also to stiffen the structure. Tension tie rods are commonly
employed to balance the horizontal reactions between the low
corners.
- ~-
Because of its architectural beauty, case of con-

struction and ability to provide large coluwn-free working
space, the hyperbolic paraboloid shell has been used for in-
dustrinl plants, churches, assembly halls, etc. In HomhasR 2 ,
the hypar structure was used as a footine on low bearing capa-
city soil. A 225-fect double cantilever hypar roof is under
construction at Los Angeles for the AJT1erican Airlines jet
hangar. The roof uses a cellular form of deck. This struc-
ture may very well prove to be the forerunner of many more
similar structures.
I. 2. LITERATURE REVIF.hl'
Like any other shell, hypars carry load hy both Jnem-

branc and bcn<ling actions. The me~branc theory, as indicated
before, results in a state of pure shear. The li~itations of
this theory verc realized by tlle most investip-ators and the
necessity of probing into the bending behavior of the shells
became apparent.
The shallow shell theory of ~.~arguerre 3 and Vlassov 4
is often used to analyze the bendinp action. Though this theory
is approximate, it is considered fairly accurate for a shell
surface where the slopes of the tangents are very small co~pared
to unity. Two basic approaches were used to formulate the

shallow shell theory.
In the first approach, two fourth-order coupled par-
tial differential equations in ter~s of normal displacement w
and Airy-stress function F are formed. ~eissner 5 using this
approach, determined the buckling load of • uniformly loaded
isotropic hypar with moment-free rigid cdgesj with the edge mem-
bers of infinite axial rigidity but negligible bending rigidity
in planes tangential to the shell. Apeland and Popov 6 reduced
these two equations to a single eight-order differential equa-
tion. Using Levy-type boundary conditions (with two opposite
edges knife edge supported) they tried to establish the effect
of cdrc disturbances in the sa~e way as that for cylindrical
shells. Their important conclusion was that the effect of the
edge moment does not die off very rapidly in the case of hypar
shells.
The formulation in terris of Jliiddle-surface displace-
ment u-v-w, results in three coupled partial differential equa-
tions, two second-order (u-v) and one fourth-order (w). Sal-
vadori and Bleich 7 using Vlassov's shallow shell equations fol-
lowed this approach. Assuming u~v•O all over the middle sur-
f ace, the fourth-order differential equation reduced to that
of a plate on an elastic foundation.

However, it must be emphasized that in order to obtain
the solutions to these mathematically complicated equations,
sirr.plifying assumptions were made. The choice of boundary con-
ditions was dictated by the possibility of obtaining solutions
rather than simulating the exact boundary conditions in a phys-
ical model.
The shortcomings of the classical solutions resulted
in the realization of the importance of numerical approaches
for the solution of these differential equations.
Chetty and Tottenha1!1 8 applied a variational method
.l l .
for the analysis of shal 10\·' shell equations ('~-F) . However,

the cnoice of approximating functions li~ited its applicability
to the specific boundary conditions. Besides, non-classical
boundary conditions presented serious difficulties.
The numerical scheme of finite difference provided a
very useful tool in the solution of these equations. Das
Gupta 9 , ~1irza 10 , Russell and Gerstle 11 , 12 using the ~-P approach
applied this method to different hypar structures using mean-
ingful boundary conditions. The edge members were also incor-
porated in their analysis. Everybody used the classical beaw
theory. Mirza solved an umbrella shell. !lowever, the magni-
tude of the corner deflections obtained for an umbrella shell
and the boundary conditions used along the line of sy1"'netry,
raise serious doubts about the validity of the method 13 .
P.ussel ·And·Gcr9tlel2 J'llainly analyzed two-corner and four-corner
supported hypars. ~on-dimensionalited desipn parameters were
provided. Tre main contribution of their work was to show the
importance of the line-load along the edge members. Croll and
Scrivener 14 , 15 used the u-v-w formulation. The effect of the
eccentric connection of the beam to the shell is discussed. One
of the important features of their work is a complete discussion
of the convergence characteristic of the solution with relation
to the relative proportions of the shell and the edge member
stiffnesses. A comprehensive review of the above mentioned
method is presented by Brebbia 16 .
The finite clement method, which is nothing else but
the matrix formulation of Rayleir.h~~itz method of variational
principle, was successfully employed by several workers. The
case with which this method can handle realistic boundary con-
ditions, made this method very suitable for hypars. It is
also believed that the variational principle used in the finite
element method will yield better results than the finite dif-
ference method because it involves integration rather than dif-
ferentiation as used in the latter method.
Connor and Brebbia 17 using shallow shell approxima-
tion, formulate<l the stiffness matrix for a thin shell curved
c1cmont, rectangular in plan. Similar formulations w~re worked
out by ncak 18 and Parker 19 • Pecknold and SchnobrichZO,ll work-
inr alonr the saMc lines, extended the work to the skewed
shallow shells. All these workers used linear displacement
field for u, v displacements. The Major difference was the
disp laccmcnt field they used for the normal displaccwent "'.
Deak 18 and Peclmold 20 both used the Birkhoff-Gara-
bedian interpolation formula. Connor 17 used a twelve-term poly-
no111ial "~hcreas Parker19 used the Lagrange interpolation formula.
Parker extended the work to orthotropic light ~a~e steel hypars
and compared the results with the experiments. None of the
Refs. 17, 18, 20, 21 mentioned any correlation with experi-
mental results but were content with the comparison to solu-
tions obtained by other nuwerical methods.
The buckling and nonlinear analysis of hypars are
also reported in the literature. Ralston 22 continued
Reissner's 5 wor}; by investigating the buckling of a hypar ·
under its ol"D weight. Dayaratnam and Gcrstle 23 presented a
-11·
solution to the buckling problem of hypars with edre beams,

simply supported at their corners. The in-plane displacements
u and v were assumed equal to zero. A double sine series de-
fleeted shape was assumed and the total potential energy of
the system was minimized using Ritz procedure. The critical
load was determined by equating the determinant of the result-
ing matrix to zero. It was concluded that for all positive
values of the ratio of bending rigidity of the edr:e ~ember to
tr.c deck, the deck buckling ah:ays preceeded the edge mefl!ber
buckling. The study in Ref. 19, indicates that for a very flex-
ible edge member, the possibility of overall bucklinr nrior to
the deck buckling does exist. The erroneous conclusion of ref.
23 was reached because of two reasons:
(l) A non-compatible displacement field between the shell
and the edge beams.
(ii) In determining the cigcnvalu~ the off diagonal terms
were neglected.
These points are discussed at length in Ref. 19. Muskat24
studied the buckling of hy1t41rs with corrugated orthotropic
deck. He proposed a method for estimating edge member buckling
based on the non-cpnser.vative buckling of an isolated member
loaded tangentially {see Section VI.8).
Brebbia and Connor 25 presented a consistent finite-

elemcnt displacement formulation applicable to the shallow
shell elements using the ;~ewton-Raphson iteration scheme, by
linearizing the incremental equation. The load deflection
curve for fixed hypar was presented.
-14--
ti. large number of experimental studies have heen re-

ported for hypar shells. f major portion of the experiments
were carried out to measure deflections and stresses on small
to medium scale models. The results were correlated with the
approximate theories and by changine the design parameters,
certain predictions on the overall behavior of the hypar shells
were rnade. Rowe and r·irza 26 tested plastic Models with two
adjacent edges fixed and the other two free. The effect of
rise to span ratio was studied on the free corner deflections
and axial strains. By changing the depth of edge beams along
the free edges, it was concluded that by increasing the depth
of the edge beam both the axial strains and the vertical de-
flections are decreased in the shell portion. Rowe 27 also
tested medium to small scale concrete models and used unsy~
mctrical loading. In order to study the ultimate load carry-

ing capacity of the umbrella shell hypars, an experimental re-
search program is underway at the Cornell University, where
s~all scale concrete umbrella shells are being tested to
. f a1lurc
. 65 .
f!uskat 24 and Leet 28 tested small scale models to

determine the buckling characteristics. Leet tested plastic
models subjected to uniform normal load. The effect of imper-
fection was studied on the fixed shells. Edge beams were used
to study the effects of different edge conditions on stresses,
deflections, shell buckling and overall buckling. The effect
of different boundary conditions was not very pronounced on the
shell buckling. By studying the effect of differcn t beam
-15:".'
siz.es Leet concluded that the cross-sectiona.l areas of the edge

beams ~'ave a significant effect on the deck bucJ,.ling.
~~ost of the experiments were conducted with iriedium

scale models. Ref. 19 has listed most of them. Bertaro and
Choi 29 tested 8 'x8' model. The mode 1 with edge beams a.nd two
diagonally opposite corners supported was chemically prc-
strcsscd by using expansive cement. The model was tested in
the inverted position using ~ir pressure loading. Deflection
profiles, crack patterns, and stresses were presented.
In the last 8 years or so, the use of light gage

steel decks as a hypar shell has gained some ~omentum.
~cDermott 3 0 tested two 8'x8' orthotropic light gage steel,

saddle-shaped (Fig. 1.Zb) models. In the first case, the
steel deck was welded to the edge members whereas in the second
case the steel deck was glued. The rubberized canvas bags were
pressurized by water in the first case whereas air was usecl in
the second case. He also tested a large-scale model with a single
layer standard corrugated deck. The edge members consisted of
built-up sections. The loading was applied with sand bag•. Strains
and deflections were measured.
Yu and
Kriz 31 tested a concrete inverted umbrella shell 24'x24' in
plan, in which upturned edge beams were used. The symmetri-
cal and unsymmetrical loading was simulated by the discrete
loads. The measured strains and deflections were presented.
Three large scale tests were conducted for hypars using
orthotropic deck as a shell surface:
Nilson 3 Z tested a lS'xlS' hypar quadrant with simu-
lated boundary conditions of the adjacent quadrant. A single
layer of cellular deck was connected to the channels, which
were used as the edge nembers, by ~eans of a warped plate. Uni-
form loading was applied by 25 jacks. The load deflection
curve and the measured m~mbrane stresses were reported.
Two large scale hypar models SO'x30' in plan, were
tested rcccntly 33 . Two different cellular orthotropic single
layer decks were used in each case. 14 WF sections were used
as the edge members, uhich were allowed to move freely in the
plane of the hypa.r but were supported against the vertical
movement. The normal uniformly distributed load was applied by
creating a vacuum in the enc.lo!:;ed chamber.
1.3. SCOPE AND "BJECTIVES OF THE PRESENT INVESTIGATION

... --- ---
The main purpose o~ the present investigation was to determine
the deflections, stresses, bcukling and collapse loads of light gage
steel hypar shell roofs and to provide design information. An analytical
procedure was developed so that it could be extended to include the
analysiB of hypar,1 with various support conditions as well as hypars
subjected to partl~l loadings.
The finite element method was selected because of its
\'ersati lity. ThE'! entire prc~ect was •PP"~oachP.d from an engineering
point of vie~. ThP. validity of the ~t·hod was established by

comparing the theoretical and experimental results for different
kinds of hypar structures.
Two separate computer programs ~ere written, (1) For
stiffness analysis; (2) For instability analysis.
(1) Stiffness Analysis
A linear elastic analysis of the structure per-
formed and the deflections, the edge member and deck stresses
~e~ v•r.A~t~
computed. The program can handle the following
(a) Different types of orthotropic decks.
(b) ~ifferent configurations of the hypar
structure.
(c) flealistic physical boundary conditions 1
suci, as eccentric connections of the edf1e

members and disc on t inui ty between the
deck and the supporting edge niembers.
(d) Different loading conditions such as uni-
form, unsymmetrical, line loads, etc.
(2) Instability Analysis r'--f ~ \'A~~i(\~~
\J4f ~ Cu\°" '1 ..
A linear load incrementation method used". The
~
effect of the in-plane forces Nx, Ny and Nxy1 included.
The experimental part of the Investigation Included the de-
termination of the effectiv~ ~~!v::l<:!"

rlgldl ties of standard corrugated
'o\Jdd~,
decks and the determination of uef J'.or:'c':f,cns, stresses, and
Q(tc( S'.\M~'--s.\\c..p~ -"\f'4fS
loads hi\ inverted umbrellas c,.mstructed of corrugated steel decking
/\
and tubular edge beams.
l 7a
I. 4 ACKNOWL~DGMENTS
1his investigation was sponsored by the American Iron and

Steel Institute. The financial assistance and the technical
guid~nce provided by the Institute is gratefully acknowledged.
We thank the Task Conunittee on Shell Structures, and especially
its chairman, Dr. J. B. Scalzi, for their interest in this
project.
Special thanks are due to Dr. George Winter, principal
investigator of the project, for his great interest in the
investigation and for the many wise suggestions throughout
the work. His insight into structural behavior often steered
the study back from tangents and away from pitfalls.
The nonconservative stability analysis was carried out
by Dr. R. Muskat. His and Mr. A. Banerjee' a help with the
experimental program was very valuable.
CrIJ'PTIJr II
DECK PR0PERTI ES
I I . 1. l1JTYHlDUCTION
A material in which resistance to mechanical actions

is different in different directions, is called anisotropic.
Orthotropy is a special case of anisotropy, where the body
possc~ses elastic properties which are symmetric about three
mutually perpendicular planes.
Orthotropy can be further classified into two cate-
r.ories namely, natural and rcoroetric. The natural orthotropy
is a result of the material property itself. A classical
example of natural orthotropy is timber, where the ~odulus of
elasticity, alonr the direction of its grain in tension, is
su~stantially hi~her than the correspondinf modulus in the
<lirection perpendicular to it.
In the reometric type of orthotropy, the difference
in elastic properties in the perpendicular direction, as shown
in Pir. 2.1, is due to the reornetrical confir.uration of the
structural eleJT1cnt even thouph it is made up of an isotropic
homogeneous ~aterial. Different types of decks helonginr to
this catccory arc shown in Fig. 2.2.
tn both the cases, the definition of elastic con-
stants in two ~utually perpendicular directions is required.
However, geometric orthotropy is of particular interest for
the hypar structure dealt with in this work.
-18-
-19-
In the following analysis the decks, shown in Fir.

2.2, arc replaced hy equivalent orthotropic plates, having the
same physical properties as the decks, such as extensional and
bending stiffnesses. It is impossible to achieve a complete
cqui valence between the actual physical system and. the ideal-
ized orthotropic plate, in all respects such as strain energy,
deflections or moments at different points under different
criteria. The properties of the idealized system can be deter-
~inc<l either b;' equivalence of stiffness or equivalence of
strain energy between the idealized and the physical syste~ 34 .
The equivalence of stiffness is established by equatine only
the deformations hetween the actual and the idealized system.
The equivalence of strain energy is obtained by equating the
work done by the internal forces in both the systems 1ihen sub-
jected to identical loading and boundary conditions.
The elastic constant for the idealized plate material
is assumed to be the same as t~at of the parent material. The
properties calculated depend only upon the direction consid-
ered and not on the position of the corresponding point on the
actual deck. The orthotropic plate theory is applicable to
the decks, sh01m in Fig. 2.2, provided the ratio of the dimen-
sions of the repetitive unit (1) and the overall span of the
deck, is very small, i.e.; <<l.
Il.2. ELASTIC PPOPERTIES Q.E DECK

In the case of a geometrical orthotropy, as present
in corrugated or closed formed decks, in order to calculate
elastic properties in two mutually perpendicular directions,
-20-
it is necessary to separate the in-plane membrane and the

bending action.
A. MEMBRANE PROPERTIES
The principal directions of elasticity in an ortho-
tropic deck are those along which the extensional rigidities
are either minimum or maximum. When the axes x and y coincide
with its principal direction of elasticity, the equation of gen-
eralized Hooke's law for plane stress-strain can be written as
x - ~xy
0
EX • °E:' CJ 2·la
x y y
E
y • ~C
x x
+ ~y 2-lb
xy -~
y 2-lc
c;ff
Solving the equations for the stresses we get,
C1x Ex vxyEx 0
1
C1
y • vyxEy Ey 0 2-2
Tf·vxyvyxl
't'xy 0 0 {l~vxy"yx)Geff
The terms "xy and "yx represent the coupling effect o the
actions (stress or strain) applied in two perpendicular direc-
tions. From Fig. 2.2, it is obvious that these coefficients
(vxy and vyx) cannot be equal. For example, consider the
cellular deck shown in Fig. 2-2c, where a uniform stress
applied along the bottom plate of the deck in the x-di~ection
will produce a negligibly small strain in the y-direction, in

the hat portion. Whereas a uniform stress applied over the
entire cross-section along the y-direction will produce a
-21-
strain in the bottom plate proportional to the ~aterial con-

stant v, (Poisson's Ratio). Though vxy 'I vyx' by Maxwcll-
Betti reciprocal theorem the following relationship holds:
Ex"xy • "yxEy • El 2-3
As explained by tl:~ physical behavior above, "xy is

equal to v and he~ce,
\)
yx
. 2-4
The value of Geff is giver by,

Geff • aG 2-5
where a is called the relative shear rigidity factor, the

value of which depends upon a number of factors. A co~plete
discussion of a, together with a description of the experi-

mental method of determination of a, is given in Section II-3.
The methods for the determination of constants Ex, Ey, "xy
and "yx' for both closed and open form decks, are given in
Appendix A.
Since the decks are ide_alized as uniform orthotropic
plates of constant thickness, it is convenient to express the
me~brane stiffness constants and the forces in terms of their
thicknesses. t"ul tip lying the first row of Eq. 2-2,
E t E1t
0
xt • (1-~"yx) £x + (1-vvyxl £Y 2-6
• iJ
x •
xt EX + Blt £y E 2-7
- ~
Fx t E1 t
where Nx • axt; Ext • (1-vv ); Elt • (l·vvyxl
yx
. 2-8
The other rows can be modified similarly (see Eq. 3·13).

Depending upon the method of connections the elastic
-22-
constants, calculated above purely on the basis of geometrical

configuration, thickness and material properties, need modifi-
cations.
The elastic constant Ex for an open deck is given
by, (Appendix A)
I0 E
Ex= r ' 2-9
y
The value of Iy' depends upon the line of application of the
load. ~ substantial reduction in the value of Ex results from
the eccentric connection. The cellular decks (Fig. 2·2c) are
usually connected to the supporting rnernher alon~ the bottom
plate leaving the hat portions free. /\.s shown in Fip.:. 2-3,
the forces applied alone the bottom plate will be partly re-
sisted by the vertical sides of the hat. Depending upon the
joint efficiency between the hat and the botto~ nlatc, the
effective area of the hat section in resisting the forces ay
will vary. In the absence of the test data, it will be con-
servative to consider the bottom plate.only as being effective
in resisting the in-plane forces ay. In the computer programs,
a provision is made to include the effective width of the web
plates of t~c hat.
B. BENDING PROPERTIES
The bending rigidities, r:x, "Y' Dxy and o1 , for a
geometrically orthotropic plate cannot be ohtained diTectly
from t~e directional elastic constants yiven in Eq. 2-2. The
relationship between the bending rigidities and the moments,
is given in Eq. 3·16. In the case of a· li~ht-gage orthotropic
deck (corrueated or cellular) DY >> so.nx. This pro~erty can
-2 3-
be used to approxiwate properties such as n1 and vyx· Since

the twistinp ripidity Dxy is not equal to ~yx (see Appendix
A), the averaRe of the directional twisting rigidities is
use<l 3 5.
In th~ case of an open f orrned deck there is a
tendency for the warping of the surface. The method of cal-
culatin~ the warping constant r, based on the assumption that
an individual unit 35 of a plate twists about its center of
rotation is well known 36 . However, considering the plate as
a whole unit, the plane of twistinr changes, depending upon
the distance of a point from the support. The extent to which
this warping restraint alters the behavior of the deck is not
very clear.
The moment of inertia used for the computation of
Dy is calculated at t~e centroid of the repetitive section.
The effect of eccentric location of the ribs or hats in the
case of decks, on the bending properties is discussed in great
detail by Massonnet and Bar~s 37 . However its use in practical
problems is difficult. Due to the local buckled form for high
wi<lth to thickness ratio for a compression flange 39 {Fig. 2.4),
the effective moment of inertia for both deflections and stress
analysis is reduced. The reduction in the mo~ent of inertia
is a function of the str~ss level. This factor may be of im-
portance in the cellular deck. Depending upon this reduction
of the effective section, a second analysis of the hypar shells
~ay be necessary though in most cases the stresses are small.
The procedure of calculating led is discussed in ~efs. 38, 39.
The mechanis~ of reduction in bending rigidities due
to the connection hetween the different panels is not very
clear. Experimental results on the determination of these
rigidities for a continuous single panel have been reported 40 •
The bending rir,idities Dx and Dxy are particularly affected by
the discontinuities between connections of different panels.
\~~~di\
Exalflination of structural problems solved shows
that the magnitude of both these constants (Dx and Dxy) is so
small for an open formed deck that the reduction even of the
order of 1/10> does not affect the results significantly
(<<5% variation).
The properties for two or ~ore decks placed per~en
dicularly can be calculated by ad~ing the corresponding direc-

tional constants.
C. ELASTIC PP0PERTIES FOR ARBITPARILY ORIENTED
CO-nPDINATE AXES
1
•rhen the structural axes and the principal axes of
orthotropy coincide, the elastic properties of the deck cal-
culated on the basis of principal axes can be directly used in
the analysis without any modifications. As explained in the
Section I.l, the structural behavior of a hypar can be broken
up into mutually orthogonal arches and cables. When the decks
are placed alonp the axes of arches and cables (Fig. 2.5), the
structural axes x and y, do not coincide with the principal
axes of orthotropy x' and y'. The elastic constants in terms
of the structural axes are expressed by the principle of work
equivalence 41 ,.; 2 . The membrane strain energy in two systems
-25-
of axes can be given by,

v• 1
! OXEX
+ 1
! OY£Y
+ 1
! TxyYxy 2-lOa
1 1 1
V' -= ! a x'cx ' + a '£ y '
!y + ! -r xy
' 'y xy
' ' 2-lOb
Expressing strains in terms of stresses in both the systeJT'ls

of axo~ and equating,
V' • V 2-lOc
one can obtain the equivalent elastic constants. The equi-
valent hending constants can be determined by equating the
bending strain energy along both axes (Eq. 3-17). The elastic
properties for an orthotropic material in terms of any arbi-
trarily oriented co-ordinate axes are given in detail by
Lckhn i tski i 41 •
I I. 3. ~HE.AR RIGIDITY OF OPTHOTROP~£ DECKS

The shear stiffness of an orthotropic plate, assum-
ing Nxy * Nyx (Section IIl.2 B), i~ given by the expression
, (~---....--Gr-:::::
Exyt =J...a. G i l t 2-11
when~ n is the numher u :".' interconnected decks, t is the thick-
ness of each deck and G is the shear modulus of the material.

a is the relative shear rigidity coefficient giving the ratio
of the shear stiffness of the actual orthotropic deck system

and the isotropic plates of thickness nt. In the case of an
isotropic plate, the vertical load on a hyp.ar is partly carried
by the membTane action in shear and theref!Jre the shear stiff-
ness of the deck is of utmost importance in its structural be-

havior. Experimental and the theoretical
was carried out by
,_.,.,
determination of a,
-26-
workers. The factor a depends upon a number of factors listed

below!
(a) The different shapes (corrugated sine form, N-type,
etc.) of diaphragms (Fi~. 2.2), show different resistances to
the in-plane shear loads. The shearing loads produce bending
and twistinp of the corrugations and also set up memhrane
stresses and shearing strains. On the basis of the assuw.ed
displaccrent field of the corrugations, energy stored due to
each of tt~esc above mentioned actions (bending, tt1!isting, etc.)
is reported in Refs. 43, 44.
(b) The spacing of the connectors. between the deck and
t:1c edge members, transverse to the corrugations have a pro-
nounced effect on the value of a, whereas the spacing along
the directions of the corrugations has a very minor effect.
The flat shear tests on a ~6-G. s. tSrndard corrugated~eck, 6'x6'
i..f\ ~\45. ~"'"t~(OA'~ VU) '-"~~ f"t('\Qt~o .
in plan, were conducted •Two tests,.._wi th the con-
nectors between the deck and the edge members at each valley
and one with the connectors spaced over every third valley.
The other factors, in the above tests, beinr, the same, the
value of the shear stiffness obtained for the forwer cases was
reported nearly three times as large as that of the latter.
(c) According to Luttre11 45 , the shear stiffness of a
panel also depends upon the length along the corrugations.
Accordion-like warping results due to the connection of the
diaphragm to the edge mel"lbers. It was found 45 that the length
of penetration of this warping is independent of the overall
span of the diaphragm and this warping reduces the shear stiff ..
-27-
ncss, particularly for short spans.

(d) According to Ref. 43, the seam-slip between the ad-
joininr deck panels and the connection of the deck to the edge
members contributes substantially to the shear flexibility.
~ot\cl ~Y\ ~~"pec-·.,W\.q (C.~\1\\).1
(c) As · for the liJT1.ited range of
thickness of panels studied, the shear rigidity increases
linearly with the effective thickness as given by Eq. 2-11.
However, consideration should be eiven to the connections be-
tween two or more decks and their connections to the edge 111em-
( V\ C.\\~ \I\\ .
bers. The results are reported -rite two layers
of corrugated decks placed perpendicular to each other were
connected to the edge members through the connection of the
lower deck (Pig. 2.6). The additional flexibility provided by
the lower deck corrugations, reduce the effectiveness of the
upper decl:. The position of the screws with res:)ect to the
direction of the applied shear also affected the stiffness of
w"s~..a-' ~
the shear panel . It · that only 33% increase in
Exyt was noted for two decks connected as shown in Fig. 2.6,
whereas the value of nt doubled..
All the factors mentioned above are important for
open form decks as shown in Fig. 2.2a,b.p In the case of
(T'l _,f. // ~v:.{ 1vJvtN0 ld:
cellular decksd (Figs. 2.2c,d) the plate of the deck
is directly connected to the edge memhers. In view of the low
shear carryinr capacity of the out-of-plane hats, a major por-
;LX
tion of the shear is caTried by the olate. Knowinr the
seam-slip characteristics between the adjoining panels, the
shear stiffness for the cellular deck can be estimated conser-
-28-
ffL~t
vatively as that of the plate.
The depenc\ence of the relative shear rigidity co-
efficient a on some other factor such as connecticn with the
intermediate purlins, etc., discussed in detail in Pef. 43.
I I. 4. DETER~1INATION OF THE EFFECTIVE SHEAR

---
P.IGinITY FOR ORTHOTROPIC HYPARS
The theoretical determination of shear rir,idity co-
efficient a, was done by equating the work done by the applied
shearing force with the strain energy stored in the deck panels
due to deformations. In Ref. 43, an excellent correlation be-
tween the theoretical and experimental results was reported.
Since a major contribution to the shear flexibility was from
the connection between the deck panels 43 and the connections of
dects to the edge ~embers, a previous knowledge of the seam-
slip characteristic is required. Based on an experimental in-
vestigation, the method for the determination of the shear
rigidity from flat shear test is given in Pef. 45.
The next question arises as to whether the relative
shear rigidity determined by the flat shear tnst can be di-
rectly used for hypar structures without any modification. To
correlate the shear riridity coefficient a determined by the
flat shear tests and that of the warped deck in the actual
hypar surface, twelve saddle-shaped hypars, uniformly loaded,
with tubular edge Pembers of 3•~ diameter and j'~ thickness )
rc...~v"
(vertically supported all around) were tested The
lower corneTS were connecte d b y 2!ln x 4
lu tie
. b ars, in
. or d er to
restrict the horizontal spreading. The tests were conducted

-29-
for both single and double decks with varying rise to span
ratios.
It was found that for the lower values of a (<0.12),
the center deflection of all-supported hypars was proportional
to the value of a. Both single and double deck hypars were
tested for three different rise to span ratios (1/8, 1/5, 1/3).
The theory developed in Chapter III correctly predicts the
effect of a and rise independently on the central defl~ction
o0 (Key Sketch, Table II.l, II.2). The value of a in the

actual hypar surface was interpolated from the structures
analyzeJ·with different assumed a values for the constant Tise
to span ratio.
The results obtained for the effective value of a
arc compared with the results given by Parker 19 , who aralyzed
the same shells on the basis of plate on an elastic foundation
(u=v=O) and did not include the effect of the tie bar. For
both single and double deck, a certain amount of increase in
the value of a was noted for the low rise to span ratio (1/8)
hut with the increase of the ratio, the value of effective a
reduces. The results obtained by direct interpolation for
sinp,le deck hypars appear quite reasonable (Table II.l). The
direct interpolation if applied to the double dbcks, results
in extremely low effective value of a particularly for the
ratio of l/S and 1/3 (Table II.2). The fallacy in the method
of interpolation can be explained as follows:
Al 1 tt.e test results marked with an index (I) ( 821 (I))
were the tests where two transversely placed corrugated decks
- 30-
were inter-connected l·rhercas in the rest of the tests, the top

deck was connected only along its periphery to the bottom deck
~1ich in turn was connected to the edge member. The center
deflections obtained for Case (I) is roughly 20-30% less than
that where the decks were not inter-connected. In the mathc-
matical model there is a complete coupling between the in-plane
displacements of u-v and the normal displacement w (see Chapter
III). Secondly when two decks are placed on top of each other
in the analysis, the in-plane stiffnesses of the top deck f xt'
Eyt~ Exyt and Elt arc assumed to be fully effective in re-
sistin~ the load.
In case of the decks only connected at the edges,
the deviation between the mathematical and physical model is
very drastic, and therefore the deflections given by the anal-
ysis are very 10\\1. '·'·!ith these considerations in view, a cer-
tain amount of discretion must be used in estimating the values
. w~
of a. The results ob taine d b y t h e analysis compared with
the test results. The resultin~ reduction in values of a
can be attributed to the warping of the surf ace and the deck
and the edge member connections. These factors are common to
both single and double decks (Figs. 2.7, 2.8).
r~ether the increase in the effective relative shear
rigidity for the lol·T rise to span ratio of 1/8, is an inherent
behavior of the hypar or whether it is the shortcoming of the
theory in t~e region of transition between flat plate and warped
hypar surface is not very clear. It will be conservative to
neglect any increase in the value of a obtained from flat shear
test. The importance of the values of a and the recommended
~
reductions further discussed in Chapters IV and v.

CIIAPTER III
FINITE ELEI'.ENT rrETHOD FOR STIFFNESS Al~D STRESS ANALYSIS
III.I. INTRODUCTION
As discussed in Chapter I, the solution of the shal-
low shell equation for realistic boundary conditions is an
extre~ely complicated mathematical proposition. This necessi-
tates the use of numerical methods. The finite element method
Has chosen because of its versatality in handling realistic
boundary conditions, different structural configurations, ortho-
tropic deck materials and any forms of loading, with ease.
The method has also demonstrated good convergence character-
istics.
The finite ele~ent method based on the stiffness
analysis uses the principle of minimum potential energy. The
total potential enerr,y of an elastic system, for a geo~etri
cally admissible state can he represented by,

ct> =u + v 3-1
1fhcre U is the strain energy storerl in the system and V is the

potential energy of the applied loads. Both rand V are ex-
pressed in terms of displacements at the joints of an idealized
structure. lJ is a quadratic function of the nodal displace-
ment. The principle of the minbmm potential energy states
that, ''The total potential energy is JT1inimurn 1 when an el as tic
body is in equilibrium" •
.!!_ •
a~.
a
ati:- (P+V) • 0 3-2
1 1
-31-
P.1 c:
-av
-- 3-3
()[\.
1
For the linear elastic analysis, these expressions

can be put in matrix form,
[P] • [K] [.1] 3-4
The eleMcnts of the stiffness matrix can be obtained by the

second differential of the strain energy,
3-5
In the case of a framework composed of linear mel'llbers such as

beams, struts, etc., the individual ele~ents are connected to
each other at their nodal points. ~ell-defined boundary con-
ditions at these joints enables one to solve the physical
problern without any difficulty. In this case, there is one to
one correspondence between the mathematical and the physical
model. However, in the case of a two-dimensional structural
medium such as a plate or a shell surface, the discrete ele-
ment approach does not give a one to one correspondonce between
the element used in the matrix analysis and the forces in the
actual surface. Jlere the entire structure is idealized into
discrete elements, curved or flat, connected to each other at
the nodal points. The displacements of the nodal points are
interpreted as those occurring at the correspondinr points in
the structure. The state of stress Bnd strain inside an ele-
ment is defined completely in terms of its nodal point defor-
~ations. The success of determinin~ the elastic properties of
an idealized structural element lies in the equivalence estab-
lished between the actual model and its equivalent discrete
-33-
model.
The linear elastic stiffness analysis consists of
four important steps:
(1) The formulation of the element stiffness matrix.
(2) The formulation of the master stiffness matrix for
the entire structure by assembling individual elements.
(3) The solution of Eq. 3-4 for the given boundary con-
ditions and loading.
(4) The interpretation of the deflected shape and the
computation of stresses and forces.
Two alternative stiffness formulation methods are
studied here:
Method 'a': The use of rectangular curved elements, based on
~
Method 'b': The actual shell surface is approximated by the
A
assemblage of flat rectangular elements.
III.2. ELEMENT STIFFNESS
The elements rectangular in plan are selected. These
elements are very simple to formulate and for the structure
under consideration, their limitation of application to the
rectilinear rectangular boundaries, is not considered to be of
any serious consequence.
As shown in Eq. 3-lc, the element stiffness matrix
can be derived from the strain energy U of an element, ex-
pressed in terms of an assumed displacement field.
A. DISPLACEMENT FUNCTIONS
The displacement fields assumed for the analysis are
-34-
as follows:
1
U = ao [(x-a)(y-b)ul - x(y-b)uz + xyu 3 - (x-a) yu 4 ] 3-6a
1
V =as· ((x-a) (y-h)vl - X(y·b)vz + xyv 3 - (x-a) yv 4 ] 3-6b
+ (3ax 2 -zx 3)(3ay 2 -2y 3 )w 3+(a 3 +2x 3 -3ax 2 )(3by 2 -2y 3 )w 4
- ax(x-a) 3 (b 3+2y 2 -3by 2 )ey 1 -a(x 3 -ax 2 )(b 3+2y 3 -3by 2 )ey 2
- a(x 3-ax 2)(3by 2 -2y 3)ey 3 -a(x-a) 2x(3by 2 -2y 3)ey 4
+ b(a 3+2x 3 -3ax 2)y(y-b) 2ex 1+b(3ax 2 -2x 3)y(y-h) 2ex 2
+ b(3ax 2 -zx 3)(y 3 -ty 2 )ex 3+h(a 3+zx 3 -3ax 2)(y 3 -by 2)ex 4
2 2 2 2
+ abxy(x-a) (y-b) exyl+abxy(x -ax)(y-b) exyZ
+ abxy(x 2 -ax)(y 2 -by)0xy 3+abxy(x-a) 2 (y 2 -hy)exy 4 1 3-6c
The same functions were used by Gallar,her 50 and

Yang 51 • but in their studies the x,y cartesian co-ordinates
rcrc replaced by the ff'.ore general curvilinear co-ordinates al
1..
and a2 . As seen here, the displaceMent field inside an ele-

men t is directly exp res 5 c·cl as the func-cion of its nodal dis-
place~ents rather than j11 terms of undetermined parameters.
/\n clement, as sholm in Fir- 3.1, has six deRrees of freedom
per nodal point and a total of 24 degrees ner elel'lent. The
displacements u,v,w,e..,._, and ey have a phyc;ical rneaninp. at each
node. The terJ11 exy .,...eprest::n-cs the tl·rist curvature ~!~y· Psing
the right cork-sere~ notation,
-35-
3-7
One may also note that the inplane rotation about the z-axis
is being omitted in this formulation.
The function for w, normal bending displacement, is
of the cubic order. The terms corresponding to the degrees of
freedom w, ex and ey are obtained by the cross product of the
corresponding terms for the beam function in x and y-directions.
In order to represent the constant strain corresponding to the
twisting term ~!ay i.e., the term 'xy', the additional degree
of freedom in the form of exy is added to the displacement.
The displacement functions are geometrically syMrnetri-
cal and include the constant strain and rigid body modes for
the flat plate. It is obvious that the assumed displacement
fields for u, v and w are not of the same order. Whereas those
for u and v are linear, as stated before, w displacement is
cubic. If the displacement fields of u and v were of the same
order as that of w, each node would have 12 degrees of freedom
thereby having a total of 48 der,rees per element. Besides
this, the linear edge member elements would have to be given
the same order of stiffness JT!atrix. The additional degrees of
freedom would involve more computational work and this effort
could only be justifi~d if good results, without sacrificing
the required degree of accuracy, could be attained with a fewer
number of elements.
Any combination of displacements which can be accoinp-
lished ~ithout straining the structure are called rigid body
displacement modes. These displacement modes can be easily
-36·
recognized in the case of linear members or flat plates whereas

their role is not easily recognized ifi the displace~ent field
for the curved shells. The assumed displace~ent field does not
include rigid body motion for the curved element. Cantin and
Clough 52 used the displacement field for a cylindrical shell
clement and added the appropriate rigid body motion terms in
the form of trignometric functions. It was shown that with the
inclusion of the rigid body motion terms, there was an improve-
ment in the rate of convergence. Haisler and Stricklin 53 and
Connor and Brebbia 17 , have shown that inspite of omitting the
rigid body motion terms with the refinement of the grid size,
the convergence is still insured.
Pecknold and Schnobrich 20 proposed the ~ost logical
method for the inclusion of the rigid body motion terms. It
~as sugyested that these terms should satisfy the ho~ogeneous
part of the strain displacement relationship used for the

curved ele1'1ent (Eq. 3-9,3-10). The inclusion of these terms
involves more computational effort but they see1'1 to have cer-
tain advantages, \·rhich are further discussed in Chapter IV.
B. CURVED ELEPENT
Strain Displacement ~elationship
The strain displacement relationship;used in the

curved element formulation are simplified according to the
shallow shell theory. The following assumptions are made:
(1) For a given surface defined by the equation, i•P(x,y),

the slopes of the surface z,x and z,y are considered negligible
in comparison with unity. In general, the shallow shell theory
will be quite accurate as long as z'x' z,y ~ 1/8, and often
accurate enough for practical purposes as long as z,x' z,y
5 1/2 5 ; though the second limit could be considered as too
liberal.
(2) The sides of a differential shell element, which are
orthogonal in the projected co-ordinate plane, are assumed to
be orthogonal in the plane of the ~iddle surface of the shell.
In other words, the geometry of the surface is approximated by
that of its projection on the co-ordinate plane.
(3) If the equation of the middle surface is of the
second order as is the case for a hypar, the assumption (1),
leads to the approximation that the curvatures of the surf ace
are constant.
The errors resulting from these assumptions increase
as the depth of the shell increases. For the shallow hypar
surface defined by the equation,
z • b xy 3-8
the linear strain displacement relationships for the thin

shells assume the following form 20117 :
The membrane strain displace~ent relationships:
3-9a
3-9b
= u ,y + v
tX
3-9c
The bending strain displ$ce~ent relationships:

Kx • -w ,xx 3-lOa
l(
y • -w
,yy 3-lOb
- 38-
21<
xy • -2w ,xy 3-lOc
Strain Energy
The strain energy for a typical element consists of
two parts: the membrane and the bendinf strain energy.
u • ub + um 3-11
The membrane strain energy for an element is given by 54 ,
3-12
The stress-strain relationship for an orthotropic material

(Fi~. 3.2) can be represented by .
TJ
·x 0
0 3-13
0 0 y :Jl:;
The method of coMputation for the above mentioned elastic con-

stants is discussed in Chapter II.
Using the strain displacement (Eq. 3-9, 3-10) and the
stress-strain relationship (Eq. 3-13), the membrane strain
energy (Fi~. 3.2) of an element can be expressed as follows:
1 2 3
4 s 6
+ Exyt[u2~Y + v2,x + Zu,yv,x]
7 8
c
- 4ExytCA!) w [u, 1 + v,x]
9
c z 2
+ 4Exyt(AB) w } dxdy 3-14
- 39-
The same strain energy expression is also reported in r.ef. 19.

The bending strain energy is given by the expression 54 ,
1
b a
Ub 11: ! : : p1x(-lt,xx)+My(-w,yy)+2r·xy(-w,xy)]dxdy 3-15
The moment strain relationships (Fig. 3. 3) are:

~:x DX Dl 0 -w 'xx
?'
·y = n1 Dy 0 -w,yy 3-16
rT 0 0 Dxy -2w,xy
xy
The bending strain energy is given by 38 t
This energy expression a$.sumes that ,.,

· xy • ~·yx· However~ as
pointed out in the Chapter II, for an open orthotropic dee~,
~;xy r/- nyx and therefore the value of ~:xy eiven in Eq. 3-16, is
to be interpreted as the average value. The energy expression
3-17 neglects any energy stored in the deck due to the re ·::. .. 1 ~
strained warping. As pointed out by Smith 35 , in an open form
deck. the twisting moment H consists of two parts:
~x
Pyx 11: -(2Dyxw'xy - Erw,xyy) 3-18
The warpinr, strain energy is given as,

1 b a 2
uwa~p = ! Er I I (w, xyy) dxdy 3-19
0 0
The omission of this energy term is not considered to be of any

serious consequence. The stiffness matrix for the curved ele-
ment is obtained by usinr. Eq. 3-5. The complete stiffness
matrix is given in Appendix B.
-40-
C. FLAT ELE~"ENT
The stiffness matrix for the flat elements can be

obtained directly by putting c • 0 in the curved element stiff-
ness matrix. The strain displacement relationships for the
flat elements are the sa~e as those for the curved elements,
except for the shear strain given by the Fq. 3MQc. For the
flat plate, the last term in Eq. 3-9c, due to the twist curva-
ture of the hypar surface disappears. The difference between
the curved and the flat element stiffness matrix can be shown
schematically as follows:
[K]m [K]~ni u
v
(K]curved • ------ -- .. ... --- - 3-20
[Y.]bb
T
[K]bm w
+[K]bbm
[K)m 0
-
v
[KJ:flat • ----·-- ·-----·· 3-21
0 [K]bb w
The MeMbrane stiffness matrix [K]m, is obtained from terrns

1 6 of the membrane energy expression Eq. 3-14. This
stiffness matrix is common to both curved and flat elements.
'
The coupling matrix terms [K]bro and (K]bbm are obtained from
terms 7 , 8 and 9 respectively, of Eq. 3-14. These terms
are zero in the case of the flat elements. The stiffness
matrix [K]bb' common to both types of elements~ is obtained
from the bending strain energy given by Eq. 3-17.
The stiffness matrix derived for the orthotropic flat
plate was checked term by term~ with the stiffness matrix for
-41-
an isotropic plate reported in Ref. 51. To establish the

validity of the stiffness matrix further, the results for the
plates shown in Fir,. 3.4a,b,c were <:becked with available class-
ical solutions. For the isotropic plates shown in Fig. 3.4a
and b, the comparison between the classical solution and the
stiffness analysis for the center deflection 68 was excellent.
The error was less than 1\ in both cases and the geometrical
symmetry in the nodal displacements was satisfied.
The third example (Fig. 3.4c) is of particular in-
terest. The 28-G standard corrugated deck can be considered
as an extrel'le case of orthotropy. Pere the bendinr, rigidity
Dy is 1678 times Dx. The bending rigidity constants were used
as given in Ref. 54. It is reported in Ref. 19, that the rec-
tangular elements proportione<l in the ratio of their bending
ripidities, in two directions, would give better results than
the square elements. A quadrant of the plate was analyzed by
using square (6x6) elements and elongated rectangular (2xl2)
elements. The results for (lx30) size elements were also com-
puted t~t are not reported since there is practically no dif-
ference between these results and those with a 2xl2 element
grid. The deflection profile across the corrugation is plotted
in Fig. 3.5. TheTe is practically no differen~e (see Table
111-1) between the rectangular and the square ele~ent solutions,
a dip in the deflection profile near the support is seen in
both the solutions. Similar deflection profiles are also re-
ported in Ref. 19. The solutions obtained by the stiffness
analysis are compared with the classical solutions given by
-42-
Timoshenko 54 in the form of a double sine seri1.>s, for the uni-

formly loaded, simply supported orthotropic plate.
00 00
w• ~ "' . mir x • n .,..L 3-22

" " amn s in a s in b
rn•l,3,S, ..• n•l,3,5, •••
where,
1
4 3-23
+ n Dy)
b4
where,
3-24
The first few significant terms in the sine series

were computed. The comparison of the results is shown in Table
111-1. Both the classical and the stiffness analysis solutions
are well within the limits of practical accuracy.
A strongly orthotropic plate such as the one under
consideration, priMarily behaves as a plate on an elastic
foundation. The stiffer beam strip near the support attracts
more load because of the presence of the plate action. In
other words, the deflected profile of the plate is the func-
tion of the assumed bending rigidities and does not appear to
be dependent upon the shape of the element (rectanpular or
square).
The non-monotonic convergence characteristic of this
problem, observed in the Ref. 19, is probably the function of
the deck properties. This analysis indicates that elements
which are square in plan can be used for the single deck hypar
structure.
-43-
D. EDGE tfEFBEP.
The conventional beam stiffness matrix 55 is primarily

based on two assumptions:
(1) The shear center and the centroid coincide, (iir,. 3.6a).
(2) The bending of the section takes place about the axix
of symmetry (Fig. 3.6a).
In the case of a symmetrical channel (Fig. 3.6b), the
shear center (S.C.) and the centroid (C.G.) of the section do
not coincide. ·n1en the section is subjected to a vertical load
1
P (Fir. 3.6c), acting at a distance e from the shear center, it

not only deflects in the plane y-z but also twists through an
angle 6 about the x-axis, passing through the shear center. In
other words, the bending displacements v and w need to be ex·
pressed at the shear center of the section. This observed fact
was elegantly expressed by Bleich 56 and Ho££ 57 , in the strain
energy of a beam of arbitrary cross-section. The total strain
energy of a beam (Fig. 3.7) is yiven as,
1 8 2 2 2 p z 2
Ubb • z0 I [Eiy'"'xx+Elzv'xx+GJe,x+..,rbe'xx+EAbu,x]dx 3-25
where, wand v are the displace:ments of the shear center (S.C.)

measured along the axes y and z parallel to the principal
centroidal axes of the section and u is the average longitudinal
displacement of the section along the axis x measured at the
centroid (C.G.) of the section. a is the angular rotation of
the section about the x-axis.
Neglecting the warping of the section, the conven-
tional beam stiffness matrix can be obtained by assuming the
-44-
following displacement fields:
3-26a
[(a +2x 3 -3ax 2)v 1+(3ax 2 -2x 3)v 2+ax(x-a) 2ezl

3
v =~
3
+ a(x -ax 2 )ez 2 ]
w z ~ [(a 3+2x 3 -3ax 2)w 1+(3ax 2 -zx 3)w 2 -ax(x-a) 2 eyl

3
- a (x -ax 2 )ey 21 3-26c
a = [-xa e2 + x
c1 - -)e
a 1
] 3-26d
Using Eqs. 3-25 and 3-5, one can obtain the conventional beam
stiffness matrix.
However, the warpine restraint is of practical im-
portance, particularly for thin-walled open sections. To in-
clude the warpinr effect, the displaceinent field for e is
assumed to be of the same form as that of v and w58 .
e c ~ [(a 3 +zx 3 -3ax 2 )e 1 +(3ax 2 -2x 3 )e 2 +ax(x-a) 2 e~ 1
+ a ( x3 - ax 2) eAZ
' ] 3-27a
3-Z7b
The stiffness matrix (14xl4) for this member can be

directly obtained from Table III-3 by substituting Yc = Zc =
Y5 • Z5 • O. The additional degree of freedom (exy) intro-
duced, does not present any problem. If the beam shown in Fig.
3.8 is attached to a shell surface or a plate along the line
passing through the shear center, one can easily find the cor·
relation between the degree of freedom for the plate and the
-45-
beam.
3-28
Omitting the rotation about tho z-axis {ez), the

plate and the beam element can be connected to each other with
a one to one correspondence in the nodal degree of freedom
E. ECCENTRICALLY COHNECTED NON-COMPATIBLE
SliPPORTI.NG EDGp HE~~BER
The above mentioned matrices can be used only in the

case of non-eccentric loads or concentric connections of the
edge members to a shell or a plate surface. In practice, con-
centric connections between the supporting edge members and a
shell or a plate are seldom possible (Fig. 3.9). The eccen-
tric connections modify the effective stiffness of the support-
ing members. The modified stiffness for the non-compatible
edge member can be obtained by the use of simple linear con-
gruent transformations similar to the one suggested in Ref. 59.
The difference between the compatible and the non-compatible
eccentric ~embers is discussed in Appendix E and the relevant
stiffness and incremental matrices for the compatible element
are given in Table E-I and E-II. The linear transformation
assumes that the beam bends about its own neutral axis (Fig.
3.10).
1 0 0 0 +Z c -Y 0 u
c
1 0 ·l s 0 0 0 v
1 +Y
s 0 0 0 w
[A] s • 1 0 0
3-29
0 ex
1 0 0 ey
0 1 0 ez
1 &x
= [[o:X] !-l 3-30
[T]s
The effective stiffness of the edge member can be given as,
[K]eh = [T]~ [K] [T] 5 3-31
Similar stiffness matrices were independently de-

rived in Ref. 21. The stiffness matrices, with the linear
variation and the non-linear variation of the twist angle are
given in Table III-2 and III-3 respectively. The convergence
characteristics and the accuracy of the stiffness Matrices, was
checked by solving three cantilever beams (Fig. 3.11). Only
axial (Fig. 3.lla) and bending (Fig. 3.llb) loads were applied
eccentrically to the beam. All displacewents were computed
along the line PQ (Fig. 3.11). The deflections and the twist
angle at the free end Q, were compared with classical solutions.
As it can be seen, the convergence in all cases is insured and
the results for the case of six ele~ents are within 0.2% of the
classical solutions. The loading II was .applied to two cases:
Ila) Free torsion with a linear variatfon of twist. IIb) In-
cluding the warping restraints at the fixed end. The classical
solutions for the case lib, were obtained from the !tef. 60.
The convergence in case of the restrained Farping is slightly
slower than in the case of free torsion. The influence of in-
cluding the warping degree of restraint can be seen from Fig.
3.13a. The free end deflection ~Q in the case IIb, is about
63.S\ of the deflection obtained in the case Ila.
F. ELASTIC SPRING8
In order to simplify the mathematical solution without
tmdue loss of generality, certain structural elements are ideal-
ized in the form of concentrated spring stiffnesses. For ex-
ample, the central column in the case of an umbrella shell, if
idealized as a physical member with its end points, will not
only c~eate an additional node point for the master stiffness
matrix but t-1 ill also disrupt the regularly arranp;ed grirl pattern
and will warrant a modification in the entire assembly routine.
In order to avoid this, the stiffness of the column can be
idealized into six discrete springs accounting for its axial,
shear, bending and tl•ris tiny stiffnesses (Fig. 3.14). These
stiffnesses are given as,
SA .-
Ab
a 8BY •
4EI.Y
a
.
> 5Bz =
4EI z
a
3-32
12E Il': 12EIZ
5sY • a
5
' sz
• a ., sT = -aC:J
These spring constants are added alonp the main dia-
gonal elements of the master stiffness matrix. This idealiza-
tion is not alway', satisfactory. In Fig. 4.1, a tension bar
connectinr, the lower corners of the saddle shaped hypars are
replaced by four springs in the u and v directions, two at each
corners f and b. This idealization eliminates the interaction
between the nodes f and b. The validity of this approximation
can only be assessed by engineering judgeMent.
I I I. 3 ~~ASTER STIFF.IESS ~~ATP-IX
The stiffness formulation presented so far is for an

individual element. The total stiffness of the structure is
-48-
<lcvelopcd by the assemblage of these individual elements. The

two different methods ('a' and 'b'), mentioned in Section III.l,
differ in tho formulation of the stiffness matrices.
a) Curved Element
The element stiffness is formulated by using an ele-
ment of the same shape as the shell middle surface (Refs. 17,
19,20). However, the fact that the shallow shell assumptions
are made in this formulation, should not be overlooked (a point
which is discussed at lenrth in Chapter IV). The assumed
shallowness of the shell does not warrant any form of co-ordin-
ate transformation. The strain displacement relationships
given in Eqs. 3-9 and 3-10 are based upon the displacements u,
v, and w which are measured along the tangent and normal to
the surface.
b) Flat Element
The middle surface of the shell is approximated by a
series of flat plat~s. The geometrical approximation of the
actual surface needs three important steps:
(1) Definition of Surface -
As pointed out in Chapter I, different hypar struc-
tures can be built with various combinations of the basic units
(Fi~. 1.2). It is necessary to express the equation of the
generated surface with reference to the chosen global axis.
The general equation of a structure using the hypar units can
he expressed as,
3-33
Figs. 3.lSa and 3.lSb show two structures and also give the
-49-
values of the constant, defininr, the surface equati~n for each

quadrant. In Fig. 3.lSb, points P,Q,~,S are the local origins
of the quadrant surface. x' and y', are the local co-ordinate
axes passing through the local origins.
(2) Element Size -
For shallow shells, the size of the elewent can be
approximated with the size of a rectangle projected on the co-
ordinate plane xy (Fig. 3.16). For example, the size of the
curved element PQRS is approximated by the projected element
P"Q''R''S".
A better approximation for the size of the element
PQRS can be made by calculating the actual lengths PO, QR, etc.,
and usinr, a rectangle P'Q'R'S', of an equal area. For the low
rise to span ratio (~l/S)~ the error introduced hy using the
projected element. is very small {Z-3%). To take advantage of
this fact, a provision iz mR~e in the computer program to
choose between the above mentioned ap~roaches. The difference
in Tesi-1.ts w!len u2ing the~e two methods was about 10% for the
structure shown in Fig. 4.3. The computation of the exact
lengths gives different stiffness matrices for each element.
(3) Co-ordinate Transformation -
It is not possible to gener~te a smooth curved sur-
face by using flat elements with rectilinear boundaries. This
results in gaps and non-compatibilities between the adjacent
elements forming an idealized uneven surface (the picture of
the idealized surface is left to the imagination of the readers).
Such gaps and discontinuities occurring at the boundaries of
-so-
adjacent elements) have been known to produce undesirable and

non-existing nodal forces, for shells of revolution 61 , which
had significant effects on the solutions. However, no such
noticabl<'' di ff icul ty was encoWl tered in the solution of sh al low
hypar shells. The solutions obtained for these shallow shells
did not shm·' any necessity of placing local tangential ortho-
gonal axes at each nodal point.
In writing the master stiffness matrix and the overall
equilibrium equations, local nodal axes can be chosen. Instead
of these, a simple anrl approximate approach is used. ft plane
tangent to the surface is drawn at a point 0 (Fig. 3.17). The
most logical point for the tangential plane is the center of
the clement. In the case of umbrella shells, the flat portion
near the free corner shows a pronounced bending action. To
estimate this bending action conservatively, the tangent planes
were drawn along the horizontal lines PQ and PS instead of at
the center point 0 (Fig. 3.17a). When using the corner point
transformation, one has to exercise proper care to retain the
symmetry of the solution. For a large number of elements, both
methods should give about the same results but the corner point
transformation is more cumbersome and therefore it is not used
in the analysis.
In Fig. 3.17a, the line OZ' is normal to the surface
whereas OX' and 0Y' are generators of the surface. The direc-
tion cosines for the lines OX', 0Y' and CZ' are given as,
.. 51-
zx
ox' • -1- 0
~x Rx
1 zy
OY' = 0
~ .A 3-34
·Z x .. z y 1
oz' •
l+Z x2+z y2 2+z 2 j+z2+z2
l+Z x y x y
where
and Zy • c x
AB o·
As discussed in connection with the shallow shell
assumptions, the angle between the generators OX' and OY' is
not equal to 90°. Hence a new set of mutually orthogonal axes
OZ, OX, and OY are obtained, where OZ coincides with OZ'. The
procedure for obtaining the direction cosines of ox, OY and oz
is a simple application of the three dimensional solid geometry
( F i g . 3 • 1 7b) •
The transformation matrix for each node can be repre-
sented as
0
3-35
0
tx ty 1:z:
p,] 1 D mx my mz 0 3-36
nx ny nz 0
The in-plane rotation ez is omitted.

From the matrix [.A] 2 it
can be seen that exy the additional degree of freedom, is trans-
formed frow the local to the global axis in t~e same way as w
- 52-
except for the assumption that there is no coupling betwe·'n the

rotation exy and the rotation ex, ey. This transformation can
be viewed as an approximation.
The master stiffness matrix for the shell surface,
using method 'b' (see Section III.l) is completed by transform-
ing each and every element stiffness from its local axes (OX,
OY,QZ) to the respective global axes (OX,OY,OZ). Depending
upon the direction cosines of the local axes of the individual
elements, every coefficient of the transformed element stiff-
ness matrix can have a non-zero value, i.e., there is a coupl-
in~ between u, v, and w displacements, expressed in terms of
the global co-ordinate.
Beam Element
The co-ordinate transformation given by ~'feaver 55 , to
transform the stiffness of the beam element from the local to
the global axis is used. The transformation matrix with a minor
modification to suit the problem at hand is given in Table III-4.
After orienting the axis x of the member, it is also necessary
to define the orientation of the principal axes y and z, since
the stiffness of the beam element is expressed in reference to
its principal axes. The angle B defines the orientation of the
principal axes. The definition of the angle a is given in Ref.
63. (Fig. 3. 18).
For the method 'a', using the curved element, the
stiffness of the edge members is added without any co-ordinate
transformation. For method 'b', using the flat elements, the
edge member stiffness is added with a proper co-ordinate trans-
-53-
formation as given in the Table III-4.
III.4 LOADING
A uniformly distributed load acting on a rectangular
clement can be replaced by statically equivalent loads of
equal intensity acting at each nodal point. This procedure is
acceptable if the size of the eleMent is small.
The alternative approach knowl' as the work equivalent
load is based on the equivalence of energy. The nodal forces
are so assigned that during any virtual displaceJ'lent the work
done by these forces is equal to the correspondinr work done
by the actual distributed load.
The work equivalent nodal loads for the rectangular
element, with unit normal load, are given 'below:
ab/4 ex:l nb 2/24

ab/4 6x2 ab 2 /24
ab/4 8 x3 -ah 2/24
ab/4 6 x4 -ab 2/24
• = 3-38
2
a b/24 6
2 2
·a h /144
xyl
a 2b/24 6 xy2 -a 2b 2/144
a 2b/24 9 xy3 a 2b 2/144
-a 2b/24
_ 2bZ/144
6 xy4 8
The nodal loads associated ~ 1 ith exy de~ree of freedom

do not have any physical significance. Tl'.c Jllornent t 1x is
associated with ey degree of freedom whereas the moment '~y is

associated with ex. For an interior point, the work equivalent
load reduces the static load for uniformly sized elements. The
effects of the nodal moments cancel out along the boundaries
.. 54-
whereas they add up in the direction normal to the boundaries.

All throughout this work. work equivalent loads arc used for
uniformly distributed loading.
In the case of a uniform load acting normal to the
surface, statically equivalent projected uniform load is cal-
culated. The work equivalent nodal loads then calculated for
the rrodified load intensity are directly applied to the struc-
ture, in terms of the global co-ordinate axis without any
trJnsformation. This is again an approximation. A more
accurate method of determining the nodal load would involve a
co-ordinate transformation from the local to the plobal axis.
Besides the uniformly distributed load, a concen-
trated force or moment can be applied to the structure by
spccifyinp, the magnitude of the load at the corresponding de-
gree of freedoM in the load vector.
111.5 BOUNDARY CON.DITIONS

The boundary conditions for a structure can be
broadly classified into two cate~ories:
(1) Force boundary conditions.

(2) Displacenient hounrlary conc~i t ions.
In the conventional stiffness analysis, the latter can te
easily satisfied whereas the former can be satisfied only in
the variational sense. A detailed discussion of this point is
reported in Ref. 21.

The typical boundary conditions for the edge where
x is constant, are:
-55-
Boundary ;..
Condit ions u v w ex xy
Hinge 0 0 0 0
Knife-edge 0 f} 0
Fixed 0 0 () 0 0 0
Free
Symmetric 0 0 0
Por the free edge, no displacement boundary con-

ditions are specified. All the boundary conditions are applied
\vi th respect to the global axes. The boundary conditions for
the member PQ (Fig. 3.19), which was supported vertically but
allowed to slide along its length, should be specified in terms
of the local axes x, y and z but instead they : arc specified
in terms of the axes x, y and z. This is an approximation and
the error due to this will increase with the increase in depth
of the shell. The procedure to express the bo~ndary conditions
- -
in the local axes x, y, z- is given in Ref. 62.
The connections of the edge members to the dect
present a proble~ in expressinr the correct boundary conditions.
Two non-compatible boundary conditions are shown in Fig. 3.2n.
In Fig. 3.20a, the deck bends freely without twisting the edge
member. This moment-free deck to edre ~ember connection is
quite common in practice. The opbn-form decks are discretely
connected to the edge members whereas the close form decks are
connected only along their bottom plates (Fip. 2.2). In both
cases there is no transfer of moments betNeen the edge member
and deck.
The other type of discontinuity can result in the rela-
-56-
tivc displaccTncnt bett·1een the deck and the edge member, normal
to the boundary (Fig. 3.20b). This type of a connection can
result because of an oversized hole, loosely connected screws,
or due to tearing of the deck. Depending upon the continuity
achieved between t!1e deck and the edge members, different
values of fixity coefficients are used. Tr and TH represent
the torsional and the horizontal fixity coefficients, respect-
ively.
In the case of the moment-free deck to edge Jl\ember
connection, Tp = O. The edre merrber stiffness matrix is modi-
fied by multiplying the columns and rows correspondin~ to the
twisting degree of freedoms (ex and exy) by TF.
The problem is further com~licated by the eccentric
connections. As shmm in Fig. 3.21a, even with a discontinuity

of the rotational degree of freedom, twisting can still be
introduced in the edge member because of eccentrically trans-
ferred vertical or horizontal load. This problerr is not solved
satisfactorily. By the method of fixity coefficients, the
twisting action introduced by these eccentric forces is elimin-
ated. There is no moll\ent transfer when two elements are inter-
connected by means of hinges and this results in the local re-
lease of the mewber forces. This for~ulation does not include
the effects of these releases. The details for the incorpor-
ation of these local wcmbcr releases are given in Pefs. 62 and
63.
I I I. 6 SOV:'TIONS OF EQPATIONS
The equation 3-4, relating the applied nodal loads and

-57-
thC' p,cncral i. zc<l nodal displacement can be solved. To ohtain

the dt~•pl11c{•f'Wflt Vt'l\'tu1·.
[ti] "' [l]-l [P] 3-39
The inversion of the large matrix [k] not only re-

quires a very long time but also needs a large storaP.c space
in the computer. The structural matrices are usually well
ban~ed about their main diagonals and are also symwetrical.
B-...-ol~
l·Irotc a subroutine wfl.ich stores only the half band
of the matrix in a vertical fashion (Fig. 3.22). The equations
are solved by the Gauss-elimination scherne 64 • ~ith the limi-
tation of the available core size and the computational cost)
a total of 486 equations with a maxiJT1um half band width of 66,
were solvecl for a normal problem (64 square eleJT1cnts). The
rectangular matrix of 486x66 was formed and stored in the
computer.
However, there are problems where the structural con-
fir.uration destroys the close-bandedness of the matrix. For
example, the tension-tie connecting the lower corners b and f
of the hypar {Fig. 4.1), creates sparse entries in the stiff-
ness matrix (Fig. 3.22c). In 'uch cases, instead of revising
the entire solution procedure, the structural element is ideal-
izcd in t 11e form of discrete springs (see Section III. 2F).
III.7 STRESS A:~ALYSIS
Since the main ai~ of the project is to establish the

behavior of the hypar shells, the physical interpretation of
the cowputer results is very iMportant. The deflections, as
well as the stresses in the different structural components
-58-
such as deck, edge mewbers, etc., represent the physical be-

havior.
In the finite element analysis, the generalized nodal
forces are related to the stresses. But because of an error of
discretization and applied joint loads, the resulting nodal
forces for the adjoining elements show deviations. In order to
obtain sowe form of average stresses, the element forces are
calculated at the mid-point of an individual element.
A. ~CK STRESSES
The deck stresses arc calculated at the center point
of an individual element. Depending upon the choice of method
of analysis, method 'a' (curved elements) and method 'b' (flat
elements), corrcspondinr strain displacement relationships are
used at the center point (see Eqs. 3-9, 3-10). The forces '.\T
i 'x'
NY and Nxy and the 1'1oments r·x and i!y are calculated per unit
length. For the complete derivation of these forces, see

Appendix C. The major difference between the computation of
stresses for a curved and a flat element is that in the case of
the former, consistent with the shallow shell assumption (see
Section III.ZB) displacements tangential and normal to the sur-
face can be used directly. But in the case of flat elements.
the displacements obtained in the global co-ordinates are trans-
formed into the local co-ordinate axes (see Section III.3) and
the relevant displacements in the local co-ordinate axes are
used. The difference between the strain-displacement relation-
ships for the curved and the flat elements was already shown
in Section III.ZB.
It must he realized that the forces are calculated on
-59-
the basis of orthotropic plate theory which can be considered

as an approximate mathematical idealization. The forces cal-
culated per unit length are multiplied by the lengths of the
basic units (Fig. 2.2). In the case of a uniformly loaded
structure, this method can be considered to be fairly accurate.
For the light gage sections witl: high width to thickness ratio
of the in<lividual components, the effective led of the section
and the location of the neutral axis need modification in
accordance with the level of the load (Ref. 38).
The stresses calculated do not include the local
bcn<ling behavior. For example, the bottom deck plate AB bends
locally between the vertical web plates of the hat (Fig. C2 of
Appendix C). The problem of deviation between the mathematical
and the physical behavior of the orthotropic deck is dis-
cussed in detail in rrefs. 46,47.
B. BE.Al'( STRESSES
The nodal forces calculated in the local.axis of the
beam can be directly used to calculate the beam stresses. The
method of calculation of stresses for the concentrically con-
nectcd beam Member is well known.
The imaginary forces calculated alonr the line PQ
(Fir. 3.11), are to be transferred to the shear center and
centroid to calculate the relevant stresses. The bea~ forces
[P]b can be calculated by,
[P]b = [TJ! [K] [T]p [A] 3-40
The pre-multiplication of the global displacement

[~],by [Tlr (Table 111-4), transforms the global nodal dis-
placements to the local axes whereas the pre-multiplication of
-60-
[K] [T]R [fi] by [Tl T9 . transforl'!'s the forces to the shear center
or the centroid of the beam.
Because of the mathematical idealization, certain
difficulties are encountered. A beam with an eccentricity in
the z-direction is shown in Fig. 3.23. The variation of t~e
axial forces is shown in Fig. 3.23b and 3.23c. Since the forces
arc balanced at point 0, the axial force also contributes to
the equilibrium of the moments at point 0. This results in the
inequality of the moments along the axis of the beam PQ. The
problem becomes particularly critical in the case of rapidly
changing axial force and a deck with strong bendin! riRidity
(e.g., concrete hypars). No suitable solution is found for
this prohlem as of this moment. In the ab~cnce of definite
guidelines, the deflected shape of the structure should be used
to decide the sign of the mnm~~t.
Experience shows that the bigger of the two reoments

O'op or ;;0 Q) is always in conformity with the correct deflected
shape of the beam. The difficulty experienced in computing the
stresses of an eccentric edge member is one of the shortcomin~s
of using the nodal points only along the shell surface.

The results obtained by this stiffness analysis are
compared with experimental and the available solutions in the
literature in Chapter IV.
CHAPTrP IV
A GENERAL C0?1PAFATIVE STUDY
IV. 1 INTrrODUCTION
As discussed in Chapter III, two methods were used

to analyze hypar shells: method 'a', uses rectanrular curved
clements based on the shallow shell theory; whereas method 'b'
approximates the actual shell surface by using a series of
flat rectangular elements. The solutions of selected problems
are presented here with three purposes~
(1) To substantiate the use of the finite element Method,

hy method 'a' only, by compariny the solutions for problews
for which analytical or other numerical solutions are available

in the literature. The comparison for flat plates and linear
beams are already presented in Ch&pter III.
(2) To compare the solutions obtained by 11'ethods 'a' and
'b', for typical hypar structures. The comparison is done pri-
marily •.'1ith a view of assessinrr their suitability in applica-
tion to the practical problems and also tq find out their
shortcomings and limitations.
(3) To compare the analytical solutions with the experi-
mental results obtained by earlier wotkers. The details of
the structures analyzed are given in Table IV-1.
All the analytical results are further used to study,
to a limited dep,ree, the effects of different structural para-
meters on the behavior of hypar shells, such as relative shear
·61-
riridity factor a, rise to span ratio, etc. The effects of
these parameters are further discussed in length in Chapter V.
IV. 2 COflPARISON OF MtTIIOD 'a I l'TITH OTHEP Nm1EPICAL SOLUTIONS
Connor and Brebbia 17 presented the centerline deflec-

tion profile for a saJdle shaped hypar structure (Struc. 'l',
Tab le IV-1) , with clamped boundaries all around. Thf!·ir results
\W\ ~, ..~.,
were based on exactly the same procedure as used
The only difference is that they used. a 12-term polynomial for
.\.-~I~~
the normal displacement w, whereas used the 16-term
displacement function as given in Eq. 3-6c. Fig. 4.6 shows
the results obtained by the author. For the grid size of 8x8,
the deflection profile along the line oa is similar with the
one reported hy Connor an<l Brebbia 17 . The deflection profile
along the diagonal ob is also plotted to check the symmetry of
the solution.
The converr.:cncc characteristics of the solutions are
checked by refininr the grid size for the above mentioned
structure. As seen in Fip. 4.7, the convergence for the center
deflection at point o (Fig. 4.1) is ~onotonic and rapid. Dy
refining the grid size from 6x6 to 8x8, an i~provement of only
2.3% is obtained in the result.
Pecknold and Scl.nobrich 20 presented the centerline de-
flection profile for the same type of a structure, with the
perimeter knife-edge supported (Struc. '2'). As pointed out
earlier, they used the Birkhoff and Garabedian interpolation
for~ula for w displaceMents. Besides that, the complete rigid
body wo~cs of displace~ents were included by solving the homo-
-63-
pcncous pnrt of the strain displace~ent relationships ~iven in

nq. 3-~. The inclusion of the rigid body motion terms des-
troyed the interelement compatibility and put additional re-
straints on the in-plane displacement fields of u and v~
u • F(ui) + F1 (w) 4-la

v • F(v 1) + F2 (w} 4-lb
The displacement functions F1 (w} and F2 (w) are the

results of the solution of the homogeneous part of the Fq.
3-9c. Pecknold and Schnobrich compared their solution with a
Navier-type (double sine series) solution, for which SO terms
in each direction were included 20 • The deflection profiles
.:,"" ~I
alonr the center line oa and the diagonal ob, obtained
s+\aA\
(grid size 8x8), are shown in Fig. 4.8. The central
;,_.,,._ ~' w..~
deflection obtained differs l)y ~o. 2% f·rom tha::
obtained by the • ~~ries soluticr1 (9.18xl0- 3 inch.); i./h~reas it
differ~ by approxi.mi:itely -lt from the fi:.:ite element solution
of Ped\'.nold and :~. 1":;brich,
The · ::bititm! oh•:a.i:i.ed by method '.:. • for hoth Struc.
'1' and '2' mentioned above, are considered to be quite good.
rv. 3 cor ·Pf\ RISON OF

----·- - f"ETHOD 1u-.sn r:rr110D
-•a' --- ---- -- 'b'
Both structures solv'?.{; hy i'°''~.:hod 'a',, ~,rere solved
again by using method 'b:. The defl·::•:,;ions obte.ined by the two
methods usinp the grid size of 8x8, are sho1·m i·1 Table IV-3.
t·!ethods ia' and 'b' show similar deflectjon profiles along both
t~."" c~nter and 1'.he diagt..n.:.:1 linr;s '" i and ob respectively. The
central deflec·... ion obtained by method 'b' for both structures
is on the higher side, as compared with the one obtained by
-64-
method 'a'. For St rue. 'l', the difference in the central de-
flcc tion is about O.Bt whereas for Struc. '2', the d~ffercncc
is about l.30t. The central deflection for Struc. '2' is only

o.si on the higher side of the deflecti~n obtained by Pecknold
and Schnobrich.
The correlation obtained by methods 'a' and 'b' is
excellent for these two structures. However, it must be
pointed out that both of these structures, taken from refs. 17
and 20, are supported all-around. From the practical point of
view, these structures are only of academic interest. The
boundary conditions such as free edges, encountered in an UJTl-
brella shell (Fig. 4.2), provides a more critical test for the
comparison of the different f!1ethods.
It was not practical to compare methods 'a' and 'b'
for all the examples, ther~fnre only a selected number of struc-
tu;-cs were chcsen for comr~ .. : son (Stru'.:.. '6' and ''J' were
used). St rue. '0 is a !;1~'<'." J l scale ccncr~te f'lode 1. In this
structure, the stiffening = ~ge members are located eccentri-
cally, on top of the shell. The idealized edge members are
considered eccentric only in the i !~rection (see Figs. 4.2
and 4.5).
Struc. '9' is also an umbrella shell hypar with 28-G
double layered standard corrugated decks placed perpendicular
to each other. Here the edge member is connected eccentri-
cally to the deck with the deck on top. In the case of Struc.
'6' there is full fixity between the edge member and the shell,
whereas in the case of Struc. '9', the connection between the
-65-
deck and the edge member is moment-free.

Tl1c comparison of the deflected profile obtained hy
methods 'a' and 'b' anJ the corresponding deflected shapes are
presented in Figs. 4.13 and 4.23. Comparing the solutions ob-
tained by methods 'a' and 'b' for Struc. '9', it is obvious
that the method 'a' underestimates the free corner deflection
Sb. !!ethod 'a' gives a good correlation between the theory and
experiment for the deflection 6a.
The deflection profile obtained by method 'b' for
the edge member ab, where a major portion of the load is car-
ried by the bendinf action, is very good when co111pared with the
experimental results. The relative deflection between the
points a and b according to experiments is 1.2 inches, method
'b' 11iving a rela.tivc deflection of 1.0 inches; whereas that
predicted by the m~t)od 'n' is 0.73 inche~
A distortion ;_I' .. ~e deflected prof !le fo~ the member

ab and the underf"stimatif:1'l of the relative deflecticn between
the points a and b, results in the underestimation of the bend-
ing an<l the total stresses at the point a. The bending stress
at point a by method 'a' is 8.44 ksi, whereas that by method
'b' is 17.14 ksi. The total stress at the point a by method 'a'
is 12.50 ksi, whereas by method 'h' it is 19.90 ksi. The cor-
ner deflection ob and the stresses at a (Fig. 4.2) are of a
~reat practical significance for a designer, from both the choice
of edge memher sizes and the overall behavior of the hypar
structure.
Another i~portant shortco~ing noted of the method 'a'
is that the statical check for the total vertical load is not
-66-
satisfied at the column. Eecause a very flexible deck was used

for the shell surface, the deck could not transfer a substantial
amount of load near the column and the resultant axial com·
poncnt and the vertical shear in the edge meJT1her oa should sum
up to the total applied vertical load; only 73% of the total
vertical load is accounted for by method 'a' whereas 98\ of the
:ipplied load is accounted for t,y method 'b'. This discrepancy
of the statical check was also noted when working
with th~ computer proflraw for~ulated by Parrer 19 .
In the case of Struc. '6', the deflection profiles
oh ta ined by mctl•ods 1
a.' and 'b' along the compress ion member
arc reasonably close. However, these two methods give entirely
different deflected shapes along the tension member ab. Accord-
ing to method 'a', the point b (free corner) instead of deflect-
inf downwards ralative to p1.~-:._n" a, it deflects upwards. The
same difficulty w~s also enc . mnt~:'."'~d r~"'!irdi ig the corner tie-
flect ion when usj ng the compute!' p 7' 1

),£' .·am of Ref. 19. Because
a p:irt of the load near the column i<. also carried by the con-
crete shell, the thickness of which is quite comparable to the
depth of the edge l"lCMbers oa and oc, it is difficult to figure
The corner deflection -·

out the statical check for the total vertical load.
of Struc. 8,.whi~h is identical
to Struct. '6' except for the fact that the edge members are
downturned, was found to be quite low when analyzed by the
method of ;\ef. 19, as compared with the expeetmental results.
Briefly, the shortcomings of method •a• can be
summarized as:
(1) The method undcrestiwatcs the deflection of the free
corner of an umbrella shell where the shell surf ace degenerates
almost to a flat plate.
(2) The prediction of the deflection profile along the
eccentrically connected tension Members ab and be is not con-
sist('!lt and leads to the underestimation of the bending and
total stresses in: the edge members, which are of practical
importance.
(3) A discrepancy for the statical check of total verti-
cal load is noted (Struc. '9').
Because of these shortcomings, it was decided to use
method 'b' in the analysis of all structures. It must he
emphasized that method 'b' does have certain shortcomings,
though none as serious as the ones associate-1 1•!i th method 'a'.
~·ethod 'b' is discussed later in Chapter V.
IV. 4 DISCUSSION Q.E f.TETnOD ~
The umbrella shells with flexible edge wembers show

a pronounced bendir <~ a>:tion near ·-~a f:ree corner b (Fig. 4. 2).
This bending action was ~-:b~erved in tests cnnducted on con-
crete shells (Refs. 31,65) and Struc. '9' (using hypar with a
corrugated deck) tested at Cornell- . The shell in this region
acts almost like a flat plate. In Ref. 1, this observed bend-
iny behavior of the shell ltas termed as secondary bending and,
based on the non-dirr!ensionali zed parameter of ~, the hencing
JTtorr.cnt coefficients for this region were given. ~'ethod 'a'
basecl on the use of the shallow shell theory fails to predict
-68-
this local izcd bending bcliavior at the free corner.

Before discussing this shortcoming of method 'a', it
is necessary to point out that in the formulation of the curved
clement, t~e nodal displacements (u,v,w) are measured along the
tangents and normal to the surface, rather than alonf the car-
tcsian axes. In other words, the strain displacement relation-
ships given in Eq. 3-9 are all expressed along the lines of
generators of the surface. The elewent stiffness J11atrix based
on these displacements eliminates the co-ordinate transf orma-
tion. In the solution of the master stiffness matrix, method
'a' gives the displacements along the generators and normal to
the surface whereas the method 'b' usinf flat clements gives
these displacefl'ents along the global cartesian co-ordinates.
However, bccau~·e of the shallo\·mess of the shells,
(see Section III.2B) the s.rface co-ordinates along the gener-
ators are approximated by :he C8rt~sian co-ordinates defining
the surface. J:;: ·"'use o.:· r~1is approxim·~.~~::~n, a constant shear-
ing strain term :~i;c is a(·ded to the sheaTing strain of a flat
plate (Eq. 3-9c}. This tt~m does not reduce to zero near the
flat corner b (Fi:. 4.2). It is believed that this term· adds
cxtr?:. sti~fness to the fre~ corne~ where the structure behaves
almost like a flat plate. This addition is probably the cause
of the underestimation of the corner deflection. The deficiency
of method 'a' in predicting the cl~flection of the free corner
n~eds further investigation. The solution could possibly be
improved by refining the grid size or by the use of higher order
strain terms 11 . But this will definitely entail additional
-69-
computational work.
The strain-displacement relationships for the curved
. c.
e 1cmen t arc d epen d ent on 1 y on t h c twist curvature AB 1rrcspect1ve
.
of the shape of the actual structure. To explain this further,
consider only the quadrant oabc of a structure of Type I, Fig.
4.1. One could build two cantilever hypars from this quadrant.
The first structure would have edges oa and oc fixed whereas
edges ab and be would be free. In the second structure, the
fixed and the free edges would be interchanged. If both these
structures are subjected to the same loadin,i conditions, method
'a' would give identical deflections and absolute values of the
stresses.
The solutions by ~ethod 'a' for Strucs. 'l' and '2'
did not show any advantage of using a 16-term displacement
function for w-displacernent, which ensures the slope compati-
bility normal to the boundaries of the adjoining element as
against the non-compatible 12-term polynomial used in ~.ef. 19.
The solution obtained for Struc. •2r the inclu-
~ith
'-"' ~·s
sion of complete rigid body rnodes 20 and that obtained
~~"\
without the inclusion, did not show ~uch of a differ-
ence (Fir. 4.6). To study t~e effects of inclusion of rigid
body modes further and also to evaluate the differences in the
solution usiny 16 or 12-tcrm polynomials for w displacement,

Struc. '15' was analyzed. The results are plotted in Figs. 4.37
and 4. 38. It may be worthwhile to note tl1at in this structure
the rise to span ratio is rather high for it to be considered
as a shallow shell (see Section V.2). The deflection profile
-70-
across the diarronal ob shows that there is practically no dif-

ference between the solution obtained by the use of a 12-ter~
polynomial for the normal displaceMent w, and the function used

by the author. The maximum difference of 2% is seen in the
~~\~~
corner deflection 6b (0.090 inches by the and 0.092
inches in Ref. 19). Even the u-v displacements all over the
shell, obtained by the two methods were within 0.5% of each
other. The striking similarity in the results tends to confirm
the conclusion that both methods give the same results for the
uniformly loaded hypars. This view is also shared by Pecknold
and Schnobrich 21 . The comparison may not be as accurate for
unsymmetrically loaded hypars 1~here the 16-term displacewent
function for the norl'lal displacement w would possibly give
hettcr results.
l!owcve.i.-, the comparison \·.rit~ results reporte~ in Def.
21 shows a di ffer·-11ce botl· in the deflect ions and stresses (Figs.
4.37. 4.38). Though the .i~flection profile and the stress
variation are alike, the added flex~~ility of the curved ele-
ment with the inclusion of rigid body modes is apparent in Fig.
4. 37, where the corner deiJi~ction is nea;.'lY 60% larger than the
·"'~~ wd~
one obtainet! \oo · · ·1s well as by the method used in Ref.
19. Though the solution obtained in Ref. 21 used a 12x12 grid
~ ~l \ ..llt.\\\~'-''OWI J \,~ ~ S
size as against a 8x8 grid size used
_...'t ~\\:~
'
that the difference in results is due to refining
of the grid size.
Analysis of the same structure hy method 'b' using
flat cleroents, results in the corner deflection ~b (0.123 inch)
heing nearly 33\ larger than that yiven by the method 'a'. As
-71-
pointed out earlier, the deflections by method 'a' are given

normal to the surface whereas for the ~ethod 'b' they are in
the global axes. However, this does not affect the corner de·
flection 6b. Moreover, the deflection profiles along the com-
pression member oa and the tension member ab, are different for
the two methods. It is quite interesting to note that both of
these methods, which give close results for edge-supported hypars
(see Table IV-3), could differ in the case of this structure
(Fir,. 4.37). The inclusion of the rigid body modes in the
solution seems to account for the correct behavior of the flat
corner but since no cow.parative results - with experiments - are
presented 21 , it is not possible to comment on
the validity of the method in Ref. Zl.
IV. s THE cot:!PARISON Qf ANALYTICAL AND EXPERIMENTA~ WC'lRK
Because of the shortcon'dngs encountered in niethod

'a', the analysis reported hereafter is carried out by ~ethod
'b'. The experimental results used for the cowparison can be

basically categorized into three types:
(1) Hypars supported vertically along the line of gener-
ators all around the perimeters. Strucs. '3', '4' and '5' come
under this category (see Table IV-1).
(2) Small scale concrete models of umhrella shells 65
(St rues. '6' , '7' and '8 ') •
(3) Umbrella shells havinP standard corrugated open decks
for the shell surface (Strucs. '9' - '13').
All the above mentioned experimental tests were con-
ducted at Cornell, except Struc. 'S' (~ef. 33). The testing of
-72-
the concrete hypars was conducted as a part of a research pro-

ject currently in progress at the Cornell University 65 .
The comparison between the analytical and experimental
results for the deflections, edge member stresses and the deck
stresses is given in Figs. 4.9-4.36. In all the analytical
solutions, the surface of the hypar is approximated by the tan-
gent planes drawn at the center of the element except in Strucs.
'6' and '8' where these planes are drawn at points along the
free boundaries ab and bc (Fie. 4.5). However, these two
structures were not reanalyzed because of minor differences
( <10\) in the results of other similar cases using
both methods of transformation.
A. EDGE-SUPPORTED HYP.A.PS
The saddle shaped hypars (Strucs. '3' and '4') were
analyzed mainly to find the effect of rise on the value of
shear rigidity factor 'a'. The values of the central deflec-
tions are given in Tables II-1, II-2, In the experiments, only
the central deflection 6 (Key sketch Table II-1, II-2.) was
0
measured. The results obtained in the analysis of these
structures are used in Chapter V, to study the effects of the
variation in the structural parameters.
Struc. 'S' was a large scale model with the plan
dimensions of SO'x~o· 33 . A single layer of a cellular deck
(see Table IV-2) was welded to the edge members usinp a warped
plate connection. The hat section was welded to the base plate
with spot welds l" o.c. The adjoining deck panels were butt
~\A.'t-
welded so as to develop the full strength of the plate.
The edge members were free to move in the plane tangential to
the shell boundaries but were supported vertically. A 3}" dia·
mct~r, high-strength steel tie bar connected the points a and
b (Fig. 4.3). A uniform load normal to the surface was applied
by vacuuming the enclosed chamber. A predeternined tension
force was applied to the tie by means of a 500 ton jack which
prevented the relative displacement between the points a and c.
How.ever, the details of the connections of members oa and oc
were such that tl:ere was no force on the member at the ends a
and c. The members ba and be were free to move at the end b.
The stresses and the deflections were measured at various lo·
cations. The complete details of this test with the instru~en
i
tation are given in Ref. 33.
In calculatinp, the membrane constants for the deck,
the stiffening effect of the hat is ner,lected. The meMbrane
stiffness calculated only on the basis of the properties of the
base plate, is on the conservative side. Since no seam-slip
was noticed during the tests, the shear rigidity factor a is
taken equal to unity. The deck is highly orthotropic as is
apparent from the bending rigidities (DY• 29,300 Dx)· The
bending constants calculated on the basis of the geometrical
shape are used in the analysis without modifications. As given
in Ref. 33, the equivalent projected load is calculated on the
basis of equating the shear force at the point o on the actual
surface and that given by the membrane theory for an equivalent
projected load. The load intensity used in this analysis is
5\ on the conservative side of the criteria given in Chapter
-14-
III, Section III.4.

The experi~ental deflection profile along the lines
de and fg (Fig. 4.3) are corrected by subtracting the vertical
displacements at points d, e, f, and g. The center deflection
c-.tlculated analytically is 5% on the higher· side of th.e exper-
imental deflection 2.30 inches (Fig. 4.9). As shown in Fig.
4.10, the axial stresses in the edge members are very close to
half tl~ values given by the mewbrane stresses. The rcported 33
strain ~easurements on the edge members tend to confirm this
observation. It is quite logical to expect the forces in the
edge members to be 101:,rer than those given by the l'l"emhrane theory
because a part of the load is carried by the flexural action of
the deck.
As shown in Fig. 4.11, the difference between the
analytical and experimental results for the shear stresses mea-
sured by the rosettes 1 and 2, is even less than 5% . 1
The var·
iation of the si.ear force all over the hy!Jar surface is shown
in Fir. 4.12. As expected, the value of the shear force over
a major portion of the shell surface is less than that given
by the 111embrane theoTy. The increase in the shearing force
noted at the corners a and c is due to the restraint offered by
the tie; whereas the value of the shearing force in the fixed
corner o is almost twice as that given by the ~embrane theory.
The connection between the edge members and the deck
should be adequate enough to carry this high value of shear.
The bending stresSE)S calculated at the center of the span on
the top of the hat, do not show good correlation with the exper·
- 7 !, "
imental results. The bendinr stresses calculated using the

effective inertia led (Chapter II, Section II.2.B) is 6.70 ksi,
whereas the measured total stress in the y-direction at the
same location is 12.50 ksi. One reason for this underestima-
tion is that the measured stress is total whereas the calcu-
lated stress is only due to bending. Since in calculation of
the membrane stiffness only the ~ '"+ plate was considered, it
is not blown as to what extent the hat portion participated at
the center of the deck in resisting the membrane stresses.
The reduction in the moment of inertia calculated on
the basis of the full cross-section, is not affected by the
calculated compression stress in the top hat plate (the vari-
ation is less than St). The change in the bending rigidity DY
does not warrant a new analysis.
B. CONCRETE Uf:'.BRELLA SHELLS
The concrete hypars differ from the hypars using
corrugated orthotropic decks mainly in two aspects. For the
of. ~s~~
loads used in the elastic analysis , the shell can
be considered as made of an isotropic n1aterial. Secondly the
bending and axial stiffness of the shell is quite comparable
with that of the edge member.
The experimental work on Strucs. '6', '7' and '8' was
conducted at Corne11 65 . Strucs. '6' and '7' were identical
except for different eccentricity of the edge members (Fig.
4.5). In Struc. '6' the beams were located on the top of the
shell surface whereas they were located below the deck in Struc.
'1'. These structures were loaded uniformly using concentrated
-76-
loads applied discretely over the surface, whereas only half

of the structure was loaded in the case of Struc. '8'.
The elastic properties of the concrete used in the
model were determined experimentally. In calculating the prop-
erties of the shell only the concrete section is considered.
The classical beam theory which assumes the linear variation of
the angle of twist is used in the analysis. The beams are con-
sidered eccentric only in the z-direction. The bea~ properties
calculated are based only on the ribs projecting above the deck.
For Strucs. '6' and '7' , the comparison between the
experimental and the analytical results are shown in Figs. 4.13·
4.19. As shown in Fig. 4.13, ior Struc. '6' the coRpression rib
deflection 68 is about 10\ smaller than the experimental re-
sults whereas the free corner deflection db is about S\ larger
than the .experimental value. For Struc. '7' (Fiy. 4.17), the
deflection 6b is about 60\ and 68 is about 80\ of the experi-
mental values. Though percentage-wise the error in 6a, in
Struc. '7' is about 40t, the magnitudes of the deflections are
very small. Except for t~e deflection profile along the dia·
gonal ob near the column support, the general shapes of the
profile agree fairly well with the experimental values. The
deflection profiles of the tension members ab in both the
structures, where the bending action in the shell dominates
over the membrane action, is very good and almost parallel to
the one observed experimentally.
To verify the idealization of the edge member, Struc.
'6' was reanalyied, but a certain portion of the deck was in·
-11-
eluded as the effective width in recalculating the beam prop-

erties. The modified eccentricity of the bea~ with respect to
the deck and properties were recalculated. It is obvious that
in doing so a certain portion of the deck is duplicated, with
the result that the properties of the edge members are over-
estimated. For the same structure it was found that the free
corner deflection ~b remained almost unaltered (0.022 instead
of 0.023) whereas the deflection oa reduced from 0.016 to 0.012
inches. This observation shows that important deflections are
insensitive to the edge member properties for this particular
structure. However, there is a redistribution of the bending
and axial stresses in the shell, which are of a relatively
small magnitude. The upturned beams used in Struc. '6' seem
to have a pronounced effect in reducing the corner deflection
6b as seen from the analysis as well as experiments. The free
corner deflection 6b for Struc. '6' is nearly half that of
Struc. 1
7' whereas the compression rib deflection 63 for Struc.
'6' is larger than that for Struc. '7'. These points are
further discussed in Chapter V.
Because of the varying size of the edge members, the
axial forces are plotted instead of axial stresses. The ratio
of the calculated axial forces to that given by the membrane
theory is 70-80% for the compression members oa and oc and
50-601 for the tension members ab and be. The analytical and
experimental values of the stresses for the tension member are
in close agreement, whereas the analytically calculated results
for the compression members are on the conservative side.
-7R·
Even though part of the vertical load near the column is car-
ried by the concrete shell, in order to satisfy the static
equilibrium for the vertical load it appears that the experi-
mentally ~easured forces in the compression rib are quite low.
The axial and the bending stresses are ~easured along
the diagonal ob at an angle of 45° with the x and y axes (Fig.
4.15). The measured axial stresses show excellent agreement
with the analytically calculated value of 72 psi. An important
point to note is that the calculated and the experimental values
are about 34% higher than those given by the ~embrane theory at
a load of 40.9 psi, the reasons for which are not readily
apparent. The values of bending stresses are very low and are
not compared here. The variation of the shearing force is
plotted all over the shell for both the structures. Though
there are minor differences in the shape of variation of shear·
ing forces, two important observations can be made. The values
of the shearing forces over a substantial portion of the shell,
are larger than those given by the membrane theory. The shear·
ing force near the coluPm is nearly twice as large as that
given by the membrane theory. This sudden increase in the
shearing force clearly indicates that the shell participates
in tr.:nsmitting a certain portion of the vertical load. The
same behavior is also noted in Struc. 'S'.
Struc. '8' is the same as Struc. '6' but it is sub-
jected to an unsynunetrical load (Fig. 4.20), where half of the
structure is loaded uniformly. Only half the structure along
the line cf (Fir. 4.20) is analyzed using 16 elements in each
quadrant. The statically equivalent load is used in one
quadrant. The central column is idealized by means of con-
centrated elastic springs as given in Chapter III, Section
111.2.F. The comparison between the theoretical and the ex-
perimental results is shown in Fig. 4.20. The deflection pro-
files appear to be quite reasonable though the magnitudes of
the deflections ob and ~e are 30·40t on the lower side of the
values obtained experimentally. A static chect for the ver-
tical load is satisfied at the center column though a dis-
crepancy in the overturninp moment is noted.
A highly irregular pattern of axial forces and moments
is obtained which unfortunately could not he verified properly
because of the difficulties encountered during the experiment.
A better solution can be obtained by using a finer grid (64
elements in a quadrant) and also by using work equivalent loads.
It was not possible to check the imp1.ovement in the solution
because of a limited comnuter core capacity. The exa~ple how-
ever, clearly showed that the theory can solve unsymmetrical
loading conditions such as wind load, etc., and can satisfac-
torily predict the overall behavior of the shell.
The corner deflection 6b in Struc. '8' is nearly
three times as large as that obtained for the uniformly loaded
Struc. '6'. The increase in deflections in the loaded quadrant
is mainly due to the twistinp of the shell about the line ah.
C. m.raRELLA SHELLS WITH STANDARD CORRUGATED DECKS
Four medium scale umbrella shell models (Strucs. '9',

'11' > '12' and '13'; Table IV-1), 12~12' in plan and with a
• Rp •
rise of 14.4 inc~cs, were tested at Cornell Struc. '10'

is a hypothetical structure analyzed to study the effects of
change of shape in Struc. '9' due to the excessive deformations.
Self-tapping screws were used to connect the adjoin-
ing deck panels and also to connect the deck panels to the edge
members. The main supporting edge member frame consisted of
circular pipes (for sizes see Table IV-1) connected eccentri-
cally below the deck.
For the structure havinp, two decY.s nlaced in a ~u
tually perpendicular manner, the decks were not only connected

along the peripheral edges but were also connected intermit·
tently all over the surface. In' the case of the two deck sys-
ten1 the bottom deck was directly connected to the edge memher
whereas the top deck was connected to the bottom deck (Fig.
2.6). All structures were supported at the center column and
a unifor~ load was applied using pressurized canvas rubber bags
with one bag placed under each quadrant: ·( ~ ~~,.\:e.c VU).
The properties of the decks used in the analys~s are
given in Table I"-1. The gage thickness of the deck was
checked by the microweter screw and the properties correspond-
ing to the uncoated decks are used in the analyses. To account
for the effect of rise, the shear rigiaity factors used in the
analyses are modified from the values obtained by the flat
shear tests (Figs. 2.7 and 2.8). The reduction in these values
of a is roughly ZS% for the single deck whereas it is about 15%
for the double decks. Zero torsional fixity between the deck
and the edge membeTs is assUJl'led for all the structures analyzed.
.. Al·
In order to have a better undcrstandinr of the be-

havior of these structures, they are classified into two cate-
gories. This cl~ssification is based on the ratio of the rel·
ative stiffnesses of the deck and the supporting edge ~embers.
Strucs. '9' and '10' are considered to have flexible edge wem-
bers whereas Strucs. '11', '12' and '13' are considered to have
very stiff edge members. The edge ~embers used in Strucs. '11',
'12' and '13' are 4.37 times stiffer axially and 236 times
stiffer flexurally as compared with the edge members used in
Strucs. '9' and '10'. This large difference particularly in
the bending stiffness alters the behavior of the umbrella
shells.
C .1. INVERTED UMBRELLA SHELL l'TITH FLEXIBLE
EDGE T!f~'BERS
Strucs. '9' and '10' were analyzed using the boundary

condition V (Table IV-2) which assumes full horizontal fixity
between the edge memhers and the deck. The convergence char·
acteristics for the co111er deflection ob for Struc. '9' are
..fh -c:. .f...,..c c Ce>.,..,..."-""<
shQl·m in Pip:. 4.22.
de.He c...-hC"'r'I ;....,.crtr
By refining the grid size,
""c.' , Th;, ;, hc.c~c °' J..c.-c(~~,..r:..~l'•· l
.. {
ld-
. c o-n-n r: c..t;--;, e..l41c:. .....,,.... ~' (f.j 3-12}
The difference
in the corner deflection 6b between 6x6 grid size and that of
8x8 grid size is less than 2%.
The deflections and the edge member stre~ses obtained
for Struc. '9' are compared with the experimental results in
Figs. 4.24-4.28. The analysis underestimates the deflection 6a
by 32t whereas the deflection 6e is overestimated by 40\. The
difference between the analytical and experimental results for
the corner deflection 6b is 10%. Comparinr the relative magni-
tudes of these deflections (oa, Bb and oe), it is apparent that
in the case of a flexible edge member the free corner deflec-
tion 6b is of utmost importance. The shape of the deflected
profile for t~e member ab and the relative deflections between
points a and b, by theory and experiments are in close agree-
ment (error ~ 1%). The reasons for the underestimation of the
compression rib deflections are discussed later in this section.
The corner deflection 6b is greater than 10% of the
rise of the hypar shell, which is 14.4 inches. In other words,
the change in the shape of the structure is quite important.
To estimate the effect of the change of shape, a very approx-
imate method was used whereby the same structure (Struc. '9')
was reanalyzed by only modifyin~ its rise from 14.4 inches to
13.8 inches. The reduction of 0.6 inches in the rise was cal-
culated by taking half the difference bett'een the relative de-
flections of the points a and b. The analysis of Struc. '10'
using the modified rise, shows an increase in deflections. The
error in the deflection 6b in particular is reduced further to
4%.
_"1., comparison between the experimental and the analy·
tical results for the axial and bending stre~scs, and the abso·
lute value of the total stresses for the edge me~ters is given
in the Figs. 4.25 and 4.26. The bending and the total stresses
show very good correlation with a maximum error of -15% for
the tension members. Comparing the analytical and experimental
results for the axial stresses, it is noted that the calculated
compression stresses for members oa and oc are on the high side
whereas in case of the tension ~embers ab and be are on the low
side. The measured axial stresses are only about }th in magni-
tude of the total stresses and therefore the deviation {-55% for
the member ab) between the theory and experiment is not consid-
ered to be a serious handicap.
To examine the validity of the solution and also to
help to understand the behavior of hypars, the variation of
the bending moment PY and the in-plane sheariny force Nxy are
plotted over the shell surface (Figs. 4.27 and 4.28). Along
the coluJ111l line 1 (Fig. 4.27), the deck bends with the tension
member like a cantilever {negative moment) whereas in the in·
terior of the span, it acts as a simply supported span between
the opposite edge members. Along the colunm line 8, the deck
has a region of negative bending moments near the supporting
colunm. The variation in the shearing force (Fig. 4.28) is
similar to that indicated for the concrete hypars (Strucs. '6'
and •7•). Near the center of the quadrant, the shearing force
Nxy is larger (by lOt) than the values given by the membrane
stresses. However, one major difference noted between the con-
crete and corrugated deck hypars is that near the column the
deck does not carry a substantial portion of the vertical load
as is seen in the case of Figs. 4.16 and 4.19. The comparison
of the axial, bending and total stresses calculated by theory
and measured experimentally at point e is given in Table IV·4.
The calculated stresses are compared with the aver·
age measured values obtained for the top and bottom deck. Be-
-C4-
cause of the very small magnitude of the stresses, the varia-

ti<>A in their measured value was extreme. The variation in
th~ ~e3sured axial stress ranges from 210 psi to 1780 psi
wh~teas that in the bending stresses ranges from 140 psi to 2840
psi. Though the calculated values appear to be in the vicinity
of t:hese measured values, a direct comparison would not be
fru:i.tful.
In the analytical solution of Struc. '9', it is noted
th9.t the deflection ~a is underestimated. Fif. 4.29 shows a
typj~al connection between the tension member ab and the com-
pression Member oa. Because of the eccentric connection be-
tween the deck and edge members, all the node points are along
the top of the edge F.embers ab and oa. The in-plane forces on
the ~e~ber ab are transferred eccentrically to the member oa
at the node a, resulting in its upward deflection as shown in
Fig. 4.29. In order to illustrate the effect of this eccentric
tran~fer of the in-plane forces, Strucs. '9' and '11' are
analtzed for the two boundary conditions V and VI (see Table
JV .. 2).
For Struc. '9', inspite of certain redistribution of
forces due to the change in the boundary conditions, there is
practically no change in the deflections oe and ob. The re·
lease of the in-plane shear of 328 lbs. acting eccentrically
at polnt a, results in an increase in the deflection ~a by
0.11 lnches (Fit• 4.30). This shear, if resisted entirely by
the cC>mpression member oa acting as a cantilever supported at
point o, produces a deflection of 0.22 inches. Except for the
bending stresses in the tension I"eJT1hers ab and he, the changes
in the stresses for both the edge members and shells are in-
significant. As shown in Fig. 4.31, the decrease in the verti-
cal shearing force due to the release of the in-plane forces
results in the reduction of bending stress at point a in me~
ber ab.
From the consideration of the maynitude of the in-
plane shear and its eccentric transfer, Struc. '11' represents
an extreme case. As shown in Fig. 4.32, the deflection nro-
file along the diagonal ob re~ains practically unaltered for
both hotmdaTy conditions for Struc. '11'. Because of the very
high in-plane rigidity of the 3" diameter pipe, the value· of
the in-plane shear developed at the junction a (Fiy. 4.29) is
quite large (742 lbs.). Thoufh small in ~agnitude, the in-
crease in the compression rib deflection c5c and 6a is almost
200t. The increase in deflection exceeded that which would
have been obtained by considering the edge members oa and oc as
cantilevers, acted upon by the eccentric shears at points a
and c respectively. A small increase in the deflection 6b is
noted and it ~ust be pointed out that the transfer of the
eccentric force also exists at the junction of the tension mem·
bers but it is of minor i~portance.
In t1:e case of con ere tc hypars ,.,here ful 1 fixity be-

tween the edge nember and the shell exists, this transfer of
ec~·.entric forces in two l"Utually perpendicular directions does
not present a problem. To get an exact solution for the dis·
continuities between the edge member and the steel deck, equa-
tions of compatibility will have to he satisfied at the addi-
tional nodal points thereby increasing the complexity and the
storage requirement for the computer program. Both cases pre-
sented here, particularly Stn1c. '11', represents an extreme
class of problems which will be hardly encountered in practice.
Besides the eccentric connection, the in-plane stiffness of
circular pipes is equal to the vertical bending rigidity. On
the assumption of full horizontal fixity between the edre mem·
bers and the deck, the horizontal stiffness attracts high in-
plane shears, the magnitudes of which raise the question of
its validity.
In practice, the rolled sections such as channels
and I-sections have very small in-plane stiffnesses as compared
with their bending stiffnesses. Secondly, these members will
be usually connected along their shear centers by weans of
warped plate connections (Fig. 3.10). One way to correct the
deflection oa is by applying the moroents, equal in magnitudes
but opposite in directions, to those produced by the eccentric
shears at the junction of two eccentric me~bers (Fig. 4.29)
and recalculate the deflections of only the supporting frame.
Since the exact amount of horizontal fixity is not
known, the other alternative is to re-analyze the structure with
a complete release of the in-plane forces (boundary condition
VI) and use the conservative results for the design.
C. 2. INVERTED UMBPELLA Sf-'ELL '\TITH
STIFF EP.CE r.ff 'BERS
Strucs. '11' and '12' used single corrugated decks
whereas Struc. '13' used two perpendicularly placed intercon-
-87-
nccted decks. The structures are analyzed using the boundary

condition VI. This boundary condition is on the conservative
side as far as the computation of deflections are concerned.
Fig. 4.33 shows the comparison of tlte ext>erimental
and analytical deflection profiles along the diagonal ob, for
all the three structures. Besides this, the comparison between
the measured and calculated deflections at points a, b, c and
e is piven in Table IV-5. Durin~ the experiinents, difficulty
was encountered in obtaining the symmetry of deflections. The
unequal rate of leakage fro~ each canvas bag, placed under the
quadrant resulted in an unequal pressure loading being applied
to different quadrants In order to show this resulting un-
symmetry in the solution, Table IV-5 shows the average, maxi-
mum and minimum measured values for the deflections. A com-
parison between the results is based on the average value. In
general, the ~hape of the deflection profile along the diagonal
ob shows a reascn&(:.ly good correlation between theory and ex-
periment. The deflection oe at the center of the quadrant for
a sinr.le deck hypar (Strucs. '11' and '12') is overestimated
by the theory whereas the deflection for a double deck (Struc.
'13') shows a difference of only lOt fro~ the measured value.
Except for the minor scatter of the deflections ~a' 6b and ~c'
the analytical results are within 15% of the average experi·

Aental values.
The axial and bending stresses are measured at five
locations (Table IV-6). For the bendin~ stresses greater than
2000 psi, the experimental and analytical values show a devia·
-RS-
tion of less than ZO~. For very sMall rnapnitudes of stresses

(such as less than 2000 psi) the calculation of the error hased
on the measured stresses 1.fill be misleading. The accuracy of
measurement for the small magnitudes of stresses is always
less. The measured total stresses also show a fair amount of
agreement with the analytical solutions.
The major discrepancy arises in the comparison be-
tween the measured and the calculated axial stresses. Based
purely on the ~embrane theory. the maximum axial stress should
be 1570 psi; as against this, the measured value of stresses
reaches as high as 2440 psi (Struc. '11 ') which is nearly 55%
lar~er than that given by the membrane theory. TI1is appears
inconsistent with the expected behavior, since a part of the
load is also carried by the bending action.
In order to understand the difference in behavior be-
tween the single layer and double layer decks, Fi~s. 4.34 and
4.35 show the variation of the axial stresses and the verti-
cal shearing forces carried by the edfe me~bers. For Struc.
'11' (which has a 28-G single layer deck), both the compression
member oa and the tension member be placed across the corruga-
tions (along the weak axis), carry high axial loads as co~pared
to the mewbers ab and oc, placed perpendicular to the direc-

tion of the corrugations. This trend is also observed experi-
mentally. Because of very lo,,, in-plane stiffness across the
corrugations, the effective area of the deck resisting the in-
plane shear along with the edge members oa and be is very s~all
and therefore the entire shearing forces are resisted by the

-89-
edge members alone. As against this, in a direction along the

corrugations a part of the deck shares the in-plane shear and
subsequently results in the reduction of the axial stresses
in the edge me~bers.
Fir.. 4.35 shows the transfer of the vertical load

to the edge me~bers. ~ith the strong axis of bending placed
parallel to the lines oa and be (Fig. 4.2) the deck basically
bends between the supportinr, lines oa and be. '~'ith Dy == 1845
Dx, practically no load is transferred directly to the edge
meMbers oc and ab. However, froITl the conditior.s of conipatibility
at points b and c, the me~ber be is supported at its end by
members ab and oc. The nerative shearine force at the point
b on the wember be and the constant shearing forces along the
~embers oc and ab confirm this expected behavior! This manner
of transfer of load for a sinrle deck is also reflected in the
bending stresses at points a and e (Table IV-6) '·!hich are
higher than those for double decks (Struc. 1
13').
The measured axial and bending stresses at the center
of the quadrant were highly erratic and did not show any con-
sistent behavior. The minimum measured bending stress was
half the value of the maximum ~easured value at the same lo-
cation. This wide range of scatter is due to two reasons, first
the magnitudes of stresses are too s~all to be reasured re-
liably and secondly there was an unsymmetry due to unequal
pressure loading. For completeness, the comparisons between
analytical and experimental values for the decl stresses is
given in Table IV-4. The bending stresses for Strucs. '11' and
-90-
'12' arc overestimated by tl.c analytical method whereas they

are underestimated for Struc. '13'.
Fig. 4.36 shows the variation of the in-plane shear
force Nxy over the entire shell surface for Struc. '11' An
almost identical variation in the in-plane shearing force is
also obtained in Struc. '12' (24-C sinr,le deck) which has 6li
larger shear and bending rigidities than those of Struc. '11'
(28-G sin{'.!lc deck). For Struc. '13' with a 28-G dou~le layered
deck, the shear force distribution is very similar to that ob-
tained for Struc. '9' with 1 1
diameter flexitle edge ~eJT1hcrs.
However, the rnaxim·1J!'I values of the shear force arc ahout 5-10\
lower for Struc. '13'. The only noticable difference for the
variation of the shear force for sinrle and double deck struc-
tures is that, in the case of the former structure, the 1"1axi-
mum value of the shearing force does not exceed the shearin~
force given by the membrane theory whereas it exceeds the 1"1em-

brane shear force in the latter case. It rnay be of inter~st
to note that the results for the deflections of the deck are
~ .. \I\\.
quite close to t~ose reported in ~ith the gtiff JT1em-
bers, as those usec in St rues. '11', '12' and '13', the def lee-
tions along the free boundaries are small and therefore the
behavior of the shell is quite close to that of an edge-sup-
ported hypar for which, as pointed out earlier, JT1ethods 'a' and
'b' give the same results.
The salient features differentiatin1! the bef\avior of
the hypar with very stiff edge meMbers (St rues. '11', '12' and
'13') and the behavior of tl':c hypars \-ri th very flexible edge
- 91 -
members (Strucs. •0 1
and 1
10 1 ) are further discussed in
dPtail in V. The effect of the edge member weight
Chapt~r
also
on the behavior of hypars is discussed in Chapter v.
The valid 1 t:r and the accuracy of thtL·f ini te element

methods were assessed. Both approaches were found to converge
satisfactorily. A grid of 6 by 6 gave essentially the same
results as a grid of 8 by 8.
For hypars with fully supported edges, both thP flat-
element and th~ c11rved-element methods yielded deflected shapes
that are identical. with those given in the litera,ture.
Satisfactory agref'm~nt was also found with experimental results
even when the eff t-cts of accentric edge members wPre included.
However, the deflections of flat corners, such as those
at the outside corners of umbrella-type hypars, 'Clre
underestimated by the curved-element method. The flat element
approach predicts the experimental deflections and stresses
of various types of hypar structures with satisfactory
accuracy.
CiiAPTL ~ V
QUA~ITATIV~ EFFECTS OF PfUNCIPAL VA"IABLES

OH BpHAVIOR OF BYPARS
V.l INTRODUCTION
Based on the analysis of some selected structures

(Table IV-1) it is possible to show qualitatively the effects
of different parameters on the behavior of a hypar shell.
Since the number cf parameters affecting the behavior of the
shell is quite lal'ge and their interaction is very complex,
attempts to sho"'' their effects on the structural behavior by
means of formulae would involve extensive cowputational work.
During the following discussion sowe of the parameters which
were not investir.ated arc ~entioned.
The structural variables affecting the behavior of

the shell can be broadly classified into four catep,ories:
(1) Geometric shape of the hypar shell.
(2) Properties of the deck used as a hvpar surface.
(3) Boundary con<litions.
( 4) Loadinr,.
V.2 GEOMETRICAL SHAPE
All hypar surfaces have a constant twist curvature

c The effect of rise to span ratio on the central deflec-
n·
tions of the saddle shaped hypars (Strucs. '3', '4', Table IV.l)
is illustrated by plottinr. the deflections ar,ainst the non-
din:ensional parameter of~ (Fig. 5.1). With the increase of
c
-93-
rise to span ratio. the curvature of the surface increases.

This increase in curvature reduces the bending action of the
shell whereas the membrane action is increased and this even-
tually leads to the decrease in the central deflection.
The cf feet of the rise can be shown by comparing the
central deflections of a simply supported 28-G square plate
(60"x60" in plan) with those of a hypar having a rise of 7.5
inches (rise to span ratio= 1/8) (Fig. 5.1). The deflections
in the latter case are nearly 40% of those obtained in the
former case.
The sensitivity of the structural behavior to the
change of rise is well de~onstrated by comparing the deflec-
tions and the stresses for St rues. '9' and '10' (Table v .1)
where the rise of !>true. '10' is only 4.3% s11'aller than that
of Struc. '9'. The increase in the bending action with the
reduction in rise is evidenced by the increase in the deflec-
tions o3 , ob, oc, and ~e and also in the hendinR stresses.
The bending stress in the center of the deck increases
from 1870 psi tc 2130 psi. According to the ~embrane theory,
the in-plane shear force is inversely proportional to the rise
g__M The same trend is also observed
to span ratio (Nxy • --ic-)·
in the increase of the in·plane shear and the axial edge meM-
ber stresses (Table V-1).
For larger values of~
c
Cf< tl• the membrane action
is reduced to a minimum and the entire load is practically
carried by bendine action. The calculation of the in-plane
shear on the basis of the membrane theory, as C approaches zero,
-94-
is meaningless. The theory given here is primarily good for

rise to span ratio of ~ i
(AB/C 2 : 15) but it can be used for
greater rise with loss of accuracy. From the construction
point of view, the choice of rise to span ratio will be also
governed by the warping of the deck.
V.3 DECK PROPERTIES
In the case of an open form deck, the membrane elastic
constants Ext' E1 t and the bending constants Dx, o1 and Dxy
(Fig. 2.1) are very small and their influence on the behavior
of the shell is insignificant (for the stiffness coefficients
see Appendix B). However, in the case of the closed cellular
decks, though the reagnitudes of Dx and o1 are small and insigni-
ficant, Ext' Elt and Dxy are comparable in magnitudes to the
properties Eyt' Dy and Exyt and therefore their influence on
the structural behavior cannot be overlooked. Since only one
structure was analyzed for the cellular deck (Struc. '5'), the
discussion given below primarily concerns the open form (stan-
dard sinusoidal) decks.
A. SHEAR RIGIDITY
According to the membrane theory, the normal loads
on the hypars are carried by the in-plane shearing force Nxy•
In reality, though a part of the load is carried by bending,
the magnitude of the in-plane shear Nxy is quite comparable to
that given by the membrane theory (Figs. 4.12, 4.16, 4.19, 4.28,
and 4.36) and even exceeds it in certain regions of the shell.
Therefore the in-plane shear resistance Exyt • Geff•nt, is very
important in the behavior of hypars. As discussed in Chapter
-9 5-
II, the effective shear modulus (Geff) is obtained by reducing

the shear modulus of the materjal by the factor a. In the
s ~1ddle shaped "'iypars (St rues. '3' and '4 ') for a rise to span
ratio of 1/5 (AB/C 2 = 25) (Fig. 5.1), the reduction in the
shear rigidity a from 0.06 to 0.04, shows an increase of nearly
30% in the central deflection. The behavior of the hypar shell
is very sensitive to the values of a < 0.10.
To illu5trate the effects of a on the behavior of
the shell, the results for Strucs. '13' and '13a' are compared
in Table V-1. Nith the increase in the value of a, the deflec-
tions (lie' ob) and the edge mernter and deck bcndinf! stresses
are reduced whereas the axial stresses in the edge members and
the in-plane shear force ri xy are increased. Except for the
axial forces in the edge me1!1bers and the in-plane sJlear Nxy ,
the response of the structure to the variation in a is siJnilar
to that of the variation in the rise to span ratio. The
optimum value of a in orthotropic hypar structures is a ~ 0.1
since larger a does not improve the behavior much. Factors
which improve the value of a ~ere already discussed in Chapter
I I.
B. THICKNESS OF THF CORRUGATED DECK

In the case of an open deck the important membrane
properties such as Eyt, Exyt and the bending riridi ty DY are
directly proportional to the thickness of the deck. However,
it must be pointed out that the bendinr rigidity of the deck
is small compared with the membrane stiffness. t•orcover it is
the change in the shear stiffness that influences the behavior
of the hypar shells and therefore the effect of increasing the
thickness is analogous to that of increasing the value of a.
To substantiate this observation the comparison between the de-
flection and stresses for Struc. '11' (28-G single deck) and
Struc. '12' (24-G single deck) is given in Table V-1.
The variation of the in-plane shear rigidity, which
is directly proportional to the thic~ness and the shear
rigidity factor a, also affects the manner in which the verti-
cal load is transferred to the supports by the membrane action.
Because of the high shear rigidities for the concrete struc-
tures ('6' and '7') and Struc. '5' using the cellular deck with
the full effectiveness of the bottom plate, the values of the
in-plane shearing forces show a substantial increase near the
supports (Fir,s. 4.12s 4.16, 4.19). The increase in the shear-
ing force indicates the participation of the deck in carrying
a part of the vertical load. As ag~inst this~ Strucs. '9'-'13'
with low shear rigidity do not show any incr~ase in the in-plane
shearing force (N xy ) near the supporting columns (Figs. 4.28,
4.36). In other words, in these structures the entire vertical
load is primarily carried by the edge members.
C. NUMBER OF DECKS
As far as deflections and stresses in a hypar are con-

cerned increasing the number of decks has the same effect on
the behavior of the shells as that of increasing the shear
rigidity factor a and the thickness. However, this observation
docs not apply for buckling (see Chapter VI). As discussed in
Chapter II, the effectiveness of the deck in resisting the
loads depends upon the l"lannel" in l·rhich two or JJlore decks are
interconnected and connected to the supportinp edge members.
However, it must be pointed out that in order to avoid chatter
and get a better structural performance, it is desirable to
interconnect the decks all over the surface of the shell.
l''hen two decks are used, they are placed in a mut •
ually perpendicular manner ancl this gives an equal bending
rigidity to the structure in both directions, thereby distri-
buting the applied loads more evenly to the supporting edge
members. TI1c comparison of the results for Struc. '11' using
a sinr,le deck (28-G) and those for Struc. '13a' using the
double dcc1~, all other constants being the same, shows that the
uniforJT1ity of the stiffness in Struc. '13a' has more even
distribution in the edge member axial stresses (Table V-1).
Though the corner deflection shows practically no change, the
center deflection oe for Struc. '13a' is nearly half that of
Struc. '11'. The change in the bending stresses of the edge
member is very small but because of the increased membrane
action the bending stresses in the center of the quadrant are
reduced by nearly three times.
In practice, the use of a double deck with two decks
placed mutually perpendicular is more desirable than a single
orthotropic deck.
V. 4. BOUNDARY CONDITI01'1S
From the practical point of view, boundaries such as

si~ply supported, knifc·edfe supported or fixed all around, are
not realistic. Boundary conditions which consider the proper·
ties of edge members and the i.a.iJmer in which they arc connected
to the deck are realistic from the practical point of view.
A. EDGE rm:BEP Pr.OPERTIES
The edge me~ber properties Ab, Iy, Iz, J and rb for

available rolled sections show variations over a wide range.
A sufficient nuF1rer of analyses could not be carried out to
formulate any definite rules by which the effect of the varia-
tion of these individual properties on the behavior of the
shell can be assessed. Except for the concrete hypars, the
analysis was carried out for zero torsional fixity and there-
fore the influence of the torsional constants .J and r is not
clearly lmown.
To get the general idea of the effect of the stiff-
ness of the edge members, one can compare the results of Struc.
'13' with very stiff edge members and Struc. '9' with very
flexible edre me~bers. The difference in the behavior of these
two extreme structures is obvious from the deflection profile
along the diagonal ob (Figs. 4.24 and 4.33). In the case of
Struc. '13' because of very high benc!ing rigidity of the edge
members, the deflections along the periphery are quite sMall
and the deck bends freely between the opposite supportinp,
edges. TJ~.e siJl'lply supported plate bending action is quite
doIDinant in t~is case. Because of the small bendinp, rigidities
of the edge members in Struc. '9', it appears from the deflec·
tion profile along the diagonal ob that it is the deck that
supports the edge JT1embers near the f rce comer and therefore
the deck stiffnesses {both bending and membrane) are quite
-99-
important for this structure.

The fact that the co~ncr deflection 6b for Struc.
'13' is not very different than that of Strucs. '11' and '12',
where single layer decks with different shear ripidities and
thicknesses are used, clearly indicates that the deflections
alone the periphery of these structures priJilarily depend upon
the properties of the edge ~embers. In order to optimize the
interaction betweon the deck and the edge members to give a
satisfactory structural performance, the ratio of the bending
rigidities of the deck and the edge mcwbers would have an
optimum value between the t·vo extreme cases (Strvc. '9' and
St rues. ' 11' - ' 13') .
B.
-
EDGE PE}'P.EP. AND DECK CONNECTION
As shown for Strucs. I 6' anc I 7 I (Table IV-1) , the

eccentric location of the edge me~bers affects the deflection
of the structures (Fips. 4.13 and 4.17). The difference in
behavior of the edge mc~bers is shown in Fip. 5.2. For umbrella
shells to reduce the vertical deflection for the compres·
sion member, it is beneficial to connect the deck on top of the
edge member whereas in the case of the tension ynembers, it is
beneficial to connect the edge meMber on the top of the deck.
The experimental as well as the analytical results for Strucs.
I 6 I
and I 7 I seeJil to confirm this conclusion.
No comparative results are presented for the zero
and full torsional fixity, though results are presented
for
the full and zero in-plane fixity (TH) between the edge JTtembers
and the deck (Figs. 4. 30 - 4. :s 2) • Though the results are very
-l'lO-
limited, it is helicve<l that providing fixity along the peri-

pheral edge wembers tends to attract More vertical load on the
edge Fiembers (Fig. 4.31).
In the case of saddle shaped hypars, the increase in
area of the tension tie bar connectinr the lower corners (Key
sketch, Table II-1) of the shell and the in-plane bending
rigidity of the peripheral edge members have beneficial effects
in reducing tJ1e bending action of the she11 19 • The effects
of these variables nay need further investigation.
V. 5. LOADDJG
All the conclusions given above on the behavior of

the hypars arc based on the analysis for the uniformly distri-
buted vertical loading. In reality the structures are also
subjected to unsym1r.etrical loads such as wind or drlftinv
snot·!. The strength of the structure under these kinds of loads
is tested more severely than under the concHtions of uniform
loads. The unsymmetrically loaded Struc. '8' shows the cor·
ner deflection o10 nearly three times as large as that obtained
for the uniform loading condition.
A. EDGE r1Et '.BER V!E I GHT
In case of some shells, such as umbrella shells, the

edge member weight is distributed along the periphery of the
shell. The customary procedure of st"earing this load uniforrr.ly
over the whole surface and analyzing the structure can lead to
a gross underestimation of both the deflections and the
stresses. To deMonstrate tnis, an umbrella shell with each
quadrant of 20'x20' in plan having a rise of 4', (Table 1''·1)
-101-
is an:.1lyzcd. The edge niembcr sizes an(~ the deflection toler-

ances used for thi~ structure represent tho values ~hich are
encountered in practice. The deflection profiles and the bend-
inr stresses for the edge Menbers, with and ~ithout the inclu-
sion of edge tnernbcr weights, are shown in Figs. 5.3 and 5.4.
The weight of the edge member is 20% of the total uniform load
of 40 psf over the whole surface. A simple frame analysis
consiaerinc only the edge me~ber weight and edge members, would
have given an increase of 0.74 inches in the deflection of
point a (as against 0.33 inches) and 0.202 inches in the defJec-
tion of point b relative to point a (as against 0.15 inch).
This sh0trs the effectiveness of the shell in carrying the
weight of the edge meMbers. The cable and the arch action along
the diagonals ac and ob is evident in Figs. 5.~ and 5.4. The
increase in the deflections ~a and ~c produces an upward de-
flection at the center of the span.
The axial stresses for both the tension and the com-
pression members sho1.; an increase of nearly 20%. This is equal
to the increase in the total load of the structure by the in-
clusion of the eclge member weight. The bending stresses for
both the tension and the compression edge Members show an in-
crease of nearly 50% in the maxi1 11 um stresses (Po in ts o and a,
Fi". 5.4). The increase in the bending stresses clearly in-
dicates the unconservative assumption of sr.>earing the edge
me~ber weight over the whole surface.
The effectiveness of the shell in carrying the weiR~ts
of the edge me~bers raises an important question as to the

-102-
method of construction. The situation is analogous to that

encountered in a composite construction usinp. steel beams and
concrete slab. There are three alternatives for the construe-
ti on. Depending upon the size and shape of the shell, ,.,i th the
decks in position, a hypar can be built on the ground and erected
in position; or it can be built in place by using an adequate
shoring for the edge me~~~rs. In these methods of construction,
the effectiveness of the shell in carrying the weight of the
edge member will be utilized. The third way of construction
will eliminate the shoring and depend entirely upon the strengths
of the edee members. The economics will obviously decide the
method of construction.
From the analysis of the different structures, it is
found that the axif.l stresses in the edge rnell'bers derived on
the basis of the wernbrane theory are always overesti~ated
(Table V-2). Because of the relatively small ~a~nitude of the

edge member stresses in comparison with the bending stresses
and with the unc~rtainty in the calculation of the exact bend-
ing stresses, the design of the edge members for the axial
stresses based on the me:rnhrane theory cannot be considered to
be on a very conservative side.
From the analysis of different structures it appears
. .
that the non-d1rnens1ona aCn t prov1. d es a goo d in
1 parameter ~ . d ex
for the behavior of the shells. The higher the value of this
constant, the more dominant is the membrane action. As dis-
-103-
cusseG earlier, the beneficial membrane action reduces both

the bending stresses and the deflections of the shell. A
second good non-dimensional parameter would be the relative
stiffness of the deck and the edge members. This however
would need further study.
C!L\PTEf' VI
!!i§_TABILITY ANALYSIS 11F HYPAP.S
VI. 1. INTRODUCTION
The linear stiffness analysis given in Chapter III does

not include the effects of middle-surface forces Nx, NY and
Nxy on the behavior of hypar shells. The omission of these
effects precludes the possibility of the analysis of insta-
bility of the individual finite elewents. The accumulation of
the instabilities of the individual elements eventually leads
to the general instability of the structure.
Jn the case of a hypar ~-rith light gage steel dee}·
used as a shell surface, the eff~ct of the in-plane forces is
manifested in three different types of instabilities:
( 1) Local BuG:kl ing • In the case of a ~ 1 -type open deck
or a cellular deck (Fig. Z.2b and c) the deck is composed of
flat plates. These individual plate components, depending
upon the thickness to width ratio and the boundary restraint
offered by the adjoining plates, rnny buckle locally when suh-
jected to in-plane compressive and shearing forces. In spite
of the uncertainty in the degree of restraint offered by the
adjoining plates, this local buckling can be approxi~ated on
the basis of the stress level in each component plate 36 . The
effect of t~e local buckling on the behavior of light gage
bea~ section is discussed in detail in refs. 38, 39.
The local buckling of the individual plate components
-104-
results in tl c redistribution of the total stiffness of the
1
shell. The theory used in thi5 chapter does not account for
the local buckling and therefore the effect of local buckling
on the shell stiffness cannot be predicted. The local buckling
can be prevented by choosing proper thickness to width ratio
for each individual plate elemont.
(2) neck Buckltne - In this mode of buckling, the edge
beams re1l'ain stable whereas tho deck, used as a shell, buckles
as a unit. To understand the deck buckling, consider the
umbrella shell in Qip. 4.2. T~e deck acts primarily as a com-
pression arch between the points o and h, and therefore it can
bucY.le along the diagonal ob; but the shell edge members re-
main stable.
(3) Qveral! Buckling - The shell and the edge members
buckle as a whole t1nit. One c-.n iroagine an umbrella shell,
foldinr down as an umbrella turned inside out. Overall
buckling could occur either simulataneously with the deck
buckling or it can happen after the deck has buckled.
According to a siwpl:lfied analysis by Parker 19 , the
possibility of overall bucklinp. for the practical size of edge
memhers is very remote. Very high values of deflections and
stresses for both edge mewhers and the deck will indicate the
possibility of overall bucklinri. T~e conclu~ion that overall
bucklin~ is very unlikely is further verified by Struc. '9'
{Table IV-1) tested at Cornell , where l" diameter standard
pipes were used as edge ~ewber1. The resulting structure was
too flexible to be used in practice. In spite of excessive
-106-
deformations (nearly one l1alf the rise of 14. 4':), the struc-
ture did not show any tendency of overall bucl<linp, though the
deck buckled.
The present study was primarily concerned with deck
buctling. However, the overall instability due to the buckling
of the edge members can also be predicted from the load deflec-
tion curve. The assumptions used during the analysis and the
li~itations of the theory are as follows:
(1) A linearized stability analysis was carried out to
predict the bifurcation point of buckling 51166 • The prebuck-
linp. deformations were within the limits of small deflection
theory.
(2) No attempt was Made to predict Dost-buckling behavior
or the post-buckliny stren~th. To be able to predict the
post-buc~lin~ behavior, one needs to retain the higher order
strain terms in the strain displacement relationships and have
. }
h ig1er or d er matrices
. 67 . It is extremely difficult to formu·
late these watrices explicitly and one has to resort to
numerical integration. The non-linear equations can be solved
by the use of methods such as Newton-Raphson scheme 25 , ener~y
.
searc h tee h nique 67 , c t c.
(3) The possibility of local buckling ~as totally

neglected.
(4) The material was assuwed to be linearly elastic.
(5) Bucklin{' was assumed to be conservative 36 •
Both curved and flat ele~ent approaches were used.
-107-
VI. 2. INCREr·lENTAL r1iATRIX FOR THE DECK AND EDGE I'!IH!BER~

------
In order to represent the instability effect in the
finite clement analysis, the change in the potential energy
due to the middle-surface forces Nx' NY and Nxy' which occurs
durinr the flexural action is to be included 66 • For the con-
stant values of Nx, NY and rJxy at any prescribed load level,
the potential energy due to in-plane forces assumes the form of,
With the inclusion of the work done by the in-nlane

forces, the total potential energy can be written as,
~ = [~j [K] {6} + 41 [N] {A} - (A] {P} 6-2
For stable equilibrium, the first variation of the total po-

tential energy is zero.
{P} = [[K] + [ll]]{A} 6-3
{P} = [K]eff {6} 6-4
The matrix (N] is called the incre~ental matrix and.it is ob-

tained by the second differential of the potential energy (Fq.
6-1) with respect to nodal displace~ents.
a2v.T
[ N] i j c: afl . '~ ~ . 6- I
l J
The coefficients of t 11e increwental matrix (N], depend only
upon the geometrical parameters of an element, such as its
len~th. The incremental matrix is identical for both ortho-
tropic and isotropic cases.
For constant values of Nx, NY and Hxy the increynen tal
-108-
matrix for a shallow shell hypc.r element and that for flat
plate elements are identical. The only difference is the ~anner
in which the in-plane forces are deter1nined (sec Chapter III).
The incremental matrix [N] for the deck is given in Appendix

D, Tables D·I to D-IV.
Due to the presence of the axial force Nx, the effec-

tive stiffness of a beam clement is also modified. Ner,lecting
the torsional mode of buckling, the potential energy due to the
axial force i~x can be obtained by putting the values of UY an«~
~xy equal to zero, in fq. 6-1. The procedure for determining

the incre~ental matrix for the ~earn is identical with that for
the deck. The incremental matrix for a beam element is given
in Appendix E. The increPental rnatrix for the whole structure
is obtained by the same procedure as described for the formula-
tion of the rnastcr stiff~ess matrix in Chapter III.
VI. 3. CHECKii~G CF THE INC~fPENTAL ~~ATRICES
Before analyzing hypar structures, it is necessary

to establish the validity of the incremental matrix P,iven in
Appendix D. The determination of the in-plane l:>uc~.ling loads
(N x, Ny, Nxy) for flat plates provides a P;ood c:;ec](. At a
critical loa<l, absolute magnitudes of the deformations are
indeterPinate and the determinant of the effective stiffness
matrix [K]eff must vanish.
I [ [K] + A [NJ] I = 0 6-6
where '- is the eigenvalue which depends upon the applied state
of me~brane stress e.g. for an uniaxially co~pressed plate,
along the x-direction it will give eifenvalues corresponding
-109-
to the in-plane force Nx (see Table VI-1). From the structural

point of view, one is only interested in the Minimum critical
load. To achieve this, it is necessary to rearrange fq. 6-6
to get the first eigenvalue corresponding to the critical load 68 .
I Cf [I] + [KJ- 1 [N]] I • 0 6-7
The negative reciporcal of the first eigenvalue of the matrix

[Q] , where,
[Q] a [[K]-l(N]] 6-8

\Jill give the critical value of the f!lembrane force. A sub-
routine named 1 ·NrtOOT 1 ' available in the IBr·~ systef!l/360 Scientific

Su~routine Package 69 , calculates eigenvalues and eigenvectors
of a real, square, non-symmetric matrix given in Fq. 6-8,
where both [K] and (N] arc real symmetric matrices and [K] is
real positive definite. In order to avoid underflow and over-
flow in the computer program, it is necessary to divide'. both
[i'l] and [K] matrices by some lar~e number, like 1000.
The first three problems solved were uniaxially com-
1•resscd simply-su~ported plates with or without stiffrncrs (Table
VI -1) . In general, the buckl inr. in -plane force l'J x ( 1 bs/ inch)
is given by,
2
~'.
i~x = 7
K'IT D 6-9
where K is a constant depending upon the aspect ratio a/b and

also on the relative stiffness of the plate and the stiffener.
The ratio of the stiffnesses of the plate and the stiffener are
given by non-dimensional parameters y and ~
36
.
a EI A
y l).b ~ • bt 6-10
-110-
I is the moment of inertia of the eccentrically connected

stiffener calculated about the junction of the stiffener and
the plate. For these problems, the torsional mode of buckling
of the stiffener is neglected. The error for the values of K
for these three cases is less than 0.2% as compared to the
classical solutions. One of the interesting observation for
these problems was that the plate with the aspect ratio of
a/b ~ 2, buckles in a double sine wave with zero deflection at
the center line. However, with the attached stiffener it
buckles in a single sine wave. V~ile analyzing only a quad-
rant of a plate, proper boundary conditions are to be applied
to account for this behavior.
Since the shearing action is of priwary iYT1portance
in the case of hypars, the shear buckling loads for a square
isotropic and for a 24-G standard corrugated flat deck were
also calculated. The values obtained for the critical shear-
ing force I! in the isotropic square is compared with Timo-
xy
shenko 36 and that obtained for the corrugated deck is com-
pared with Pcfarland 48 • error bet1•reen the classical solu-
The
~"' ~s ~~
tions and that obtained · for the shear bucklinp:
(N ) is more than that for the uniaxial compression (Nx).
xy
One reas0n for the greater error is that the assumed displace-
~ent field for the displacement w (Herwitian Interpolation)
closely approximates the buckled surface for an axial compres-
sion. To approximate the bucklect wave form due to shearing
load, a greater number of elements is required to achieve equal
accuracy. The error for the critical shearing force for the
-111-
orthotropic deck is 7.46% on the high side compared with a

simplified formula 48 . However, according to the authors 48
their formula widerestil'lates the critical load by as much as
by 5%. Therefore the actual error may be considerably less
than 7.46%. The error for the isotropic plate with only 6
eleMents is 4.65\ (Table \'I-1).
The correlation between the classical solutions and
the solutions obtained here is considered adequate to sub-
stantiate the incremental matrices for the beam and the plate.
A further check ui 11 be presented in the Section VI-6, for
the case of an isotropic hypar for which a classical solution
is available 5 .
VI.4. INSTABILITY OF HYPARS
The increl'lental matrix {N] is a function of the in-

planc forces Mx, NY and Nxy· In the case of plate bucklin~
problems there is a complete uncouplinp between the flexural

and membrane action. This enables one to formulate the [N]
matrix from a piven distribution of the in-plane forces which
a.re predetermined, independent of the flexural action. ln the
case of a hypar, or for ttat matter any curved shell surface,
the values of Nx~ Ny and Nxy arc dependent on the deflections.
Nith the change in the applied loadinp, the magnitudes of the

in-plane forces also change. In other words, there is a
coupling between the JT1ernhrane and flexural behavior. The in-
crease in the in-plane forces resulting from the corresponding
increase in the load causes some of the elements to undergo a
marked decrease in the effective stiffness. This reduction in
-112-
the effective stiffness of an element will adjust the incre-

mental force distribution. The accumulation of these local
element instabilities will eventually lead to buckling. The
discussion of this membrane and flexural behavior for the shell
structures is given in detail in ~efs. 51, 66.
The incremental [N] matrix used for both curved and
I
flat elements is identical and can be represented by,
6-11
The only difference is that u, v, and w for the flat

clements are ~easured along local axes ~1ercas those in the case
of the curved ele1!1ents are measured along the tangent and nor-
~al to the surface. As pointed out in Chapter Ill, the transfor-
mation frow the local to the global system for the flat cle-
ments can result in non-zero entries in all clements of the
(N] matrix and the matrix assumes a general form of,
-n11 n1~
--------- _;l 6-12
[N]global = w
n21 nzz
whereas its basic form remains unaltered in the case of the

curved clement.
VI. 5. DE°TEI'J'INATION OF
- -TI-IE BL'"Cl\LING- _,,..,.,__.
WAD
The linear eigenvalue forwulation for the determin-

ation of the eigenvalue and thereby the lowest buckling load
is well documented in the ~efs. 51, 66 and therefore it is not
repeated here. Because of t~e lack of a reliable eieenvalue
subroutine for the large-order systems and sufficient computer
-113-
storage, the author could not use the direct eigenvalue approach.
Instead, a linearized load incrementation Jnethod had to be used.
In the load incrementation Jnethod, as used by the
author, the assumption is made that the in-plane forces Nx' Hy
and I~xy are constant during an incrcrnental step and are equal
in magnitude to the value at the end of each step. The proce-
dure of the solution can be demonstrated by the use of Fig. 6.1.
In the incremental step I, only linear analysis is
carried out by solving the linear part of the equation assum-
ine [N] as a null matrix.
[~] • [K]-l [P] n·l3
From the kno~m values of displacement vector [A], corresponding
in-plane forces Nx, try and Nxy are calculated and. the incre-
mental matrix [N] is formed. The effective stiffness matrix
[K]eff is used in iterative cycle II.
The iterative cycles are continued till convergence
is obtained for the nodal displacements and consequently the
incremental matrix [N] is consistent with the deforJnations.
It is found that for swall incremental loads, convergence of
displacements is obtained d thin three or four cycles.
1 An
incremental load op is applied on the l'lOdified effective stiff·
ness matrix and the increase in the displacef"lents 6 and A is
calculated by finding new values of the in-plane forces at the
end of the step II by the iteration as clescribed. before. The
analysis is continued by applying the increment of the load 6p
on the previously determined effective stiffness ~atrix.
Any sudden change in the load deflection curve be-

-114-
tween any two load levels indicates the occurrence of bucklin~.
In this method it is possible to predict only the range within

which the buckling occurs. Since the solution near the un-
stable configuration is very sensitive, it is not possible to
determine the exact point of buckling.
VI. 6. NUMERICAL RESULTS AND THEIR DISCUSSION
Reissner 5 analyzed the case of a simply supported

isotropic hypar with edge wembers having infinite rigidity
along their axes but having zero stiffness in a plane tangent
to the shell surface. The saddle-shaped hypar, Struc. 'l'
(Table IV-1) was analyzed for these boundary conditions. The
deflection profiles at three points along the compression dia-
gonal bf are plotted for the load level of 0.20-0.so kgms/cm 2
increased by the interval of 0.1 kgm/cm 2 . The deflection pro-
files along the tension and the compression diagonals are sh?wn
in Figs. 6.2, 6.3. The sudden change of deflection profiles
(Fies. 6.2, 6.3) between the loads 0.40 and n.so kgms/cm 2
clearly indicates that the bucklinr occurs between these two
limits of loads and moreover close to 0.40 kgms/cm 2 • The
analysis based on tl\e curved element for the saine structure,
also predicts the load between the same range though some dif-
ferences in the deflection profiles are notec.
The load thus predicted is slightly higher than that
given by Reissner (0.38 kgms/cm 2) but this is because of the
fact that Reissner used linearized membrane analysis. To ver-
ify this fact, analysis was carried out where predetermined
membrane shearing force Nxy • ~ was used to establish the
-115-
incremental matrix. From the deflection profiles not shown

herein, the buckling of the shell occurred between uniform
loads of 0.35-0.4 kems/cm 2 which is in the range of the loads
predicted by Reissner. One of the interesting points is that
for the linear elastic analysis, the normal deflections w were
symmetrical about the lines bf, dh, cc and ag (Fig. 4.1).
Si~ilar observations were also made by Deak 18 . However, after
the inclusion of the instability effects ci1e symmetry of the
normal displacement is still retained about the diayonals bf
and dh but there is no symmetry about the lines ce and ag. This
is because of the readjust'Plent of the effective stiffness due
to the in-plane tension and compression forces.
In order to assess the effect of edge deflection and
the stiffness of the edge in.embers on the buckling of the hypar
deck, an umbrella shell with h ratio equivalent to that of
Struc. 'l' (Table IV-1) was analyzed for two different sizes
of edge members. For all edge ~embers of size 6x3 ems. {18
sq.cm. cross-section area), it appears (Fig. 6.4) that buckling
occurred between the loads 0.20 to 0.30 kgms/c1"1 2 , which is
nearly ~ne
half that of the all-supported case discussed before.
According to Leet 28 , the buckling of the hypar shell primaTily
depends upon the axial stiffness of the edge me~bers and not
on the edge deflections. The boundary conditions used in
Rcissner's solution correspond to the infinite axial stiffness
of the edee member. To verify Leet's conclusions, the same
hyp~r was analyzed with the edge bea~s having the sa~e proper-
ties as those of the 6x3 size beam except for the cross-sec-
-116-
tional area, wilich was increased froJT' 18 sq.cm. to 108 sq.cm.

Though the deflection profiles did not show the buckling very
clearly up to 0.50 kgms/cm 2 , there was a small deviation in
the deflection profile at about 0.40 kgms/cm 2 (Fig. 6.5). It
appears that the buckling depends upon the cross-sectional area
of the edge member and not so wuch on the edge deflection. It
is believed that for the stiff edge members used in Strucs. 'll'-
'13' the deck buckling load can be predicted by calculating the
bucklin~ of a single quadrant of the umbrella shell with all-
supported edres.
Two hypars, Struc. '13' uith double 28-G corrugated
decks and Struc. '12' with a single 24-G corrugated deck were
analyzed to determine the buckling load. In the case of a
double deck structure, the deck buckling load is between the
ranee 200 to 243 psf. (Fig. 6.6). Experimentally the struc-
ture was tested up to 145 psf and no deck buckling ~as obser-
ved, though deck tearing along the lines of connection was
noticed.
Struc. '12' with a 24-G single deck was analyzed
usin~ both curved and flat elements. The experimentally ob-
served deck buckling load for this structure was in the vicinity
of 75 psf, but from Figs. 6.7-6.9 ~oth curved and flat ele-
ments analyses predict a lower buckling load. One of the pos-
sible reasons for this is that the pressurized canvas bags
used as loading devices, tend to offer some restraint to the
buckling of the shell. It is known that even a small external
restraint to the shell surface is adequate enough to raise·the
-117-
buckling load substantially. Comparing the deflection pro-

files in Pips. 6.7-6.9 it appears that the flat elements pre-
dict the load in the vicinity of about 46 psf whereas the curved
clements Predict the load in the vicinity of 60 psf. fts
pointed out in Chapter IV, the curved and flat elements, for
hypars with all edFes supported give very close results for
the linear elastic analysis. It appears that the flat ele-
ments give very conserv:•tive results for the buckling of single
decks. As pointed out in connection with Eq. 6-12 for the flat
elements, the transformation of the incremental matrix froJF1
local to the ~lobal axes results in the modification of the
flexural as well as membrane stiffnesses. Since the in-plane
membrane stiffness for a corrugated deck is very small in the
weak direction, a premature buckling could possibly be trig-
gered by the reduction in the me~brane stiffness. As against
this, in the curved element formulation because of the
assumptions used in the analysis, the membrane stiffness is not
modified.
For a double deck or an isotropic deck, because of
hi!~h in-plane rigidity in both directions, hoth methods pre-
dict the buckling loads in the same range. As pointed out in
Chapter IV, the bending action is very dominant in t~e flat
portion of the shell. But in the buckling analysis, the curva-
ture of the hypar in the center of the quadrant may be more
critical. This fact occurs even more in the case of a single
deck and therefore the curved element, which accounts better
for this curvature effect, probably predicts a hi~her load.
-118-
Peissner 5 in his analysis of buckling of isotropic

hypar shell, has indicated that except for a difference in
numerical coefficients, the critical in-plane shearing force
for the shell differs from that of a flat plate by the fact
that a thickness square factor is replaced by the product of
shell thickness and shell rise. However, in the case of an
orthotropic shell, the increase in the critical shearing force
does not appear to be as high as in the case of an isotropic
shell. Because of equal bending rigidities in both directions,
the buckling load for a double deck hypar shell appears to be
three or. four times larger than that of a sinr,le dect. This
is also observed in the case of the critical shearing stress
for a flat plate.
Struc. '12' was also analyzed using the identical
boundary conditions as the structure analyzed in Fig. 6.7, but
allowing the edges to deflect& As pointed out previously,
though the exact point of deck buckling is not known, the deck
buckling occurs between the loads of 43.0·SO.O psf. The in-
teresting point to note here is that even though the deck
buckled (Fies. 6.10, 6.11), the deflection of the edge JT1embers
"as still quite linear. The same trend was also observed dur-
ing the experill'ent. The bucr.ling also showed a similar trend
as ~bserved in the case of an all·supported hypar. The deflec-

tion profile alonF, the tension diagonal did not show any
huckling (Fig. 6.10).
It may not be convenient to analyze every structure
in practice by this load incrementation method. According to
-119-
Reissner 5 , the buckling load for an isotropic hypar shell is

given as,
qcr & 4(C/AB) 2/D Et 6-14
where D is the bending rigidity and Et is the membrane stiff-
ness. In the case of a double deck, the bending rigidities of
the shell are equal in both the directions, however, the mem-
brane stiffness is affected by the shear rigidity factor a. If
the equivalent thickness in the formula 6-14 is replaced by
at, the estimated buckling load will be very conservative be-
cause of the fact that it will also underestimate the in-plane
axial stiffness. In order to eliminate this underestimation
of the axial stiffness the shear rigidity factor a is arbi-
trarily multiplied by 2(l+v) (the ratio of E/G for the parent
material).
2 6-15
qcr Double Deck • 4 Ch) ) 2 Dy E(l+v)at
This formula is a conservative approximation of the buckling
load of hypars with double decks.
V \ . , • DECK BUCKLING BY THE ENERGY METHOD
The buckling of isotropic hyperbolic parabmloid shells

was inves•igated in o. classical paper by Reissner 5 . He
developed the ~eneral shallow shell theory
and reduced it to two differential equations in the disaplacement
w and a stress function F. From these Pquations he studied
i~ot("O(>\'.. '-
the bucklin~ of simply-supported
. "-
hypars 11nder
•miform loadin~. The assnmed displacement was a double
sine Sl=!ries and the form for the stress function F was assnmed
to be a donble sine series with an additional term containing
xy. The substit11tion of these series into the two e4•1tions
result in homogenesous algebraic equations beca11se the sinP
terms drop out. The condition of non-trivial sol11tion yields
the buck Uing load:
2E
each of
This solution was poslible becausel\..the t'fo differential
one .:."' Cl4.t. I,... e,. ~~c--\' 1 ~ '.
eq11ations contained only elastic constant:.. • F
in one equation and D in the other. Thus the sine terms could
drop out,
In the case of orhbotropic shells these eq11ations
contain several elastic constants and therefore
the sine terms would not drop out. This means that
~Wvl-t
one cannot use title shallow shell equations to obtain reasonable
buckling load. It may be possible to find • very simple
-\ff-
assumed functions for w and F but the accuracy of such a solution
would probably be very poor.
Another alternateve approach uses energy
principles. The potential en~gv of an orthotoopic hypar

19
shell is :
"' "'"
\j =- l ff [t>)( J.S:
0 0
T 2 D , ,L,)>'l< J.S;:
~1)
+ I)'?! ~ ~ -t- 4 D'l-'Q J.f ~ ,_,1
-t- 4 r,; e.(f "C7Pl~ ,.w-1. + 2 i.i ~a
l. )..S)<. µr-i ~ d.,,. ~
The selection of an assumed buckling shape
is ~ A'4'Uilt"
requires very careful attention, A double sine series ·
h>
use because of the complexity' of the resulting arithmetic.
Furthermore, the direction of the buckles is at an angle
with the coordinate axes in the case of orthotoopic shells
and the deflection function must contain a factor which accounts
for this fact, Several types of deflected shapes were tried,
By far the best results were obtained usingi the folb1ing
shape:
where the factor s represenas the tangent of the angle of

~ "'- ~s. \\..a ~~\.Le c~ ~~~ .AJeW-es .
the buckles measured from the 't
axi~ This function
36
was used by Timoshen~o and Gere
48
and by Easley and McFarland for the buckling of
shear diaphragms.
This deflection assumption corresponds to a simply
supported shell (or diaphragm) where ~(and they axis) is

~easured along'the deformations (corrugations). Actually
this expression does not satisfy w ~ 0 along the y axis, but
comparlions weth more exact analyses for the buckling of diaphragms
showed~B that this diecrepancy is not serioas.
Substitution
of the assu•ne4i w into the potential ener~y expression and
integration yields: ~
~ab ~
v s ;01 + Q2 + Q3)
where ~ \-:. ~tt+2 D, a'\..~T 2.D l a<\-.s.1.+ D'i \tt' D'"a&.4 ~
+ 2~'-a a,,. 8,_s-i.. -t 4- o ,,.'-a a<L s i. + c'
Qt. -= 4 D "a Cl'"\.. la"\. s.' ;- 4- D~'o Q,..a'"
Q l =-2 \j Y-<o ~:i.. s a = "l"/f'i la" 'ifa d =+G~~ t
Since V represents the change of potintial enefgy under B
a deflection w, the condition V =0 corresponds to the
buckling load. Thus
Q \ +- Q'- '\-- Q J ~0
or
From the membrane theory N

xy
= {>4.8/( 26), thus
\J'1;-=- ~~YT~L\i. -t 2 t>t(H-Ll...s'-)+ °'a (Y4"tll's~~, ..~)

+4 D~'a ( l+ ~1..) + CVi..t,,.~
where
This expression has to be minimized with respect to

the direction of the buckles (s) and the number of bukkles (n).
This can ea~ilt done by trial and error, using a computer.
- \&.~ -
2..4
The analysis of the single-layer Sage structure (~o.t'l)
gave 61 psf for bee buckling load as compared with the
experimental value of about 75 psf and the finite element
analysis of about 60 psf
--
using curved elements. Minimum
s
occurlul for n~ 6 and the buckles formml at an angle of 11.3
(,::..·r.: ~.lZ)
dggrees with the y axi~ The buckling load was not very
sensitive -\.a n anJ. therefore the prediction
I
of the numbe~of buckles may not be accurate.
For structure No.~ (double-layers, 28 Gage) the above

analysis resulted in a buckling load of 176 psf at n ~ 2
and at an angle of 36·.s degrees. The finite etement analysis
gave about 200 psf and the test structure did not buckle
up to a load of its psf when loading was discontinued.
The main advantage of this eifrgy analysis of the
buckling load is that it is very fast as compared with the
finite element analysis. Simple-supports are assumed and
therefore the actual buckling load may be somewhat higher
if some fisity is present. The deflection of the edge
members is not considered, but the f:tmi te J.ement analysis
showed \ec..~""' "'·" that edge member deflections do not
affect the deck buckling load appreciably. The deck
buckling load is influenced by the area of the edge members.
If the edge members are very slender then deflections will
controls on the other hand, the above analysis shou4d give
conservative relults for structures with heavy edge members.
V\•1 STABILITY OF ISOLATED EDGE MEMBERS
The edge members of a hypar structure

axial
receive unifoJ:tin loads along their lengt~from the
deck. Some or all the edge members are in compression.
The loading remains axial during
the deformation of the edge members and therefore
it constitutes a non-conservative force field.
To obtain an approximate value
of the buckling of the compression edge members,
the membrane shear f or11ce was applied to isolated
edge members with val!"ious idealized boundary
conditions. The buckling loads determ:lned in
this manner are conservative since the stiffening
effect of the deck is neglected.
- \S.,. -
The marginal member was analyzed both as a fixed free

and a fixed pinned member. The equation used to solve the
non-conservative force field problem was:
EI y IV + S(L-x)y" + My • 0
where S • the shear force, L • the column length, M • mass
per unit length, y •deflection , and ( 1 ) means differentia-
tion with respect to time.
The solution or the differential equation was obtained
by using Galerkin's Method for solving differential equations
with non-constant coefficients. The numerical solution tor
the fixed free case was found in an article by v. H. Leipholz;
"Die Knicklast des Einseitig Eingespannten Stabes mit
Oleichmassig Vertielter, Tangentialer Llngsbelastung", pub-
lished in 2AMP, 13, 6, 1962.
The solution for the fixed pinned case was determined
using the same method as described above, but a computer
program was written to facilitate the trigonometric integra-
tions and the mathematical solution.
The numerical solutions are:

2 for the fixed free case
(ql)cr • 40.7 EI/1
2 for the fixed pinned case
(ql)cr • 122.6 EI/1
where q • shear per unit length transferred from shell to
marginal member. These values and some comparison values
.r;::; I 9', <a. I~•
are given in
The results show that the possibility or buckling of
members
~<~
loaded by tangential shear forces the buckling is very
much smaller than in members under gravity load or constant
direction parallel to the original, undeflected axis or the
member.
Thus, if case IV is compared with V, with loading and
end conditions the same except tor the direction or the load,
it is seen that in the elastic range the edge member or a
hypar will buckle at a load (1.12/0.49) 2 • 5.2 times that
which the same member would carry under uniform gravity load.
Similarly, comparing cases VI and VII, it is seen that the
corresponding ratio is (0.436/0.284} 2 • 2.4.
If the total load which will make the edge member or a
hypar buckle, is compared with that which the same member,
loaded as a hinged-hinged Euler column (basic case) would
carry, it is seen that the edge member in the fixed-free
condition (case V) will carry (l/0.49) 2 • 4.1 times the
buckling load of the basic Euler column, or, for the f1xed-
h1nged condition (case VII) (l/0.284) 2 • 12.4 times the
basic Euler load.
The above results do not represent a complete analysis
or the problem, which is really one or buckling interaction
between the edge member and the shell. However, if the
shell action is close to that of a membrane, as in the case
ot light-gage steel hypara, the approximation should be
- \1;7 -
reasonably good. Thie approximation shows that, while a

buckling possibility exists, these edge members are very much
more stable than they would be if they were part of ordinary
1
framing subject to gravity loads.

EXPERIMENTAL INVESTIGATION
VH.1
INTRODUCTION
An extensive experimental investigation was carried out. The
purpose of the testing program was twofolds (1) to study the
behavior of light gage steel hypar roofs subjected to vertical
loads and (2) to provide a comparison with analytical approaches.
The following tests were carried outs
a) Sixteen flat shear tests - Properties and dimensions of all

VII-I
the specimens are presented in Table and a picture of one is
"'1. I
shown in Fig ~ It was necessary to carry out these tests to
determine the shear rigidity G' of the decking used for the hypar
models. Twelve specimens were 6' x 6' in plan, three were 5 1 x 5'
in plan (all referred to as ''medium scale tests"), and one was
l' x 1 • in plan ("small scale test").
b) Twelve saddle shaped hypar specimens - All of them were

,,...._ 'c>\« V\\ • t. -'11\-:f\ ·
5' x 5 1 in plan with various ri~es"-. A picture of one ti shown in
1.t.
FigUffe • The specimens are designated by three numbers: the
f lrst one indicating the rise/span ratio, the second the number of
layers of decking, and the third whether it Is an original or
duplicate. For example, for test no. 512, "5" indicates a 1/5
rise/span ratio, "l" Indicates one layer of decking, and "2"
Indicates that It is a duplicate test (each specimen was duplicated).
c) Three Inverted umbrellas with very stiff edge beams~ll were
12 • x 12 • In plan wt th a 14.4" rise. A picture of one ls shown in
36
\l~~s u, It 115 ~t\ Tc:..W,.
\V- \ . .
1:~
Figul:-8 • They are referred to as "medium scale inverted umbrellas
with very stiff edge beams".
d) 8ne small-scale inverted umbrella with very flexible edge meems
The model was 2' by 2' in plan with a 3" rise. The edge
members were maGe of 3/16 11 o.o. and 0.014 " thick brass tubes,
two ~ere used for the interior compression beams. Two la~ers
of corrugated decking of 2 mil thickaess formed the shell.

The deck was soldered at every valley to the Elege members.
~, .... od-..(
ft picture of ls shown
,.\.
in Figure 3t4.
e) One medium scale Inverted umbrella with very flexible edge
beams - The model was 12' x 12' In plan wt th a 14.4" rise) ~+rv<....\o~
V\\Jv..~~r ct , T0\.9\c. \\1-1. ._
A picture
of it ls shown In Figure -r.s.
The tests are described in the follo,wing sections.
'Ill.
JilR .2 FLAT SHEJ\R TESTS
1) INTRODUCTION
A series of flat shear tests were conducted in order to de-
4S
termine the shear rigidity G' of corrugated steel decking. Luttre14
Investigated the shear behavior of light gage steel diaphragms.
Based on the results of numerous tests he concluded that the
primary variables influencing the shear rigidity are length of
diaphrasm parallel to the corrugations and spacing of connectors
to the edge members in the direction perpendicular to the corru-

- \1"0 -
gatlons. It was also determined by Luttrell that the only reliable
means of determining the shear rigidity of a given diaphragm with

frame was by experiment.
The formula for G' developed by Luttrell accounts for the de-
f lectlon due to shear alone by subtracting from the measured de-
flectlon the bending deflection of the cantilevered structure due
to axial deformation of the edge members. The value of G' obtained
from a given test can be applied to the following case: Any set-up
with exactly the same spacing of diaphragm to edge member connections
perpendicular to the corrugations. and the same diaphragm. The size
of the edge members should have little or no effect.
It may be applied with simple modifications to the following
cases a (1) everything the same exce.pt diaphragm has different
thickness - G' ts (approximately) directly proportional to the
thickness; (2) everything the same except length of frame parallel
to the corrugations Is different - new G' may be obtained from

4S-
Figures 4-23 and 4-24 In Reference for box-rib and standard
corrugated dlaphrasms. respectively.
,2 EDGE !-EMBER FRAME
For all tests except two the edge members were light gage
channels. Two different stze channels were used. One, designated
as "heavy frame"• consisted of 6 11 x 1%" x .1046" channels and the
other, designated "light frame", consisted of 6" x 3/4" x .1046"
channels. For the other two tests. the frame was made up of tubing.
- \ ~' -
\
3: DEClSlNG
All the medium scale tests employed standard corrugated
decking; including 24, 26, and 28 gage. Either one layer or two
layers, with the second layer placed with the corrugations perpen-
dlcular to those of the f lrst layer, were used.
The small scale test had 2 mil corrugated steel foll. This
material was produced by United States Steel Corp. as a reinforcing
center for cardboard cartons. Two layers, running transverse to
each other, were used.
4 CONNECTIONS
For the medium scale tests, ii8 x %" self-tapping screws were
used to connect adjacent sheets to each other along the seams.
#14 self-tapping screws with aluminum-backed neoprene washers were
used to connect the decking to the edge members. For double layered
decking, the second layer was attached to the f lrst layer around
the perimeter only with iiB x %'; self-tapping screws. The spacing
of connections for each test ls given in Table "''·'

•
In test no. 14, the effect of the relative positions of the
screws connecting the top layer to the bottom layer and the screws
connecting the bottom layer to the edge members was noted. Re-
ferrlng to Figure
1.' one possibility ls shown in (a) where the
1
shear flow causes a region of compression between.screws 1 and 2
and the bottom layer distorts considerably between screws 2 and 3.
The other possibility ls shown in (b) where the shear flow now
causes a region of tension between screws 1 and 2 and very little
distortion of the bottom layer occurs. However, in this case,

-\~Z-
because of the tendency of the tensile region between screws 1 and
2 to flatten out, eventually screw #1 tears out of the bottom
layer.
For the small scale test, the top layer was soldered to the
bottom layer at every point of contact around the perimeter and
the bottom layer was soldered to the edge members at every point
of contact. One full panel was used for each layer.
Si LOADING APPARATUS
The medium scale frames were placed in a horizontal plane
and load was applied by means of a 50 ton hydraulic jack. Vertical
support along the loaded edge was provided by rollers on beams.
Reactions were provided by a fixed wall beam to which the frame
was pinned. Steel bricks were placed on the loaded edge to prevent
out-of-plane warping.
The small scale traDMt was attached to a wooden frame with
metal guides to prevent the specimen from warping out-of-plane.
The entire set-up was placed in the vertical position In a Tinius-
Olsen hydraulic testing machine and load was applied by the machine.
6; DEFLECTI<l'i 1'£ASURE?-£NT
For each test, deflections were measured by .001 11 dial gages.
At first, .0001n gages were tried but it was discovered that they
are too sensitive because the readings obtained from them were
erratic. Sufficient accuracy was obtained with the .001" gages.
If possible the stem of the gage was placed In contact with the
specimen, otherwise a thread was used to attach the stem of the

gage to the specimen.
,7 DETERHINATION OF SHEAR RIGIDilY G'
G' was determined by the use of the expression given in
Appendix c. The initial linear portion of the load-deflection
curve was used to obtain the slope, The customary testing procedure
was to initially load the specimen to a predetermined value so as
to seat all connections and then to conduct as many loading cycles
as necessary to get good agreement between the load-deflection
curves of successive cycles. Usually, only two or three cycles
were required. The dimensions, properties, and the values of G'
and of a a ""·•
G'/Gnt for each test are presented in Table - ·• The
-y.1(
load-deflection curve for test #14 is shown in Figure •
SADDLE SHAPED HYPAR TESTS
,1 INTRODUCTION
A series of light gage steel hypar models, 5' x 5' in plan,
with various curvatures were tested in an effort to determine if
the shear rigidity as determined by a flat shear test is valid for
the plate on elastic foundation approach. If an exact analysis of
light gage steel hypars were possible, then it would appear that
the experimental spear rigidity could be used with validity in
the analysis. However, the assumptions which were introduced into
the approximate plate on elastic foundation approach may lead to
errors in the results if the shear rigidity as determined by a flat
shear test is used in the analysis. Better results may be obtained
if a "fictitious" shear rigidity were calculated backwards from

- \ "!.4
experimental deflections. Thus, it might be possible to determine
the fictitious shear rigidity as a function of the curvature.
For the above reasons, the only variable In this series of
tests was the rise, i.e. curvature. Three different rise/span ratios
were employed, 1/8, 1/5, and 1/3# one set for single layer decking
and the other for double layer. Two tests, one a duplicate of the
other, were carried out for each rise/span ratio and number of
layers of decking. The tests were designated according to the rise/
span ratio, number of layers, and first test or duplicate as ex-
plained in Section
\Jn.'·•
2. DETAILS OF TEST SPECIMENS

/
The edge beam frames with supports were the same for all the
tests. They consisted of 3" o.D. x ~" wall thickness cold rolled
steel tubular members welded together to give the correct rise/span
ratio for each specimen. The diagonally opposite corners, two low
and two high, along with the midpoints of each side were rigidly
"1-2
supported In the vertical direction by steel bricks (Figure ).
This support system together with the fact that the tubes have a
large bending rigidity was considered to offer continuous rigid
support In the vertical direction.
A tie bar, 2%" x -k" In cross-section, was used to llml t the
spreading of the two low corners due to the vertical load.
The decking was made up of 28 gage standard corrugated steel
sheets with 2• cover. Three sheets were used for each layer, with
the middle sheet cut to flt the 5' width. Single layered decking
was connected to the edge beams by #14 self-tapping screws with

- l ~5" -
aluminum-backed neoprene washers @ 8" spacing while the two seams
were fastened together with trn x lz" self-tapping screws @ 2-2/3"
spacing. For the double layered decks, the top layer was fastened
to the bottom layer, around the perimeter only, with 118 x lz" self-
tapping screws @ 8" spacing. The two seams of the top layer were
fastened together exactly the same as those of the bottom layer.
Uniform vertical loading was achieved with sand held in place
by wooden sides and screeded to a uniform depth for each 400 lb.
(about 15 psf) load Increment.
For each one of the models, subsequent to the uniform load
test, a partial load covering an 8" x 12" area was applied at the
center of the same decking. These tests were designated with a "C".
In addition, for each one of the two duplicate models with double
layered decking and subsequent to the tests with unconnected decks,
the two layers were fastened together with #8 x ~" self-tapping
screws on an 8" square grid and subjected to uniform loading f lrst
and then the partial loading on an 8 11 x 12" area. These tests were
designated with an "I".
Deflections were measured with .001" dial gages. Strains were
measured at the center of the decking with SR-4 strain gages with
l" gage length. The strain gages were placed in a valley on the
top side and on the adjacent crest on the bottom side, and in the
direction parallel to the corrugations. They were also located at
mid-length of the tie bar, top and bottom, on some of the models.
The purpose of the gages on the decking was to determine the axial
and bending stresses at the center of the deck and on the tie bar
was to determine the axial force in it.
The experimental deflections and stresses at 40 psf for the

\I \\12. Vll -3
saddle shaped hypar tests are presented in Tables and • The
method of determining the shear rigidity versus curvature (or rise/
span ratio) by using the experimental deflections in the plate on
elastic foundation approximation is presented in

S•c. ~'o""
NEDIUN SC.ALE INVERTED UNBRELLAS WITH VERY STIFF EDGE BEA?-5
1 INTRODUCTION
( ~<. ~t \\ J \ t J l ~ / T~~, l V ~I)
Three medium scale modelsA...were designed and tested with the
main purpose being to check the theory for the buckling ot ortho-
tropic hypar shells. Therefore, edge beams with a large bending
rigidity were chosen so as to remain stable when the decking
buckled. Vertical deflections as well as strains were measured
for the decking and the edge beams.
The same edge beam frame was used tor all three tests. It
consisted of 3 11 O,D, x ~" wall thickness cold rolled steel tubular
members welded together. The dimensions of the full inverted um-
brella were 12 1 x 12 1 in plan, centerline to centerline of the edge
beams, with a 14.4" rise (1/5 rise/span ratio). Each interior edge
beam consisted of two tubular members side by side (spot welded to-
gether at l~' intervals) so that decking could be fastened to one
of the nembers in a given quadrant and decking in an adjacent
quadrant could be fastened to the adjacent member. Therefore, the
plan dimensions of each quadrant were 70.5" x 70.5", centerline to
centerline of the edge beams.

- \ ~., -
2) DETAILS OF TEST SPECINENS AND PROCEDURE
Two of the models had one layer of standard corrugated steel

~oeA.J
decking, 28 gage for one and 24 gage for the other.
The third model had two layers of 28 gage standard corrugated steel
decking.
For both of the models with one layer of decking, #14 self-
tapping screws @ 8 11 spacing fastened the decking to the edge beams
and li8 x 5/8" self-tapping screws @ 2-2/3" spacing fastened the
seams together. Three panels, each with 2' cover, were used per
quadrant. For the model with two layers of decking, the bottom
layer was attached to the edge beams exactly as described above.
The top layer was fastened to the bottom layer, around the perimeter
only, with /i8 x ~" self-tapping screws @ 8" spacing. The seams for
the top layer were fastened together exactly the same as those in
the bottom layer as described above.
For all three models, uniform normal air pressure was applied
to the inverted umbrella shape by four rubberized canvas bags
placed between the floor and the hypar in the inverted position.
A water manometer was used to measure the pressure.
Vertical deflections were measured by level sightings on
meter sticks held at each location. Strains in the decking were
measured with SR-4 gages with 1 11 gage length and those on the edge
beams with SR-4 gages with ls" gage length. The vertical deflections,
- \ ~Q-
experimental axial stresses, and experimental bending stresses at

"7.g -"7.fO
40 psf normal pressure are presented in FigureE
vu.s
INVERTED mmRELLAS WITH VERY FlEXIBlE EDGE BEAMS
1 ·, INTRODUCTION
J
Two models were tested with extremely flexible edge beams.
The purpose of these tests was to determine the mode of failure as
well as to check how closely the failure could be predicted by
theory.
2 SHALL SCAlE HODEL
One of the inverted umbrellas with very flexible edge beams

,,.4-
(Figure ) had overMall plan dimensions of 2' x 2•, centerline
to centerline of the perimeter edge beams, with a 3" rise (1/4
rise/span ratio).
The edge beam frame was made up of 3/16" O.D. x .014" wall
thickness brass tubes brazed together, The decking consisted of
two layers of 2 mil corrugated steel sheets soldered together at
each point of contact around the perimeter. The decking was soldered
to the brass tubes at each point of contact.

Uniform vertical loading was simulated by 64 discrete weights
(16 per quadrant) hung from strings passing through very small
holes in the decking and attached to 1" square pads.
Vertical deflections were measured with .001" dial gages.
The dlal gages were positioned such that the stems did not touch
the shell. To take a deflection reading, the ste~ of the gage was
pressed until contact with the deck was made; avoiding the
application of any force by the stem as much as possible.
Experimental deflections at 40 psf load are presented in
Figure
Collapse of this model occurred at 73 psf. It was initiated
by splitting of the decking along a row of holes through which the
weights were suspended. This splitting caused the brass tubes to
fail in bending very close to the central supporting column.
3) MEDIUM SCAlE MODEL
The test set-up and procedure for this model were almost the
same as for the 12 1 x 12 1 inverted umbrella with two layers of 28
gage decking described in Section

- vu.•
except for the size of the edge beams.
The edge beams were l" nominal diameter standard weight black
steel pipe with a SO ksi yield point.
The vertical deflections were measured with .001" dial gages.
The stems of the gages were attached to the shell by means of thread
which was glued to the structure.
Experimental deflections and stresses at 40 psf load are pre-
sented in F6a -, . 'a___

I 40
\)\l. '-
DISCUSSION OF EXPERIMENTAL RESULTS
G'
.1) SHEAR RIGIDITY G' OF STANDARD CORRUGATED DECKS
Comparison of the results for the flat shear tests, which are
\J\\-t
presented in Table , reveals that the variable having the
largest effect on~ls the spacing of the connections to the edge
members. In particular, the spacing transverse to the corrugations
has the predominant effect whereas the spacing along the corru-
gatlons has little or no effect on G'. For example, screws at
every third valley as in test no. 3 produce a G' approximately
one-third that for screws at every valley (one-third the spacing)
as in tests no. 1 and 2.
According to Luttrell• another variable which affects the
shear rlgldlty ls the dimension along the corrugations; the larger
this dimension the larger the shear rigidity. From the results
presented ln Table 3.1, it ls not possible to verify Luttrell's
finding conclusively. For test no. 8, with plan dimensions of
5' x 5', G' ts less than that for 6' x 6' test no. 5 but slightly
more than that for 6' x 6' test no, 6. Thus, lt appears that
scatter in the test results masks any effect of the small
variation in stze,
The spacing of seam connections and number of seams have no
dlscernable effect on G' based on the results presented ln Table

'6\\-1
, Also, it appears, at least for a limited range of thicknesses
of panels, that the shear rigidity increases linearly with the

l 4-1
thickness. For example, compare the results for tests no. 5 and 7,
Tests numbered 10 - 16 had two layers of decking r~ning
transverse to each other. In each case the top layer was connected
to the bottom layer along the edges, rather than being connected
directly to the edge members. The effect of this arrangement of
connections ls to make the top layer less effective in shear than
the bottom layer, which ls attached directly to the edge members.
The reason for this ls that any shear deformation which occurs ln
the bottom layer along the edge members occurs also ln connections
for the top layer which ln turn reduces the effective shear stiff-
ness of the top layer below that obtained by connecting the top
Vil-I
layer directly to the edge beams. From Table ~ tests no. 10 and
11 versus 1 and 2; 12 and 13 versus 4, 5, and 6; and 15 versus 8
and 9 show that the shear rigidity for the double layer tests ts
less than twice that for the corresponding single layer tests. It
appears that connecting the second layer ln this manner results in
only a 33% increase in G' over that of the single layer~ although
as mentioned previously the relative positions of the screws appear
to affect the shear rigidity. If all the connections between the
,.
two layers were such as to create the condition shown in Figure
~(•)
) then the shear rigidity should be larger than for the case
, , , (q)
shown ln Figure •
'1:-J
Figure shows the load-deflection curve for test no, 14,
The linear portion extends to about 1800 lbs, load which corre-
sponds to a shear Nxy • 25 lbs. per inch. Beyond this point~ the
curve tndlcates that the shear rigidity decreases with increasing

load.
1.8 ,,,
Figures and indicate that the shear rigidity of
standard corrugated steel decked hypars with a rise/span ratio
~ 1/5 is somewhat (20 - 303) less than that of the same flat deck.
\Jl\. '2.
However, as mentioned briefly in Section. , this Information
may be misleading because of the approximate plate on elastic
foundation approach used in conjunction with the experimental
deflections to obtain these curves.'
t·B
Two factors were omitted in the theory used to obtain Figures
-r.~
and , the inclusion of which indicate that the shear rigidity
obtained from a flat shear test may be valid for the hypar shell.
One of the factors omitted was the middle-surface deformations
u and v. The other was the spreading of the low corners of the
saddle shaped models. Even though a tie bar connected the low
corners, tension in the tie barcausedby the shear forces resulted
in its elongation. Duplicate tests no. 521 and 522 were chosen to
illustrate the effect of these two factors. If the u, v displace-
as shown In Figure
,.,
ments are included then the resulting a ls .046 Instead of .042
for the 12" rise. If, In addition, the
spreading of the low corners ls included by introducing the stiff-
ness of the tie bar Into the analysts then the resulting a ls
approximately .o5 which ls the same as that obtained from the flat
shear test. These results for only one example are not meant to
be conclusive but the indication ls that the shear rlgldlty ln a

hypar ls substantially the same as that determined from a flat shear
test •
..
2 HYPAR DEFLECTIONS
\, .·SADDLE SHAPE SUPPORTED ALL AROUND
Table
v.- 42.presents the maximum deflections at 40 psf for the
models tested by the author. The results reveal the decrease in
maximum deflection with increasing rise as well as the fact that
the maximum deflection of a given model with two layers of decking
ls more than half that of the corresponding model with one layer
of decking; the reason being, as mentioned previously, that since
the top layer ts connected only to the bottom layer lt ls not as
effective in shear as the bottom layer and thus the rigidity of
the single layered deck is more than half that of the double
layered deck.
\J\\ - t.l'•)
Table shows the effect of interconnecting two layers
all over and not just around the perimeter. Test numbers with an
"I" indicate interconnected layers. The reduction in maximum de-
f lectlon due to interconnecting ts seen to be only 10 - 20%. How-
ever, lt appears that interconnecting would be necessary in
practice to prevent chatter.

vu- J
Table presents the maximum def lectlon due to a partial
load covering an 8" x 12" area in the center. A comparison of the
maximum def lectlon due to a 100 lb. load on a single layered deck
with that due to a 200 lb. load on a double layered deck reveals
that a double layered deck ts more than twice as rigid as a single
layered deck for carrying a concentrated load; the reason being

144
that a concentrated load on a single layer ls carried mainly by a
few beam strips whereas on the double layer it ts spread out and
thus carried by more beam strips in each layer. For all the models
tested, the maximum deflection due to a 100 lb. load on a double
layered deck ls approximately one-third that for a single layered

deck.
INVERTED mmRELLAS WITH VERY STIFF EDGE BEAN.~

., .•~".'J.1.•H!') I 1·l"~.(A)
Figures • and show the measured vertical
def lectlons at 40 psf normal pressure for three inverted umbrellas
tested at Cornell. It ls seen that difficulty ln obtaining symmetry
was experienced. The lack of symmetry was probably due to unequal
air pressures ln the rubberized canvas bags. The bags were not
entirely air tight so that wtequal rates of leakage from the bags
could have caused relatively large percentage variations in the
pressures.
6) Inverted Jmbrellas with Very Flexible t:dge Beams

figures 7.13 and 7.14(a) show the measured deflections
at 40 psf for the 2 ft by 2 ft small-scale test and test
No. 9, respectively. Figure

14S -
INVERTED mmRELLAS WITH VERY STIFF EDGE BEAMS

i_.g~~.1-l'~~'} , .• i(•)
Figures . and show the axial and bending
stresses obtained from strain measurements at 40 psf load tor the
three models tested at Cornell. The axial stresses In the strong
direction of the corrugated decking are seen to be quite erratic
and do not Indicate any consistent behavior. It appears that their
magnitudes are too small to give reliable results.
The bending stresses In the decking at the quadrant centers
vary from 6200 to 10200 psi for the 28 gage single layer, from
4800 to 5700 psi for the 24 gage single layer, and from 5100 to
10200 psi In the bottom layer of the 28 gage double layer decking.
The wide range of these values for each model indicates a con-
siderable departure from symmetry which was probably caused by
unequal pressures applied to each quadrant.

1.r~
reveals that rotation about one diagonal apparently occurred
; . '4(.Y
during testing of the small scale model. Figure shows that
a reasonably good condltlon of symmetry was obtained for the medium
scale model. For both models, lt ls apparent that the deflection at
the free corners is much larger than that at the center of each
quadrant.
"!, . HYPAR STRESSES
Q.)
SADDLE SHAPE SUPPORTED ALL AROUND
\J\\- 2.
The experimental stresses presented ln Table - reveal the
ditference tn structural action between a very flat hypar (rlse/span
• 1/8) and one with a much larger curvature (rlse/span. 1/3).
The bending stresses in the models with a 1/8 rise/span ratio are
much greater than those in the models with a 1/3 rise/span ratio.
The decrease ln bending with lncreaslng rise/span ratlo ts accom-
panted by an increase in membrane action as seen ln the last column

vn ...z
of Table The experimental force ln the tle bar for the models
with a 1/8 rise/span ratio is much less than that given by membrane
theory whereas the force in the tie bar for the models with a 1/3
rise/span ratio ts almost as large as that given by the membrane
theory. Thus, it is demonstrated experimentally that for very flat
hypers, shear stresses are much smaller than predicted by the mem-.
brane theory and that bending constitutes the major part of the
structural action whereas for deep hypars (rise/span ~ 1/3) mem-
bran• stresses predominate with values close to those given by
membrane theory and bending stresses are lnslgnlflcant. The axial
stresses in the deck, for all models, are seen to be insignificant.

\JV-3
Table · shows that a concentrated load produces &ignlf lcant
bending streaees, especially in the flat hypars, but very little
Mmbrane action.
Experimental edge beam axial and bending stresses as well as edge
beam axial stresses from the membrane theory are also shown In
, • 10(1') , . ll{lf)
1
,. rz fl,)
Figures , , and • It ls difficult to observe con-
ststent trends in the experimental axial stresses in the edge beams.
In some locations they are very close to those values given by the
membrane theory while at other locations the difference ts quite

'
large. It Is to be expected that the experimental axial stresses
would be less than those given by the membrane theory because of
the tact that part of the load Is carried by bending. However,
the results for the three models do not give any Indication as to
whether or not this Is the case.

-r. 1o(i,) 1. \\("I
1 l·\1(\i)
Again, referring to Figures , , and , the
bending stresses at the mid-lengths of the perimeter beams for the
two models with one layer of decking are larger In the beams para-
llel to the weak direction of the decking. This ts to be expected
because the bending which occurs in the decking transmits a vert-
teal shear loading to these edge beams. The bending stresses in the
interior edge beams for all three models are small although the
strains in these beams at points of maximum stress near the column
were not measured,

- 148 -
CiiAPT.~~ X
su.,1MA·'-Y, coNcLusroNs AND ~lSC(MlENDATioNs
The behavior of thin steel hypars was studied

experimentally and analytically to determine the stresses,
deflections, and buckling loads of such stt"uctures under
various conditions.
Th~ exp~rimental proAram consisted of tests on1 a) Five

inverted umh'!'."ella typP. hypat" structures, b) Sixteen flat
shear tests which were made to determine the shear rigidity
of corrugated steel decks used on the hypar models, c) Twelve
saddle-shaped hypat"s with different rises and with rigid edge
supports which were tested in order to evaluate the effect
of rise or warping on the shear rigidity of decks.
Several snall-scale and single-quadrant exploratory tests

preceded the above tests, the results are not included in
this report because of experimental difficulties or
uncertainties experienced with those models.
- \~~-
Two different approaches

based on the finite clenent method ,·:ere used in the analysis
of hypars. These approaches basically differ in the stiff-
ness forJT1ulation for individual elements. In the first, a
curved element rectangular in plan was developed on the basis
of shallo~ shell theory. The displacements u, v, and w used
in the forrnulation of t"ds ele1"ent are measured along the
tangerts and normal to the surface. In the second approach,
the actual shell surface was approximated by a series of flat
plates Rsscnbled in the Rlobal co~ordinate systeJI". The stiff-
ness matrices for the eccentrically connected edrc rne~bArs
were developed. The effect of a tension tie rod and a colu~n

wC..l!.
support incorporated in the analysis by replacing the phys-
ical structural members by idealized equivalent springs (see
Section III-2F). ( o~\~.< \'V?c~a~ ~ ci..J_~r ~
\\. c;x.__,,~'\~'s c~ ~..--- s~~ ~ ~~"""'~~.
' ~ The· linear elastic analysis is ap!llicable for various
types of hypar structures (e.g. umbrella shell, saddle-shape,

etc.) and also for different loading conditions such as uni-
formly distributed load, unsy~metrical load and the load due
to edge member weight, Veflections and stresses for both deck
and edge members '"-.l~calculatcd.
-120-
-\So-
The linear elastic analysis was further extended to

include the instability effect introduce<l due to the in-plane
forces Nx , iJ y and Nxy The load incrementation method was
used to predict the deck buckling and the overall (edge member)
buckling of the structure.
The solutions available in the literature for both -+~
linear elastic analysis (Refs. 17,19,20,21) and for the in-

stability analysis 5 were compared with the
\~ ~' $~
solutions obtained by (Pigs. 4.6, 4.R and 6.2).
The analytical results were further compared with the experi-
..,.-~
mental work conducted at Cornell and also wi th results avail-
11
able in the literature elsewhcre 33 , 65 •
Based on the finite element analysis of several
structures and the comparison of analyses with available ex·
perimental and analytical results, the following ohservations
evolved:
In the case of hypars with all edges fully supported,
and for hypars with very stiff edge members (Figs. 4.6,4.8 and
4.33) the results of curved and flat elements show excellent
correlation. However, in the case of hypars with flexible
edges (Fir. 4.23), the curved element formulation tends to
underestimate the bending in the flat portion of the shell
(e.g. free corner of an umbrella shell). In the case of a
uniformly loaded hypar. the fully compatible 16-tern Hermi-
tian polynoriial used for t! e normal displacement w, does not
1
show any improveJT'ent over a 12-term non-compatible polynoriial

displacement field.
_, $ \ -
The relative stiffness of the edge members to the

decks is quite important from the point of view of behavior of
shells. In the case of moment-free connections between the
deck and the edge members, the type primarily investigated in
this work, for very stiff edge members the deck tends to bend
freely between the opposite sides. In the case of umbrella
shells with flexible edge members, the deck supports the edge
Me~ber at the free corner. The change in behavior of shells
with stiff and flexible edge members is quite noticable from
the deflection profiles for Strucs. '9' and '13' (Table IV-1).
,..,:,,.; .<..'.,.. . . . .
The in-plane shearing of a hypar shell is
of utmost importance in its behavior. The effective value of
the shear modulus of the corrugated decks is given by Geff •
aG. Depending upon the rise to span ratio of the hypar sur-
face, the value of a, deter~inedby the flat shear test needs
-.o f......,- -.. 1c..r· rc..,r·
..u-•..:--.1
modification. It is recommended that a be reduced by 25% for
single deck structures whereas it should be reduced by 20%
. (ft:Oi'·"·'-2.g)
"
for a double deck. Since the deflections and the stresses of
I\.
the hypar shell primarily depend upon the value of a, the
conservative estimation of its value will provide a significant
factor of safety for the structure.
Since a part of the vertical load is carried by
bending action, tt~ value of the meFbrane shearing force given
by the Jl'embrane theory is on the conservative side. However,
the analysis of several structures (Chapter IV) tends to in-
dicate that over a major portion of the s,!:~1 value of the
in-plane shear is very close to and at places1 hirher than the
-\St.-
Membrane shear (Figs. 4.19, 4.28). (onnections between

the adjoining panels designed on the basis of the membrane
shear, are considered adequate.
The axial forces calculated in the edge member on
the basis of the membrane theory are on the conservative side
(see Table V-2). The axial stresses form a relatively small
portion of the t0tal stresses (axial + bending) in the edge
members. Therefore the computation of the axial stresses in
~i~ fl..._ '4--L.-;1 /~,
the edge member on the tiasis of the ~ernbrane theoryA cannot be
considered as conservative.
The eccentric connections of the edge ~embers to
the deck have a pronounced effect on the deflections of the
shell (Firs. 4.13, 4.17). With a proper choice of the
eccentricity, the influence of the eccentric connection can be
used to advantage (see Chapter V)~ f. (~ '1.lc.J'. .,,._,.
As indicated in the analysis of Struc. '8' (Table
IV-1) produces considerably larger de-
uniform loading. The unsymmetrical load due to wind or

drifting of snow should be given due consideration in selec-
tion of decks and edge member sizes.
The weight of the edge members is partly carried by
the shell nction. To average it over the whole surface of the
shell will underestimate the bending stresses in the edge mem-
bers (Firs. 5.3, 5.4). The effect of the edge meinber weight
on the behavior of the shell will depend upon the construction
procedure.
- lS'Z. °'--
INSr=RT
The effect of the amount of restraint against

spreading of the low cornwrs of saddle-shaped hypars
on the deflection of the center of the shell was studied
by means of num•rieel examales. It was found that ...
tie bar connecting the points of support is very
effective in restraining the outward movement of the
supports and thus in reducing the center deflections.
Edge members with large bedding rigidity in • horizontal
plane alee restrain the spreading of the supports
effectively.
Partial loading on hypars was studied to a very

limited extent experimentally. Loads were applied on a""-
8 in. by 12 in. area in the center of the saddle-shaped
models w4lth rigid supports around the perimeter. It was
found that such a loading Qn a single layter of corrugated
decking produces a maximum deflection three times that for
two layers of corrugated decking. Therefore, two layers
~'-V
of decking recommeded for hypars whihc will be subjected
to significant loads on small areas.
-\S l -
The linear elastic analysis adequately represents

the behavior of the shell for low levels of loads. However, as
the load level increases it is necessary to incorporate the
effects of instability in the analysis.
The buckling in hypar shells using a light gage
corrugated open deck is manifested in three different forms.
The individual plate elements composing the deck may show
local buckling when subjected to in-plane shearing or axial
forces. This hm·1ever, can be prevented by a proper choice of
thickness to width ratio for each.plate element. Jteck
buckling takes place w'1en the deck actinp as a unit buckles
along the compression arch (Chapter VI) ~--- lt.; the edge mem-
bers are still stable. ~erall bucklinv of the shell is
defined as the one when the edge members along 1·d th the deck
buckle simultaneously. For practical sizes of edge reef!'lbers
there is only a very re~ote possibility of overall buckling 19 .
Thou£h the ~agnitude of the deck stresses is quite
low, the possibility of the deck buckling must not be over-
looked. Two decks placed perpendicular to each other for the
same geometric configuration of the shell ~ill increase the
deck bucklinr load roughly three to four times compared with
that for a single deck. From the analysis it appears that the
axial stiffness of the edge members is more important than the
o..s ~(J....('" ~'> ~~ 'o\lc)il\.1.f'.'\ ~~ Q::) ....~
deflections of the edges. It also appears that the deck buck-

"
ling load is roughly proportional to the area of the edge
members.
The membrane action in the shell increases with in-
A
crease in the rise to span ratio, thickness t, and shear
rigidity factor a. A'.non ..-dir:1enslonalized parameter aA:ic can be
used as a good index for the behavior of the shell. A higher
value of this non·dimensionalized parameter indicates reduc-
tion in tl·c vertical deflection as well as in the bending
stresses for both the deck and the edge members. Furthermore,
it will increase the buckling load.
Tilough this investigation resulted in acceptable and
reliable methods of analysis, the following recommendations
~,·,.d..
can be made for futureAwork:
The computer program using flat elements for the stiff-
ness analysis gave consistent results. It is not clearly es-
tablished as to whether the deficiency of the curved elements
in predictinr deflections and bending stresses near the flat
free corner of an umbrella shell is due to ele~ent stiffness
formulation or whether its due to the use of stiffening
eccentric edge members.
As for the response of the structure, for the variations of
different parameters (rise to span ratio, a, etc.) both curved
and flat elements in general give the same pattern.
The assumption of st,allowness of the shell surface
is used in applyin£ boundary conditions. This limitation can
be eliminated by a suitable choice of local co-ordinate axes
along the boundaries. This will need additional computer

progra111111ing.
The computer program can in-
corporate the beams built along the lines of generators. How·
ever, their influence on improving the shear rigidity and re·
-1ss-
ducing the deflection of the shell is not studied. The local

release of forces such as hinge connection (moment-free) was
not incorporated in the solution.
BIBLIOGPAPHY
1. ''non., •:Elementary Analysis of Hyperbolic Paraboloid

Shells, 11 Portland Cement ftssociation, 1960.
2. Sondhi, J. JL, and Patel, i'~. N., ''Design for Hyperbolic
Paraboloid Shell Footings for a Building in Pomhasa,
Kenya," Indian Concrete Journal, June 1961.
3. f!arguerre, K. , "Zur Theorie der gekriimmtea Platte grosser
Formanderunr,," Proceedings, Fifth International Congress
(.n Applied Hechanics, 1938.
4. Vlassov, V. Z. , ' General Theory of Shells and its Appli-

1
cations in Enyineering, 11 tlASA Technical Translation F-99.

5. Feissner, E., "On Some Aspects of the Theory of Thin
Elastic Shells,,. Journal of the Boston Society of Civil
Engineers, Vol. 42, No. 2, 1955.
6. Ape land, K. , and Popov, E. , "Analysis of Bending Stresses
in Translational Shells," Procel'!dings of the Colloquium on
Simplified Calculation r~ethods, Brussels, September 4-6,
1961.
7. Bleich, IL H., and Salvadori, l. i. G., ''Bendinp. f!oments on
Sl1ell Boundaries," Journal of the Structural Di vision, ASCE,
0 ct ob er 19 59 •
8. Chetty, S.H.K., and Tottenham, 11., ''An Investigation into
the Bending Analysis of Hyperbolic Paraboloid Shells,''
Indian Concrete Journal, July 1964.
9. Das Gupta, N. C. , .. Edge !Jis turbances in a Hyperbolic
Paraboloid,n Civil Engineering and Public T"~orks r>eview,
February 1963.
10. Hirza, J. F., flStresses and Deformations in Umbrella
Shells," Journal of Structural Division, A~CE, April 1967.
11. n.ussel, rt. n., and Cerstle, K. H., "Bending of Hyperbolic

Paraboloid Structures," Journal of the Structural Divi·
s ion, /.SCE, June 196 7.
12. Russe 1, R. R.. , and Gerstle, IC. H. , "Hyperbolic Para·

boloid Structures on Pour Supports,'· Journal of the
Structural Division, ASCE, April 1968.
-\S1
13. Schnobrich, t1. C., and Gustafson, W. C. , ';Dis cuss ion on

Reference 10," Journal of Structural Division, ASCE,
January 1968.
14. Croll, James G. A., and Scrivener, J. C., "Fdpe Effects
in Hyperbolic Paraboloid Shells," Journal of the Struc-
tural Division, ASCE, March 1969.
15. Croll, James G. A., and ScriveneJ", J. C., ''Convergence of
Hypar Finite-Difference Solutions, 11 Journal of the Struc-
tural Division, ASCE, r•ay 1969.
16. Brebbi~1,Carlos A., "A Critical Review of Bending Solu-
tions in Hyperbolic Paraboloid Shells," University of
Southampton Report (En.gland).
17. Connor, J. , and Brebbia, Carlos A., 1 'Stiffness ~"a tr ix for
ShalloN 'lectangular Shell Element,'' Journal of Engineer-
inf! t'echanics Division, ASCE, October 1967.
18. Deak, Alexander L. , "Analysis of Shallot\' Shells by the
Displacement tfethod," Report 67 .. A, University of Washing·
ton, College of Engineering, Department of Aeronautics
and Astronautics.
19. Parker, John E., ''Behavior of Light Gage Steel Hyperbolic
Paraboloid Shells," Ph.D. Dissertation, Cornell Univer-
sity, 1969.
20. Pecknold, David A., and Schnobl'ich, W. C., "Finite·Ele·
ment Analysis of Skewed Shalloi-1 Shells,'· Journal of the
21. Pecknold, ti. A., and Schnobrich, W. C., "Finite Elell'ent
Analysis of Skewed Shallow Shells,'' A Technical R.eport
to the Office of Naval Researc}'l, Department of the Navy.
22. Ralston, A., ' 0n the Pro bl en 0£ Buckling of a Hyperbolic
1
Paraboloid Shell Loaded by Its O\illl l'teight, '' Journal of

~.:athernat ical Physics, 19 56.
23. Dayaratnam, F., and Gerstle, I<. 14., "Buckling of Hyper-

bolic Paraboloids," World Conference on Shell Structures,
0ctober 1-4, 1962.
24. !1uskat 11ichard, "Buckling of Light Gage Steel Hyperbolic
Paraboioid Poofs,'· Ph. fl. I'issertat ion, Corn el 1 l'ni versi tY,
1968.
25. Brebbia Carlos A., and Connot, J., r:r;eol'l'etrically Non-
linear Finite-Element Analysi~, '; Enr.ineering ~.lechanics
Oivision of ASCE, April 1969.
_,~.
26. Row~, R. E., and Pirza, J. F., nAn Exl)erimental Iii.vesti-

gat1on of Small Scale Hyperbolic Paraboloids,'' Proceedings
of '1'orld Conference on Shell Structures, San Francisco,
October 1962.
I
2 7. Rowe, R. E., "Tests on Four Types of Hyperbolic Shells,''
Proce~dings of the Symposium on Shell Research, Delft,
August 30-September 2, 1961.
28. Leet, Kenneth t·~., "Study of Stability in the Hyperbolic
P~raboloid, Journal of the Engineering ~~echanics Divi-
0
sion, ASCE, February 1966.

29. Bertero, V., and Choi, John, "Chel'lically Prestressed Con-
crete Hyperbolic Paraboloid Shell Model,'' Proceedings of
World Conference on Shell Structures, San Francisco,
October 1962.
30. r··cDermott 1 John F. , ''Single-Layer Corrugated-Steel-Sheet
Hypars," Journal of the Structural Division, ASCE, June
1968.
31. Yu, C. !'/., and Kriz, Ladislav B., tlTests of a Hyperbolic
Paraboloid Reinforced Concrete Shell, 11 Proceedings of
Octorer 1962.
32. Ni ls on, A. H. , '·Testing a Light Gage Steel Hyperbolic
Paraboloid Shell," Journal of the Structural Division,
ASCE, October 1962.
33. ''Reports of Shear Proprietary Tests of I!ypar Metal Deck-
ing," (Private Communication).
34. Huffington, N. J., Jr •• "Theoretical Determination of
nigidi ty Properties of Orthogonally St iffencd Plates,,,
Journal of Applied r1echanics, t'arch 1956.
35. Smith, Geoffrey E., 1 ;Elastic Buckling in Shear of In ..
finitely Long Corrugated Plates with Clamped Parallel
Edges," A Thesis presented to Cornell University, for a
?Jaster of A,)ronautical Engineerinr, 1957.
36. Timoshenko, S. P. , and Gere, J. r'. , 'Theory of' Elastic

Stability," McGraw Hill Second Edition 1961.
37. Bar.Ss Richard and l~assonnet, Charles, ''Analysis of Beam

Grids' and Orthotropic Plates," Crosby Lockt*lood and Sons,
Ltd., SNTL-Publishers of Technical Literature, Prague, 1968.
38. Anon., 1 Light Gage Cold·Formed Steel Design tTanual,"

Edition 1961, AISI.
- \'SC1·
39. Winter, G., "Commentary on 1961 Edition, Light Gage

Cold-Formed Steel Design Vanual, 11 AISI.
40. Kinloch, H., andIIarvey, J. n., "Analysis and nesign of
Anisotropic Plate Structures with Particular Pe£erence to
Decking Systems," An Article from "Thin ·''!alled Structures,"
Edited by Chilver, A. H., John ~Hley and Sons, Inc., N.Y.
41. Lekhni tskii, S. G. , ''Anisotropic Plates," Translated from
Second Puasian F.dition by S. w. Tsai and T. Cheron,
Gordon and Breach Science.
4Z. Zi cnkiewicz, O. C. , and Cheun e!~ Y. K. , "The Finite E le-
ment ttethod in Structural and Continuu"' r'echanics, '' ~·cGraw
Hil l Publication 1968.
43. Bryan, E. R., and El-Dakhakhni, ''!. rr., "Shear Flexibility
and Strength of Corrugated reeks," Journal of Structural
Division, ASCil, November 1968.
44. Falkenberg, J. C., "Discussion on ~eference 43, ,. , Journal
of the Structural Division, ASCE, June 1969.
45. Luttrell, L. D., 11 Structural Performance of Light Gage
Steel Diaphragms,'' Ph.D. Dissertation, Cornell Univer-
sity, 1965.
46. Troitsky, M. S., ''Orthotropic Bridges Theory and Design,"
The James F. Lincoln Arc Welding Foundation.
47. Anon. , "Design ifanual for Orthotropic Steel Pl ate Deck
Bridges,'' AISC Publication 1963.
48. f::~~~ J. T., and T·'cFarland, D. E., "Buckling of Lip-ht
Gage Corrugated !fetal Shear !liaphragms, ,. ASCE Journal of
the Structural Oivision, July 1969.
49. Wright, R. N., et. al., "BEF Analogy for Analysis of Box
Girders," Journal of StructuTal Division, ASCE, July 1968.
so. Gallagher, R. H., "The Development and Evaluation of

''a tr ix tTetl;ods for Thin Shell Structural Analysis," Ph.D.
Dissertation, State University of ~Jew York, Buffalo,
June 1966.
51. Yang Henry T. Y., ' J. Finite Element ForJnulation for
1
Stability Analysis of Doubly Curved Thin Shell Struc-

tures," Ph.D. Dissertation, Cornell University, 1968.
52. Cantin, P.• , and Clough, P., "A Curved, Cylindrical Shell
Finite .Element," AIAA Journal, Vol. 6, June 1968.
-\60-
53. Haisler, t·1., and Stricklin, J., "'Hgid.-Body l)isplacements

of Curved Elements in the Analysis of Shells by Patrix
Displacement f.'ethod," AIAA Journal, Vol. S, No. 8,
August 1967.
54. Timoshenko, s. P., and Woinowsky·Kricger, ,.Theory of Plates
and Shells,'' J<tcGraw·Hi 11 Book Company, Second Edition 19 59.
SS. Weaver, William, Jr., "Computer Programs for Structural
Analysis,'· D. Van Nostrand Company, Inc.
56. Bleich, Friedrich, "Buckling Strength of ~'etal Structures,"
r1cCra't1-Hi 11 Book Company, 1952.
57. Hoff, N. J., 11 The Analysis of Structures," John l''iley
and Sons, Inc., 1964.
58. Gallagher, R. and Barsoum, rt., "Finite f.lement Analysis of
Torsional and Lateral Stability Probelms,'' Int. Journal
for Num. I '.eth. in Engrg. , Vol. 2, No. 2, April ·June 1970.
59. Hall, A. s. and Woodhead,'°'· l''., "Frame Analysis," John
Wiley and Sonr., Inc., 1961.
60. Anon., "Torsion Analysis of Rolled Steel Section,u Steel
Design File Design Data from Bethlehem Steel, AIA, File
No. ·13A-l.
61. Jones, n. r. and Strome, P.R., t:A Survey of Analysis of

Shells by the Dis\)lacement Jtethod," iTatrix r!ethods in
Structural rechan1cs Procee<iings of the First ~lfright
Patterson Airfield Conference, October 1965.
62. Gere, J. H. and l'/eaver, 1'!., "Analysis of Framed Struc-
turt"s, '' D. Van Nostrand Company, Inc., 1966.
63. Anon., ''Stress, A User's ?!anual," MIT Press.
64. Brooks, D. F. and Brotton, D. ~·.,'"'Formulation of a
General Space Frame Program for a Large Computer," The
Structural Engineer, Vol. 44, No. 11, November 1966.
65. '·Behavior of Hypar Shells,'' National Science Foundation

Grant GK·l282, A Project under progress at Cornell Uni-
versity.
66. Gallagher, P. H., et. al., "Discrete Element Procedure
for Thin ~hell Instability Analysis, AIAA Journal,
January 1967.
67. '!allett, ~. H. and r•arcal, P. V., '·Finite Element Analysis
of l!on·Linear Structures,'' Journal of Structural Divi·
sion, ASCE, Septe~ber 1968.
- \&I-
68. Kapur, K. and Harz, B. J., "Stability of Plates using the

Finite tle!'!ent i(ethod," Journal of Engineering i·~echanics
Division, ASCE, Vol. 92, April 1966.
69. Anon., "Systcm/360 Scientific Subroutine Packape
(360A-Cf·I·03X} Version II, Programer' s r~anual IBJ~ Applica-
tions Program Hanual HZ0-0205-2, 1967.
70. Gustafson, W. C. and Wrieht, R. N., ".Analysis of Skewed
Composite Girder Bridges," Journal of Structural Division;.
ASCE, April 1968.
71. Argyris, J. H. and Scharpf, D. l•.r., "A Further Sequel to
Technical Note 14 Puga Pamily of Elements for Folded Plate
Structures," The Aeronautical Journal of the Royal Jt.ero-
nautical Society, Vol. 73, July 1969.
72. Ojalvo, I. v. and New.aan, H., ''Natural Vibrations of a
Stiffened Pressurized Cylinder with an Attached 1'ass, 11
AIAA Journal, June 1967.
73. Scniggs, !'. r•. , ''An Analytical and Experimental Study of
the Vibration of Orthogonally Stiffened Cylindrical
Shell, 11 AIAA/.f\SME 9th Structural Dynamics and Patcrials
Conference, rs.11!' Springs, California, April 1-3, 1968.
APPENDIX A
DECK PnOPERTIES
The elastic properties along principal directions of

orthotropy are given for both close and o~en decks. A detailed
computation of the elastic constants for an orthotropic deck

are reported in Refs. 34, 35, 36, 37, 40, ·41, 46, 47, 48.
A.- I fORl?UGATED Q!:1W. DECK

Fig. A 1 (a) shows an arbitrary cross-section of unit
length, in y·direction, for an open deck.
A- la t'eJ11.brane Constants
: : E • The extensional elastic constant for the equi-
1 x
valent orthotropic plate, is obtained hy equating the extension
flx bet\1een the physical and the idealized system.
F rorn F i g • A (a) ,
1
A-1
For cnnstant thickness t of the decl<.,
Ax• fit;:
Pt s z2t ds
-r A-2
s 2
0
J y ds • I '
y
A-3
where Iy' is the moment of inertia of the cross-section, per

unit length, about the line of action of load P, which coin-
cides with the x-axis.
For the equivalent flat plate, (Fie. A1 (b)),
~x • ~
.ext
A··4
Bquatin~ equations A-2, A-3 and ft·4, we get

Io
E • p F: A-5
x y
The other membrane constants are given in Table A-I.
A-lb Bending Rigidities
Bendinr rigidities for the x and y-directions are
given in Table A-I. The method of determining the constants is
given in ~ef. 35. For the orthotropic plate, the twistinf con-
stants Dxy and Dyx are not equal and hence the average values
of these constants are used in the analys:i s 35 • t'inor devia-
tions in the properties of nx and nxy are noted in nefs. 48,54,
however, the small deviations in these relatively uni~portant
properties are considered of no consequence at all.
A-II CELLULAR _cL;;..o....s.....E_D ~
The Fir,. A3 shows a typical C<·llular deck where a

trapezoidal hat section is connected to the base plate.
A-Ila t'embrane Constants
The principle for the computation of membrane con·
stants for the closed decks is the same as that for the open
decl--. However, because of the continuous base plate, certain
modifications in the constants, such as rxy and vyx' are re-
quired. Ex · Assuming full fixity between the base plate and
-
the hat (Fir. A3), for BC,
A-6
Prom Eq. A·S, [xh' the elastic extensional constant, for the
hat only,
= Eiot2 A-7
Exh I 1
y
l)r' is the IT'oment of inertia of the hat portion about the base
plate. By compatibility,
A-8
The equivalent orthotropic plate is assumed to be of the thick-
1.-9
For all practical purposes, the denominator of the above equa-

tion (A·9) is eQual to Ul".ity.
Ex • E A-10
~nd vyx • If the full cross-section in Fir. A3 is effective

in resistiny the axial force along the y-axis then,
A-11
where Ay is the cross-sectional area of the section sh~·wn in

Fig. A • Dy the riaxwell-Betti reciprocal theoref!'I,
3
A-12
Ax
vyx. vry A·l3
The shear rigidity for the closed deck can be considered equal
to that of the base plate which is usually directly connected
to the edge me!"bers. Even a thin plate interconnecting the
free hat portions will increase the effective shear rigidity of
the section. However. the cost of construction will have to be
taken into consideration in this approach.
A-lib Bending Properties
These properties are given in the Table A-I.
A-Ile Torsional Rigidity
Here again the values of Dxy and Dyx are not equal.
As shown in Fig. A 5 , the value of Dxy is negligibly small as
compared to that of Dyx· The value of Dyx can be obtained by
using Brendt's formula 46 •
4A 2
c A·l4
where Ac is equal to the area enclosed by the cellular hat

portion.
However, the magnitude of Dyx tacitly assumes that
the shape of the deck does not deform (dotted lines in Fig. A4} 47 •
The in-plane shear forces set up due to the twisting, produce
secondary bending ~oments in the individual plates. Besides,
the bending of the overhangi~g plate, outside the cellular
portion, further adds flexibility to the cellular section.
In order to account for this reduction in torsional
stiffness, an effective torsional rigidity of the cellular deck
is defined as the torsional rigidity of the ideal system (shown
by the dotted lines in Fig. A4), free from individual plate
flexural deformations. The work of the deformations due to the
torsion of an idealized system is equated with the work due
to torsion and secondary flexure. The modified twisting
rigidity can be given as,
- \t..~ -
A-1·5
The author checked the reduction factors given in Ref.

47. The method of calculating the reduction factors is given
in the Ref. 46. As pointed out in Chapter II, the warping 49
stiffness for both close and open form decks is neglected in
the analysis.
-le.;-
z
lay
_________. . . .__ x
z •
p
x
u !1x_ j y
(a)
r ~~x.I •
P --------1c:===========::::========::::j~~~-
(b)
I
p Fig. Az
Fig. A Equi~alent Orthotropic Plate

1
x
t;T
p2
... B f tl c 2-1 p
y
fl·
p
a -1;~
J.
Fig. A3 Cellular Deck

Assumed Hinges
Midway Between
Ribs
Rotation in
\"'-.an Ideali Zf::d z
System
Actual
Rotation
Fig. A (Ref. 47)

4
Fig. As
T-wisting of a Cellular Section

TABLE A-I
Pr.OPEPTY: DECKS
CPEN CELLULAR ISOTY10PIC
:·embrane
Io
Ex E y-r F. E
y
A
s H
F.y F
r F ,;
"xy v v
1o 1 Ax
vyx "rs· v r..y v
y
l
El vE o vE vE
i;-r
2 Ax
l·vyx"xy c L fl 1-v A l-v 2
y
a·: aG iG
Exy
Bending
Et 13 Et 3
1Et 3
DX !T:-·
0
i2~~1-v 2 ) 12(1-v 2)
n··y F.Iyd EI :,'ti Ft 3
12(1-v 2)
vEt 31
nl 12s " vEt 3
12(I-v 2 )
Et 3
!t GIO 0 24(1+v)
·xy
2
GIUs 4Ac Et 3
Dyx -r- a•. 24(1+v)
RrcdCrr)r,
APPENDIX B
STIFFNESS LATRICE$
The stiffness of the curved elerent is given in three

parts: I: r!embrane Stiffness (u,v, displacements); II: Bend-
ing Stiffness (w, ex, ey' exy); III: The coupleil terms of
u, v and w.
I: f\'TEl·1BRAME STIFFNESS f'ATRIX [K]m
TA:BLE B-I
Term Ext Elt Eyt Exyt

b a
'1 Ji 30"
1 1
~2 4 4
--·--
rt> 1 .. 1
'7
.:J 4 4
~4 - b3a a
6b
--· a
~s - b6a - ITT)
--o a
,6 Oa ~
a b --
4>7 30 Ji
-- -·-
a b
<t>a 6b a
~--o-
a
419 - m;- - 6a
----
'10 . a
~
b
Oa
e.g. •1 • !b i E xt
+ a E
3b xyt
-\t. -
I I: BENDING STIFF~ESS P~TPIX [K]bb + [K]bb~
All terms multiplied by c 1 are resulted because

shear strain depends upon the w terJ11s as well (see Eq. 3-~c).
These terms follow the same pattern as the bending stiffness
terms derived from the Eq. 3·17.
TABLE B-II
Coeff. DX ny !\y lJl cl

156 a .--1 'll 1 72 I
f1 156 " '"'TI" ib 25ib 676 ab
~;y 35 b3
22 b 2 78 a 12 1 3f' 1 286 ab2
ez n-~ Ilb"T l! i l! i -r
63
78 b
n-;r - n22 3z-
-"'!
12 l
n n
36 1
- IT o
286
-r 8
2b
.. b
11 b 2 11 a 2 1 11 121 a2b2
64 n 17 IT !O g-
~~
156 b- ---- 54 a r44 1 72 1
as --n- ~ n- b! . -rr aD - Bib 234 ab
22 b 2 27 a 12 1
- n- 36 1 33 ab 2
~it:,
1· - ~ """"'!
a Ilb"f a ·rra:
-·-·- 13 a 2 12 1 6 1 169 a2b
a, -
78 b
35 Ilb'J - 25 f:) - IT S -r
3 2a
11 b 2 13 a 2 1 3 143 2b2
Bs n;z - '°b! n rr 18 8
54 0 54 a 1~~ I 72 1
69 - 35 ~
- 3Sb3 --ZS ib 25 ab 81 ab
13b
2 27 a 12 1 6 1 - 2!7 abz
alO n~ llb! rr a IT a
13 a 2 12 1 6 1 217 2b
611
27 b
- !! ,. - Il"bJ rr r> !St> -r 8
a
Continued
- \11-
Continued
Coeff. DX Dy D·xy Dl cl
13 b 2 . 13 a 2
B12 - 70
..
;r 70b2 1
25 so
1 169 a2b2
36
54 b' 156 a 144 1 12 1
6 13
35~ - 353
b 25 ifi - rr ao 234 ab
2
6 14 - --
13 b
35 a3
78 a
35 b2
12 1
rr a
6 1
fS a
169 ab2
3
2 2 a2
8 1s --27 b
35 - 2
8 35 b3
12 1
rr r;
36 1
25 b 33 a 2b
i3T-- 11 a 2 l 3 _ 143 a2b2
6 16 - 70 ;r 35 b2 B IT 18
--r-
4 b 52 a 16 b 8 b 52 b3
8 11
35 ~ '31' o 2s a ff i 3a
11 h 2 11 a 2 1 61
8 1a - --
35 ~
a
- Eb! - IT so
121 a 2b2
- g-
2 b3 22 a 2 4 4 g a2b3
8 19 -rr;r 105 b 75 b 25 b 9
6 20
4T 18 a 16 b
fl a
8 b
- IT i 6 ab 3
- 35 a3 35 fi
3 --·-:-z-
621
2 b
35 8 2
13 a
- 105 b
4
75 b rs2 b - 13
9a
2b3
3 b3 9 a 4 b 2 b
8 22 35 8 3 35 F rr i 25 i 29 ab3
3 b
3 1:5 a 2 b b !l a2b3
823 - 70 ;z - ITff.fJ. 75 - 150 12
3 b3 4 b
624 - 35 ~
26
35 0
a - fla - rs2 ab 13 ab 3
3 b3 11 a 2
82s - 'f7J za ms- - b
75
b
- IT
11
6 8
2b3
52 b 4 a3 16 a 8 a 52 a3b
8 26 ~a n-b3 fl o IT o 3
22 b 2
3 4 a3b2
-ma- - rs;r
2 a 4 z~
6 IT a - -25 a 9
--21 Continued
Continued
Co cf f. D Dy D nl cl
x xy
8 2a 26 b 3 a3 4 a 2 a 3 ,)•
35 a J5b3 25 0 25 b
.'- 13 ,l
11 b 2 3 a3 a a .!! a3b2
S29 -ma- 70 b! ff IT 6
---
9 b 3 a3 4 a 2 a 9 a3b
~30 na- 3SbJ IT b IT o 2
-·-· 3 ar-- a
13 ·bz a l12l a3b2
a31 210 a 70 bi"
75 150 _,
18 b 4 a 16 a 8 a
'332 ~- 35 a 6 a 3b
35 ~ 25 0 Eb"
13 b 2 - 2 a~
4 2 13 a3b2
633 ma- Eb! 11! Cl. ~a 9
4 b3 4 a3
B34 ma- mo m16 ab mR ab 4 a3b3
~
2 b3 1 a3 4 2 a3b3
6 35 mano - m ab - fl! ab 3
1 b3 - 1 a3 1 1 ! a3b3
636 - 10 a 701> m ab 450 ab 4 i.
1 b3 2 a3 4 2 a3b3
8 37 - na- mo- - m ab - ITS ab - ---r
II I: COUPLING r!ATRIX [K]bm
These terms are formed due to the coupling of u, v

and w displacements (see Eq. 3-9c)
Cz = Exyt <AA£gB)
The stiffness matrix for the curved element can be
reduced to that of a flat plate by putting c•o; whereby c 1=o
and c •o. The schematic representation of these two matrices
2
is shown in Eqs. 3·20 and 3-21.
- \i4- -
TABLE B-III
Coeff. c2 Coeff. Cz Coeff. c2 Coeff. c2
7 a 3 7 b __! b2
el m es 20 a 69 20 8
13 30
7 1
- 1 b2 1 2
02 ITlf ab 06 iO ab 610 20 8 14 lPO ab
1 2
03 -21fl a2 67 30 a 0 11 m1 ab2
l 2 3 b
04 m1 a2b 68 -180 a b 8 12 20
SYMMETRIC
!ABLE B-IV
CURVED ELEMENT STIFFNESS MATkI! [K]
APPENDIX C
DECK STP.ESSES
A typical deck element 1-2-3-4 rectan~ular in plan is

sholm in Fie. c1 • The deck forces and the deck moments with
notations shown in Figs. 3.2 and 3.3 are calculated at the center
point o. The deck in-plane forces and moments are calculated
for a curved clement using the strain displacement relationships
eiven in Eqs. 3-9 and 3-10. The forces for the flat elements
can be deduced from the expressions given below. The two major
differences are: (a) c•o; (b) The local no~al displacements
in the local axes for each element are to be used.
c
l'1 xy • (u ' y + .r ' x - 2 Alf w) EX)' t
• Exyt{(-u 1 -u 2+u 3+u 4)/2b+(·v 1+v 2+v 3 -v 4 )/2a
~ [(w 1+w 2+w3+w 4 )/2+(0x +ex -ex -ex ) ~

l\L' 1 2 3 4
+ (-0
Y1
+e
Yz
-ey -ay )
3 4
l• (exy -exy +exy -exy
1 2 3 4
)~)}3
b
w,xx • [(eyl-ey2-ey3•ey4)/2a•{-exyl+exy2-exy3•exy4)'Sa] 4
w,YY + [(·8x1-ox2•&x3•8x4)/2b+(-9xyl•axyz-exy3•exy4llo1 s
The moments per unit length can be obtained
i· r
x • - [D xw'xx + D1w,yy1 6
l'.1 • 1
'y -CD1"'•xx + Dyl"·r j yy]
Fig. C2 shows the bending stresses which could be

in trcduced due to the local bending of each individual plate
elements. These stresses are not calculated here.
1 I 2 _ _ _ :ic
I! .
b 1:x,Ny
! , xy
(Mx, My
-·-·-10-·-·- .....
b
! I
-- I
'L~3
l- 'f 'f
Fig. C,... Deck Forces at Point 'o' .
Fig. c2 Local Bending of Plate AB

APPENDIX D
INCREt~NTAL ~iATRIX FOR THE DECK
For si~plicity, the ~atrix is split up into two

parts. [NJ 1 which contains the terms corresponding to the in-
plane forces iJx and NY whereas the matrix [N] contains the
2
terms correspondinr to the shearing force Nxy
[N] =- [N ] l + [N] 2 D-1
TABLE D-1
Coeff.
- Nx Ny Coeff. Nx Ny
-
78 b ' 78 a: . 27 b 78 a
411 ma ffib 4>13 rrr a - !ff 0
11 b 2 13 a ... 13 b 2 13
- •2 ma- m 4114 ma- 350 a
13 11 az
- ms- - m9 11 a 2
41-., - 350 b 4115
b
mo
11 2 11 2 13 2 11 2
2100 a 4>16 - 4200 b 2100 a
. 4> 4 2100 b
-----
78 b 27 a 2 h3 26
415 - 175 i mo- ~17 11s a 525 ab
11 b 2 11 2 11 2
,6 - ma- m9 a
tis - ~lQO b - 2110 a
13 az 1 3 11 2b
•1 - m13 b
m-o- 4>19 1050 b Il'1! 8
4>a
11 b2
-~
13 a2
4'zo
. rrfS b3a mab
s
. ~
- 27 b
- ma- 27 a
- fill> 1 b3 - 13 a2b
99 •21 m1f 3ITO
13 b 2 9 3 b3
- 3 ab
4110 3so a ..
1150 a 4>22 noa- 'Tm)
-
13 a 2
•11 - 700 b
9
- 350 b 4>23
1 3
- Itf olJ b
IS
t2l>oo a b
2
3 ... ~
•12
. 13 b2
mlf
13 a2
- 42lrn"
- 4>z4 - ~ J.·.:. - ~ab
Continued
- 'to--
Continued
Coeff. N H Coeff. N N
x y x y
--r:-3 11 a2b "' 2 a'
-- ?: L _ _=_ I41fO b rnra- IP•
,) 17s ab - 1750
2 a3 13 2 1 3
~26 m26 ab
~o 4>33 3150 ab - 1050 a
11 alJ2 1 3 2 ab3 2 a3b
4>27 InS° . ION a ~34 rsn rn"!
-·-·-.-· 3 a 3-
~28
13 ,,.1
- 1050 ::...;.) 3so b 4>35 - m! ab3 - ~ a 3b
11 2 '.{ a3 1 ab3 1 a 3b
~29 6300 ab ma 4>36 mo 42~0
-
3 a 1 3 1 3
~30 - 100
3 ab
mo- ~37 1050 ab - 3150 a b
13 bt 1 a3
cfi31 ·12600 a TI1fO
}]_~LI D-II
Coeff. Nxy Coe ff. N r.oeff.

xy
1 a
------- ---·-----
2 1u e·1
·----- -----···. -~---
ab
so
------·- - --------·-·-··------
b
IO
·--~- ·--····--------
wl 8x1 9 w2 8x2 0 2 9xy2 0 e~., e>:y:; w . 6x4
-•1 '2
•11
eil
4>3 <1>4
-----
•1s 4>19
xx:1
4>5
¢if
 6
l
4>7
-ct> 8
<l>s
4>21
"'3
4>9
-•10
$10
x3
4>2 2
4>11
•12
ri2-
4>2 3
•13
-4>14
'?
4>14
924
el'.:4
~15
e•I4
•16
?iz s
Wl
e;1
•20 •16
•26 •21
cf>34
-·-· 7
8
's
-4i21
•2s
•29
4>z9
$35
-•11
•12
•12
-+23
4>3(\
-~:;1
4>31
.36
•1s
-¢>16
-•16
cfi2 5
4>32
- <1>3 3
4>33
4>3 7
er1
6
x:rl
•1 4>2
•11
- 4' 3
·•1s -·
·•19
4 •13
-~14
$14
<1>2 4
-411s
·•16
-~16
-•:s
cf>g
-$10
•10
4iz2
-•11
·•12
-•12
"'.~23
Wz
8x2
•2c, <1>2 7 -4>15 •16 4'32 4>33 •11 -•12 4>10 4>31 e:rz
--·
+34 •16 ·cf>2s -+33 4'37 -4>12 •23 -4'31 .36 ijXI2
svn·ETRIC
4>1
-·•11
2 -<1> 3
•1s
 4
-.P19
•s -·
-·_,•1 -·
6
•s6
•20
-cfi 7
8 -•21
W3
6x3
\
cfl26 -•21
+34 8
cfls
cfl21
ct>zs
-+z9
-<1>29
<1>3 s
el3
eXI3
-
DC
•..
•1 -4>z 4'3 ... 94 W4
TABLE D-IJI
(N]
1
t•ATRIX
4>12 <l>p7
t26
-4>1 s
- ,7.1
; ,,6x4
r4
4>34 exr4
. - ·---···· ---·-8 ----
~
8 ex3__ ~~ exy3
. e1 dx1 e,1 ex2 ey2
"'1 xyl w2 W3 \•'
·4 ex4 0,4. exy4
..,·.
..
-e .. .... -8 ...) 0 e2 -e 1 83 -84 -e 2 - 0 -n
.... 84 ez w·
1
0 -62 0 e3 ~ e2 0 -e 3 68 -e 2 -6 6 0 0 ez 06 e
xl
0 0 0 02 0 -e s 94 -e 2 e7 e5 -e 4 e2 0 0 e·y t
0 e2 0 es 0 -e 2 96 -e s -e 9 62 -e 6 0 0 e,1-!_
-a 1 0 0 -e 2 0 0 e4 e2 el -e 3 -a 4 -e 2 w·
2
(J _n
- ,,. \J () 0 ez e6 83 -e 8 -e 2 -eg e72
0 lj -0
4 82 0 0 64 -0 2 -e 7 -6
5 eyz
0 82 -e 6 0 0 -e ... es e(}
2 96 exy2
el 0 0 -e 2
0 e3 0 02 W3
('.; -ez 0 -e 3 0 02 02 6x3
0 0 0 e2 0 es ey3 ,.\
0 e2 0 -es 0 6xy3 .,e
-e 1 0 0 -e 2 W4 "'
SYt-!f. iETP.I C
0 -6
z 0 ex4 '
0 (J
eI4·
.. 0 ~xy4
TABLE D-IV
[N] 2 PATP.IX
APPErlDIX E
STIFFNESS AND INCREt~NTAL PATRICES FOR A
FULLY COiiPATIBLE ECCF.NTRIC STIFFENE~
INTRODUCTION
In the case of light gage hypars the orthotropic deck
is connected at discrete points to the supportinp edge mell'lber.
Because of this discrete connection, there is a certain amount
of non-compatibility between the edge members and the deck. The
stiffness matrix used in Chapter III was developed by the use
of direct co-ordinate transformation. The co-ordinate trans-
formation assumes that the neutral plane of bendiny for the
original beam cross·section rel'f1ains unaltered.. l'Tith the type
of connections used for the light-gage hypars, the assumption
made above represents the tru~ behavior.
However, in the case of stiffeners which are rigidly
connected to a plate or a shell or a monolithically cast con-
crete beam, there is full coll'lpatibility between the strains at
the junction of the deck and the beam. In this compatible case,
a part of the deck also acts along with the stiffener (effective
width concept). The interaction of the deck and the stiffener
results in the adjustment of the neutral axis of the section.
To account for this change of neutral axis, the bending property
of the stiffener will have to be ~odified by arbitrarily assum-
ing the effective width of the deck actin~ along with the beam
(see discussion on Struc. '6' , Section IV- 51') • Depending upon
the relative stiffness of the plate, this kind of arbitrary
adjustment in the stiffener property will present a problem.
This arbitrariness is removed by formulatinp. the compatible
stiffener element.
STIFrNESS PATRIX
A typical eccentric member is shown in Fig. 3.10.
It is assumed that the mer.iber is uniform i.n size and its local
axes i, i and i, throug~ the shear center and the centroid, are
parallel to t~e global axes x, y and z. Let Uc be the average
axial deformation of the section measured at the centroid of
the section (C.G.) and V5 and ~·\ be the bendin rr. deformations
measured at the shear center (S.C.).
Assuminrr a rieid connection between the stiffener and
the reference structural node points, the displacements at the
shear center and the centroid can be expressed as follows:
'~--c = u
,, (aw) . y (av 1 E-la
ax ax
~
"'c c
"s = v - z a 5
E-lb
\'I
s =w + Yse E-lc
aw The total strain energy of the beam ele~ent can
where e • ~y·
be given as
a a21·r 2
~
EAb a 3U 2
-r'0 ( a~> dx I (
ub • + s) dx
0 7
2
Elz a a v·s 2 l
+ -2 r,J
a
I aidx + !2 Er I
a
o" 2dx E-2
+ 2 J C-::-rl
ax
c1.x -·
0 0
All the member properties are expressed with refer-

ence to their local principal axes. The displacement fields
-\lf -
for u, v, wand e are given by Equations 3-26, 3-27. The

stiffness matrix (neglecting the in-plane rotation e ) is given
i
in Table E-I. Though not tested on the hypar structure, the
element when checked for simple cases such as simple and canti-
lever beams, t.·onverged to the correct results. The element can
be used for a rigidly connected eccentric member and correctly
locates its own neutral axis under different types of loading.
It may be worthwhile to note that because of the
coupline of the u, v and w displacements, the resulting displace-
ments for u are no loneer linear. The use of a Permitian
polynomial of the same order as used for the v and w displace-
ments will give more rapid convergence characteristic than with
a linear u displacement field. --
'"'°
-
'41C.S
--
~ ~t.~~\'4t.
use u as
°"
..~ ~t \•c.k. •'f
non-linear because a corresponding degree of
freedom in ~ formulation for a plate or a shell element.
Similar stiffness matrices are reportec in Refs. 70, 71, whereas
the theory of the element is developed in detail in Refs. 72,
7 :s.
INCRErTENTAL MATRIX fOR UNIFOrJiLY corrPPESSED

DCCENTRIC §TIFFnNER
The load acting on a fiber whose cross-sectional
area is dA, is ac'l• and the change in the distance between the
end point of a fiber is given by 6u. Tile increase in the po-
tential energy due to the change of fiber is given by
6V • .. I at.udA E-3
A
For the section which shm~s both r.iodes of buc1(ling (flexural

- ' iC. -
as well as twisting), the change in the length of a fiber is

given by 57 , (Fig. E )
1
1 a 2 2
du• 'J" I {(~+ye') + (av - ze') }dx E-4
£ ax 0
ax-
For the uniforlll.ly loaded section, a is the constant
quantity and aA a P. Therefore the change in the potential
energy
a 2 2
~v a !2 p f { (~) + (av) + zy {aw_) 9 ,
0
ax ax c ax
E-5
where Zc and Ye are the distances of the centroid of the sec-

tion (Fip. 3.10) and p is the radius of gyration for the polar
moment of inertia of the cross section about the point o.
r-6
The incremental matrix can be obtained by inte~rating
the expression E-5 and then obtaining the second derivative of

the total change in the potential energy. The incremental
matrix (8x8) is given in Table ~-II.
The stiffness and the incremental matrices for the

eccentric stiffeners can be used to predict the buctling load
for the stiffened plate. Because of the non-availability of a
large-order eigenvalue solution routine, the formulation could
not be checked with other available solutions. However a
simple case of twist-buckling of colunm section about the
forced shear center for a T-section and I ·section ''las compared
with the classical solution for hinged end condition given by
- \ 11-
Bleich 56 .
2
.2E Z
• • ~ [ s Iy _+ rb
1
+ G
2
a J E-7
ac 7 pc ,..2E I~~]
I pc • Polar moment of inertia about the forced shear center o.
The only difference for both the cases is that the T-section
has rb • o whereas rb for the I-section is 1590 inch(. In
both the cases, for simply supported ends, the twist buckling
load was within <<11 in error, from the values given by the
classical solutions. A negligible error is observed in one
element solution. The buc~ling stress predicted with
2 and 4 number of elements showed an ~lmost insignificant im-
provemcnt.
z
t:.y = ·-Ze
t:.Z • Ye
{\ z
z
y
,___ _ _ _ _ _"""", Ay
y
Fig. E1 Displacement Due to Twisting of a Beam
Re-;; train--
,,' '
\.
,,.' ,.."'.")
a u zoo•· \
,,
\\
\\
~'-;/ \ \ .........,,,,. ,,,,. ...
,,,, "
,..-"'
~Tl2 Fig. E2
I 39-95
Twist Buckling
12 WP 40
- ,.... -
\
U. I 'JI w, a')(., I 0 (f I I 9 1'..:( I Lli. lI vl- W::z_ e 1(. :a- e(!-:a- e")(.aJ..
"
A'bE~
AbE.
0...
-
0 0 0
0...
0
-
-AbE.
G,..
0 0 0
-AbE l.,
0..
0
1 2E l
Oa
12EAi.'1'c.~
·-
-12E r.,, ls _,lA!,"'l'c.~ -~er.
a.j.
i-s 0 _,_,
~I U
o l-t'J.£Ai;fJ
l 1.1E1 1 l 5 i-(,EAi.'fc:!J-,Eiil-1.S
"-a o.l ~l o."' I o.'1

~l Q,. ~ 0...
;J. a,.l.
11.EI 01 1:z.E L
1
r, -C. El 0 y 'e.r'4'fs
u
0
'-12.E~"'l'c.~ -11E IO'/ -12Eii1's -(.E 101 'Eict"fs
c..- 1
~·
Cl) a.3 a..3 0..3 0.. ~ 0.. 'I. CL~
-·-- -----,:-;--
11u3 l's i!o.3 ,EI,_!s'3./J 1 i-11e11 Ys~ :,ui1 "fs/ ~~
+ 11E11i:/o: -(.ElJ~ HEl~l;/d
0
I I
/ZE!i ~s ~Lt.l'J 'Ii i-12.U 1 1:~ -H !~rj +' Eli'~/d
+ ~ ~-vo. o.:~
+ c,J/ro l
! o..l o.~
- ~ GJ"/a_ J.. + GJ/10
5
+1Hr~/~ + ~ ~1;/c..i -12.ErJo.3 C\. t (,EC'._/d°
lfE Loy -4U't 'r's i-A&,.: tc.. 'EA~Y, ~ 6 Eloy

0.. 0... d I a.'-
- 0.:Z.
'EI.4Ys 2. E tO'f'
o..1.
-- :_1!: Il s
0... 0...
"f
'IEl~ ..,•lo: _,EI 1 rt/d, ~-- ').E.I'J Y5 /a.

;_H !1 !sj'Ei~'f, _,E [!!~ -2.E 1aY,I_.. i.Eiz l.s/~
!
~'4EI 1 !~ ~ l.
I
0
+- ! GJo.
15 : (l~ i (),,_ - GJ/10 -~+ 2. EI't./o.
+ ,, f!'j,,/(1 I ~ ,J
I , - 'ET'i.-ld -~ u 0..
AbE I AbE~c.i
---
v...
Q I
'
j
0 0 ---1 I
0
SYMMETR.IC l I
-- - Q.
-
12.E I
011
:,l.EA'"Y,i! -1.z.£I 1 ls i' EA'-'l",!J 'E Ia~ S
2 A.~ '
'
~3 o.3 d.~
I
o.~
Ioy • I y + AbZc I
I
12.E I 0 ., 11.E I-;, 'f's ' ' E r.,,. -,a let 'Ys
2
1oz • Iz + AbYc o.~ a. s I o..a. a.. J.
12.i t 1 Y.sf<l.l; -·' E l;J i~~/J
1
- --
TABLE E-I If. 11. E 1~1~~' eL'1T,~ t. E.t~t.:;&-
s TIFFNESS MATRIX FOR A COMP A';' I Br~.
~}Gj/o.I ... ~~GJ/,o
ECCE~1'RIC STJJ™ER i... 11£rb /Q.3· ._,Er~/o.2.
4 E I 0 T -4£ !"' y~
Q. <>-.
~E.l~ Ys/ o..
'I.'
it-'~ El 1i&/ a.
~Ji. GJO..
-1- 4E l'1o/0..
-\C\O ., , .
wl 6x1 eyl exyl Wz 8x2 ey2 exy2

6 6 a a a
s ! Gl I1f In" Gl - r6 6
) Gl TIT IT Gl wl
6
) Gz -m Gl
a
IO" G2
6
- ! Gl
6
- ! Gz - Tir Gl a
1lf Gz 6x1
..
2 az
I! -h azGl I
·nr a G1
' . l1f
1
Jcr 8
2
-ba2Gl eyl
2 2 a a az
I l a Gz - Ilf Gl - 11f G2 !lf Gl • ka2G2 6xyl
6 6 a a
SYHHETP.IC r ) Gl nr - IO Gl Wz
Gl • ye 6 a a
! Gz IO Gl - i1f Gz 6xz
I
G2 -- ...E£
ft.
Note Common Factor i

p
z
r~ a
2 haze 1 eyz
J.
TABLE E-II INCREt1ENTAL ffATRIX FOR ha2G2 8xyZ

cm1P~TIBLE Et;CENTRIC STtf~ENE~
APPPJIDIX F
COMPl'TEP PROGPAPS
Computer programs were written for linear stiffness

and instability analysis using both curved and flat ele~ents
(rtethod 'a' and rrethod 'b'). Since the input data for both
curved and flat elements is nearly the same, a general des-
cription of both stiffness and instability programs is given
here. All information given in this appendix, is pertinent to
!ID' 360/65 wodel available at the Cornell Universit~ Double
precision is used in all programs.
STIFFNESS Al-JLAYSIS
Hypars have mostly been analyzed for a uniform grid

size 8x8 (64 elements). Dependin?. upon the nvailable core
storage and required accura~y of the solution, the grid size
can be varied. For the flat plate ele~ents, the program re-
quires ~ core size 365K and has a compilin~ time roughly of 55
secso The coJttf~Jtation time fc·r one problem is roughly 70
secs. The time requirement f~r the curved element pro~ram is
somewhat less.
The i~put information can be catef?orize~ as follows:
L Properties of the ]eek
Properties of deck::: .;hm.'l'l in Fig. 2. 2a, b and c can
be calculated by a subroutine ''PDEC" in which the geometrical
properties such as thickness and other physical parameters can
be specified. For decks other than those mentioned above,
elastic constants E,
x E,y
..• etc., are to be calculated and
read in as input data. Besides the type of deck, the orien-
tation and DUJl'ber of decks also need to be specified. The
geometrical properties calculated on the basis of shape can
be further modified by the use of coefficients determined ex-
perimentally.
2. Geornet.!L.Q..f..!he Surface
The geometry of the surface is defined by specifying
the numher of hypaT quadrants and the property of each quadrant
in terms of its rise, spans, local origin (A 1. , B.,
1
C.,
1
i.1 and
yi, Chapter III). The groupin~ of the elements in each quadrant
also needs to be s1,ecified. Structural shapes other than
umbrella shell, can be handled with ease.
3. Spring Cata
As pointed out !A Chapter III, i~ order to retain
the close-handedness of t~e stiffnes~ ~atrix, ~embers such as
a supporting column and :.en~ion tie .rod are idealized by
springs hllVing equiv.::_; ''nt stiffnesses. 1h.e number of such
stiffness constents ,.1s to be specifie.'..
4. Beam DatJ!_
Beams can h-.· si.dded al01q~ any l int:s of generators.
The beam data is given in the usual way with jts location
specified by the start and end points and their co-ordinates.
Beam properties such as area, moment of inertia, warping con-
stant, eccentricities and distance of extreme fibers for the
calculation of stresses is required.
S. Loading
The solution can be obtained for three different
loadinp, cases or any combination of the three. The three
cases of loading are: a. uniformly distributed load over the
~1ole surface; b. weight of the edge member; c. discretl)
applied forces at any nodal points.
OUTPUT
The output is given in the following order:

{i) Six components of displacements at each node point
cu , v , w , ex , eY , exy) .
{ii) Axial bending and total stresses for beam.
{iii) Deck forces per unit length {Nx, NY, Nxy' r~x' and
~ry) .
{iv) Reaction forces ir. each idealized spring.
INSTABILITY ANALYSIS
Instability analyJis is done only for uniformly dis-
tributed loading for the grid size of t5x6 with 36 elements
though results can be obtained for any general type of loading.
As pointed out in the stiffness analysis, the grid size can be
varied to meet the requirertents. The compiling til!'e is roughly
so secs, whereas the time for individual iterative cycle
ranges a?ong 20 to SO secs.
The input stream is basically the same as that for
the stiffness analysis, the addi.tional information needed here
is the s tartinp; load point (I S.t. increment t see Chapter VI)
subsequent increl'l'ental loads and the number of iteration
-\ C\4- -
cycles required at each step.

The output gives only the load level and the corres-
ponding displacements at all node points.
The program using the flat plates can be easily
modified to solve any shallow shell problems by defining the
equation of the shell surface in the global co-ordinates.
TABLE II-1
SINGLE DECK
EXPT. AVERAGE THEORETICAL

TEST INCH 6 INCH a INCH
No.
60 0 60 aint~x-polated
811 1.14 0.01 1.23

812 1.14 1.14 o.os 1.45 0.077
Sll 0.80 0.06 0.68

S12 0.77 0.78 0.04 0.11 o.oso
311 0.33 0.32 0.06 0.28 0.052
31Z 0.32 0.04 0.38
TABLE II-2
DOUBLE DECK
2 1/2" x
1/4"
TEST EXPT. AVERAGE THEORETICAL

60 INCH 6 INCH a 40 INCH 0
No. 0 interpolated
821 0.86 0.75 0.07 0.71 0.080

822 0.65 0.06 0.77
o.os 0.84
82l(I) 0.65 o. o.i 0.93
521 0.59 0.58 o.~6 0.38 0.040
522 0.58 0.04 0.48
521(1) ·o.4e
321 0.24 0.26 0.06 0.14 0.037
322 0.32 0.040 0.20
321(1) 0.21
Test No. 521(1) ~~ans:

S • The curv~ture of the hypar, salbe as that of a
quadrant having rise/span, ratio of 1/5.
2 • Two decks.
1 • Tes.t No.1
(I) • Both top and bottom.deck&· were;, l!iterconnected.
TABLE III-1
COMPARISON OF CLASSICAL AND FINITE

ELEMENT SOLUTIONS FOR S n~PLY-SUPPOP.TED
AND UNIFOPJ~LY LOADED 28-G
ORTHOTROPIC PLATE
FINITE ELEMENT SOLUTION

DOUBLE SINE
LOCATION GRID SIZE SERIES SOLUTION
6x6 12x"'2
'5 INCH 6 INCH 6 INCH
A 0 0 0
1/6 6.11? 6.106 6.31

1/3 7.39? 7.397 7.36
1/2 ·7 .137 7 .139 7.09
2/3 6.961 6.962 6.96

5/6 6.944 6.944 6.98
'8 6. 952 6.952 6.97

-\C\, _
"1 a"'-1 e-. 1. e e<J"1.. 6~2-

"l.
'U'
Ai. E
- 0..
0
W1
0 0
e(f 1
A.i,E2c. ~ A1:iE'1'c. .... Ai,t

0.. 0.
_'U.2..
(L
0
w'l.
0
'7(.
0
2-
,_AL,E\ AbE 'Ye.

a... a. I
12.E 11
0
"""12.EI._ i:s
0
'E I i 0
-12EI~
0
1'2. E. I a. -C s
0
"€.Ie
.,..
a.' a.a l.
0.. o.." o.?> ()..
12.EI I 1%.EI}f'l& _,£I~ i-12.EI31-1zEI~ Ys ,_'El~

0 0 0 0
o..' a,.l
'l.
0. a..3 a.a
•1:a.. u 1 1-i/!
a!-
•1u1 1VJ ;a:.
~
-6'EJ~1 _,E~~ t~EI.,
-12.UJ 4 "" ._,El~~ _, EI~l..s
+1s.na~
2.
GL
I. 0..1.
0
0.
.. o.'"6
-12.E•i y~1
I a, o.2. C.,.
i. I
I
+ G;/.:L
.....__. - Go/"o. j
'J.
""'Aa.E·Zc.
.. I
t..1:,Ei,
G. '-A'1Y,1rAi,E. l,c
0
i;>EI;J 'f I~ Ya c.\. tA.oE"Y, "1i:.
I
i- 4El~ .:L CL o.2- o.'"
!f-2.EI;r ,".\..
0. a..
A&e 'Yc.
1 '<-AloE l',-:
~i>E "f, -,EI ~ A.F~~~
- 'EI~ls
(l.. C)I,..
0
.. .o.£li
C1. 0..1. a. J. .... :e.ri:
,. ()..
0... '
I
A0E A1,£ :,, -A 0 E l', :·

SYMMETRIC CL
0 0 0 ...._._
0... 0. I --1
111H~
0
'""12.!:I~l:c:.
0 - "e.ri! I
TABLE III-2 f
a,1
o..' c..:i.. I
STIFFNESS MATRIX FOR NON- l:Z.EI 11 IJ.El;('fs 4'E Ia_

0
I
COMPATIBLE ECCENTRIC BEAM .,.,3 o..' (). 2.
ELEMENT CLINEAR VARIATION OF ~2.II~ !.o/'J

ANGLE OF TWIST 0) 10E1j1J ,u,?,I
11>t1UiYVc;
o..~ !
0.:- I
+ GJ/o. I
- Ar.E i.t'° l
a.. i-A&,EYc.1,! I
~!-·4'Ii I a.
CL. I
'-•''ft I
ct I
+ 4:_11 II
--·---
e., ~I~ e_1t._:.z..-+l_e_<1_J.-+-s_1 • J G
,+--v-:l..--'--w·-J.-+___ -~J
0
A&,E Z.~
--..- •At/Yr..
_..._-...:;.... 0
__t]_,
- A i l
0 0 0 ·Abf'. '!( AbEYe! 0
0.. a. OI. 0.. 0.. I
-
-12.E.Ii 12f.I1 Zs
0 0 0 0
a. 3 0. '3
SYMMETlUC
_____ _
._..._........-
TAB LE I 1I - 3
--
STIFFNESS MATRIX FOR
.- -
NON ·COMPAT IBLI:l ECCENTRIC
BEAM r:tEMENT (CUSIC
~.:.----
V.l\RIATION OF AN(U,!_ OF TWIST, ~l

TABLE III-4
ROTATIONAL TF.ArJSFOP'·!ATit"N
........... .........
., - ·--,----
PATPIX
FOR A BEAiT ELE!~~ENTSS
_......,_......,. --
[).] 0 ]
(T] P. - p.] •
[ 0 (AJ
CZ
-CICZcosS+Cxsin~
(A] 1:
}~2x + c2
'z
CyCzsinB+Cxcos~
j crx + c2
z
CJC 0
-C C cosa-czsinB
•-.l- -·· - '
rcr:
../ "':ic
CZ
z
t:r:::zcos B
J"'x""""z
0
0 0
Xz . Xl
ex • L
Y2 - Y1
Cy • L
z2 . z1
CZ • L
See Fip. 3-18.

- 2,.oQ
-
TABLE I"-1
STR~CTVRES ftNALYZED
STRUCTURE GE TRY DECK EIJGE ~-'EHBERS

BOUNDARY LOA
CONDITION~ IN'
··~ I d Exl0 6
A E c NC. OF TIIICKNESS
TYPE PSF
i~o. TYPE Ir;Qi INOl INCH I'YPE DECKS INCH INCP? /INOI a LBS/INCH 2 Bl B2
-
x
"
1 I xso xlO I 1 0.80 - 1.0 28.! 0.40 - - I 10.
.. -
2 6.46 6. 46 1. 30~ I. 1.
0.25 - I
0.5 0.39 - - iI 144
3 30 30 var- II 1 28-G 0.00047 var- 29.S 0.30 3''0.D. 3"0.D. III 40

ies 0.0149 ies J.:;':Thk ~" Thk
pipe pipe
4 ~; II
"
,, II
2 " II II '; I!
" . I! II
-s IV 360 60 0 100. IV 1 See 1.53?· 1. 0 I' II

14 WF 14 WF II
80
69 Fig. 43 68
4.3 .....
- '.
6 III 24 24 6 I 1 0.25 - 2.77 0.15 +See Fig. IV 40
9'
4.S +
1·
- - --··-· r: I. ,. II
7 I. I
" It
" ':
"
8 •• '· ll p !1 II II ll Ii II
" VIII *28
8
9 II 72
., 2 14.4 II 2 28G 0.00047 0.04 29.S 0.30 111~ 2-1 ·~
1
v 40
' Std Std VI
0.0149
pipe pipe
,, ,.
10 II
13. " Ii II
" ii ll
" " "
80
- Z.01 _
TABLE IV-1 C'lllTINUED
- BOUND.ARY LOA
STRUCTURE ~E0~:1:TRY DECK EDGE t~ErBERS CONDITION~ UJ·
i~O.
11
TYPE
It
A
Ii~ O!
72
J:rn 72
c
INOI
14.
TYPE
II
NO. OF THICKNESS
DECKS
1
INCH
28G
I d
INCHt/INOI
0.00047
a
0.06
Ex lo'
LBS/INCH 2
29.S
"
0.3
Bl B2
3 ''C. D. 2-3"0.D.
TYPE
v
PSF
40
40 0.0149 's"Thk l:("Thk VI
pipe pipe
--··
12 •; t;
" 1; It
" 24G 0.000753 i· fl fl
" " VI "
.. 0.0239
- .
,. II 0.00047 o.os " " "
ll II t. ti
2 28G f:
" "
0.0149
t3a
--·- I'
0.06 fl
"
,,
" fl f I
ti
" ti !I
" "
' - h ti
12[40 Z-12 v ti ti
14 1:
240 240 48 ti
'' 16G 0.00192 If
0.060 [40
--
15 i.
12. 12. 5.216 I 1 0.25 - 1.0 I). so 0. 39 +NO BEAMS+ IV 1.: A
92 92
--
x Dimensions are in ems
+ gm/cm
2
* Only t.alf the structure was loaded

• Two leading conditions are analyzed:
1 With 40 psf only 2 40 psf + weight of the edge ~embers.
For types of structures see Figs. 4.1 - 4.5.
TABLE IV·2
DECKS
I • Isotropic deck of uniform thickness
II • Corrugated sine-form. Fig. 2·2a.
III • Corrugated trapezoidal. Fig. Z·2b.
IV • Cellular trapezoidal deck. Fig. 2·2c.
BOUNDARY CONDITIONS
I • Bdges x • ! A and y • + B are fixed.
II • Edges x • ! A and y • ! B lnife-edge supported.
III • TF • 0 for edge members along x • ! A and y • ! B.
(i) w • o along x • ! A and y • + B
(il) a tension bar connects low corners. (f and b -
Structure I and a and c • Structure IV).
IV • TF • 1 and TH • 1. For all edge members. For the
eccentric locations of edge member with respec~ to the
deck, see Pig. 4-S.
(i) x • y • o Lines of symmetry.
(ii) x • + A and y • + B Free edges.
(iii) x • y • o rigid support.
v • TF • o TH • 1 for all edge members. For the eccentric
connection of edge members see Fig. 4-2. For
(i) x • 0 u • o.
(ii) y • 0 v • o.
(iii) eya • exya • eyh • exyh • exc • exyc • exf •
exyf • o.
(iv) x • y • o rigid support.
VI • Boundary conditions i<lentical with V except TH • 0
for all edge members.
VII = Tr • 0 TH • 1 for all edge members. Edge members are
connected concentrically. Boundary conditions (i),
(ii) and (iii) arc same as in the case V.
VIII • Boundary and fixity conditions. Same as given in IV.
The line cf (y=o, Fig. 4.2) is the only line of symmetry.
Also see Fig. 4.20.
cortPARISON OF DEJ-·LECTIONS BY
l ETHODS 'a' and 'b'·z FOR STPUCTURES land 2
-----
STRUCTURE 1 STRUCTURE 2
LOCATION -nethod I a 1 Method 1 b' !~ethod ta I r·riethod 'b I
~2
l(} ems io· 2cms 10 ~ 3 irich 10 • 3 inch
c 1. 228 1. 235 s.so 5.552

2.352 2.384 8. 398 8.466
2.574 2.602 9.149 9.262
0 2.531 2.551 9.196 9.322
----
6
0.742 0.750 3.271 3.251
2.218 2.248 7.526 7.561
2.613 2.647 9.081 9.184
0 2.531 2.551 9.196 9.322
NOTF.: See Figs. 4-6 and 4-8.

-"?..O~ -
TABLE IV-4
DECK STRESSES IN PSI AT THB CENTER

OF A glfADRANT (Point e in Fj.g. 4. 2)
AXIAL STRESSES BENDING

STRESSES lTOTAL STRESSES
STRUCTURE
NO.
EXPERI·1CALCU-
DESCRIPTION MENTAL LATED I
EXPT CALCU· EXPT CALCU-
LATED LATED
9 28-G
Double Deck 832 848 1460 1870 2292 2718
11 28-G
Single De<.:k 890 62 6820 10700 7710 10762
12 24-G
Single Dede 69 5600 6200 6269
13 28·G
Double Deck 2780 22 6505 4510 9385 4532
For the Structure numbers, refer to Table IV-1.

TABLE IV-5
-
DEFLECTIONS IN INCHES OF INVERTED mfBRELLA
SHE_!.L WITH STIFF EDGE r~EMBERS
EXPERIVENTAL
LOCATION ANALYTICAL AVEPAGE J r•AXIPm,• lr,,If'lIMUt'
STRUCTURE 11
c5 a 0.133 0.20 0.26 0.15
c5b 0.29 0.29 0.48 0.18
c5 c o.16 0.15 0.15 0.15
c5e 0.91 0.70 0.92 0.59
STRUCTURE 12
c5 a 0.12 0.11 0.180 0.04

ob 0.26 o. 2 3 0.26 0.32
cS c 0.13 0.095 0.150 0.040
c5 e 0.57 0.42 0.62 0.26
STRUCTtmE 13
--
c5 a 0.18 0.15 0.26 0.040
c5b 0.31 0.31 o.ss 0.15

6c 'J .18 0.15 0.26 0.040
60 0.56 0.54 0.92 0.26
See Fig. 4.2.

TABLE IV-6
EDGE BEAt' STRESSES FOR INVERTED tT'·tBPELLA SHELL
WITH STIFF EDGE ~4'I!HBERS (g • 40 PSF)
----- --·-
AXIAL
--·----------------
BENDING TOTAL (Absolute)
LOCATION EXPT. ANALYTICAL EXPT. ANALYTICAL EXPT. ANALYTICAL
1 STRUCTURE 11 28·G Single Deck cx•0.06

a 2440 1070 4850 4481 7290 5551
b 1400 456 3120 3990 4520 4446
c - 580 - 586 1660 1167 2240 1753
d - 900 2320 2496 3396
e 1620 856 3020 3050 4640 390G'
---
2 STRUCTURE 12 24-G Single Deck a•0.06
a 1740 1093 4140 4385 5880 5478
b 530 420 3500 3801 4030 4221
c -1060 - 609 540 955 1600 1564
d Ml070 ·1075 1100 2165 2170 3240
e 1100 874 2300 2668 3400 3542
3 STRUCTURE 13 28-G Double Deck a•0.05

a 22SO 435 4110 4228 6360 4663
b 930 435 5260 4228 6190 4663
.-
c - 900 Sl8 2040 1650 2940 2168
d . 560 734 2020 2106 2580 2840
e 1110 357 2320 2561 3430 2918
All stresses are in psi

... -
d
- ~
-e
c
TABLE V-1
SELECTED EXAMPLES SHOWING THE EFFECTS OF CHANGE OF STRUCTURAL PARAMETERS
DEFLECTIONS EDGE MEMBER STRESSES PSI

VARIABLE IN INCHES MAX. AXIAL
•structure
Type Value oa oc ab be
9 g 10 Rise 'C' 14.40" 0.38 1.57 0.38 0.69 -5329 -5329 2747 2747
13.80 11 0.41 1.69 0.41 0.75 -5518 -5518 2752 2752
13 g 13a Shear 0.05 0.18 0.31 0.18 0.56 - 856 - 856 43~ 43~
Rigidity -0~06-~~o-.15-~-o~.2~ai:--_,,,.o~.1~s----=o-.4~s~~---....-ss=1.--~---....-8s~1=--~~4~91..--~~4~8-1~
£actor a
8
11 & 12 Thick- 0.0149 0.13 0.29 0.16 0.91 -1129 - 802 456 1070
ness
o. 02 39 0.116 0.26 0.13 0.57 -1150 - 870 420 1093
No. Single
11 & 13a of Deck 0.13 0.29 0.16 0.91 "'.1129 - 80 2 456 1070
Deck Double
Deck 0.15 0.28 0.15 0.48 - 881 - 881 481 481
EDGE MEMBER STRESSES PSI DECK
MAX. BENDING BENDING STRESS Shear force at
*Structure --~--~----~~--~--~~--~--~~~~-----------------
PS I Center for Nxy
oa oc ab be lbs/inch
52.75
9 fl 10 19,077 19,077 18,179 18,179 2,130 54.96
6,648 6,648 4,228 ~,228 4,510 50.0
13 g 13a 5,899 5,899 4,095 4,095 3t780 51.28
11 & 12 5,225 4,634 3,990 4,480 10,700 36.96
---------~~--------4~,_7_2_3_____3~,_9_7_4~---3~,~8_0_1~~-4,385 6 zoo 37.75
11 & 13 a 5,225 4,634 3,990 4,480 10,700 36.96

5,899 5,899 1.095 4,095 3,780 51. 28
For locations of deflections and stresses see Fig. 4-2.
* For the type of structure according to number see Table IV-1.
-7oc;-
TABLE V-2
THE co•·PftRISON OF AXIJlL STJIBSSES IN THE EPGE ~"E~!BER AS GIVEN

Er ?·E;'BP,~NE THEORY AND BY ANALYSIS
--- Ar\ALYTICAL PSI OF ~'fEMBRANE THEORY
STRUCTURE
!>··E~·IBR.AN E
THEn~y
STP.ESS
'
NO. a. PSI oa oc. ab be 0'11 oc ab be
--- -
SINGLE !'EC!
- s 1.0 *19660 10700 907(1 ~440 11~40 54.4 48.? S'l.6 56.1
18640
11
..... ~-
0.06 1570 1129 snz 45f

-----
1070 71. 8 51.S 29. 68.2
12
- 0.06 -- 1150
--
870 420 1093 73.2 ss.s 26.7 69.6
lS70
--- OOPBLE nECKS
--.. - 75.2 38.8 38.8
9 0.04 7100 5329 5329 2747 2747 75.Z
-
13 o.os 1570 856 856 435 435 54.40 54.4 Z7.1 27.7
13a
- - --· 0.06 1570 881 881 481 481 56.Z 56.2 31. 30 31. 3
14 0. 06· 3420 1673 1673 1575 1575 49.0 49.0 46.0 46.0
* 1-:ember siz.es are different, see Table IV-1

?ABLE Vl·l
PLATE AND STIFFE~rnr BUCKLING PR<'BLENS
-· DI~NSIONS
LOADING a·ineh b .. inch t•inch y ·4 I t Error
: '
b !OS 108 S/8 . - 4.0
~
... 2
106 )08
,_
5/8 $ 0.10 :it. to

""/, -
.
....- 108 108 5/8 .. - 4.0

b <0.2
""! r 108 108 5/8 s 0.10 20.S
a/2
-
.....- ..
. b
108 54 S/8 - 4.0
• f
~ 108 54 5/8 s 0.10 7.69
. a/2 . - -
~a...._
t
t ........ •'b - -
16 16 0.10 9. 34 4.65
__.
-- a ... ""'
t t 70.S . 10. 5
24-G
0.0239
- - 0.670 7.46
Jb
t
-- .
Properties nnd DiD.ensiotts cf Flat ~he~r Tests
Test Dicensions Steel Edge No. of Seam Deck - Ed.nc G'

a (1 ......, ,.ln 'J
-No.
1
(ft)
6 x 6
Decking
26G s.c. *
Hembers
6 "xl ~"x. 1046"

Panels
3
Connections
screws @ 8 11
Cor-.r.c ct ions
scre'«s @ every 33100

.163
l layer channels valley and @ 4"
2 6 x 6 26G S,C. 6 11x3/4'7:.. .1046 11 3 scr;;;.ws @ 8 11 scre-ws @ cw:.ry .is; 32~CO

1 layer channels valley and @ 4"
-'
3 6 x 6 26G S .c. 6 11x3/4 11x. l046 11 3 screws @ 8" screws @ every .056 11400 r->
1 layer channels 3rd valley and @8" ,
4 6 x. 6 28G S.C. 6°xl12 1'xe 1046" 4 screws@ so screws @ every .011 12000
1 layer channels 3rd valley and @S"
5 6 x 6 28G s.c. 6"x1;•tx.l046" 3 screws screws @ every .011 13000

1 layer channels @ 2-2/3" 3rd valley and @8"
6 6 x 6 28G s.c. 6"xl%''x. l046" 3 screws screws @ every .066 11200

l layer channels @ 2-2/3" 3rd valley and @8"
7 6 x 6 24G s.c. 6''xl% 0 x.1046" 3 screws s~rews @ every ,078 21100

1 layer channels @ 2-2/3" 3ro valley and @8"
8 5 x 5 . 28G s .c.
1 layer
6 1'xl':t"x. l046"
channols
3 screws
@ 2-2/3 11
screws @ ev\!ry
3rd valley and @8"
.068 llSOO
* Standard Corrugated
v \\ - \
v 1\-\
Table Z1 Continued
Test Dimensions Steel Edge No. of Scam ~eek - Edge G'

:\o. (ft) Decking ~·~crr:ba rs Panels Con."1actio::ls ,..
Conncctio~s '-' (lb/in)
9 5 x 5 28G s.c. 6 "xl.!;z"x.1046 11 3 screws screws @ every .068 11500

1 layer ch<in.."1e ls @ 2-2/3 11 3rd valley and @8"
10 6 x 6 26G S.C. 6''xl~"x.1046 11 3 screws screws @ every .098 39800

2 layers channels @ 8" valley and @ 4"
11 6 x 6 26G $.C. 6"x3/4"x.1046 11 3 screws screws @ every .ll4 46300

2 layers channels @ 8" valley and @ 4"
12 6 x 6 28G s.c. 6''xl!.z"x .1046 11 4 screws scre'W's @ every .056 18900

2 layers channels @ 8" 3rd valley and @8 11 \
13 6 x 6 28G s .c.
2 layers
6 11x1!.zr:x. l046"
channels
3 screws
@ 2-2/3"
screws @ every
3rd valley and @8"
.040 13500 -
r"'
t"J
14 6 x 6 28G s .c.
2 layers
l'' std weight
pipe
3 screws
@ 2-2/3"
screws @ every
3rd valley and @8 11
.045 15200 '
15 5 x 5 28G s.c. 6''xl1i"x.1046 11 3 screws screws @ every .oso 15400
2 layers channels @ 2-2/3" 3rd valley and @8"
16 1 x 1 2 mil corr. 3/16"0Dx.014"t

2 layers brass tubes
l - soldered@ every
valley and cont.
.030 1360
-t. \' -
V\\-2
T.:i.ulc }:xpc~d1;1,~nt.1l Rc.r;ults for S.1dcllc Shaped Hypars
Supported 1\l l Aroun<.l (q ra 40 psf)
(n) One Layer of Decking
Test in
w1:i.:i.x Stress nt Center in Strong Dir. Axinl Force in
No. inches BcndinB (psi) Axial (psi) Tie i3a r ( 1 bs)
811 l. it. 20f•OO -640 1820 (5660)"'*
Sl2 1.14 19100 -1700 1630
511 o.ao lf•70o'"
512 0.77 15900* .
-
311 0.33 2~;00 1260
312 o.32 3.:60 810 1710 (2120)
(b) 1wo Layers of Decking
Test w1,1ax in Stress at Center in StronP, Dir. Axial Force

l\o. inches Bcndinf2 <esi) A..-<ial <esi) in Tie nar
l3ot. Layer TC?P Layer Bot. Lnxer Top Layer (lbs)
321 0.86 12!.oo 10000 -290 1810 1560
822 0.65 13000 13000 -820 740 1310
S21I 0,65 10300 7600 -680 1750 1250
521 o.59 11300 7700 1660 650

522 o.53 11000 7800 1380 360
522I o.48 10000 5200 80 2060
321 0,24 4000 ~500 570 920 1720

322 0,32 5560 2310 -560 1220 1570
322I 0.21 3990 640 -310 1670 1570
* Based on one strain gage at extreme fiber, axial stress

assumed to be zero.
** Value in ( ) ls calculated from membrane theory.

- t\4 -
\J\l- 3
Tnblc Expnriracntnl Results for Saddle Shaped Hypars
Supported All Around (8 11 x 12" area loaded)
(a) One Layer of Dcckin3, Load ,,. 100 lb.
T1!!>t "'1:1;1x in Sti·css ;1t_Ccnt:cr in Stt·ong Dir. Axial Force in

No. inc~1cs ncndln,I?, (psi) Axial (psi) Tie Bar ( 1 bs )
BllC 0,63 ----- 17000 4300 --·--- 150
812C 0,67 17400 2300 130
512C 0,61
311C 0,39 13900 1990

312C 0,37 12400 2050 0
(b) Two Lnyers of Decking, Load a 200 lb.
Test "'1;i,1x in Stress at Center in Strong Dir. Axial Force

No, inches 3cndlng (psi) Axial (psi) in Tie Bar
Dot. Laye£ ToE..J:.~yer Bot. Layer Top I.ayer (lbs)
U21C
-o.so
-- 13400 20000 . -160 -1380 470
822C 0.42 11300 17000 -400 210 530
32 lIC o.39 13600 10900 -1650 -1300 390
S21C o.44 14000 16500 1060 -1090

522C o.41 15300 11300 460 360
522IC 0,32 13000 ..9300 860 -350
321C 0.24 15700 10100 1220 750 140

322C 0,28 10200 9900 10 -40 180
322IC 0.11 6800 7700 1060 -340 110

0
- t1tt -
,......
3 !/)
c::
0
.....
+-'
al
,....
::J
bO
•l"'I
'+.
s::
0
u
,....
al
p.
....
~
N
I
.-1
-'t.\j -
0
0
s
A
s
A I
J
y
y,Ey,Dy
Fig. 2-1 Typical Orthotropic Deck
a) Sinusoidal
Corrugated
-
b) Trapezoidal
I
J-- 1
... I I
t[J L ·1 1 l .l'\-1
~
1
--1'
c) Cellular
t
:1_ _L I
J- i
•l
-.r l
d) Stiffened
Panel
; 11 JI
,.. 1
11:
~I
I[ : e) Box··type
Fig. 2-2 Section S-S.Showing Different Types of Decks

-1..tl-
Effective Width
Varies
z
Fig. 2- 3 Effective Cross-Sectional Area
of a Hat for Axial Force
Max.
Stress
- ]Hean
..1-~1--~------~--------
Stress
e/Z
Effective Width
.I
Fig. 2·4 Effective Width of Compression
Flange in Bending
Principal Directions
Fig. 2-S An Arbitrarily Oriented Orthotropic Deck
----Lower
Deck
Edge ~embe~ ~~
---·--
Fig. 2-6 Edge Member and Decks Connection

0.10
8 The Author
A Ref. 19
0 .. 08
...0 A.
+'
u
Flat
"'>.
fJ,,
Shear
.,..+' Test
....
"CS G
A
N
'
,.,,
.,..
b(,
A.
°'•
~
... 0 .. 04
aS
u
.a=
(/)
..
. ti
s ..... c"
""
o.oz
c in inches
=3c
4 8 12 16 20 Z4 28
0
Fig. z.. 7 Shear Rigidity Factor r (l I Vs Hypar Curvature

(One Layer of Decking)
0.10
•
'\
\
- - - - T h e Author
0.08 Ref. 19
...0
+1
u
QS
µ..
- - Flat Shear Test

..
. tS
0.02
C in inches Jc
4 8 12 16 20 24 28
0
Fig. 2-8 Shear Rigidity Factor 'a' Vs Hypar Curvature
(Double Layer of Decking)
e
xl
r,
ox
1
'"'
ul Uz
wl exyl
exy2
ex ex
?\ "3
71
3
U4 U3
] ey
3
w3 6xy3
W4 axy4 V3
Fig. 3-1 Nodal Displacements
N My
Nxy . r Mxvr
0 2.LJ ___ 0
\ . x
x
1
-~
N
x ( )Mx
Mx
Nxy J
u~?Axy
. Nxy
~
y
My
Fig. 3-Z In-plane Forces Fig. 3 ... 3 Moments

!s-s
a) a = b = 80 inches
t = 0.05 inch
E = 29.Sxl0 6 lbs/inch 2
b/Z s-s 'V= 0.30
-·
s-s
,___. ___ IB - x
q • 1 psi
No. of Elements = 64
(Full Plate)
~ 0.00406qa ..
b/2 UB = D -
s-s
l- aL2 '
a/2
y
s-s b) a = 96 inches
b = 144 inches
t = S.O inch
E = 3x10 6 lbs/
F.E. s-s inch 2
v = 0"30
IB
--·-·...,..--------1---x q = 200 lbs/ ft 2
'No. of Elements
• 18 (Half
Plate)
:1 1>.0064ai.
0B . · D " ·
s-s
a. 2 a/2
I Y
;S-S
I. c) a= b = 70.50 inches
t "' 0.0149 inch
s-s 28G Std. Corrugated ·
b/Z ~
I ..
) - Deck · ·
E = Z9.Sxl0 6
lbs/inch 2
v .. o.. 1a .
·q • a• 3 o. p s i.
r{o .. of El~ments
. • Varies (Quadrant)
b/2 ....
o • .S~
• :~
B ·n-4 EI·y
I J ,)
a/2 a/2
1•
y
Fig. 3-4 Plate Bending Problems
Fig, 3-5 Deflection Profile Across Corrugations
0 (see Fig. 3-4C) for Uniformly Loaded
simply Supported 28G Standard
Corrugated Steel Deck
q = 0.30 psi
2 Ill
G>
.i::
u
s::
1-4
s::
•r-4
""r''
4 0s::
•r-4
+J ~
u
G>
r-i
~
G>
Deflection of
'
Q
6 Beam Strip
... 6.96 11
B
A
8 0 The Author
A Series Solution
0 1/6 1/3 1/2 2/3 5/6 1

p
e
C.G s.c.
z z
(a) (b) (c)
Fig. 3-6 Typical Cross-Sections of Beams
- - - - - - - - s. c.
y,v
-
x
x,u
z ,w
fig. 3-i An Arbitrary Cross-Section of a Beaw
wl, 8xyl
Fig. 3-8 Nodal Bending Displacements

of a Typical Beam Element
z,w
-
z
-
y
-
x
~y,v
zc
y
x,u
e-' (
Fig. 3-9 Typical Eccentric Edge Member
e Plate
-·--· . . .,__ ___

c. G. I C.G. s.c.
s.c.
(a) (b)
Fig. 3-10 Eccentric Connections

- ·z.:i.;-
0.25K/"
(a) Case I (b) Case Ila and IIb
Fig. 3-11 Loading on Eccentric Edge Members
'C - 4
.....~·
0
- ~~~-il~~-~~~--~~-
No. of Elements
Fig. 3-12 Convergence Characteristics for Vertical Deflection

'oQt for Case I (Fig. 3-lla)
(Ila)
"---. ~Warping Restrained (Ilb)

6.0
-·· - ,- -· - -- -
\_ 5.48"
::::S- LiiWWW
2 4 6
No. of Elements
Fig. 3-13a Effect of Restrained Warping on
Vertical Deflection 1 oq'
. 8. 0
6.0
(Ila)
1-4
0
1-4
1-4
'-1.l
4.0
2.0
0
No. of Elements
Fig. 3.13b Convergence Characteristics for
t:. 0 Q t and t eQt
-LLC\,.
x,u
·--"E-e_z_....,;__ z, w
Column
Fig. 3.14a A Cantilever Column
Fig. 3.14b Idealized Spring System

-'l..!.O -
A A
A =A =A =A =A
1 2 3 4
B1•B 2=B 3=B 4=B
C =C =c =C 4=C
1 2 3
- - - - =o
x •x =x =x
1 2 3 4
- - - -4=0
r1=r2=r3=r
(a)
A A
A =A =A =A =A
1 2 3 4
3 4
B •B •B =B =B
1 2 3 4
x c1 =c 3 =C
C2=-c 4•-C
x 1•x 4 •A
z
--- x' x 2:ax 3 =-A
/R(A,B,O)
y
y• Y1·r2·B
(b)
Y3-r4=-B
Fig. 3-15 Geometrical Definition of a Hypar Surface

- c...;.' -
Q
pt Qt
Q" x
S' R'
PO+RS
P'Q' • R'S' = 'z
P'S' ... Q'R' • PSiQR
y
(a) (b)
Fig. 3-16 Element Size
P_ _._ _ _Q.,.__...,...x 0 -
,_:::::::::::::-----..--::::--~.._.. x
---..,.... x '
x'
- z'z t
y y'
(a) (b)
Fig. 3-17 Co-ordinate Transformation

Fig. 3-18 Co-ordinate Transformation for
a Beam Element
z
-
x
Fig. 3-19 Boundary Conditions in Global Co-ordinates

Deck
Edge
Member
(a) Moment-Free (b) Sliding-Connection
Connection
Fig. 3-2) Deck and Edge Member Connections
{a) (b)
Fig. 3-21 Twisting of an E.ccentr.ic Edge Member for TF=O

1 I Half Band Width1 r
.....,.....,,_..,,
' !'.. '' )(
....
'
' l"I ,., '·
' "... - ~
,.., '' ''
' ' .... ''
.... ' ,, ,, ... '' ' ' ....
' .... ' .... ....
'"' ....
' ... ... x .....
'
Ca) Stiffness MatriK [K] (b) Vertically (c) Stiffness Matrix

Stored Half- with Sparse entries
Band
Fig. 3-22 Solutions of Equations
0 , ... 1
0
x
z
·--· ·--· ;-r--~Y
z
-
x z
(a) M.:. 1
Pxl .. I }7 °~xl
p.x2 (1 l----Px2 EM
0
&'I 0
MoQ, M-oP + (P x2 -P xl)
i:b J ZS
- M- • 0
oQ
Ip xl l MOP ... M-oQ = (P xl -P x2) z,.
~x2
(c) MoQ ~ MoQ
Fig. 3-23 Computation of Forces in Eccentric Edge Members

-21.r-
A A
--i h
t
c
b
Fig. 4-1 Structure Type I
Fig. 4-2 Structure Type II

B/2
Deck
'C >
Fig. 4-3 Structure Type IV

Note: 1 & 2 Refer to the Locations of Rosettes
1 2 3 4 s 6 7 8
b
I I I I I I I a
A -
. '
B
-
c-
D - e
E-
F-
G -
H -
c - 0
Fig. 4-4 Quadrant of a Hypar

-z.~1-
g h 1.... z
1" (TYP
B Varies
p
1a.-1 A
-
PA
f
- ti
·_01
I
c
Strucs.
8t
'6' & t
B 1/ 4" Th·c
Sh·!ll
d a b
A A
1
Plan
Strucs.
'6' & tgt
Z-1/2" square---
column __)----'
Struc. '7'
Section P-P z
Fir.. 4-5 Structure Type III

a
v-· Symmetry
1.0
2.0
f2.53
A*Z.55
3.0
Scale _
1" = 10
2 cm
·lo -.... b
I Symmetry
I 1.0
0 x 10 2
2.0
r:2. 46
3.0
* Method ta'
** Method t b t- .
+ Ref. 17.
Fig. 4-6 · Deflection Profiles
(Structure I 1 I)
0.035
E
u
s::
\3349
·~
c
'O
=
0
Scale
·~
~
1" = 0.0025 cm
,'
u
G.> f')
r-4 0.030 w
~
11>
~
r-4
C1S
'
....E
0
z
0.02588
0.02530
0.025
2x2 4x4 6x6 8xB
Grid Size
Fig. 4-7 Convergence Characteristics for Deflection ' <5
0 ' (Structure t 1 t)
a
.. I
.
v-Sy:nmetry
4I9. 20"'
10
Scale
l"=Sxl0- 311
Symmetry
s
~9.18+
9. 28 ... 10
* Method 'a'
•~ Method 'b'
+ Ref. 20
Fig. 4-S Deflection Profiles
(Structure '2')
-z..q. ·-
d e
\
\.
\
.o "' ·~
' .
............. ·- /:rs.-·
~ .
2.41 Scale
1"=1"
f g
.o -·Ar·-- Experimental
e Analysis
Fig. 4-9 Deflection Profiles

(Structure '5')
_ , 'l-1..-
9.32
7.50
b ~-~------'-----i---'--__.a
9.83
7.50
Scale
1"=7.50 ksi
1/2 The
9.32
7.50
9.83
7.50
a o
Fig. 4-10 Axial Stresses in Edge Members

(Structure '5')
100 jt\,
\
'\_Rosette No. 1
Max. Reading
.
..... Gage Broke When
en Local Buckling
p.. 80 of Plate
....
~ 7150 Occurred
"O
~
0
...4
"O
G>
60
I/I
0
p.. \
........
e; N
~
Q)
p..
""
::l
[/)
40
'
e Ana~ysis
-·-·- Experimental
2000 4000 6000 8000 10,000 12.000 14,000

o'
Shear Stress in psi Along Plate to Plate Welding
Fig. 4-11 Load vs Shear Stress (Structure 'S')
1-be-----.------ c
c--·-----------=i
z-L: ::J
v-------
3 - -~
4_L -- :sJ Scale
,~----------;::--- - - --- -·] 1"=1000

lbs/"
s-~- ~
[----:-:-- ----:------1
6_L ~
7- L:----~--_:_:::---1
____ L~::_::~ Mem.brane Shear
8-a o
Fig. 4-12 Shear Force Nxy lbs/inch
_(Structure 'S ')
-l-4r,
0.016
0.02 0.018
a b
o.o
--c-· Q..-• - 9 ·--..;:) 0. 0038
_ _.c _ _ . ca-. _...Q.
.!..
----.-----~:-=-= 1 e . s . Q- · lt: · ~ . o• a2 z
• ~ 0.023
Scale
1"•0.02"
. --
0• or~~-=-~·A::=:::;=::·=4s: - -
·e-- ·--eJ-. - . '""""""
:--s-.:.. ~
~ .-- ·t3--- •--El

b
I,...
0 .. 02
'·"
- ·-A- Experimental
~·-&·~ Method 'a'
o Method 'b'

(Structure '6')
324
300
200
c::::
.....
100
a b
Tension Member
Sea le l" .., 100 lbs
324 .. 0
300
Mern.brane
Theory
~·-4r·~ Experime~tal
200 ---A-e-- Analysis
'
100
''
''
0
--~~ a
Compression Member
Fig. 4-14 Axial Forces in Edge Members

(Structure '6 ')
-·-6·- Experimental
0
c
Analysis
80
Axial Stress
A.
. aJ
~
Cl) 60 ~
~
p..
s::
•rl
Otj
~
'
_,)
0
~
40 MtJmbrane
Theory '
20
4 6 s9
Stress in psi at Point 'e'
Fig. 4-15 Load vs Stress at Point 'e' (Strucutre '6')

Scale
l"=ZS lbs/"
Membrane Shear
,. 13. 48 lbs/" ~
c------~--
8-------~~~~----~~--~---
a o

(Structure '6')
0 a
. --· -6-. - - · - - . --- 0.009

..............0.015
0.025
·-- ........~ ............

o.ozs ............ ~
°"'· .............
A
o.oso ""'· 0.045

"-...
~ 0.056
Scale
l"=0.025"
0.025
o.oso
-·-A-:-·- Experimental
e Analysis
Fig. 4•17 Deflection Profiles
(Structure '7'1
324
300
' ' ,,
en
,Cl
' '' Membrane Theory
' ',/
~
200
....s:: 160
4)
u
$-.
'
''
0
u..
100
+ ''
''
a, Tension Member b
j
Scale
1"=100 lbs.
300
''
en
. '''
,Cl
''
'' , f
.....
....s:: 200 Mcmbrane
Theory
~
u
...
'
0
u..
100
0
Compressio11. Member a

(Structure '7 ')
-
-zs1-
Scale
1"=25
lbs/l'
Membrane
_____ .L:~48lbn"_. __ _
Shear
a o

(Structure 1 7 ')
-~ S-1. -
0.10
4
0.08 --~.
0. 0 s G- ""'--...A.._
dr1~-+~~+-~-+-=~~.:-.~~~~-4-~--+-~--1V
+
~. "'·~ 0.068
0.10 "--£0.098
0
0.] 0
+
~.~
~ ~0.052
" .. o.o9o
0 • .10
~
- · 6-.·- Experimental
Analysis

(Structure '8')
50
40 - · -&-- . - Experimental
Anaiysis
....
0.10 0.20 0.30 0.40 o.so 0.60 0.70
Deflections in inches
Fig. 4-21 Load vs Deflections (Structure '8')
Grid Size
4x4 6x6 8x8
1.60
II)
Cl> 1.574
.£::
u
s::
•"'4
s:: 1. so
·~
.tl
co
c::
0
..,
•"'4
u
Q)
..... 1. 372
~
Cl>
Q
1. 30
Fig. 4·2Z Convergence Characteristics for Corner

D~flection ·~b· (Structure '9')
0
,....
a
o.so
cS ,
a-------
0. 3 8 <1---A-
O. S1 Scale
0.56 1 11 :110.SO"
1.0
1. so
1.57
0 r"b
- ·/Jr.. - Experimental
e Method 'b'
o.s - -c- -- Method 'a'
1.0
1.5
Fig. 4·23 Deflection Profiles
(Struc:. 1
91) ~
--z.$'t..-
0
- , 1• a
c5
o.s
Scale
l"•O.SO"
1.0
1.S
- ·~·- Experimental
- - - o - - Structure '9 t
o.s ---o--- Structure '10 1
1.0
1.50
'.Fig .. t·24 Deflection Profiles
(St rues. '9 ~ & '10)
-2.'SI-
Axial
Stress
. -
S.35 1 tt,. S ks i
S.52
Bending
Stre.ss
-
1"•10 ksi
a
0
--.-- --·A
-G
1• a
Total ..
Stress
10.0 1"=10 ksi
--·--A·--- Experimental
Structure '9 t
20
---iC:I---- Structure '10'
23.70
24.60
a
Fig. 4-25 Stresses in Compression Member
(Structures • 9 t and '10 1 )
--i. s: l-
...-- .. ----- ·----- 1"=-5 ksi
a
a
Bending
10.0 Stress
l"slQ ksi
17 .14
a
a---- ------,--
:;;-o"
Total Stress
10.0
~- l"s.10 ksi
7.K -·-A· - Experimental
~·~ / Structure '9'
~ . _...I).__
19.90 ~ / - - .;i.- - · Structure '10'

zo.o .
20.9~
z3~sr
a
Fig. 4-26 Stresses in Tension Member
(Structures '9' and '10')
i-r------ :;ac
3-c----------~
4 __ C_ _______.A;g~ Scale
l"-10
--
6.75
lbs -inch/
inch
S- , , _ _ - - - - - - - - - - - · - - - - - -
,,,,----
/ 7.69
6 ___ /_~-·--------·~----~~
7------
S-a
~-+-------·--
- ~~o
Fig. 4-27 Bending Moments My (Structure '9')

r--------------1
1-[
b
= . c
r=--- ~- --..::--]
z_t:: ::::J
t:=_±C_-_-_-_--____-_-- -j
3_
Scale
4-E::~_:._ ______ ~
1"=100 lbs/"
s- ~-=----~---_--_=-_ _:_~-_]
6_E::_-_---_-_-
_ ~: =:J
7_E..;:::
Membrane Shear• 49.50 lbs/"
r--------------=i
s-C
a == ] o
. (Structure '9')
-'Z. C., I -
Deck
/
(a)
l ==- ar- - -------....;....;..=:....----·---+'

A1- - 4=----
'' . ·- -
-- -
..........
·--..:.-:-=....:..
.......... ._.__.
~------·------=.-...----~-~
0
A• Mt 2 ~--L--·-~
ZEI
(c)
Fig. 4-29 Effect of Eccentric Transfer of Force at

a Junction of Tension and Compression
Edge Members
-'Z.Go2...-
()
t5 -o 0. 38
0. 50
o.ss
0.56
...-~~~---~~~~--~~~~----~~~,~~-b
a
o..3&r---..,..
0.5 Scale
0.5
l"=O.SO''
1. 0
1. so
()
0 ... b
-·A-· - Experiinental
- - e - - B.C. V
0.50
--e--- B.C. VI
1. 0
1. so Fig. 4-30 Comparison of Deflectio~

Profiles fer Boundary
Condition, With (V) ar.d
Without (VI.) in-plane fixift'"
(Structure tg t)
10.0
Bending Stress
13.4 l.,==10 ksi
Bending Stress
Shear Loads
1"=25 ibs
zs.o - · -h:-· - · B.c. vr

_ __,e,.__- E.C. V
so.o
75.0
Shear Loads
Fig. 4-31 Effect of in-plane Fixity on Bending Stress and
Vertical Shearing Loads on the Tension Member
(Structure •gt)
-2.tA- -
0 a
=-
~~-,~~--~-1r~
-~Q...- .!.--~--a0.043
~0.133
0.20
c5
c
-~~g._-..oo.os3
o._.._............ ~ l iif"
------------.0.156
0.20
0
Scale
l"=0.20"
C----,-r-----,--.;....__-.------r ..b
--a.- ---&.-. -Q..- '"'€1
..................
~ .....
0.20 '19-.,.-a 0.24
0.29
b
I =-
Scale
1 "=0. 50"
0.50
Experimental
1.0 9 B.C. \'I
__...._ __ -
cS
----- B.C. V
F ig. 4-32 .
Deflection Profiles
'llt)
(St TUC tu re
0 I .,. b
0. so
"· 0.29
0.29
1.0 0.91 Scale

0 1"=0.50 11
Structure '11'
o.so
0.57
Structure '12'
0
I , ... b
0.50
0.56 - · - 6 · - Experimental
Analysis
Structure '13'

(Structures r11','12 1 ,'13')
a
667
+
c b
Scale
1"=667 psi
0
J. c
<1
Max. Membrane Stres~
• 1570 psi
667 --eld'---- Analysis
o a

(Structure '11 ")
-u;--
75.0 61. 5 -4--· e 9 --Q

e
+
Q
"' I
a b
150.0
-t-
0 _ _ _ _ _ _ _J _ _ _ _ __.. _ _ _ _ _ _ _ _ _ _ _..__.._
c
\_
300.0 Scale
l"=lSC lbs
150.0
~
+
0
-----~~-~l----~----
a
Fig. 4-35 Vertical Shear Force Di~grams

(Structure • 11 •)
-z.-, -
IA 1B I c I D IE IF IG IH I
[- - - -- - - - ---- - - -]
1-v:::~- ·~
b c
c---------------·---
2-k:::::. - ~
c-------·--------
3_v...-:~ ~
-J
c------------
4-~
---:.---1
~ -
Scale
1 11 :. 100
lbs
c---------------'1
s-v::= ~
6-V_;:::;=-=::--:------- ~
7- r:-:::::::-=-=-- - --- - - -=-:;J
c---
s-k::::·
---- -------J
~Membrane
~
Shear = 48 lbs/"
e o
Fig. 4-36 Sh~ar Force Nxy lbs/inch
(Structure '11')
0
------------------,.- a
.-.,.,.~. 4. __ Q • -~
~·-A- -=--eo.029
.......A.
-.... ~o. 046
o.os
0
a------.,-------------------b
0.05
0.092*
0.10 0.090
0.123
... b
0.05
--·-4·-- Method I aI
-- • Method 'b t
0.10
---#--- Ref. 21
* Ref. 19
0.123
0.144

(Structure '15 ')
-u.ro-
Section y =~
10-0
Scale
l" = 10
lbs/n
2 (}()
(a) Shear Force Nxy lbs/inch

~--
/ .,,. 2.05
"'2.41
.
. \
I
/I \
•
\
~
Scale
1" = 1.0
Section ~. lb• -incft'
a
.~.
x • 2
+
e ~fethod r a'
-·-·-Ref. 21
(b) Bending Moment My lbs~inch/inch
Fig. 4-38 Shear Force and Moments

(Structure '15')
for c =o
60 = 3.38" Single Deck
1.67
60 = 1.69 Double Deck
1. 33
.o
<C
---e-- Single De~k
-·-.!!-·- Do1ible Deck

1. 0
~·A
/
~~·..
o.__...
_,,,,...
0 r.. .,,,,,...,..
b '
..A '
rJ
-
.J
0 • 61 . ~.
~
c;_;.-.().o.:.----- -~ '
._,.,- ,,,,.,,,.,,,,,,.-
•..,...,
.~ ~ .
~..,,,,,. .... ~·
/ ,__,,,,,
/
0. 3 3 / .,,,,..,--
. ~-
~
.~./
·
er
AB
-.~· 10 20 30 40 so 60 70
2
Fig. 5-1 Central Deflection '\S ' vs AB/C (Structures' '3' and '4')
0
-2..11..-
~1
a
Dec
(a)
--- ~~=::::!;:;~~~--~..-!=:-=-=--~·==:l=l._
--·-·-·
_..,,.____....a...~~----·--·-· ~
0
..... -------- ---
(b)
Fig. 5-2 Effect of Eccentric Connections

Between Deck and Edge Member
0.72
Scale
J"::0.50"
0.4
0. 8111----e----
1.20
0
- e With Edge Member wt
-·l:r·- Without Edge Member wt

o.so
l. 0
l.50
Fig. S-3 Effect of Edge Member Weight

on Deflections (Structure '14')
-42..14 -
0.48
Sym.
0.81
Scale
t
! 1. ZS --.._
l"=O. SO"
t;·;-;·-·A--.-___..--.. . . . . -
Def l ec: t icn Along Tension Diagonal
a
., b
...
Scale
3 .. 2 7
l"=S ksi
5.06
Tension Member Bending Stresses

o a
.- __.£>:---4
~·
.k"·
~
S.O / -·-6:-·- Without Edge Member wt
6.7 With Edge Member wt
Compression Member Bending Stresses

Fig. S-4 Effect of Edge Member Weight on
Deflection and Bending Stresses
Line at·
Elastic
Analysis
1st Trial
op II
Value of De-
fc!'ma t ion at
the end of p I
First Step
Fig. 6-1 Load Incrementation ~ethod

t
a.so /
'
/
/
I \ Buckling
/ I \ Load
/ I I
0.40
~.l!_~~c;:m
N
·-#--· ·-· h j
& 6
u
.........
E 3
bl)
.:.:: 0.30
Reissner7
~
......
.,,
QS
f
1
0
c \
r-1
~
0 .....
"'
Point t 1I I Point ' 2 t Point ' 3 '
d a
Ke}'.: Sketch
•D
2.0 1.0 1.0

·------~---_._I_....,,..._ o in cm.
Fig. 6-2 Buckling of an Isotropic Hypar (Structure '1' Table IV-1)

0 --- • b
z.o
4.0
Scale
C~mpression Diagonal 'ob 1 1"=2 cm
a--~~--~·~--,.--~/"'--><--,-~--------------,·~ r"- c
I
/ '
I
/
/
,/
,,,,,,."""
2.0
- - - - -Buckled Scale
Tension ...
Dia~onal 'act l "• l cm
Fig. 6-4 Buckling cf an Isotropic Umbrella Shell

(Edge Member Sizes 6cmx3cm)
-, '>" b
kgm/cm
z
1.0
2.0
I~
Compression Diagonal 'ob~
Scale
1 11 :.:: lcm
----r---.....,..-----.-·---- c
0.?. 0
Scale
e 1tt=0 • Scm
Tension Diagonal tac'

Fig. 6-4 Buckling of an Isotropic Umbrella Shell
(Edge Member Areas ~ 108 crn 2 }
1!1
Z43
/
I \ Buckling
I I I Load
I I
202 I I
200
I
~
fA
• 150
I b a
p.
....
"'co'
~ 1
"O 2 0
~
0
...::I
3 '
i k'
Ill ..
~
0
Point 'l' Point '2' Point '3'
_ _2.i..-.0- · ~ 2.0 il 2. 0
_,.__ _ _ _,.....,,.._ ~ in inches
Fig. 6-6 Buckling of a ZSG Double Layer All-Supported Hypar (Structure '13')
0
~ • 23.psf e ___/ / ~ b
\ /
so\\ ,,--.......
\ I
/ ------
46 psf _____ /
1. 0
'" C~mpression Diagonal 'ob'
Scale
l"•O.SOtt
1.0
c5
Tension Diagonal 'ac'
Fig .. 6-7 Ruckling of a 24G Single Deck All-Supported

Hypar (Structure '12')
Note: Results With Flat Elements
. b
1.0
Compression Diagonal tobt
Scale
l"•0.50°
1. 0
Fig. 6-8 Buckling of a 24G Single Deck All-Supported

Hypar (Structure tlzt)
Note: Results with Curved Elements
t
60
~
. i
~~e--- Flat Element~
Cl)
i:l.
....=
't:S
45
I
.I
"
I Laa
Note:
Curved Elements
For Key Sketch

C1S
~
0 see Fig. 6-6
""w""'
'
Point 1
11 Point 'Z' Point '3 t
1.0 1.0 1.0

I
i-----__.._•_ _,__ . 6 i.n inches
fig. 6-9 Buckling rif a 24G Single Deck All-Supported Hypar (Structure '!2')
c----- --.....-----,----r---y----.......- b
o.2s
36 psf
o.so
0.75
Cvmpr.ession Diagonal rob'
6
--.-..., ---.- ......
0.25
Scale
o.so P'•O. ZS"
0.75
Fig. 6-10 Deflection Profiles Along Tension and

Compression Diago11als (Structure '12 '1
60
so
-~
.3 30 Scale
l"=0.10" Deflection
20 I l",.,10 psf Lo2d
II
0.10 0.20 'l.3J 0.40
.l
o in inches
Fig. 6-11 Load vs Free Corner Deflection I "

ob T
(Structure '12 ')

;-11r·t11 ,f/¥j1-~.4 • "?1JJ 1IA
?~'' ., //'9'4..U a "t1J} 1E
~6/' '(/> 1fl'7.r?r.d -~'(1IJ .A
T?r/ '/ ,(T~;J>L, ~ ""(1),) XI
61" ·t? ., (/'o/_z-2¥.u :: d' .m

Q•"T
.r11'1.r.:7r4 a "':t_ :rr
o·/ .//?If:r.;;.,.~
'
• "":t I:
>t (\I 0 I ..L bl n "4> ~ •t1.,.,

\-X\ .:t ,,.~
l1I I
I
I
•'
.' I
I
I
,
I
,•' I
I
''
f
I
,
I
..
•
f
I
I \
''. l \
di" 1
\
gurc 3IA F test
1-Tl
' I.
I "'
.. hs I 1:: :ii> , 1'.I·
'
1.~
Figure~ no.
- Z.'t'I
I
; - . -- ----·- -- -of shear . r. -----·------·---------··-~op
!.·\ 1-, +.t'\··- A·.·-· :/-~-

-- ,;· ;:\.-----~-,_-.-.:_\-~~~-::rx--:;;*·-n:--
- - --~ r.y---;;;F,~~~r
; 7 .\:' / -•\. ,J/· - ~ i/ '.
J · 12-···
I _;! \,: /
-'"······-····~---- .
"
3
___ ..... ..!...•. ----'="-·--t· '\ - •/
----"'··---~~il<l:ycr
.-. !-
F.d~o
- --- - ·-· ·------·- ·-----------·-
Coam
--------, -1
-I · -· .
I l
'\.··~··
'
I
j
·-1\'.
. ,---z·
,
I
1
\
'/
I
--
\
'
1-- - i-- -·-- · · ······ ···-·-···--· -3-------- ··-. -
•· "'
I
l)Tr:-Of siwa:r;.
,..----..r-~---~"-·-
--~- I \
\
J
)'
I
--
~ :rs----~.
\ I; \
...
.
·' .
I
I
,
I
•rop
' ~l.:1:1cr
'
' I
,'
I \
'
'A
I '
:'\,,.
. '
.~ \
'
·- ,
---•
~
\ ,
'
'r"
(layer
l) 0 t •
<
(1:) J.ocal Tension ~ tcnring
-1 ...... I
Effect of Position of Screws

MJOO ----------.---- -- --·J -- ··--
--3:>00
-3000
-2500
-<WOO ·o
....r::
rl
I
II
0
·-l
-l '.>00
(.1
F'i[jtlt'e ~ J.oacl versus Deflection
! Flat Shear Test Nn. 14
[<)()() !
- j i
~6 in inches
o. t 0.2 0.3 0.4 o.s
____ L·--·-----·-- _ L ____ _J_ - - -
-0.10 16900
'
.....1t;
l
I
~0.08
Figure 'li;l" Shear Rigidity G' a cGnt i
Flat She~r Tests Versus Hypar Curvature for I
No. 3., 9 Cne Layer of Decking
No.
i-o. 06
l
10140 Jl
! I
!
\, Hypnr Tes ts No. Hyp..-ir Tes ts No. I
C!. C' '--s-511,, 512 311., 312 l
'
0
J
i
!
!
t-0.04 6760 I
I
' i
I
I
i I
I
i
I
!
I'
3380
Tc
--1.
t-Oo02
! I
I
I C in inchas
3 12 16 20 24 28
I
I
-0.10 33300 J
! 7. c-,
Figure r;::;::2 She~r Ri~idity G' a r.Gnt
i
I
Versus Hyp.:ir Curvature for
Two Layers of Decking
I
1-o.oa
I
21040
P.yµnr Tests No.
S21, 822
a. G'
0.06
$hear Test
No. 15 No.
Hypar Tests No.
-----~3~21, 322 5r
r-0.02 6760
re
--L
l'
c in inches I
4
I
8
I
12
I
16
I
20
I
24
I
28
I l
--i.,s-
•2~ -------·----~~.?___ _ _ _• fi_6_- -
:t .ss
• 11 -.. ----~~-. +· 59· .
-·. + .22.
•l5 • 11 • O7 -• Ot. -• 01.
··-=.:..T:.:::.=t- t=. +
O
. . *-.
__ __.,......,·:- --- -·--:·--_-----~·04 - ·- ~ ·---
.15
l
.• l}
('!.
.
• 55 .66
••. •
.48
.15-f-
• Ol.-----. 26 · · ·
-t- ;- +·- .::.-.-:±..~.
-. 07 .59 • 66
+ +
• lt
.s4 .68 •ot. _·:. :. -- ___ •6_6 __

• 15 +_ + -- ----· .. ------------..+.....
• 77
+ •n: :-------~-- --~ --~-· -----~~-~ -~-
.18
.33 . ·- ..• 11 - - 5-~~;~<lLi-;.~-~
+ ... -
·-
--1--1 I
.22 .1s .22 .is !i 5 .11 .15 .18
figure :::xpcrir1t<?ntal Vert lcnl Deflections

in inches at 40 psf
Test l~o·. l .. 28G•Cl-Tt
-1060 (+1550)
+2860
-u,10+.
+i
-140 c=77of·.-. -~: __·.-
·.;4so· . +
+6200
300
+T. 5130
·~236o
.
--····- · ··- ·-·· -+10200 ···
-~---·::-~.: ·_:_-_++ 2290 ·-·~- ..
-·. - . +7690----
+2710 (1550)
·-· +4110
·i-120 ( ~ 1160)-~~-- -. -. . . +162§,1160)

;·2320 +302
+301.~. - ++2500_
-;·637U' . +4030
-2990 + -1501' (-770)
-l·t300 . - . .... . -230 +-
+ 7c,50 _ - . ·- - . -+7070
.. - --- - -- -
+360\520) ------- -- - - -- --·· -- -- ..
+3of - -- - - .- -
- -- strong dli;,.
-i-s'd)( s20) +93o( io?o) . ----.;lone( io3o)

--+-------f + l4oob sso) -
+3370
F 1s.;urc i:xpcriJ:1cntal Stresses in psi

nt 40 psf Loo<l for Test No. l-28G-Ct-Tl
Axinl stresses nrc directly ~hove bending.

v.,lucs in ( ) nre nxlal stresses from
uc1abrclnc theory.
.26 - ·122 _______2'.·~l~-----------------"t· 22
.13 .01 ·l_f

• .'35
+
.33
+
• 0 7 • 1 l • Ol1 • GL1
.l) --+--c: ~-1· ... - i·.. --1- cc- .04
-.01
.07 +
.22 ,29 .18 .04 .22
+ + + +
.• 11
·-17 ·¥
.15 • 7.9 ,110
·43 o_
+ +
, .13 .33 0
.22 + +
.26
·~
·+9 .l10 .o4
f +
• 18 0
+ .15
--··--- -t-----------1'.·--·-----··------·
.22 •oi. •04 .04
!·:xperincntal Vertical Dcfl!.!ctions

in inchos at 40 psf
Tost No. 2-24G-Cl-Tl
·:·530(1550)
+3
--- - ·-- -- ... "-----------~----·-·'-------·-·------·----··------
(-770) -1030
++.3660
+1270 -1110
·rl1 lJ '.)(1 ·-t- +5080 +
·- --- --~_7)_2~;_7-~~~~-:.-:::::.-,.}_1?:~~IS5 0)
.. _ -·
J
-------~---
- +12t.
-10'/0, (-1160) .;. l 1 00 ( 116 0)

+~ao~ 1010) l 100
-i· .;. 230
+ L 770
++38600 .., t:
·h>)
-8SO
++2seo
-30 - l 03 ( ( - 7 7 0) + +1 tSO
.;.!;.000 t ;· 1 sc +5620
-10.;0\520)
-:-1n -)!~O (-300)
+ 70 i
+9flfl()'.~O)
strong dir.
H>30(1030) j +700( 1030)

·I - · · · ··· I·-· - -·---+----·-
Figure I::xr;cri:iental Stresses in ps l

at 40 psf Lond for Test ~:o. 2-2l1G-Cl-Tl
Axinl stre:.s0s :ire <liructly above hondin:~ strc-ssc.?s.

V,1lucs in ( ) LlrC .:1Xinl str~SSCS f1·om r.1e;1lirnnc theory.
-'LC\q-
• 55 .1~0 • 6 .37
--- --·- -·t· ---·-;--7·-.
• 37 -.01
.33 .40
+:.
t
.29 :
+ .
• ll .n • 1l .ot~ 0 .11
+· + + +
• ta .29 .04 15
.l'... + +
• ta
·~ >¥·
.Ol1
.lH
.2G
+
.11
' '
'f3
.15 .26 .22 -.04
+ + +
.22 : 1I .01
i 0 I . . ; ;. .
.
.... . ·l -·-t ---·-I· + --·-·-!-'---:__,. ·-.----~~-----·-----~·-~-----t
• l v') 0 .15 .01 .01 .01 .04 .15
1.\2(°')
~xperiraontnlVertical Deflections
in inches nt 40 psf
Test No. 3-23G-c2-Tl
- 'SOo -
Axial stresses are directly above bendinc stresses.

V;1lucs in ( ) aro n:<lal stresses from r.1C1nbrnnc theory.
+930(1550)
+5260
i.
·I· -112oll(-770)
.' +3960
+290011 ' ++t0150 I;
.+--1120 ' '

+l1370
I I
; !
i
-~!t? C-3~~-?- ~s~o~-~!_?_Q)] __ _..

- 1--~io(-/idf::..
1
r--..- - - ··--·~---.-~
- 100 -.~IQ) _____ .. :i:J.2~::q-< t sso)
.;:,_ -- ---·--·-+;;cv\
- l'
' '
- 5 6 0 ,'( - 11G0) ' +11 l<J(ll60)

+2020 ~
·:·:>»o ( l0'.30) I
'·•·212-
+2/.()() ·1'2300 ' ' -!-3330
+ +790 . +2970
·•·2090 ' -400-~-770) +
.;.)Ol>CJ + +l07
+4560
-1·61)10
+'.')!1( ')~'.())
+<}!)(.
I
I
l.\l(~)
Fi;·,urc :·>\.porh1cntol Strcsse>s in psi
nt 40 psf Load for Test No. J-2~G-C2-Tl
_.,.. f -
.300
--------.135
.026
+
. ~
. ~
···"-~·----·
. -·-
.. ----...........----. ,. ..... --·---·-··--
~~ ,
~--- -- ...·---
' '
- - .020
.029 .015
.063
+ + .032
+· It- .057 +
· .os2
• 315
:·~xpcrhKmtnlVertical Deflections
in inch0s at 40 psf
Test No. 4-2N-C2-T2
1.73 1.01 .48 .33 1.56
--·- -----1----,,...._..---,1.---i-_-__-.-.:-..-.-i------..
.92 1.25
t .J
'
).
.14 .....f2
: .- -.
__ ~ "
1.01
.. ~.
••. • • •
i - . 1
!
'J
' I . I
.04: . : ... .76

' . II- -: . I . : : ' i
,'Ji .1J; :.~o
. ------1-
":-~----~1~--.-
-
'.; .: .'01 · ·
__, _ _
"----~~=--~~~~
: ...~_c~;. ~26;
-=--=====::..l:f::::::=::::::::::-.'
=- ~ 1---
.53
-·-===:::1-
.60
' '
.05.: .79
I;
1
i ' ' t
'
I
: ' ' '

1.01. ~/18 .13 : I .64 . 1.06
+ .
: ,
'
.I ! • +~.
.I !
'
.32
''
'
. ·.36
I'
I:
~33 t.46
' ' ! •
I
+. ' I
I t:
'
----~--~-1---'--'--------'l'------''-l1~--'---ll,_._ __ . 1 . - - - - 1
1,83 .99 .59 .67 1.00 1.39 1.84
1. \4-(c:\j
r;xperi:i1enta 1 Vertical Deflections
in inchas at t~o psf
tast No. 5-2SG·C2-T3
,\:~i.11 :;tt·,~:;:;'-·:; ar•: directly above bcndinp, stresses.
;),•ci-; :;ti-c·:;:~l':; in h<Jtto!'I L1ycr <1rc to the left of those
in lop 1;iy,~r.
+5J20 +6450 +3550

+lliOGO +l(,000 +5970
--- -----I ·I-~--------!------
..;. :~ 1 0 -i-1280 -i-l 7SO

-;- ri :~ o I- .;. 516+ ... 2060
- ·-~----
-- -
. -'
.
-;-111~0 +570
+201.. 0+·1·2270
, I
I ,
'-·--~---- --~--·-
;·.'1C''1<) ;-(,1)()() +·~t,)O

,.rJ4 70 -;-;;:;;'.Q -.-11:wo
I . \ if 'c,» .
i~i:~ut·c · 1 .J::x;x•ri:.\~ntnl Str.e:.scs
in psi <1t t+O psf
Trst No. 5-2SG-C2-T3
PROF. IR. c. G. J. VREEDENBURGH
THE SHELL WITH DOUBLE CURVATURE

CONSIDERED AS A PLATE ON AN
ELASTIC FOUNDATION
U.D.C. 624.074.4
From the differential equation of the doubly curved shell in terms of the
displacement components u, v and w it is apparent that, in so far as the
.flexural phenomena are concerned, the shell can be considered as a .fiat plate
on an elastic foundation, at a~y rate in a region - of sufficiently small size
to be regarded as quasi-Euclidean - around the vertex of the osculating
paraboloid. The modulus of the elastic foundation is dependent on the shell
thickness and the principal curvatures of the middle surface. The actual load
on the plate must, however, be increased by a deformation load depending on
the displacements u and u in the middle surface.
Various aspects of this approach to the problem are verified by reference to
simple examples and to the results of tests on a large model of an equilateral
hyperbolic paraboloid shell.
0 Introduction
It is a well-known fact that a curved shell can resist loading acting perpen-
dicularly to the middle surface by the agency of membrane forces. In those
cases where the membrane reactions can be resisted and the corresponding
deformations of the shell can freely take place, a force distribution pattern
constituted solely by membrane forces is a good approximation of the trans-
mission of forces that actually occurs - for a statically possible stress distri-
bution in which bending and torsion are avoided will come close to producing
the minimum strain energy in the structure. The usual procedure, therefore,
is to begin by calculating the distribution of forces in the shell according to
the membrane theory. Then a set of correcting forces will have to be applied
so as to take the best possible account of the boundaries (edge beams or sup-
ports) of the shell, where the deformations due to the said state of membrane
stress are prevented from freely developing. Let us suppose the edge member to
be detached from the shell, thus enabling these deformations to develop freely:
the edge of the shell and the edge member would then no longer fit at their
junction. A proper fit can be obtained only if the edge members exerts forces
(normal and shear forces) and moments (bending and torsion moments) on
the edge of the shell and if, conversely, the shell exerts opposite forces and mo-
ments on the edge member in such a manner that the additional deformations
associated with these forces and moments enable complete adaptation (struc-
tural fit) to be achieved. The calculation of these edge disturbances, which
must be superimposed upon the state of membrane stress, is done with the aid
of the so-called flexural theory of shells.
2
Simplified flexural theories are already available for shells shaped according
to simple mathematical surfaces such as the spherical and the cylindrical shell.
A. L. BouMA has applied a theory of this kind to elliptic and hyperbolic para-
boloid shells bounded by four principal parabolas.I)
Even these simplified theories undeniably involve a fairly considerable
amount of arithmetical computation; and for doubly curved shells of arbitrary
shape or combinations thereof (which may or may not be bounded by elastic-
ally deflecting and/or elastically rotating edge beams), such as are rather
frequently employed in modern architecture, there is as yet no serviceable
theory for the analysis of flexural disturbances.
The distribution of the edge disturbance moment from the edges towards
the interior of the shell surface is known in general to have the character of a
damped wave. An obvious question is whether it would not fundamentally
be possible to calculate the approximate magnitude of this edge disturbance
moment by means of the theory of the elastically supported beam - for in a
beam of that kind the edge disturbances display a damped-wave distribution.
Thus J. W. GECKELER 2 ), as long ago as 1926, gave an approximate analysis
for the edge disturbance in a spherical shell which leads to the theory of the
elastically supported beam. Furthermore, in 1953 K. HRUBAN 3 ) analysed the
edge disturbances along the curved edges of a number of hyperbolic para-
boloid north-light shells with the theory of the elastically supported beam,
while in 1958 W. S. WLASsow 4 ), with reference to the spherical shell and the
equilateral hyperbolic paraboloid shell, called attention to the analogy with
the elastically supported plate.
Finally, for the estimation of certain edge disturbance moments associated
with the structural research on the Philips Pavilion at the Brussels World
Exhibition in 1958, the present author also proposed a method of analysis
based on the theory of the elastically supported plate.5)
If permissible, the method of analysis based on the elastically supported
plate or beam undoubtedly constitutes a considerable simplification. This is
because the differential equation of eighth order on which the flexural theory
of shells is based is replaced by an equation of fourth order, which provides a
simpler means of gaining an insight into the anticipated edge disturbances,
this being of great importance to the designer.
1
) BOUMA, A. L. Some applications of the bending theory regarding doubly curved shells.
Proc. Symposium on Theory of Thin Elastic Shells (1.U.T.A.M.), North-Holland Pub!. Co.,
Amsterdam. 1960.
2
) GECKELER,j. W. Uber die Festigkeit achsensymmetrischer Schalen. Forschungsarbeiten
Ing. wesen, No. 276, Berlin, 1926.
3
) HRUBAN, K. The general theory of saddle-shaped shells (Czech), Techn. Univ. Brno, 1953.
') WLASSOW, W. S. Allgemeine Schalentheorie und ihre Anwendung in der Technik, Aka-
demie-Verlag, Berlin, 1958, pp. 330 and 372.
5
) VREEDENBURGH, C. G. J. The hyperbolic-paraboloidal shell and its mechanical proper-
ties, Philips Technical Review, Vol. 20, No. 1, 1958/1959.
3
On the other hand, when applying an approximate method of analysis it is
essential to have a clear conception of the quantities that are being neglected
and of the assumptions made, in order to be able to judge whether useful
results are likely to be obtained in any particular case.
1 The doubly curved shell and the elastically supported plate

Consider a point 0 in the middle surface of a doubly curved shell ofarbitrary
shape. This point is the origin of a rectangular co-ordinate system XYZ. Let
the plane OXY coincide with the plane tangential to the middle surface at
0 and let the axis OZ be directed along the normal (Fig. 1).
In the immediate vicinity of 0 the shape of the middle surface can, as we
know, be replaced by a paraboloid, the so-called osculating paraboloid, which
has the following equation:
x2 2xy y2
2z = - +- + -ry ................... (1)
or
2z = kxx 2 +2kxyxy+kyy2 • • • • • • • • • • • • • • • • • (2)
In these equations rx, ry and rxy respectively denote the radii of curvature
of the middle surface at 0 in the x and y directions and the radius of torsion for
these directions, while kx, ky and kxy respectively denote the curvatures in the
x and y directions and the warp for these directions.
Fig. 1. OZ = normal at point 0 of the middle sur-

face; Sx and Sy = normal sections; r x and rY = radii
of curvature of Sx and Sy at 0 (positive if the centre
of curvature has a positive co-ordinate z); u, v and w S,
= displacement components (positive if the displace-
ments take place in the positive directions of the co-
ordinate axes).
From (2) we obtain by differentiation:
kx = a2z l
ax 2
a2z
kxy = -- . . . . . . . . . . . . . . . . . . . . . (3)
axay
a2z
ky = ay2 J
The equation of equilibrium for a shell element at 0 in the z-direction is:

a2z a2z a2z
nx -
ax2
+ 2nxy -axay
- + ny -
ay2
+ Pz = Kf:,.f:,.w . . . . . . . . . (4)
4
For P 0 we furthermore have:
au w nx
Ex
ax rx Eo
av w ny
(5)
Ey
ay Ty Eo
au av azz nxy 2nxy
Yxy -+-
ay ax 2w--
axay Go Eo
or:
Eo, lau w azz]

l
nx -
ax ax2
ny Eo lav
- w a2z] (6)
ay ay2 f
nxy = Eo l~ (au+ av)-w ~ J

2 ay ax axay
Notation
x,y and z co-ordinates of a point in the middle surface

u, v and w displacement components of a point in the middle surface in the directions
x,y and z respectively
rx' rY and rxy radii of curvature of the middle surface in the directions x andy respectively,
and the radius of torsion for these directions
1 az 2
1 az
2
1 az2
- - k - -2
rx - x - ax ' -
ry
= ky = -ay- Q and -
r xy
= kxy = -aa:
x ry
curvatures of the middle surface in the directions x and y respectively, and
the warp for these directions
principal rad.ii of curvature of the middle surface
1 1
k, = -, k, = - principal curvatures of the middle surface
T1 T2
nx, ny and nxy shell normal forces in the directions x and y respectively, and the shell shear
force
m shell moment
shell shear force (transverse shear)
strains in the directions x andy respectively, and angular deformation due to
shear (deviation from right angle)
E,G elastic modulus and shear modulus
v Poisson's ratio
('j shell thickness
£(j3
K = ( -v') flexural rigidity of the shell
12 1
modulus of the elastic foundation
characteristic length
load per unit area of the middle surface in the direction of the normal
deformation load
a, a,
-
ax2+ay- 2
5
By substituting the relations (6) into (4) and having regard to (3), we obtain:
Kjj,_jj,_w = Pz+Eo [kxau+

ax kxy (au+ av) + ky~~1
ay ax
2 2
ay - Eo(kx +2kxi+ky )w (7)
Essentially this equation is valid for the shell in the immediate vicinity of 0,
but it is approximately valid also for points of the shell situated not too far
from 0. For practical purposes we may consider this region to be bounded
by a circle having 0 as its centre and having a radius equal to half the smallest
principal radius of curvature at 0. Then the ratio between the rise and the
chord of all the normal sections of the osculating paraboloid will be smaller
than 1 /7 so that the corresponding region of the shell can, as a rule, be assumed
to be quasi-Euclidean.
Putting:
Eo [kx au + kxy (au + av) + ky av] = Pz (8)

ax ay ax ay_
. . . . . (9)
we can now write equation (7) as follows:

(10)
The latter equation is seen to be the differential equation of a flat plate,

having a stiffness K, supported on an elastic foundation with a modulus of
reaction (or modulus of foundation) c, this plate being subjected to a real
load Pz and an additional load Pz which is dependent on the differential quo-
tients of the displacement components u and v (the displacements in the middle
surface of the shell).
This last-mentioned load will be called the "deformation load". The values
of Pz and c are invariant, i.e., they are independent of rotation of the co-ordi-
nate system XYZ about the origin 0.
Choosing the axes OX and OY so as to coincide with the principal directions
of curvature of the middle surface at 0, we obtain with kx = k1, ky = k2 and
kxy = 0:
Pz = Eo [k1 au + k2 av] ( 11)
ax ay
c = Eo [k12+k22] ••• ( 12)
If the factor v is also taken into account, the relations ( 11) and ( 12) become:
( 13)
( 14)
6
while the plate stiffness in (10) becomes:
1 Eo 3
K=--- ( 15)
12 l-v 2
Summarising, it is apparent from the foregoing that the displacements w in

the direction of a normal at a point of a quasi-Euclidean portion of an arbi-
trary shell are the same as those of a flat plate on an elastic foundation, the
edge support conditions of the plate being the same as those of the shell. The
modulus of the foundation is determined by the local principal radii of cur-
vature and the local thickness of the plate according to formula (12) or (14).
Besides the normal load Pz it is, however, necessary also to apply an additional
load fJz (the so-called deformation load) in the direction of the normal, accord-
ing to formula ( 11) or ( 13).
In contrast with the load Pz, which is known, as given quantity, the de-
formation load is, generally speaking, unknown. Often, however, if the latter
is not neglected, it can be approximately determined; some examples of this
will be given in due course. In the design of a shell it is, of course, advisable to
endeavour to keep the displacements u and v as small as possible by employing
sufficiently rigid edge members of the shell - so that, in particular, no exten-
sionless deformations 1 ) will occur - or by the use of appropriately chosen
pres tresses.
In some cases these displacements are even zero, e.g., in a spherical shell or
a cylindrical shell subjected to a constant normal load and supported in the
manner indicated in Fig. 2, so that no edge
disturbances occur. Another example of a
case where the deformation load is zero is
afforded by the cone of revolution subjected
Fig. 2. Spherical shell or circular cylindrical shell

subjected to uniformly distributed radial load,
without edge disturbances.
to axially symmetrical loading when 11 = 0. If the X-axis is made to coincide
with a generating line of the surface, then obviously k1 = 0 and - = 0, so

ov
ay
that Pz = 0 according to (13).
The application of the plate analogy becomes particularly attractive in a
case where the edge disturbance moments are confined to four strips along the
1) For the significance of"extensionless deformations" see: BouMA, A. L., Stijfheid en sterkte
van schalen, Waltman, Delft, 1960.
7
Fig. 3. Hyperbolic paraboloid shell, bounded by
four generating lines and subjected to uniformly
distributed load, considered as an elastically sup-
ported plate. The edge members are very stiff
beams. The shaded area is the edge disturbance
zone. In the region surrounded by this zone the
deflection w of the plate is constant and the bend-
ing moment is therefore zero in all directions. The
edge disturbance zone has a maximum width
of H. B
~!!11, ID P .11Llll~
~ '-,y~~"-''-"''"'·''-Y//"'-'-.Y/O''-'y'/ ,
A-B
edges, as indicated in Fig. 3 for a rectangular hyperbolic paraboloid shell

bounded by generating lines.
Consider a "beam" cut from the plate perpendicularly to the edge and
passing through the zone free from disturbances. Equation ( 10) can then be
simplified to:
d4w _
K~ = Pz+Pz-CW (16)
dx 4
if the X-axis is chosen in the direction of the axis of the beam.
Having regard to (12), we may put:
c 4 Eo(k12+k 22)
- -
K Jc4 1/12£03
Hence:
152
,14 = - - - - - ( 17)
3(k12+k22)
o,76 Vo
( 18)
The quantity A, which is known from the theory of the beam on an elastic
foundation and has the dimension of a length, is sometimes termed the char-
acteristic length. It can often be assumed that, depending on the mode of
support, the edge disturbances will, for practical purposes, no longer be per-
ceptible beyond a strip having a width ranging from A to 4A along the edges of
the shell.
8
2 The foundation modulus c
Consider a circular cylindrical shell rectangular on plan, which is subjected
to a uniformly distributed radial load p acting over its entire surface and
which is so supported at the edges as to be free from disturbances.
Let the radius of the circle be r (Fig. 2). Since, as already stated, in this case
the displacements u and v are both everywhere zero, the deformation load is
zero. Furthermore, the displacement w is constant, and - having regard to
( 10) - we therefore find:
p
W=- (19)
C
1
Introducing ki = - and k2 = 0 we then obtain, according to ( 14) :
r
Eo
c = ------
(l-v2)r2
so that finally:
( l -v2 )pr2
W= . . . . . . . . . . . . . . . . . . . . (20)
Eo
In this case the displacement can also be calculated directly, as follows. The
compressive (tangential) "hoop" stress in the shell is:
pr
(J =- (21)
0
As the lateral contraction is, in the present case, prevented in the longitudinal
direction of the shell, the compressive stress in the direction of the generating
line will be:
(J ' = ')l(J • • • • • • • • • • • • • • • • • • (22)
The specific compressive strain along the circle is therefore:
(l -v 2 )a (l-v 2 )pr
E=--- (23)
£ Eo
Since this strain is equal to ~ (inasmuch as u = v = 0), we arrive at the result

embodied in (20). r
Now let us consider a spherical shell, with a circular base, subjected to a
uniformly distributed radial load p acting upon its entire surface. Let r be the
radius of the sphere and let the shell be again so supported at the edge as to
be free from disturbances.
In this case too all the points on the sphere will undergo only (constant)
radial displacements w, so that u and v, and therefore also the deformation
load, will be zero at all points.
9
The displacement w is obviously again expressed by the formula (19), but
the foundation modulus c for the sphere is different from that for the cylindrical
shell. 1
Introducing k1 = k2 = - we now have, according to ( 14) :
r
Ea [-+-+-
1 2v 1] 2Ea
c=~-
l-v2 r2 r2 r2 (l-v)r 2
so that:
(l-v)pr 2
W=--- (24)
2Ea
By direct calculation we obtain:
~ w(1-0pr
a = - and s = = -
2a r' 2Ea
i.e., w is found to have the same value as that given by (24).
These simple calculations may be regarded as providing a check on the
correctness of the formula (14) for the foundation modulus c if the shell is con-
sidered as a plate on an elastic foundation and the deformation load is zero.
Since - as will be apparent in due course - both the actual load and the foun-
dation modulus of the analogous plate are modified as a result of taking the
deformation load into account, the foundation modulus according to (12) or
( 14) (which therefore relates to the case where the deformation load is zero)
will be called the "primary" foundation modulus. The modulus modified by
the deformation load will be called the "secondary" foundation modulus.
3 The deformation load

Consider a circular cylindrical shell having an elongated rectangular shape
on plan and subjected to a uniformly distributed radial load p acting upon its
entire surface. In this case, however, the shell is assumed to be completely
restrained at all its edges. Let the Y-axis be chosen in the longitudinal direction
of the shell, and the X-axis in the transverse direction (Fig. 4).
Evidently, at a sufficiently great distance from the transverse edge AB the
disturbance emanating from this edge and proceeding in the Y-direction will
ow
have been practically damped out, so that we can write here - = 0. Now
oy
consider a transverse strip of unit width cut from the shell in this region. For
1
the deformation load we then obtain, on introducing k1 = - and kz = 0 and
having regard to ( 11) : r
_ Ea ou
Pz =-·- (25)
r ox
10
C-D Fig. 4. Long circular cylindrical shell rectan-
gular on plan, completely restrained at all the
edges and subjected to uniformly distributed
radial load. Outside the edge disturbance zone
of the short edge (shaded area) a transverse
strip cut out of the shell no longer behaves as a
beam on an elastic foundation, but as an arch
fixed at both ends, in which the line of action
of the horizontal tensile force passes through
the elastic centre. This is caused by the defor-
mation load.
I
11
11
11
~_.::-_.::-_.::-~~~~r~~~~~~~ D
11
L_ _ _ _ _ _ _
and having regard to the first equation ( 5):
Pz = ~o [;~ + ~J . . . . . - (26)
As the effect of bending on nx is very slight in the present case, we can write
nx = -pr, so that equation (10) becomes:
Eow
Kl::i!::iw =P-P+--cw
r2
which, on introducing
Eo ow
c =-and - = 0
r2 oy
gives:
(J4w
-= 0 (27)
ox
4
In this case the loading of the analogous plate and the foundation modulus
have both become zero in consequence of the deformation load, and the dis-
placements w of the cut-out strip CD are governed only by an edge load at
the points of restraint C and D of an unloaded beam. This restraint reaction
is not difficult to determine. It is a horizontal tensile force H passing through
r sm rp
the centre zb of the arch, zb
being located at a distance - - - from the
rp
centre M.
For v = 0 the magnitude of H can be calculated from the relation:
12r3 ( 2 sin 2 rp) 2pr2 •
H - - rp+l/ 2 sin 2rp - - - - = E'~ sin rp . . . . . . (28)
Eoa rp u
11
Fig. 5. General view of test arrangement for a hyperbolic paraboloid shell in the Stevin
Laboratory. Overall dimensions of the model on plan 5 m X 5 m.
If His known, the displacement w 0 at the crown of the arch can be calculated.
By considering an elastically supported beam strip with its axis along OY we
then obtain - with }c = 0.76Vro according to (18) - the following expressions
for the moment and for the shear in the shell at 0 1 ) :
Eb 3 wo
mo=---- (29)
5,1c2
qo (30)
Mention should be made of an interesting case where it was possible to check

the effect of the deformation load against the actual behaviour, namely, the
research conducted on a large model of a hyperbolic paraboloid shell in the
Stevin Laboratory, Delft, by C. VAN DER SCHENK in 1958.2) The structure in
question was a system of four equilateral hyperbolic paraboloid shells, each of
1
) See also: Randstoringen bij axiaalsymmetrisch belaste omwentelingsschalen, IBC-Mede-
delingen, Vol. 6, No. 1, January 1958.
2
) VAN DER SCHENK, C. Onderzoek naar de spanningsverdeling en de sterkte van een hyppar-
schaal, Part II, Techn. Univ., Delft, 1958.
12
Fig. 6. Underside view of hyperbolic paraboloid shell with loading devices. Shell thickness
3 cm. Cross-sectional dimensions of hangers 10 cm X 10 cm, other edge members 15 cm X
15 cm.
-60
~
c c
-40
-20
80 100 1 I
140 1 0
+20
______.... x in cm
Fig. 7. Reinforced concrete hyperbolic paraboloid shell with all four bays carrying a uni-
formly distributed loadp = 1000 kg/m 2 ; A= 22.4 cm at B.
EF considered as a beam on an elastic foundation, completely restrained at E and elastically
restrained at F.
o measured shell moments
calculated without deformation load
calculated with deformation load equal top
13
which was square on plan and which were interconnected by stiff ridge
beams, while the inclined edges were also provided with stiff beams (see Figs.
5, 6 and 7).
Consider the case where the four bays carry a uniformly distributed load
p = 1000 kg/m 2 over the entire surface.
Having regard to the characteristic length (22.4 cm at the vertex), it was
anticipated that the case envisaged in Fig. 3 would occur. Hence the elastically
supported plate was locally replaced by an elastically supported beam. Con-
sider a beam strip of unit width cut out of the bay ABCD along the generating
line EF (Fig. 7). This beam can be regarded as completely restrained at the
ridge beam. On the other hand, it is elastically restrained at its junction with
the edge beam at F. Let x be the distance measured to the completely restrained
end of the strip. The distribution pattern of the moment in the elastically
supported beam for a uniformly distributed load p will then, as we know, be
given by:
mx = - 1 /2PA 2 e-i V2 cos (i + ~) ............ (31)
Neglecting the deformation load, we obtain withp = 0.1 kg/cm2 and).= 22.4
cm the following value for the restraint moment at E : m; = -25.1 kg. The
n
maximum positive moment occurs at x = A and is 5.2 kg.
2
In Fig. 7 the distribution curve of the moments calculated m this way
emanating from E is shown dashed.
The moments emanating from the elastically restrained support F can
similarly be calculated and are also shown dashed in Fig. 7. On comparing
the values calculated by neglecting the deformation load with the measured
actual moments (indicated by the small circles in Fig. 7) we see that there is
indeed good agreement in the pattern of the distribution, but that the calcu-
lated values are much too low. If, to allow for the effect of the deformation load,
we consider twice as large an actual load on the shell (load factor = 2), we
obtain the curve drawn as a full line in Fig. 7; this curve is in better agreement
with the measured values. As emerged from a further investigation of the
problem, the order of magnitude of the deformation load could, in the present
case, be explained with the aid of the measured strains of the edge members,
au av
from which it was possible to calculate average values for - and-.
ay ax
As regards the peak values of the moments at E and F it is to be noted that
these are usually not so serious as they may appear to be.
For one thing, acute angles are always more or less chamfered in actual
practice. Furthermore, ifthe limit of proportionality of the material is exceeded,
a stress peak will be levelled down, while a residual stress of opposite sign will
be produced on unloading. Also, if the load on the shell under present consid-
14
Fig. 8. Underside view of a
fractured bay, showing cracks
extending approximately par-
allel to its edges.
eration is gradually in-

creased, the first cracks
are likely to occur along
the edge members. With
further increased loading
these cracks will function
as linear hinges. The elas-
tically supported beam
EF (Fig. 7) will than have
hinged bearings at both
ends, with the result that
the maximum positive
moments will, as we know,
. n
occur, at distances - A.
4
from the ends. In the ul-
timate condition (i.e., at
failure) a second set of
yield- lines, parallel to the edges, will therefore have to develop at these points.
In the test this was actually found to be so (see Fig. 8).
We can utilise this knowledge for calculating by means of the yield-line
theory the ultimate load of the shell loaded over its entire surface.
For the sake of completeness it should, finally, be mentioned that the total
failure of the shell in question did not occur until a load of over 6000 kg/m2
was reached !
4 Consideration of a diametral beam strip of the circular

elastically supported plate
Let us again consider a spherical shell circular on plan and subjected to a

uniformly distributed load p. The edges are completely restrained, however.
If the rise/chord ratio of the meridian section is not too large we can, for cal-
culating the deformations w, replace the spherical shell by an elastically
supported flat circular plate with a radius equal to that of the base of the shell
(or, preferably, with a radius equal to the generating line of a cone tangential
to the spherical shell at the base circle), this plate being completely restrained
15
at its perimeter. Let r be the radius of the sphere and neglect Y. Then, according
to ( 12), the primary foundation modulus will be:
2Eo
C=-- .
,2 ............ . (32)
Supposing this plate to be subjected to a load p, the displacements w and the

shell moments can be found either analytically or experimentally.
We may ask ourselves whether it would, in this case too, not be sufficient
merely to consider a diametral beam strip (see Fig. 9).
Fig. 9. Spherical shell, circular on plan, con- A

sidered as an elastically supported plate. The '-'---11'3&7,G;=-==
edge is completely restrained. The shaded area
is the edge disturbance zone. AB is a beam strip
cut out of the plate. As part of the plate, how-
ever, the beam strip is subjected along its sides to
shear forces and torsion moments (acting in the
regions AC and BD) which have a relieving effect
on the free beam.
Neglecting Y, we obtain - according to formule (24) - for the displacement

of the spherical shell outside the zone of edge disturbance:
pr2
Wo = - - .
2Eo
The deformation equations for calculating the restraint reactions mo and q0
of the elastically supported beam of unit width will then be:
:;: ~~~ = 1. . . . . . . . . . . . . . . . . .
qo).3 - mo).2 Wo
(33)
2EI 2El
Hence:
2Elwo
mo= - - . . . . . . . . . . . . . . . . . . . . . (34)
). 2
4Elw0
qo (35)
16
With formula ( 17) we obtain for the sphere:
i5r
;.b2 = V3V2 (36)
and for the circular cylinder with radius r:

or
;. c2 -- V3 . . . . . . . . . . . (37)
Neglecting the deformation load, we must, for calculating the restraint mo-
ment mo, substitute Ab 2 according to (36) for ?. 2 in formula (34).
It should be noted, however, that by considering a detached diametral
beam strip too unfavourable a distribution of forces is found, because in the
edge disturbance zones AC and BD, as part of the complete circular plate,
shear forces and torsion moments are actually present which have a relieving
effect.I)
Consequently the calculated moments will be larger than those occurring
in reality. A further reduction will in this case be caused by the deformation
load.
On substituting for ?. 2 in formula (34) the value Ac 2 according to (37), which
relates to a cylinder, we obtain the restraint moment according to GECKELER's
approximate theory, which is, however, smaller than the correct value. 2 )
Hence, if the deformation load is neglected, it is safer to apply the characteristic
length based on the primary foundation modulus of the sphere. The fact the
characteristic length of the cylinder of revolution occurs in GECKELER's ap-
proximation can be explained with the aid of the deformation load. Because
of axial symmetry, we have:
ov = 0 hence p= Eo OU = Eo [_ii:_ + ieJ] = nx + Eow (38)

oy ' r ox r Eo r r r2
Eo .
For -
r2
= 1/2c (where c is the pnmary foundation modulus for the sphere)
n
and putting --"' = - 1 /2P, which is an acceptable assumption if the central
r
angle of the spherical shell is not too small, the differential equation for a
meridian strip will be:
d 4w
K- = p- 1 /2P+ 1 /2cw-cw
dx4
1
) In this connection it is appropriate to recall that, for example, in a circular plate uniformly
loaded over its whole area (freely supported or completely restrained at the perimeter), the
moments in a diametral beam strip, as part of the plate, are only 3 / 8 of those in the detached
beam subjected to the same load.
2
) Cf. TrMOSHENKO, S. Theory of Plates and Shells, McGraw-Hill, New York, 1940, pp.
472-475. See also HETENYI, M. Spherical Shells subjected to Symmetrical Bending, Puhl. Int.
Ass. for Bridge and Struct. Eng., Vol 5, 1937/1938, pp. 173-175.
17
or: (39)
2
· 1ar so1ut10n
W e see t h at t h e part1cu · of t h"is equat10n · w0 = p
· remams pr-
- = -
c 2Eb
whereas the value of the foundation modulus has become half that for the
sphere, i.e., equal to that for the cylinder of revolution.
In his plate analogy for the feebly curved shell WLAssow also arrives at a
foundation modulus equal to that for the cylinder of revolution, this evidently
being - as in GECKELER's method of analysis - correct only for axially sym-
metrical loading.
It should, however, be borne in mind that the load which has to be applied
to the analogous plate or beam may differ considerably from the actual load
on the shell.
In more general terms it can be said that, for a shell of revolution with
axially symmetrical loading and a meridian section of arbitrary shape the
secondary foundation modulus will be Eb , where r2 is the second principal
r2 2
radius of curvature (length measured along the normal up to its intersection
with the axis of revolution).
For ov = 0 and neglecting P, we obtain from ( 11) and (5) :
ay
_= ou = -Eb ( -ni + w)
Pz Ebk1 -
ox r 1 Eb r1-
which in combination with ( 10) and ( 12) gives:
Kt::,.t::,.w = Pz + -n1 - -Eb W

r1 r2 2
(40)
If 1! is not neglected, we obtain:
Kt::,.t::,.w = Pz + (-1 + -j!) ni -

r1 r2
Eb
-
r 22
W (41)
5 Conclusions
The conception of the doubly curved shell as an elastically supported plate
has much to commend it. In highly complex cases it would, for the present,
appear to be the only method available to the design engineer for estimating
the approximate magnitude of the edge disturbance moments. It should be
borne in mind, however, that a doubly curved shell can reasonably be re-
placed by a flat plate only if we confine ourselves to a so-called quasi-Euclidean
region, i.e., a region around the origin of the chosen system of co-ordinates
where the principles of plane geometry are still reasonably valid. This can be
18
assumed to be so up to a rise/chord ratio of about 1/7. For calculating the pri-
mary foundation modulus at any particular point both the principal radii of
curvature must be taken into account.
This foundation modulus is valid only if the deformation load is zero or
negligible. If the deformation load is taken into consideration, then the actual
load on the analogous plate or beam as well as the primary foundation modulus
will, generally speaking, be affected. This gives rise to a secondary foundation
modulus. In shells of revolution with axially symmetrical loading this second-
ary modulus is dependent only on the second principal radius of curvature,
this being in agreement with the foundation modulus as applied by GECKELER
and WLAssow. In the case of a cylindrical barrel vault uniformly loaded over
its entire surface, presenting an elongated rectangular shape on plan and
supported on all four sides, the secondary foundation modulus may, for a
shell strip extending in the transverse direction, even become zero.
The analysis as a plate, or a beam, supported on an elastic foundation
appears to be capable of yielding useful results in dealing with spherical shells,
hyperbolic paraboloid shells bounded by generating lines, and elliptic para-
boloid shells rectangular on plan.1)
It will have to be further investigated whether this analysis can be applied to
other cases and also whether it can be applied if the deformation load is neg-
lected but suitable load factors are introduced at the same time.
1
) GIRKMANN, K. Flachentragwerke, 5th ed., Springer-Verlag, Vienna, 1959, pp. 462-464.
See also the publication mentioned in note I on p. 229 of that book.
19
DOUGLAS FIR
USE BOOK
STRUCTURAL ·oATA AND PESIGN TABLES
1410 S. W. MORRIS.O N STREET

PORTLAND 5, OREGON
Price $5.00 per copy

276
HYPERBOLIC-PARABOLOID SHEW
DOUGLAS FIR USE BOO~
1
•••1
",.
.- '
•:r
tJ~
Doubly curved, hyperbolic.paraboloid roof Figtire 85 is an isometric view of a simple

structures of lumber are easy to design and to shell where: .
construct, and hav~ the added benefits of
economy and striking appearance. The name I}. = length of one side,
stems from the mathematical formulae de· a' = length of the horizontal projec-
scribing the curvature of the decking surface. tion of length a,
of
The curve the decking is a convex para· c - total compression force in peri-
bola when viewed parallel to one axis ·drawn meter member,
through opposite comers, and a concave para- c - principal compressive force in
bola when viewed from an axis 90 degrees to sheatliing per foot,
the first. Curves formed by the intersection
of the surface of the shell with a horizontal F - resuitant of the vertical reac-
plane are hyperbolic. The horizontal projec- ti on R and the horizontal
tion of such a shell may be square or diamond thrust H,
in shape, with the four sides of equal lenRth.
When a roof structure consists of a number
H = ho.r:izontal thrust,
of hyperbolic-paraboloids, the horizontal pro- h = vertical distance frolll a sufu-
jection of an indi'vidual shell may he rec- port to the highest point of t e
tangular in shape with a~jacent sides of shell,
I,
J unequal lenRtb. In elevation, the opposite k - inclined distance from a sup-
corners are elevated an equal distance above cort to the mid-point of tlie
the other .two. (See Figure 84). ength I,
The principal forces to be considered in l1 = length along longitudinal axis,
ii the design of a hyperbolic-paraboloid shell 12· = length along transverse axis,
are the reactions., tlie compression forces in
ji the perimeter members tlie shear forces at R = vertical reaction,
the 1unction of the sheathing and the peri- t principal tension force in
~ ·',
meter members, and the direct tensile and sheathing per foot,
compressio~ ~orces in ~e shea~. These
forces are easily· determined by statlcs. Once v = boundary shear force per foot.
the forces have been determined, the member
l. f:.
1 1.= .
sizes and connections ar-e designed by stand-·
ard engineering procedures. The method of
Lo1dlng Conditions:
. : .\r . analysis described here can also be applied Because of curvature, the dead load is not
i .....
) .i
to a series of hyperbolit-'paraboloids . uniform ever the projected area and the vari-
ance increases as the rise increases. However,
it· While there are several methods of com-

puting the forces in a hyperbolic-paraboloid
shell, the simplest is to resolve the reactions
into component forces in the surface and the
for hyperbolic-paraboloid .shells built of lum-
ber the dead load is likely to be very small as
compared to the live load, and the non-wri-
form distribution of the dead load is therefore
perimeter members. This method of analysis
i:' can best be illustrated by the following ex- usually neglected.
i : ample: Unsymmetrical concentrated loads or un-
I;.
.- .~.
-~
·i •.· .
t
Ii
·I·
Outer
·concave parabola
.
1)
Inner 1htathin9
Convex parabola
F11un 84-S11111t Hyperbtllle·Pmltolold Shell.

iJ
1.:
F
DOUGLAS FIR USE BOOK 277
HYPERBOLIC-PARABOLOID SHELLS
· balanced live loads could cause the structure

to tip unless restrained by:
1. A tension tie or strut from foundation
to each high point. (Wall framing may
be designed to-serve this purpdse~)
2. A combination tension and compres-
sion tie to one high point.
3. Connections at the two principal points
of support especially designed to resist
rotation.
4. Multiple hyperbolic-paraboloid struc-
tures, inter-connected at their adjacent
high points.
Res4"aint of the high points that prevent nor-
mal deflections of the shell under live load
will cause secondary stresses in the structure.
However, the load, and therefore the deflec-
tions, at the high points are very small. It is
suggested · that these secondary stresses be
neglected, since the unit stresses in the shell
are normally very low.
Since wind, snow or other live loads are
normally unbalanced as they occur on struc-
tures, the restraint against tipping must be
provided even though the design assumes bal·
anced loading. Because the structure's dead l
:~
wei~ht . is small, it should be well tied down
dunng erection to resist wind forces.
·:·
,.
Force In Perimeter Members:
For balanced live and dead loads the
vertical reactions are one-half the sum of the Figure 85-Geometry of Hyperbollc.P1r1boldld.
vertical loads. The. horizontal thrust, H, at
the reactions is derived by sii!Jple proportion,
and for this condition (see Figure 85), the Dividing the force, "F," into components
thrust is determined as follows: parallel to the perimeter members, tlie com~
R H press~on force, "C," in the perimeter mem-
= bers is:
h l/2 c
F/2
=
solving for H; k a
Rl
H=- (1) solving for C;
2h
aF/2
The force, "F," which is in the direction c =
of the line kin Figure 85, is the resultant of k
the horizontal thrust, "H," and the vertical
reaction, "R." substituting the value for F in equation (2)
Rk
"F" is solved by proportion as follows: a-
F R h Ra
k
= h
c (3)
2k 2h
solving for F; The foregoing is based on the assumption
Rk that the reactions will be at the two low
F = (2) points of the structure. If the supports are
h placed at the two high points the forces will
Fl '
I
278 D'OUGLAS F,IR USE BOOK . ;
HYPERBOLIC-PARABOLOID SHEUS
be the same, except that the perimeter :mem- ing act in the direction parallel to the grain,
bers will be in tension mst.ead of compression which provides the most efficient use of the
and the horizontal thrusts will be in the op- sheathing.
"
·,:,:
posite direction. For small shells, the second system can be "'
used to advantage as the sheathing boards do ·J;.·'
Boundary Shear: not curve and nave only a slight twist from
end to end. If the shell is small enough so
1
:·.··
There is a shearing force along the len~ that the sheathing is without undesirable de- l
'
ol the perimeter members at the junction of flection\ the falsework can be omitted. This
the sheathing and the ~rimeter members. system has one disadvantage since the joints
This boundary shear, "v, ' in pounds per lin- between constant width sheathing boards
eal foot, is found by dividing the compressive would leave gaps varying in width from a
force, "C," in the perimeter member by the maximum at the ends to no gap at the mid-
length of the perimeter member. point. · U this is undesirable, the sheathing
boards may be tapered to fit snugly. ·
Arrengtment of Lumber Sh11thlng:
Stress In Sheathing:
Lumber sheathina is ideal for shells be·
cause the comparativily narrow widths of the Since the stresses in the sheathing result
boards permit easy adjustment to the doubly in boundary shears along the perimeters, the
curved surface without .special cutting and bounda.r y shears can, conversely, be resolved
fitting except for end trimming to length. to determine the stresses in the sheathing.
Lumber sheathing may be applied to hy- The principal forces in the shell are tensile
perbolic-paraboloid shells in two ways: · forces, ''t,' parallel to the direction of the
concave parabolas; and compressive forces, /.
1. One larer of sheathing boards placed "c," paral1el to the direction of the conyex
paralle to the transverse axis where parabolas (see Figure 86): When the hori-
each piece bows to fit the curve of the zontal projection is a diamond shape, the ,1
,,
convex parabola with a second layer principlil tension and compression forces par-'
placed parallel to the longitudinal axis allel to the longitudinal and transverse axes
where pieces bow to fit the curve of can be resolved by proportion as follows:
the concave parabola.
2: One layer of sheathin~ boards placed Referring to Figure 8(), the principal ten-
parallel to two opposite sides of the sile force per foot of width, "t,' is: ·
structure and a second layer _placed
parallel to the other two sides. In this t li/2 Liv
system, each sheathing board twists = -- and t =
slightly. The total amount of twist v a' 2a'
from perimeter member to perimeter
member depends on the slope of the The principal compressive force per foot
perimeter members. of width, "c," is:
If the horizontally projected shape of the c l2/2 12v
shell is a square, the layers of roof sheathing =-·-and c =
are at right angles to each other for both sys- v . a' 2a'
tems ,of placement. If the shap_e of the struc-
ture is that of a diamond, the double layer of When the horizontal projection of a hy-
sheathing boards will be at right angles to perbolic-raraboloid is square in shape, the
each other when placement follows tlie first princif a tension and compression forces per
system. When placed by the second system, foot o width are equal in magnitude to the
the angle between layers of boards will de- boundary shear forces per foot of length of
pend on the angles between perimeter perimeter members.
members.
The unit tensile stress in the sheathing
As· there js a slope, but no curvature, to a lumber is equal to the principal tension force,
series of straight lines from one edge to an "t," per' inch of width divided by the thick-
opposite edge and parallel to a side, falseworlc ness of the sheathing in inches that parallels
placed in diis manner will automatically gen- the longitudinal axis. The unit compressive
erate the doubly curved surface and serve as stress in the sheathing which acts at 90 de-
support for the placing of the sheathing. grees to the tensile stress is equal t!) the
The first system of placing sheathina principal compressive force, "c," per inch of
boards is advantageous beciluse the principal width divided by the thickness of the sheath-
tension and compression forces in the sheathing in inches paralleling the transverse axis.
providing interaction between the layers.

Where butt joints occur in the sheathing,
additional fastenings should be used to trans-
fer the fo!ces across the joint.
Another method of increasing the stiffness
of the shell is to nail-glue the layers of
Component sheathing together in a zone _a rowid the
of principal perimeter of the shell. The width of the zone
compreHion varies depending on the size of the hyper-
force normal bolic-paraboloid, but is generally about one-
to perimeter tenth of the length of a side. This method
member.
should be -used with extreme caution as there
are many uncertainties conn~ted with field
gluing.
Component
. ~-.i....of principal Sht1thlng P1r11lel to Sides:
enslon force
nonnql to When the sheathing layers are placed
perimeter
.w ith the boards parallel to the sides of the
. Transverse axis member.
(Convex parabola) structure, each layer of boards is at an angle
to the direction of the principal tension and
compression forces with each layer resisting
a portion of the princifal tension force and a
portion of th.e pnnciJ>a compression force. Al>
the join1s between adjacent boards in a layer
Figure 86-Stre1111 In Sht1ffil119. represent a discontinuity, these forces have
to be transferred across.the joints through the
The component of the compressive stress adjacent layer by means of the fastenings
normal to the perimeter member exerts an connecting the layers. This resul1s in a shear
outward thrust on the perimeter member and between the two sheathing layers which must
the component of the tensile stress normal to - be resisted by the lastenings. The shear force
the perimeter member exerts an inward pull is equal in m~gni\ude to the .boundary she.a r
on the perimeter members: These com- · stress per mut of length which can be con-
ponents, being equal and opposite' in direc- verted to a shear stress per unit of surface
tion, as can be seen from Figure 86, will area.
cancel each other, with the result that the
perimeter members are subjected only to
axial compression forces. Design Considerations:
Light concentrated loads produce localized In the design of the c;;omponent members
membrane stresses and tena to cause ·local of a hyperoolic-paraboloid. structure, the
buckling in the vicinity of the load. These sheathing resists the principal tension and
effects can be ignored in design since they compression forces parallel to the longitu-
dissj_pate rapidly a short distance away from dinal and transverse axes. As the unit stresses
the load. Also the unit stresses in these shells in the sheathing are· generally quite low, ~co
from the usual loadings are q_uite low. nomical, one-inch nominal thickness lumber
. . can be used for the sheathing material.
Where the underside of the sheathing is
Sheathtng Parallel to Diagonals: le! t exposed to serve as the finish ceiling, the
With the sheathing· placed in the direc- nails or staples should be of e length that will
tions parallel'-to the longitudinal and trans- not penetrate c?mpletely through the bottom
verse axes of the hyperoolic-paraboloid, each layer of sheathing: They should be of large
layer acts independently; the layer parallel enough diameter, however, to develop ·the
to the longitudinal axis transmitting the required lateral and withdrawal resistance.
principal tensile forces to the perimeter mem- The boundary shear in the sheathing is
bers and the layer parallel tQ the transverse transferred to the perimeter members by
axis transmitting the principal compression means of the fasteners used to connect the
forces to the perimeter members. Nailing or sheathing to the perimeter members. In this
stapling the layers together, although not re- respect the design of the perimeter connec-
quired for strength, is required to prevent tions is the same as for lumber sheathed
buckling of the comP.ression layer and will diaphragms. A detailed discussion of this
impart additional stiffness to the shell by procedure is given on pages 262 to 274.
....':ii.C ti
280 DOUGLAS FIR USE BOOK
Unloaded end-
designed accordingly. If the sheathing is
cumulative comprHtion sandwiched into the perimeter members with
force varl11 uniformly half of the perimeter members above and half
from ztro ot ll'le unloaded below, there is no eccentricity and the peri-
end to ma•lmum ot ll'lt
btarln9 1nd . meter members are subjected to axial com-
Pet mtter pression stresses only. The latter method
mtll'lber
permits a somewhat smal)er perimeter mem-
ber to be used, but increases the number of
pieces to be framed and handled in erection.
. .. i
As a hyperbolic-paraboloid shell becomes
flatter, it becomes more flexible .with an in-'
1.
- ~
creasing tendency to buckle. For this reason
it is desirable to place a limita~on on the
flatness which can be expressed as the ratio
t, of the rise to·the length of a side. A minimum . ..
..
rise-length of side ratio of 1 to 5 is suggested. ' '
. ~~·
B1ori"9 end Method of Determining Twist In Perimeter
:ro1a1 cumulative
comprntion force Members:
at blarlnia end. Becaufie a hyperbolic-paraboloid shell is a
doubly curved.surface, the slope of the sheath·
ing at the junction with the perimeter mem-
Fl1ure 87-Column Action of bers is con.~tantly and uniformly: changing
Perimeter Member.
•'
along. the length of the perimeter mem'Oers.
As the surface of the perimeter members
The perimeter members transfer all loads must be tangent to the slieathing where they
to the bearing point and must have sufficient connect, the contacting face of the eerimeter 1 1 i
cross section to resist the cumulative axial members must be shaped aJ>propriately. ll ;~
compressive forces. As the boundary shear the perimeter memben are glued laminated ).
forces are distributed uniformly along the members the .changing slope can be obtained I
length of the perimeter members, the com- .by building in a twist to the whole member. ' :
pression force in the perimeter !llembers For each shell, two right hand and two left
varies as a cumulative sum of the boundary hand pe~eter members are required. ,'i
shear from zero at the high point to maxi-
mum at the sup.P.orts. The perimeter members
can be tapered if desired. The sheathing pro-
vides lateral restraint to the perimeter mem-
bers in the direction parallel to the plane of
the sheathing. In the din!ction perpendicular
to the plane of the sheathing the j>erimeter 90• onat1
members receive no lateral support, and the
Angl1 ABC
slenderness ratio related to this plane must
be considered. As the compressive force varies
·" uniformly from zero at the peak to a maxi-
mwn at the ~mpport, the peruneter members
are considered as a series of columns varying
.,;
~· :
from a long column with no load to a short
~. column with maximum load. At any point
i along .the length the induced compression
1: · parallel to grain stress due to the accumulated
load must not exceed the allowable unit stress
as determined by the standard column formu-
la for a column length equal to the distance
t: from the support to the point being consid:
h
·''
ered (see Figure 87). If the sbeath~ng is Perim1ttr member
!
placed on the top or the bottom of the
perimeter members, the boundary shear for<:es Angle of twiat
cause bending stresse.s in the perimeter mem-
bers due to eccentricity. Hence, the perimeter
members 'a re subjected to combined bending
and axial compression stresses and must be F11ure 88-Twl1t In Perimeter Member.
,.
The total change in slope, or rotation of the . mond shaped horizontal projection, the
contact surface of the perimeter members formula for the angle of twist is ·
from one end to the otlier, is the ang1e of ha
twist. As the ratio of the rise in respect to the tan of angle of twist =
length of the perimeter member increases, · (a') 2 cos ABC
the angle of rotation also increases.
where angle ABC is the angle shown in Fig-
The procedure for computin~ the angle of ure 88.
twist is applicable to h)'l'erbolic-para bOloids
having square or diamond shaped horizontal When the horizontal projection of the hy-
projections. perbolic-paraboloid is square, the angle ABC
Using the mid-point of the length 'ot a becomes zero and the formula reduces to
perimeter member as a convenient reference, ha
and considering the twist at this point to be tan of angle of twist =
zero, the angle of twist from the reference (a')Z
point to either end of the perimeter membeJ'
as shown in Figure 88, can be determined The total angle of twist from one end of
from the following formulae : the perimeter member with respect to the
other is twice the angle determined from the
For a hyperbolic-paraboloid having a dia- preceding formula.
f
['
•'
,;
•1
t'•I
;
.r
Slven Graceful 75 Foot Sp~l'I DouglH Fir Hyptrbollc·

P1rabolold1 form a 24,000 Square Fovt Pnlllon.
i """
iX~9A
J>6~~·
f C f1Vll ENGINEERING STUDIES
0
r r, . ,- STRUCTURAL RESEARCH SERIES NO. 304
THE ANALYSIS F SHALL SHELL STRUCTURES

BY A D!SCRETE ElE ENT SYSTE
by
B. MOHRAZ
and
W. C. SCHNOBRICH
A Report on a Research
Program Carried Out
under
National Science foundation
Grant No. GK-538
UNIVERSITY OF ILLINOIS
URBANA, ILLINOIS
MARCH, 1966
THE ANALYSIS OF SH.ALLOW SHELL STRUCTURES
BY A DISCRETE ELEMENT SYSTEM
by
Bo MOHRAZ
and
Wo Co SCHNOBRICH
A Report on a Research
Program Carried Out
under
National Science Foundation
Grant Noa GK ... 538
UNIVERSITY OF ILLINOIS
URBANA, ILLINOIS
MARCH, 1966
A'.cKNOWLEDGMENTS
The 'results reported he;rein were· obtained in the course of a research
study conducted in the Departm~nt of Civil Engineering, University of Illinois~
This report is. based on a thesis by Bi.JC!I1 M()hJ'."El.Z tl.119.~:r tli~ d:i.ir<=ction Qf
Professor Arthur Ro Robinson~ which was submitt~d in partialijfulfillment of
the requirement for the degree of Doctor of Philosophy in Civil Engineering
in the Graduate College of the University of Illinoiso
rhe investigation was made possible by the National Science
Foundation, under Grant NSF GK=538o
The authors express their appreciation to Dro John Wa Melin for
his valuable suggestions regarding the development of the computer programo

TABLE OF CONTENTS
Page
ACKNOWLEDGMENTS . . iii
LIST OF TABLES . vi
LIST OF FIGURES. vii
JNTRODUCTION . . 1
l.l. General . . . . . . . 1
l. 2. Objective of Study. 3
l.3. Nomenclature. 3
METHOD OF ANALYSIS . . 6
2.1. General . . .6
2.2. Description of the Model. . . . . 7
2·. 3. Coordinate System . . . . . . . . . 8
2.4. Displacements . . . . . . . . . . . . . . . . 9
2. 5. Strain-Displacement Relations . . , . 10
2.6. Foroes and Moments . . . q •••••••• l3
2.7. Equilibrium:Equations . . . . . l4
BOUNDARY CONDITIONS. . • . . . . . 19
. 3.1. General . . . . • . . . 19
3.2. Simply Supported Edge . . . . . 19
3.2.1. Roller Support. 19
3.2.2. Hinge Support . • r . . ' .. 24
3.3. Free Edge . . . . . . . 24
. . 3.4.
'(',t
Other Boundary. Conditions . 31
NUMERICAL RESULTS. . . . . . . . . . . 32
4.l. General . . . . . . . . . . . 32
4. 2. All Edges Simply Supported . . 34
4.2.1 .. Uniformly Loaded Rectangular Plate . . . 34
4. 2. 2. Uniformly Loaded. Cylindrical Shell. 34
4.2.3. Uniformly Loaded Elliptic-Paraboloid. 36
4. 2. 4. Uniformly· Loaded Hyperbolic Paraboloid. 36
4. 2. 5. Uniformly Loaded Hyperbolic· Paraboloid -Bounded by
· Characteristic Lines of the Surf ace . 37
'· 4. 3. Two Opposite Edges Simply Supported and the
RemainingTwo Edges Free. . . . . . . . . . 38
4. 3.1. Uniformly Loaded. Square Plate . . . 38
4.3.2. ·Cylindrical Shell Subjected to Sinusoidally
Varying Edge Load . . . • . . . . . . . . . . . . . 38
-iv-
-v-
TABLE OF CONTENTS (Continued)
Fage
4.3.3. Cylindrical Shell Subjected to Sinusoidally

Varying Load. . . . . . . . . . . . . • . . 39
4.3.4. Elliptic Paraboloid Subjected to Sin~soidally
Varying Load, . . . . . . . . . . . . • . .. . . 40
4.3.5. Hyperbolic Paraboloid Subjected to Sinuso~dally
Varying Load.. • . . . . . . . . • . . . • . . • ·42
CON~USIONS AND RECOMMENDATIONS FOR FURTHER STUDIES. 43

5.1.
5. 2.
Conclusions . . .. . . .
Recommendations for Further Studies . . . •
43
. 4.4
J3IBLIOGRAPHY ... .... .... .. .. 45

~rABLES .. .... ... 47
J~IGURES. .... 51
APPENDIX A. EQUILIBRIUM EQUATIONS . . . 76
APPENDIX B. A GENERAL DISCUSSION OF THE COMPUTER PROGRAM. .. 79
LIST OF TABLES
Number
l Solution of Simply Supported Rectangular Plate . . . .
2 Maximum Positive and Negative Values of w, N , and M

at the Midsection of Simply Supported Cylind~ical Shgll. 48
3 Solution of Sq_uare Plate, Two Opposite Edges Simply

Supported, the Other Two Edges Free . . . . . , . ~ .
.4 Maximum Positive and Negative Values of w, N , and M

at the Midsection of Cylindrical Shell, Two OppositexEdges
Simply Supported, the Other Two Edges Free . . . . . . . . . 50
-vi-
LIST OF FIGURES
Number
1 Typical Model of Positive Curvature . 51
2 Typical Model of Negative Curvature . 52
3 Rigid Joint Connection . . 53

4 Locations of Displacements and· Loads. 54
5 Grid Point -Indentification. . . . . • 55

6 Transformation of u Displacement Into Tangential Plane . . 56
7 Effect of w.Displacement on Extensional Strain. • q 56

8 Effect of w Displacement on Shear Strain
Due to the. Twist of the Element . 57
9 Rotation of Rigid Bar ij,i+2j , 58

10 Effect of Rotation of Rigid Bar on Flexural Strain • . 58
11 Simply Supported Edge 59

12 Location of Auxiliary Rigid Bars at the Free Edge . 60
13 Free Edge . 61
14a Rotation of Bar i-lj+l,ij+l . . 62
14b Equilibrium at Free Edge . . . 62
15 Se:raration of the Two Networks iri. Flat Plates .
16 Deflections Along Diagonal of a Uniformly

Loaded, Simply Supported Rectangular Plate. 64
17 The Displacements w, the Forces N , and the Bending

.Moments Mx at the Midsection of aYCylindrical Sb,eli,
All Edges Simply. Supported. . . . .
1
. . . .
18 Tbe Displacements w, the' Forces N, and the Bending

Moments M at the Midsection of afi Elliptic Paraboloid
Shell, -Alt Edges _Simply 'Supported . . . . . . . . . . 66
19 The Displacements w, The Forces N , and the Bending.
Moments M at the Midsection of ayHyperbolic Paraboloid
Shell, Alt Edges Simply Supported . . . . . . . . . . .
-vii-
-viii-
Number
20 The Displacements w) the Forces N ) and the Bending

Moments M at the Midsection of aXSimply Supported
x
Hyperbolic PaTaboloid Shell Bounded by Characteristics. 68
21 The Displacements w) the Forces. N ) and the Bending
Moments M at .the Midsection of ayCylindrical Shell)
Two EdgesxSimply Supported and the Remaining Two Edges Free . 69
22 The Displacements w) the Forces N , and the Bending
Moments M at the: Midsection of aYCylindrical Shell,
Two EdgesxSimply Supported and the Remaining Two Edges Free . 70
23 The Forces. N and N at the Supported Edge of a
Cylindrical fil;.e11, ~o Edges Simply Supported and
the Remaining Two Edges Free . . 71
24 The Displacements w) the Forces N ) and the Bending
Moments M at the Midsection of afi Elliptic Paraboloid
Shell) Tw3 edges simply supported and the Remaining
Two Edges Free. . . . . . . . . . . . . . . . 72
25 The Forces N . at the Midsection of an Elliptic Paraboloid
Shell, Two Efiges Simply·Supported and the Remaining Two
Edges Free. . . . . . . . . . . . . . . . . . . . . . 73
26 The .Plot of the Forces N at Free Edge VS Inverse of the
Square of Number· of Spac:rngs for an Elliptic Paraboloid Shell,
Two Edges Simply Supported and the Remaining Two Edges Free .
27 The Displacements w, the Forces N ) and the Bending Moment M

at the Midsection of a HyperbolicYParaboloid Shell) Two Edge~
Simply Supported and the Remaining Two Edges Free 75
B-1 General Flow Diagram of the Computer Program. . . . . . . . . 81

INTRODUCTION
1.1 General
Shell structures, both from the point of view of structural efficiency
and esthetic value, are fre~uently used to cover large, column-free spaces.
Determinations of the magnitude of stresses and their distribution into the
shell is of necessity in the design of these structures.
In the interior regions of a shell, away from the supports, the loads
are carried mainly by in-plane forces. The membrane theory provides a good
estimate of stresses for design purposes. However, in the regions close to the
supports, bending stresses' are developed as the results of the sharp change .in
the stiffnesses from that of the shell to that of the edge member.
For shells with simple geometry, such as(qylindrical and spherical
shells, bending solutions are available for a variety of loadings and support
conditions. However, for shells with complex g~ometry, such as translational
shells, bending solutions exist only for a limited number of cases with simple
loadings andsupport conditions. The support conditions for which solutions
have been found are rarely those in practical applications. Usually the govern-
ing e~uations are reduced to two e~uations in terms of· a stress function and
the radial displacement [ 13] * . A Levy type solution can be obtained when at
least two edges are simply supported. But these supports are usually a poor
idealization of actual supports.
One shell surface of particular interest is the hyperbolic paraboloid
bounded by the characteristic lines of the surface. A variety of shape$ can be
*·· Numbers in brackets refer to references in the Bibliography.
/
-1-
-2-
formed by different combinations of hyperbolic paraboloid units. Due to con-
struction economy and the inherent beauty of these shapes, they have been widely
used in recent years. The difficulties encountered in the analysis of these
shells and their extensive use have led to numerous studies of their structural
behavior. The edge disturbance from a straight boundary pe~etrates into the
shell further than the edge disturbance from a curved boundary [4]. Therefore)
a bending analysis which takes into account the effect of supporting the sbell
by an edge member is very important in the design of long span hyperbolic
paraboloids. Because of the asymmetric relation between the stress function
and the radial displacement in the governing equations of hyperbolic paraboloid,
a Levy type solution such as the one used for cylindrical shells leads to
unrealistic boundary conditions [5]. This is due to the absence of suitable
orthogonal functions.
Approximate methods have been used whenever the exact methods have
failed to yield the desired solutiona Variational procedures, such as the
meth9ds. of\-Rit'.?, · .(ialer®:in;:L~antorov.ich, etc]:;:.are. ·avaiJ!.ab1e: ill'b:ti.ithe solution
of boundary value problems. A combination of Kantorovich 1 s and Galerkin's
methods has been used to obtain a solution for a hyperbolic paraboloid with
simply supported and clamped boundaries [5]. For shells of arbitrary shape,
the selection of approximate functions is usually difficult and the integration
of these functions is very involved.
Numerical procedures, such as finite differences, have been used to
solve the complex differential equations encountered in shell structures
[9,10,17]. The use of the method of finite differences in shells frequently
results in a set of inconsistent difference equations. This inconsistency can
be avoided if the tangential displacements are properly specified [6,14,16].

-3-
A still different approach to the problem is the use of models to
simulate the shell structure. Hrennikoff [12] replaced the continuous structure
by a system of elastic bars. Yettram and Husain [19] used the system of elastic
bars to obtain solutions to plate problems. Parikh [15] and Benard [2] applied
this method for cylindrical shells. A different model which replaces the
continuum by a system of finite plate elements has been used by Clough [7,8] .
.Zienkiewicz and Gheung. [ 20, 21] have applied the method to orthotropic .slabs
and arch dams. Discrete models consisting of rigid bars and deformable nodes
have been used by Ang and Newmark [1] to obtain solutions in plates.
Schnobrich [16] used a discrete model for cylindrical shells. One of the
advantages of the discrete model is that a consistent Bet of equations is
obtained.
l.2. Objective of Study
.With rapid advancements in computer technology, one can now obtain
solutions of complex shell problems which previously were not possible because
of computational work involved. The objective of this study i.s to develop a
discrete model to simulate a variety of shell structures. The classical shell
theory will be used as a guide in the development of the model. The generality
of the model is illustrated by presenting numerical examples and comparing
them with existing solutions whenever possible.
l.3. Nomenclature
The symbols used in this study are defined when they first appear.
For convenience they are summariZJed below:
a shell span in x direction
b = shell span in y direction

-4-
c = rise or sag of hyperbolic paraboloid bounded by the characteristic

lines of the surface
d twist of element i-lj-1, i+lj'--1, i+lj+l, i-lj+l
E modulus of elasticity
h shell thickness
L grid length in x 'direction

x
L grid length in y direction
y
M ... moment about y~axis in deformable node ij
XlJ
M .. moment·about x-axis in deformable node ij
YlJ
M .. twisting moment in defonnable node ij
xylJ
N .. membrane force in x direction in deformable node ij
XlJ
N .. membrane force in y direction in deformable node ij
YlJ
N. . . = in-plane shear- force in def6rmable node iJ'
xylJ
q =· external load·
Q vertical shear- force at a section perpendicular to x-axis

x
Rx radius of curvature in x direction
Ry_ radius of curvature in y direction
t distance between the. ·extensional~ ele~ents
ui+lj = displacement of point ·i+lj in x direction
vij+l = diS:pla cement of point ij+l in y direction
w .. - displacement of deformable node ij in z direction

lJ.
x external load in x direction
y external load in y direction
z = external load in z direction
a = angle· between rigid bars in xz plane
angle between rigid bars in yz plane

-5-
E .. extensional strain in x direction at deformable node ij

XlJ
u extensional strain due to u displacement in x direction at
E ..
XlJ
deformable node ij
w
E .. extensional strain due to w displacement in x direction at
XlJ
deformable node ij
E .. extensional strain in y direction at deformable node ij

YlJ
E •• shear strain at deformable node ij

xyiJ
u,v shear strain due to u and v displacements at deformable node ij
E •.
xylJ
w
E .• shear strain due to w displacement at deformable node ij
xylJ
eXl+
. l"J rotation of bar ij, i+2j in x direction
8 rotation of bar ij, ij+2 in y direction

yij+l
v Pois son's ratio
a .. extensional stress in x direction at deformable node ij

XlJ
b
a .. = flexural stress in x direction at deformable node ij
XlJ
'T
xy shear stress
cpx :::;: opening angle in x direction
cpy opening angle in y di.rec ti on
xXlJ
.. flexural strain in x direction at deformable node ij
Xyij flexural strain in y direction at deformable node ij
x_xyij twisting strain at deformable node ij

METHOD OF ANALYSIS
2 .1. General
Although specialized methods of analysis, such as Levy solution)
Ritz and Galerkin methods) etcoJ have certain advantages for a particular
problem) it is desirable to have a general method of analysis which can be
used for a variety of shell surfaceso Different boundary conditions) different
values of Poisson's ratio) and different combinations of loading should not
present any difficulty in the method of analysis. The method should be adapta-
ble to non-linear problems, orthotropic shells J and shells with vari.able ,;
thickness.
One such method is the use of a discrete model consisting of rigid
bars and deformable nodes similar to that proposed by Schnobrich [16]. Since
the material properties of the shell are concentrated at the deformable nodes,
the method can be used to study the behavior of shells with different material
properties in the two directions. It can also be used to analyze shells with
variable thicknesso The forces and the moments are constant across each node,
thus, non-linear material properties would not present any difficulty in the
use of the model.
Because of the complexity of the equations governing the behavior of
the model, it is advantageous to generate and solve the resulting set of
equations within a digital computer. Due to the effort involved in the de-
velopment of individual computer programs for various cases) it is preferable
to develop a single general computer program which can be used and can easily
be extended to a variety of shell problems. Although such a computer program
can become very complex) once it is completed many shell surfaces with various
-6~
-7-
loadings . and support conditions can economically be -investigated without much
additional programming.
The equations governing the behavior of the model are formulated
fo.r a general doubly curved surface using orthogonal coordinates. However,
the coordinate lines are not necessarily the set which coincides with the
line·s of :prin·cipal curvature.
2.2. Description of the Model
The dis.crete model employed in this study cons is ts of rigid bars and
deformable nodes arranged as shown in Figs. (1,'2). The nodes have extensional
and shear properties similar to those of t.he real material. At the midpoint
of each rigid bar there is a.groove with a circular hole at its center, Fig. 3~
The rigid bars are connected tO each other at these grooves by a pin which is
inserted in the holes. With this type of connection, the two bars move
independently of' each other in the radial direction. However,·the rotatton of
oria bar would cause a twist in the other bar.
This model is a modification of the one used previously for cylindrical
shells [16]. By specifying· the tangential displacements at the same point, the
extensional and shear behaviors are no longer separated from each other. The
model can thus be used to study the nonlinear material behavior of shells
based on accepted yield theories. It also makes it possible to formulate the
governing equilibrium equations of hyperbolic paraboloid shells bounded by
characteristics as well as other shells of negative curvature.
When the model is used for the analysis of shell structures, the
deformable nodes are placed at the intersections of the surface generators,
Fig. 1. For shallow shells the generators are usually approximated by circular
-9-
vary at each point. It is therefore convenient to adopt a moving triad of
axes. When the origin of the coordinate system is placed at the deformable
node ij, the x and y axes are set along the bisectors of the angles formed by
the rigid bars in the xz and the yz planes respectively. The z axis is
directed along the perpendicular to the plane containing the bisectors (i.e.,
the tangential plane). The positive direction of z axis is into the plane
of the paper, thus, resulting in a left handed coordinate system. By rotating
this coordinate system through angles ~/2 and a/2 about the x and y axes
respectively, the desired coordinate system at points ij-1, i+lj, ij+l, and
i-lj is obtained.
Throughout this study, the strains, the forces, and the moments are
formulated for deformable node ij. Node ij may be a typical interior node, a
node near the boundary, or a node on the boundary. The equilibrium equation
in the x direction is formulated for point i+lj which is located one-half
space~ from the node ij, in the positive x direction. Similarly the equilibrium
equation in the y direction is formulated for point ij+l which is located
one-half space from the node ij, in the ·posi.tive y direction. The equilibrium
equation in the z direction is formulated at point ij. Points i+lj, ij+l,
and ij may be typical interior points, points near the boundary, or points on
the boundary .
2.4. Displacements
Displacements are defined in the following manner:

11
a. The tangential displacement u" is defined at the intersection
of rigid bars and is directed along the axis of the bar in the x direction.
b. The tangential displacement "v" is defined at the intersection
of rigid ba~s and is directed along the axis of the bar in the y direction.
-10-
11
c. The radial displacement w" is defined at the deformable node
and is directed along the perpendicular to the plane containing the bisectors
of the angles formed by the rigid bars (tangential plane) .
The positive direction of u, v, and w is the positive direction of x_,
y, and z axes respectively. Figures 1, 2, and 4 show the manner in which the
displacements are defined. The above displacements are only the components
of the total displacements at the specified points. The total displacement
of any point in a given direction is obtained by proper combination of these
displacements.
2.5. Strain-Displacement Relations
To find the extensional strain in the x direction at the deformable
node ij, the u displacements of the rigid bars adjoining the node must be trans-
formed to the plane containing the node (tangential plane). The extensional
strain due to u displacements .(see Figi~ 6) :is ·
u
E .. (2.2)
XlJ
The extension~l strain due to w displacement of the deformable node ij
(see Fig. 7) is.
w -1
E ..
R Cos a/2 wij (2.3)
XlJ
x
Therefore, the total extensional strain in the x direction at the node ij is
1 1
E ..
XlJ L lbs a/2 (11i+lj - ui-1) - R Gos· a/2 w ..
lJ
(2.4)
x x
Similarly the extensional strain in the y direction at the deformable node ij
can be written as
-11-
The in-plane shear strain due to tangential displacements surrounding the
node ij is
(2.6)
Because of the twist of the surface there is also a shear strain due to the
normal displacement (see Figo 8).
w d
E ..
xyiJ 2~wi.
x y J
Therefore) the total shear sttain at the deformable node ij is
(2.8)
It should be noted that these strains are the average strains through the
thickness.
The above strains are due to extension and shear. There ar.e also
strains due to bending and twist which result from the rotation of the rigid
bars. The rotation of the rigid bar ij J . i+·2j in the x direction can be seen .
from Fig. 9 as
(209)
Similarly the rotation of the rigid bar i-2j, _:_i.J: 1,i'.l the.:U<>di:rec~;i_on•is
-12-
The flexural strain in the x direction at the top and the bottom elements of
the deformable .node ij is obtained from Figo 10 as
1 t
xxi· J. - ·L ( 8 Xl+
· 1 J· - 8 Xl-
· 1 J· ) 2 (2.11)
x
Substitution for B and B . . yields:

Xi+lj Xl- 1 J
t t
X . . 2R L Cos a} 2 ( u . 1 . - u . 1 . ) + 2 (w . 2 . - 2w . . + W• 2 .)
XlJ x x i+ J i- J 2L Cos a/2 i+ J lJ i- J
x
(2.12)
Similarly the flexural strain in the y direction at the top and the bottom
elements of the deformable node ij is
t t
2R L Gos t3/2 (v ij+l - v ij-1) + 2 A/ 2 (w ij+2 - ~ij + wij-2)
y y 2L Gos f--'
y (2.13)
The twisting strain is obtained from.the relative rotation of the four rigid
bars surrounding the deformable node ij. Hence)
(2.14)
Expressions for B's similar to Eq_. (2o9) can easily be obtained by proper use
of subscripts. Substitution for B's yields
t ) t
~ij 2R L
x y
Cos a/2 (uij+l - uij-1 + 2R L
y x
Gos t372 (vi+lj - vi-lj)
t 1 1 )
+ 2L L (Cos a/2 + Cos t3/2 (wi+lj+l - wi-lj+l - wi+lj-1 + wi-lj-1)
xy
(2.15)
-13-
2.6. Forces and Moments
As was mentioned in Section 2.2) it is assumed that the deformable
nodes are in a state of plane stress" The in-plane forces and the bending
moments are concentrated at the deformable nodes. The in-plane force in the
x direction at the deformable node ij is obtained by multiplying the exten-
sional stress a .. by the area. Thus)

XlJ
L
N .. a .. ( 2Y) h (2.16)
XlJ XlJ
From . Eq. ( 2 . 1)
E
a .. - - - ( E .. + VE .. ) (2.17)
xiJ l-v 2 xiJ yiJ
Substituting for a .. in Eq. (2.16) we obtain

XlJ
L
N ... -_ Eh i (E .. + VE .. ) (2.18)
xiJ l-v 2 2 xiJ yiJ
Similarly
Eh Lx .
N .. = - l
2 2 (EYiJ' + VEXiJ°l) (2 .19)
YlJ -v
The in-plane shear forces are
L
Eh i E
N (2.20)
xyij 2(1+v) 2 xyij
L
Eh x
N E (2.21)
yxij 2(1+v) 2 xyij
The expressions for E . ·J E . ·J and E .. are given by Eqs.

XlJ YlJ xylJ
(2.4)) (2.5)J and
(2.8), respectively.
-14-
11he bending moment in the x direction at the deformable node ij is
obtained by taking moments about the mi.d-de:pth of the deformable node. Therefore_,
b _][_ h ] t
L
M .. - 2 [ crxi/ 2) 2 2 (2022)
XlJ
b
wµere a .. is the stress due to bending and is given by
XlJ
(2.23)
Thus,
L
M .. ~ Eh _]£_ ~ ( ) (2. 24)
XlJ 2 2 2 Xxij + vxyij
1 -v
Similarly
Eh Lx t
M .. - --2 -2 -2 (xyi" J. + vxxi" J·. ) (2025)
YlJ 1 -v
The twisting moments are
L
Eh 'y t
M
xyij 2(1+v) 2 2 xxyij (2.26)
Eh Lx t
M
yxij 2(1+v) 2 2 xxyij (2.27)
The expressions for X .. , X .. , and X .. are given byEqs. (2.12), (2.13), and
XlJ YlJ XYlJ
(2.15), respectively.
To obtain the distribution of the forces and the moments, Eqs. (2.18)
L L
through (2.21) and (2a24) through (2a27) are divided by . ; or ~ .
2. 7. Equilibrium Equations
The :principle of virtual displacement is used to formulate the
equilibrium equations governing the behavior of the modeL By the :principle

-15-
of virtual displacement, if the system is in equilibrium the total work done
by the internal forces plus the total work done by the external forces is
equal to zero for any arbitrary virtual displacement. For example, to obtain
the equilibrium equation in the radial direction at a specified node_, the
node is given a virtual displacement while all other displacements remain
fixed. The sum of the works done by the internal and the external forces
caused by this virtual displacement is then set equal to zero. The results
are a set of linear algebraic simultaneous equations in terms of the three
displacements. These equations are solved and the obtained displacements are
used to compute the forces and the momentso
The equilibrium equation in the x direction is obtained by giving
the intersection of the rigid bars a virtual displacement 6u while keeping
all other displacements fixed. Referring to Figo 5, a virtual displacement

6u. ., results in a change in extension and flexure of the nodes ij and i+2j
l+1 J
and a change in shear and twist of the node·s i+lj+l and i+lj~l. The internal
work is equal to the negative of the change in the strain energy of the four
deformable nodeso HenceJ
Wint = - l N 6E L
where N and 6E are the force and the strain due to ·extehs ibn, shear;,: ;bending)-
and twist;~i.1 .Subst::Vtutfon ;for ·ws and ;6.E' s ·yields
u Eh L
~ L [ (.
.w.int E • • + VE . . )6E . . + ( X , . + VX . . ). tsx , .
1-y 2 2 - XlJ YlJ XlJ XlJ YlJ XlJ
+ (E.
Xl+ 2·J +VE.
yi+ 2.)6E.
J Xl+ 2·J +vx~-·
Xl+ 2·J + (x. ·-yi+2J.)6X.
Xl+2j
1-v
+ - - ( E • • 6E . . + . . 6X . .
2 xyi+lJ-1 xyi+lJ-1 xxyi+lJ-1 xyi+lJ-1
. l"J+ I 6 Exy1.+
+ Exyi+ . l"J+ I 6 Xxyi+
. l"J+1 + Xxyi+ . l"J+ l)J
-16-
where 6E' s and SX' s are the incrementa 1 strains, i.e.,
1
f:.E ..
XlJ L Cos a,/ 2 t:.ui+lj
x
1
f:.E . . L Cos a/2 (-t:.ui+lj) (2.30)
Xl+ 2 J
x
etc.
The external work of the component of the load in the x direction at point
i+lj is
Wu X b. (2.31)
ext = i+lj ui+lj
For equilibrium
(2.32)
Substituting the strain-displacement relations into Eq. (2.29) and then sub-
stituting the results i.nto Eq. (2.32), the desi.red equilibrium equation is
obtained. Thus, the equi..librium equat.ion i.n the x direction at poirit i+.lj is
2
LxLy [(. 1 t 1
-2- 2 + 2 0 ) I 2 ( u • ".') • - 2u ' 1 • + u • 1 • )
· -cos a/2 4Rx cos'- cx/2 1 i+ _.,J i.+ J i- J
.x
·2
+ 1;':' ( 1 + t ) l ( u . l . ,-. - 2u . l . + 11 . . )
4R2 Cos2 a/2 L2 . i+. J+~ i+ J i+lJ-2
x ·y
2
(1-v · v l+v t 1
+ 2 + Cos a/2 Cos f3/2 + ~ 4R R Oos a,/2 Cos f3/2) LL (v i+2j+l -v i+2j-l
x y x y
1 1
- v. · 1 + v. · 1) - (
lJ+ lJ- . 2
R Cos a/2
+Ry Cos aJ2 Oos f3/2) Lx (wi+2J' - wiJ.)
x
-l7-
1 d t2 1
(11-v) 1 ) -3 (w . .4 . - 3wi·+2J·
4R Cos 2 aI 2 '·1 L
(w.i +l"J +1 - w.i +l"J - 1) + (
x y y $
l+ · J
x x
2
t (1-v 1 l+V 1 ) 1
+ 3wij - wi-2j) + 4Rx Cos a/2 ~Cos a/2 +~Cos ~/2 2 (wi+2j+2
1 1
x y
- 2w. . + w.
l+ 2 J
.
l+ 2 J- 2
- w ..
lJ+ 2
+ 2w .. - w ..
lJ lJ- 2)·_] = 0
(2.33)
Similarly the equilibrium equation in the y direction is obtained
by giving the intersection of rigid bars a virtual displacement 6v and then

satisfying the relation
WVext + WV 0 (2.34)
int =
Finally when the deformable node ij is given a virtual displacement
6w the internal work of the surrounding nodes can be written as
1 1
Eh x y [
w.wint = - --2 --2- ( Ex l. - 2.J + VEy l. - 2.J )6Ex l. - 2.J + (x x l. - 2.J + vxy i. - 2.J )6xx l. - 2.J
1-V -
+ ( E .. + VE .. )6E .. + (x .. + vx.__ .. )6x ..

XlJ YlJ XlJ XlJ ·-ylJ XlJ
+ ( EYlJ+
.. 2 +VEXlJ+
.. 2)6EYlJ+
.. 2 + (x.__ .. 2 + VXxi· J"+2)6Xyi· J"+2
'-YlJ+
1-V (
• 1 . 1. 6 E
+ - 2 · Exyi- . 1 . 1 + X . 1 6 Xxyi-
. 1 J-. . 1 +
. 1 J- E • 1 . 1 6.E . 1 J-
. 1
J- xyi- J- · xyi- xyi+ J- xyi+
+ xxyi+lj-1 6.Xxyi+lj-1 + Exyi+lj+l 6 Exyi+lj+l + xxyi+lj+l 6 xxyi+lj+l
+ Exyi-lj+l 6 Exyi-lj+l + Xxyi-lj+l 6 Exyi-lj+l + Exyij 6 Exyij)J

(2.35)
-.18-
The external work is
Z .. 6w ..
lJ lJ
For equilibrium in the radial direction
WW WW 0
ext + int
Thus, the third equation representing equilibrium in the radial direction is
obtained.
The complete set of equilibrium equations for typical interior
points is given in Appendix A. When the equilibrium equations for all grid
points in the model are formulated, they constitute a set of linear algebraic
s~multaneous equations in terms of the three displacements. These equations
are mathematically consistent with the finite difference expressions of the
bending theory of a general shell. If one :radius of curvature is infinite,
the equations are those for a cylinder [16]" If both radii are infinite,
the plate equations result [1].

BOUNDARY CONDITIONS
3.1. General
One of the advantages of using models to study the behavior of ·shell
structures is their adaptability to various boundary condi ti.ans. As was
mentioned in Section 2.7J the internal work of the system corresponding to
a virtual displacement can be formulated in terms of strains) Eqs. (2.29 and
2.35). Since the geometric and the force boundary conditions can be expressed
by strain-displacement relations J the internal work of the deformable nodes
on or near the boundary is easily obtained by the use of appropriate strains.
The procedure for obtaining the equilibrium equations of points on or near
the boundary is similar to the procedure for obtaining the equilibrium equations
of typical interior points, i.e. the desired point is given a virtual displace-
ment and the sum of the internal and the external works of the system associated
with the virtual displacement is set equal to zero.
Whenever the strain-displacement relations at the deformable nodes
on or near the boundary include displacements of points outside the model_,
new strain-displacement relations must be formulated. These expressions are
obtained according to the restraints at the edge and include only displace-
ments· of points within the model.
3.2. Simply Supported Edge
3.2.1. Roller Support

0
This_ type of support) usually referred to as diaphragm support 11 is
rigid in its plane but offers no resistance in the direction normal to the
plane. Assuming that the edge parallel to the y-axis is supported by rollers)
-19-
-20-
the following conditions must be satisfied~
v 0
w 0
N 0
x
M 0
x
With the above relations, the strains at the deformable nodes on or near the
edge can be easily formulated.
Node on the e~dge. Using the first two condit±ons of Eqs. (3.1), the
extensional and the flexural strains in the y direction at the deformable
node ij located on the edge are obtained from Eqs. (2.5 and 2.13)"
,€:yij 0
(}.2)
0
Xyij
Using the last two conditions of Eqso (3"1) and substituting E ..

YJ..J
and xYJ..J
,.
into Eqs. (2.18 and 2.84) we get
E .• 0
XlJ
Since v =0 all·along the edge, the deformable node ij cannot displace in
the y direction. Thus,, at the deformable node ij located on the edge, the
shear strain is found to be '(Fig. 11)
(3.4)
Since w = 0 all along the edge, the points of intersection of rigid bars
cannot displace in the radial direction. This can be accomplished by
removing the dowel pin (see Fig. 3) and rigidly connecting the two bars to
-2l-
each other at the edge. The rotation of bar i-lj+lJ ij+l in the x direction
{Fig. 11) can be written as
1 Lx 1
8
xij+l = r::J2.
X
(2R
X·
cos a/2 uij+l - cos a/2 wi-lj+l) (3.5)
Similarly) the rotation of bar i-lj-1, ij-1 in the x direction is
(3.6)
The rotation of rigid ~bar i-lj-1, i-lj+l in the y direction is
(3-7)
The twisting strain at the deformable node ij located on the edge can now be
obtained by substituting Eqs. (3.5 - 3.7) into the following relation:
xxyij 1 (e :_ e ) t 1 t
8 (3.8)
- L
y
xij+.1 ' xij-1 2 r::J2.
x
yi-lj 2
Thus)
t t
2R L Cos a/2 (uij+l - uij-1) + 2R L Cos ~/2 (- 2vi-lj)
x y y x
(3-9)
+ t ( .l + l )( 2w 2 )
2L L Cos a/2 Cos ~/2 - i-lj+l + wi-lj-1
x y
Node one-half 'space from the edge. The strain expressions E •• '
XlJ
E •. ) and i .. at the de,formable node ij located one-half space from the edge
YlJ ·-yiJ
are similar to those of 'a typical interior node. The shear strain E .. and
xylJ
the twisting strain xxylJ
.: . are obtained by substituting Eqs. (3.l) into
. Eq_s . ( 2 . 8 and 2. 15) . Thus )

-22-
. l· ·1 " d ;
E -. · = L- (u.··+1 - ufJ·~i) + Lx (-vi-lJ.} - 2 LL wi ·
XYlJ y lJ x y J
t t
x ..
. ~lJ 2R L ,Cos a/2· (uij+l - uij-1) + 2R L Cos f3/2 (-vi-1)
(3 .10)
x y y x
t 1 1
+ 2L L,· (Cos a/2 +Cos f3/2)(-wi-lj+l + wi-lj-1)
x y ·.
The flexural ·strain Xxij is obtained by substituting expressions similar to
Eqs. (2.10. and 3.5) into Eq. (2.11). Therefore)
= 2Rx Lx ~os a/2 (u:

i+
r · ·- u · 1 ·) + · 2
J l- J ~L
t
Cos a/2 (-3w · · + w · .2 ·)
lJ i- J
x
(3.11)
.. Node one. space from the edge. At the deformable node ij located
one space from the edge) all strains except X are similar to those of a
x
typical interior node. The flexural strain· xxij is
2R L
x x
~os.a/2 (ui+lj - ui-lj) +
21
2 t
Cos ·a/2
( -2w ij + wi-2j._)
x (3 .12) .
Equilibrium equations. The equilibrium equations for points on
or near· the edge can now···be obtained by the principle ·of virtua·l displacement.
For example, to obtain the eq~ilibrium equation in. th.e y .d;frection at .point
ij+l loca·ted one:half .~p&ce from the edgey.. the ·point is given a vfrtual dis-
placement 6v .. . The internal work of the system is°' thus given .by
lJ'.1- -
LL,;• .
v
Wint = l~y 2
Eh x y.
2_.
[c EY i· •
J
+ VE .. )6E. . . + ( :L , . + VX .. )f:.X__ ..
XlJ .Y:!-,J -· ~YlJ . XlJ -yl,J
+ (E .. + VE .. )6E .. · + (X .. + VX .. )6x._ ..
YlJ+ 2 XlJ+ 2 YlJ+ 2 YlJ+ 2 XlJ+ 2 , . YlJ+ 2
1-~J
+ -2~ (Exyi-lj+l
6
Exyi-lj+l + Xxyi-lj+l 6 Xxyi-lj+l
·1 . 6 1 )'1 (3.13)
+ 2 Exyi+lj+l Exyi+lj+l + 2 Xxyi+lj+l 6.Xxyi+lj+lj
-23-
The coefficient 1/2 which appears in the last two terms of Eq. (3.13) is due
to the reduced width of the strip of the shell at the edge) Fig. 11. The
external work of the system is
(3 .14)
For equilibrium
v v
wex t + w.in t = 0 (3.15)
Substituting the appropriate strain-.displacement relations into Eq. (3.13) and
then substituting the result and Eq. (3.14) into Eq. (3.15), the desired
equilibrium equation.is obtained. Thus,
. L XLy [ ( 1 + t2 ) _2:__ ,r {/ . )
2 2 2 2 2 \ ·ij+3 - 2 vij+l + vij-1
- Cos f3/2 4R Cos f3/2 L
y y
2
1-v t 1 (1-v v
+·2(l+ 2 2 ) 2 (-3v .. l+V/. 2·1)+ - 2 +Cosa/2Cosf3/2
4R Cos f3/ 2 L lJ+ i - J+
y x
l~ ~2 1
+ 2 4R R Cos a/2 Cos f3/2) LL (ui+lj+2 - ui-lj+2 - ui+lj + ui-lj)
xy xy
( 1 v ) 1 ( ) (1-v ) d ( )
- Ry Cos f3/ 2 +·Rx Cos a/2 Cos f3/2 Ly wij+2 - wij - · LxLy Lx -w i-lj+l
2
t 1
+ ( 2 ) 3 ( w . . 4 - jw . . 2-~ 3w . . - w . . 2)
4R Cos f3/2 L lJ+ lJ+ lJ lJ-
y y
1-v 1 l+v 1 ) 1
+ -4R--C-os_f3_/_2 ( 2 Cos f3/2 + 2 Cos a/2 L 2L ( - 3w lJ+
. . 2 + w. 2 . 2
l- J+
y x y
2
l-v - 0
+ 3w.lJ. - w.l- 2;·J )] + -Eh y (3.16)
-24-
3.2.2. Hinge Support
The difference between the hinge support and the roller.support is
that the former cannot move in a direction perpendicular to the edge.
There~6re, the boundary conditions, when the edge parallel to the y axis is
hinge supported, are
u 0
v 0
w 0
M 0
x
Taking these conditions into account the strains at the deformable nodes on
or near the edge are obtained and the equilibrium equations for displacement
points near the edge are formulated. The procedure is similar to that of
roller support.
3.3. Free Edge
.l\ssuming that the edge parallel to the y axis is completely free,
the boundary conditions are
N 0
x
N 0
xy
(3.18)
M 0
x
R 0
x
where Rx is the classical reaction composed of the shear force Q and the
variation of the twisting moment M along the edge, i.e.

xy
R
x
(3.19)
-25-
Since the classical shell theory is used as a guide in the develop-
ment of the model, the reaction R must be defined at poi.nts along the edge
x
which best simulate a behavior similar to the continuous shell" The most
logical points are the points of intersections of rigid bars along the edge.
For example, the reaction at point ij+l, Fig. 12J is composed of the shear
i
force Q .. which results from the bending moment M . . ; and Q . . which
XJ.J+1 Xl- 1 J+1 XlJ+1
results from· the two adjacent twisting moments M .. and M . . .
xylJ xyJ.J+ 2
As was pointed out in Section 2.2; at the points of crossings of
the rigid bars; the two bars displace independently of each other in the
radial directiono Therefore, the forces Qv resulting from.the twisting
moments along the edge must be transferred to the ends of the bars intersecting
the edge. This is accomplished by means of auxiliary rigid bars (Figo 12)
which connect the nodes on the edge to the points on the edge where the bars
cross~ To eliminate any additi.onal twisting moment at the nodes one-half
space from the edge which may result from the displacements of nodes along
the edge, frictionless hinges are inserted at points of connections of
auxiliary rigid bars and the crossing bars.
By the use of Eqso (3.18),the strains of the deformable nodes on
or near the free edge can be easily formulated.
Node on the edgeo Since the strain-displacement expressions are
geometric relations, both the extensional and the flexural strains in the
y direction at the deformable node ij located on the edge are similar to those
of a typical interior node, .Eqs. (205 and 2013). The expressions for the
extensional and the flexural strains in the x direction at the deformable node
ij located on the edge are obtained from the first and the third of Eqs. (3018).
Thus_,
-26-
.E .. (3.20)
Xl.J
rl t t --,
Xxij -v 2R L Cos f3/ 2 ( v .. 1 - v .. 1) +
- y y lJ+ lJ- 2L
2
Cos f3/2
( wij+2 - 2w ij + wij-2) J
y
At the deformable node ij located on the edge; the shear strain is found from
the second of Eqs. (3.18) to be
E •• 0 (3.22)
xylJ
To obtain the twisting strai.n, we first have to consider the rotation of rigid
bars i-lj+l, ij+l and i-lj-1, ij-1. The location of these bars is as shown
in Fig. 13. Since at the crossing points on the edge, the two bars displace
in the radial direction independently of each other, additional unknowns,
5 ',s (radial displacements of the ends of the crossing bars); are introduced.
Additional equations necessary to solve for the extra unknown o's are formulated
later from the equilibrium of a portion of the model near the edge. With the
extra unknown 5 1 s, the rotation of bar i-lj+l, ij+l in the x direction is
(see Fig. 14a)
1 1 L
8
xij+l = r::J2. (oij+l - Cos a/2 wi-lj+l + 2R C~s a/2 uij+l)
x x
Similarly the rotation of bar ij-l in the x direction is
L
1 1 ·x
eXlJ-
.. 1 r::J2. (oij-1 - Cos a/2 wi-lj-1 + 2R Cos a/2 uij-1) (3.24)
x x
The rotation of rigid bar i-2j, ij (which is equal to the y rotation of bar
i-lj-1, i-lj+l) in the y direction is

-27-
The rotations of the two auxiliary rigid bars ij-1; ij and ijJ ij+l in the
y direction are
L
1 1
e~. =~(Cos f3/2 wij +
2R
y
Clos f3/2 v ij - 5 .. 1)
lJ-
lJ y y
(3.26)
L
1 1 y
e~. = ~ (oij+l - Cos f3/ 2 wij + 2R cos f3/2 vij)
lJ y y
respectively. The displacement v .. of the node ij in they direction is equal

lJ
to v. .. The average rotation at the node ij is
l-1 J
_!
8 ··· ·· -2 ( 8 · ·
~J . lJ
1
+
2 _
el· J· ) -
.l
Ly [.,. Ry
·y
cL0 s f3/ 2 v l· J· +. ( 0 l· J+
· 1
_
5 · · 1)
l J- .
J
The twisting strain is defined by
1 t 1 t
xY\T i· J. = -L ( e · · 1 - e · · 1) -2 + Ltn2 ( e · · - e · 1 · ) -2 (3.28)
~~ y XlJ+ XlJ- Dx/c YlJ yi- J
Substituting Eqs. (3.23; 3.24; 3.25; and 3o27) into Eq. (3.28) yields
t (u ) t ( l
2R L Cos a/2 -__ ij+l - uij-1 + 2L L ·cos a/2
x y x y
1 2
+ COS f3/2)(..- 2w.l- l"J+ 1+2w.l- l"J- l)+L Lt .. 1 -
(b lJ+ olJ-
.. 1) (3.29)
X y
Node one-half space from the edge. At the deformable node ij
located one-half space from the edge; all strains except xx are similar to
those of a typical interior node" The flexural strain xXlJ
.. is obtained from
expressions similar to Eqs. (2.10, 2.11 and 3.23). Thus,

-28-
2R L
x x
~os a/2 (ui+lj - ui-lj) +
21
2 t
Cos
a/
2
(-3Wij + wi-2j)
x
t
+ L- 2 o.i+l"J
x
Considering the equilibrium of a T section near the edge, Figo 14b,
the following relations can be written~
1
= ~ Mxi-lj+l
x
i 1
Qlxij+l = ~ Mxyij
y
I 1
Q2xij+l = r;J2.
y
Mxyij+2
where
Eh Lv t
-- .-JL ~ (x V L )
MXl-
. i·J+ 1 - 2 .2 2 xi-lj+l - yi-lj+l
1-v
L (3.32)
Eh ·y t
M ..
xyiJ 2(1+v) ~ 2 xxyij
L
Eh y t
MxyiJ+
.. 2 2(1+v) ""2"" 2 xxyij+2
For the equilibrium in vertical direction
I i
0 (3.33)
Qxij+l + Q2xij+l - Qlxij+l
Substituting Eqs. (3032) into Eqs. (3o3l) and then substituting the results
into Eqo (3. 33 ), we obtain
L 1-v
LY (xxi-lj+l + V"Xyi-lj+l) + 2 (xxyij+2 - Xxyij) 0 (3.34)
x
-29-
Equation (3.34), after substitution for strains, yi.elds the additional
equilibrium equation at the free edge. Thus,
L
Ly
·x
[
R L
x x
~OS a/2 (uij+l - ui-2j+l) +
L
2
l
Cos a/ 2 (-3wi-lj+l + wi- 3j+l)
x
+
1
l
x2
r· 5
\:? ij+l
) v
+Ry\ Cos fJ/2
(
vi-lj+2 - vi-lj) +
,
t/ v
Cos f'>/ 2
(
wi-lj+3
- 2wi-lj+l + wi-lj-1)] + l;v [RxLy ~os a/2 (uij+3 - 2uij+l + uij-1)
l 1 1 l
+ R·L Cos ~/2 (- 2vi-lj+2 + 2vi-lj) +LL (Cos a/2 +Cos ~72)(-2wi-lj+3
y x x y
+ 4w. 1 , 1 - 2w. 1 . 1 ) + L
.l- J+ l- J-
{ 45. ,
lJ+ 3
~L
- 85.lJ+
. 1 + 45.lJ-
. 1 )J 0
x y
Nodes one space from the edgeo At the deformable node ij located
one space from the edge, all strains are similar to those of a typical
interior node.
Equilibrium equations o The procedure for obtaining the equilibrium
equations for points on or near the edge is similar to that of Section 3.2.
For example, to obtain the equilibrium equation in the z direction at point ij
on the free edge, the point is given a virtual displacement 6w .. a The

lJ
internal work of the system is thus given by
w.wint
-30-
1 1
+ - (E • • + V E • • ) l::,.E • • + - ( X__ . . + 'VX . • ) 6;x__ . ,
2 YlJ XlJ YlJ 2 .. -YlJ XlJ . -yiJ
1-v (
+ ~2 . 1°
Exyi- J- 1
6E
xyi-.l"J- 1 + Xx-,yi- 1°J- 1 l::,.X xyi- l"J- 1
0 0 0
1 E.
+ ~
2 .xylJ
•• 6E .. )
xylJ
J (3-36)
The external work of the system is
For equilibrium
Substituting the appropriate strain-displacement relations into Eq. (3a36) and
then substituting the result and Eqa (3a37) into Eqa (3a38)) the desired
equilibrium equation is obtained a Thus,
LL ,... 2 2
x2 Y 1·_- 1~2v R 1 1 (V 'l - 00
V ••
, 1-v
l) + ~2~ 2 1 2 Wl J
0 0
Cos 2 ~/2 L y lJT lJ- R · Cos ~/2

y y
-31=
+ v l- · 1 + v_ lJ=
· 2 J+ · · 1 - v l=
· 2 J-
· 1)
2 1 2
4 + 6 4 + ) + l=v L (Cos 1 a/2 + )
- wij+2 wij - · wi.j=2 wij-4 2 4. Cos f3/2
1
L 2L 2
( -w l J +2 + 2w lJ
0 0
. . 2 + w i. - 2 J+
. . - w lJ- . 2 - 2w i~
. 2 J. + w.•i - 2 J-
. 2)
x y
3.4. Other Boundary Conditions
The procedure for obtaining other boundary conditions is similar
to that of the simply supported and free case. Fbr a shell clamped along
the edge, all displacements and slopes along the edge vanish. For a shell
continuous across a diaphragm support J the displacements vani.sh while the
slope vanishes only if symmetrical geometry. and loading conditions exist.

NUMERICAL RESULTS
4 .1. General
As was mentioned previously the objective of this study is the
development of a discrete model for the analysis of shallow shells of double
curvatureo It is not within the scope of this study to consider the effect
of various parameters on the behavior of a multitude of shellso Solutions
for a variety of shells with different boundary conditions are presented.
One.of the advantages of the discrete model is that the various types of
loading can easily be.handled. Several examples of shells subjected to uniform
loads, sinusoidally varying· distributed loads:; and sinusoidally varying edge
loads are given. Some solutions for shells having different values of Poisson's
ratio are also presented. The obtained results are compared with existing
solutions whenever possible to demonstrate the applicability of the model to
a variety of shell problemso
. Two types of shells are of particular interest as test problems for
the model. One is a shell of negative Gaussian curvature.? the hyperbolic
paraboloid bounded by characteristic·· lines of the surface. For this shell
the coordinate lines do not coincide with the lines of principal curvatureo
The external loads are transmitted to the supports mainly by the in-plane
shear forceso The other shell of particular interest is a shell of positive
Gaussian curvature'} the elliptic· paraboloid with two opposite edges simply
supported and the other two edges freea For this shell the magnitude of N
y
forces (edges parallel to the y-axis being free) varies very :rapidly across
a section normal to the free edgeso If the model can predict the behavior
of the above two shells 3 then it can be expected to give good estimates of
-32=
magnitude and distribution of boundary disturbances for general shell
problems.
Except for the hyperbolic paraboloid bounded ·by the characteristic
·lines of the surface J the shells considered herein have both s;~nnmetrical
geometry and loading. ThereforeJ by using only one quadrant of each shell
in the analysisJ one could reduce the number of unknown displacements to
one-fourth of the original number. The spacing between the two extensional
elements, t, is obtained by equating the bending stiffness of the model to
that of the real materiala Thus,
2
12(1-V )
which gives
h
t = -
0
For flat plates the equilibrium equation in the z direction,
Kea.. (A-3) is independent of the other two equilibrium equations J Eqs. (A-l
and A-2). Furthermore, the set of equ.ilibriumequations in the z direction
reduces to two sets of completely. uncoupled equations. One set contains the
w displacements of the deformable nodes marked as solid circles (see Figo 15).
The other set contains the w displacements of the deformable nodes marked as
hollow circles. Although, the two systems act independently of each other,
the solutions for the two sets were found to be in good agreement with one
another. For example, the plot of w displacements along the diagonal of a
rectangular plate is given in.Fig. 16. Alternate nodes belong to one systemJ
the remaining nodes to the othera The agreement between the two systemsJ as
-34-
seen from the figure, is excellent. For shells whose coordinate lines
coincide with the lines of principal curvature, such as cylindrical shells,
elliptic paraboloid shells, and saddle shaped hyperbolic paraboloid shells,
the two sets of equations a~e also completely uncoupled fTom each other .
. For shells whose coordin~te lines do not coincide with the lines of principal
curvature, such as hyperbolic paraboloid bounded by the characteristic lines
of the surface, the two sets of equations are strongly coupled through u,
v, and w displacements.
4.2. All Edges Simply· Supported
4.2.1 . . Uniformly Loaded Rectangular· Plate
TWo rectangular plates of .same dimensions, uniformly loaded and
having a Poisson 1 s ratio of 0.3 are considered. The dimensions of the plates
are given in Table 1. Because of symmetry, only one quadrant of each plate
is used in the analysis . . In tpe first plate, the quadrant is divided into
three spacings in each direction . . In the second plate the number of divisions
in each direction is doubled . . The values of the deflections and the bending
moments at the center of the two plates are given in Table 1. The solutions
are in good agreement with those given by Timoshenko and Woinowski-Krieger [18].
Reasonable accuracy is obtained by using a 3 x 3 grid. . By increasing the
number of divisions in each direction, the values of the moments are slightly
improved.
4.2.2. ·Uniformly Loaded Cylindrical Shell
A cylindrical shell with dimensions corresponding to those given
by Bouma [3] is considered. The shell is u,niformly. loaded and has a Poisson's
-35-
ratio of O.Oo The dimensions of the shell are given in Fig. 17. Due to
symmetry only one quadrant of the shell is used in the analysis. The
quadrant is divided into six spacings in each direction. The plots of the
displacements w, the forces N ,and the bending moments M at the midsection

y x
of the shell are also given in Fig. 17. For a modulus of elasticity of
2 2
2 x io 5 kg/cm and a loading of 0.019 kg/cm , the- maximum positive and
negative values of w, N , and M at the midsection of the shell are shown in

y x
Table 2. The results are compared with those given by Bouma [3]. In general
the agreement is good. The small differences between the results could be
due to both the manner in which the two shells are supported along the edges
and the manner in which the two shells are loaded. The edges of the shell
in Ref. [3] are supported by vertical diaphragms and the loading normal to
the surface, is given by
Z(y) z1 Cos a y
1
(4.1)
where
4 q 1(
z1 1( al - b
Although the transverse edges of the shell in this example are supported by
vertical diaphragms, the longitudinal edges are supported by diaphragms which
are perpendicular to the surface. The loading is considered as the weight of
the shell. The most significant difference in the transverse bending moment,
M , is to be expected because of the manner of supporting the longitudinal

x
edges.
4.2.3o Uniformly LoadedElliptic·Paraboloid
The dimensions of the elliptic paraboloid used in this example are
given in Fig. 18. The dimensions of the shell in the x .direction are the
same as those of the cylindrical shell,. Fig. 17. The dimensions in the
y direction are so chosen that the curved length and the rise of the shell
are 1800 ems and 200 ems respectively. A 6 x 6 grid on a quadrant of the
shell is used in the analysis. The plots of the displacements w, the forces
N , and the bending moments M at the midsection of the shell are shown in
y x
Fig. 18.
Significant forces exist at the corners of this shell. The
in-plane shear force and the twisting moment are
N
xy
.:...2. 23 qa
2 2
M -0.069 x 10- qa
xy
respectively. These forces produce principal· stresses in the diagonal
directions which are as large or larger than those which exist in the center
regions of the shell.
The plots indicate that except near the edge the membrane theory
provides a good estimate of stresses for design purposes. Due to low values
of deflections, forces, &.'1d bending moments, this ,shell is very suitable for
roof construction.
4.2.4 .. Uniformly Loaded HyPerbolic,Paraboloid

The dimensions of the hYPerbolic paraboloid used in this example
are given in Fig. 19. The dimensions of the shell in the x direction are
kept the same as those of the cylindrical and the elliptic paraboloid shells,
·while the dimensions in they directions are so chosen that the curved length
~3T-
and the sag of the shell are 1800 ems and 200 ems respectivelyo A 6x6 grid
on a quadrant of the shell is used in the analysiso The plots of the dis-
placements w, the forces N , and the bending momentsM at the midsection of

y x
the shell are shown in Fig. l9o
The plots indicate that the membrane .theory can not be.used to
estimate the stresses. The magnitudes of deflections, forces;i and bending
moments for· this shell are much higher than those for an elliptic paraboloid
and a cylindrical shell of the same dimensions.
4.2.5. Uniformly Loaded Hyperbolic Paraboloid Bounded by Characteristic

Lines of the Surface
A s~uare hyperbolic paraboloid bounded by the characteristic lines
of the surface is used in this example. The shell has dimensions corresponding
to those used by Chetty and Tottenham [5]. The dimensions of the shell are
given in Fig. 20. The-Poissonvs ratio and the modulus of elasticity of the
6
shell are 0.16 and 3 x 10 psi respectivelyo The shell is subjected to a
2
uniform load of 50 lbs/ft . Since the shell is not symmetric, an 8x8 grid
on the complete shell is used in the analysiso The plots of deflection w, the
force N , and the bending moment M at the midsection of the shell as obtained
xy x
with the model together with the corresponding values obtained by Chetty and
Tottenham are shown in Fig. 20. The agreement between the deflections and
the shear forces is very good while the maximum bending moment obtained by
the model is slightly higher than the one in Refo [5].
As was mentioned previously, this shell is of particular interest
in the test of the model. The results indicate that the model can indeed be
used to study the behavior of shells of negative Gaussian curvature.

4.3. Two Opposite Edges Simply Sup£orted and the Remaining Two Edges Free
4.3.1. Uniformly Loaded Square Plate
A square plate uniformly loaded and having a Poisson 1 s ratio of 0.3
is considered. The dimensions of the plate are given in Table 3. A 5x5 grid
on a quadrant of the plate is used in the analysis. The values of the
def+ections and the bending moments at the center and at the free edge of the
plate are given in Table 3o The results are in good agreement with those
given by Timoshenko and Woinowsky·-Krieger [ 18].
4.3.2. Cylindrical Shell Subjected to Sinusoidally Varying Edge Load
A cylindrical shell with the transverse edges supported by diaphragms
and the longitudinal edges free is considered in this example. The dimensions.
of the shell are given in Figo 21. The Poisson es ratio and the modulus of
2 respectively.
elasticity of the shell are Ool5 and 4o32 x 10 5 kips/ft The
shell is loaded along its longitudinal edges by a loading of the type
:rr
Z(y) ·- Cos b y
A Quadrant of the shell is divided into 10 and 6 spacings in the x and y
directions respectively. The plots of displacements w, the forces N , and

y
the bending moments M at the midsection of the shell are shown in Fig. 21.
x
The analytical solution which was obtained by a Fourier series solution of
the differential eQuation is taken from Parikh [15]. The fact that the
results are in good agreement with the analytical solution indicates that the
model can be used to study the effect of edge disturbances in shells . . Parikh 7 s
model, which is an extension-of Hrennikoff 1 s framework [12] to cylindrical
shells, does not demonstrate the same accuracy as the model presented in this
the displacements w, the forces N • and the bending moments M at the mid-
Y' x
section of the shell are also shown in Figo 220 The plots of the in-plane
shear forces N and the forces N at the support for the two shells are
xy Y
given in Fig. 23.
The magnitudes of w, N , and M at the midsection of the shell

y x
having hinged support are lower than those of the shell having roller support.
The magnitudes of N and N at the hinged support are much higher than those
xy y
at the roller support. This is to be expected in view of the fact that the
shell now behaves in a manner like a fixed ended beam. However, as pointed
out below computations based on considering the shell as a fixed beam are not
adequate.
A comparison of the hinged support case with the stresses that are
obtained by multiplying the N stresses for the roller case by the ratio of
y
the bending moment in a fixed ended beam to the center moment in a simply
supported beam is not conservative. The error also may not be insignificant.
Furthermore, the· substantial modification of the N stress distriabution is

xy
not predicted by such an approximationo
The equilibrium of the model is checked by considering one-half of
the shell and applying the usual equilibrium equations. The summations of
the forces in. the horizontal and the vertical directions vanish and the
summation of the moments about any axis is found to be zero, ioe. the internal
2
forces produce moments equal and opposite to the simple beam moment l/8 wL .
4.3.4. Elliptic Paraboloid Subjected to Sinusoidally Varying Load
The dimensions of the elliptic paraboloid shells used in this
example are .similar to those of the simply supported case and are given in
~42-
reaction of 0.95 qa which is l.7 percent of the N force at the free edge.
y
Reducing the value of the N force at the free edg~ obtained from the model)
y
by l.7 percent gives a N equal to 2l02 kg/cm. As was mentioned previously,
y
this shell is of particular interest in the test of the model. The results
indicate that the model can be used to predict the rapid variation in forces
and bending moments.
Due to large values of deflections, forces, and bending moments
and also due to rapid variation of N forces, this .shell is not an ideal
y
shell for construction.
4. 3. 5. Hyperbolic Paraboloid Subjected to Sinusoidally Varying Load
The dimensions of the hyperbolic paraboloid considered in this
example are similar to those of the simply supported case, and are given in
Fig. 27. The shell has a PoissonYs ratio of 0.0 and is subjected to a
loading of the type
rr
Z(y) q Cos by
A 12x6 grid on a quadrant of the shell is used in the analysis. The plots
of the displacements w, the forces N , and the bending moments M at the

y x
midsection of the shell are given in Fig. 27. The magnitude of the displace-
ment at the free edge is very amall which indicates a behavior somewhat
similar to that of the simply supported case.

CONCLUSIONS AND RECOMMENDATIONS FOR
FURTHER STUDIES
5.1. Conclusions
The discrete model developed in this study can be used to analyze
a variety of shallow shellso The model gives good estimates of the magnitude
and the distribution of deflections and stresses in shells with different
edge conditions, different types of loading, and different values of·Poisson s· 1
ratio" The comparisons of results indicate that the model can be used to
study the behavior of shells of different Gaussian curvatures" It can also
be used to study the effect of edge disturbances in shells" Rapid variations
in forces and moments through the shell can be predicted by the model,
however, for·better results larger number of spacings should be used . . In
general, the obtained results are sufficiently accurate for design purposes.
In view of the construction of the deformable nodes and the
existance of constant forces across each node, an approximate analysis which
includes the plastic behavior is possiblea Partial loadings and the exten-
sion of the·model to include elastic supports, columns or point supports
along the edge, different material properties at different nodes, and shells
with variable thickness should not present any difficulty"
For elastic shells whose coordinate lines coincide with the lines
of principal curvature, the model can be redaced to the on~ in Refo [16]"
The extensional and the flexural behavior are then defined at the nodes of
every other row, while the shear and the twisting behavior are defined at
the remaining nodes. Radial displacements are specified at the extensional
nodes, and only one tangential displacement is needed at ea'ch intersection
-43-
-44-
of two rigid bars. ·The number of equations are thus reduced to one-half,
permitting an increase in the number of spacings in the two directions.
Although the method of analysis is straightforward, the equilibrium
equations governing the behavior of the model are very complex. An efficient
method of computation which eliminates the possibility of human errors is to
generate and solve the equilibrium equations within a digital computer. With
such a computer program for the model it is possible to ecconomically investi-
gate a variety of loading cases and support conditions which are encountered
in the construction of shell roofs.
5.2. Recommendations for Further Studies
In future studies, the behavior of a portion of the model subjected
to a complete set of edge disturbances should be investigated. The analysis
of non-shallow shells, continuous shells, different combinations of hyperbolic
paraboloid units which are bounded by the characteristic lines of the surface,
and finally arch dams can then proceed by proper combinations of various
portions.
The model should be extended to include elastic supports, so that
the behavior of shells with edge beams or supported by colurnns could be
investigated.
The governing equations of the model should be modified to include
generalized a's and ~·s. Solutions of shells with variable curvatures such
as elliptic and parabolic cylinders and conoids can then be easily obtained.
The model should be generalized to non-orthogonal coordinates, so that shells
which cover a non-rectangular planforms could be considered.
Finally, the application of the model to non-linear problems in
shells should be investigated.

BIBLIOGRAPHY
..1. Ang, A. H. S. and Newmqtrk, N. M., 11A Numerical Procedure for the
Analysis of ContinuousjPlates, 11 Proceedings of the Second American
Society of Civil Engineers Conference on Electronic Computation,
Pittsburg, Pennsylvania, September 1960.
2. Benard, E. F.,. 11 A Study of the Relationship Between Lattice and

Continuous Structures, 11 Ph.D. Thesis, . University of Illinois,
September 1965.
3. Bouma, A. L., 11 Some Applications of the Bending Theory Regarding Doubly

Curved Shells," Proceedings of the Symposium on the Theory of Thin
Elastic Shells, August 1959.
4. Bouma, A. L., "On Approximate Methods of Shell Analysis: A General

Survey," Proceedings of World Conference on Shell Structure,
San Francisco, California, October 1962.
5. Chetty, S. M. K. and Tottenham, H., 11An Investigation into the Bending

Analysis of Hyperbolic Paraboloid Shells," Indian Concrete Journal,
July 1964.
6. Chuang,.K. P. and Veletsos, A. S., "A Study of Two Approximate Methods

of Analyzing Cylindrical Shell Roofs,rr Civil Engineering Studies,
Structural Research Series No. 258, University of Illinois, October 1962.
7P Clough, R.· W., "The Finite Element Method in Plane Stress Analysis, 11
Proceedings of the Second American Society of Civil Engineers Conference
on Electronic Computation, Pittsburgh, Pennsylvania, September 1960.
8. Clough, R.. w. and Tocher, J .. L., "Analysis of Thin Arch Dams by Finite
Element Method, 11 International Symposium on the Theory of Arch Dams,
Southampton University,. Pergamon Press, 1964.
9. Das Gupta, N. C., "Using Finite Difference Eq_uations to Find the Stresses
in Hypar Shells," Civil Engineering and Public Works Review, . February
196L
10. Das Gupta, N. C., "Edge Disturbances in a Hyperbolic Paraboloid," Civil
. Engineering and Public· Works Review, February 1963.
11. ·Design of Cylindrical Concrete Shell Roofs, American Society of Civil

Engineers, Manuals of Engineering Practice, No. 31, adopted 1951.
12. Hrennikoff, A. , 11 Solution of Problems in Elasticity by Framework Method,"

Journal of Applied Mechanics, Vol. 8, No. 4, December 1941.
-45-
-46-
13. Langhaar, H. L., "Foundation of Practical Shell Analysis," Department

of Theoretical and Applied Mechanics, University of Illinois, June 1964.
14. Noor, A. K., 11Analysis of Doubly Curved Shells, 11 Ph.D. Thesis,

University of Illinois, August 1963.
15. Parikh, K.· S., "Analysis of Shells Using Framework Analogy," Sc.D. Thesis,
Massachusetts Institute of Technology, June 1962.
16. Schnobrich,·W. C., 11A Physical Analogue for the Analysis of Cylindrical
Shells," Ph.D. Thesis, University of Illinois, June 1962"
17. Soare, M., Application des Equations aux Differences Finies au Calcul
des Coques, Editions de L'Academie de la Republique Populaire Roumaine,
Bucarest, 1962.
18. Timoshenko, S. and Woinowsky-Krieger, S.,. Theory of Plates and Shells,

McGraw-Hill Book Company, New York, 1949.
19. Yettram, A. L. and Husain, H. M., "Grid-Framework Method for· Plates in

Flexure," ·Proceedings of the American Society of Civil Engineers,
June 1965.
20. Zienkiewicz, 0. C. and Cheung, Y. K., "The Finite Element Method for
Analysis of Elastic,· Isotropic and Orthotropic Slabs," Proceedings of
the Institution of Civil Engineers, August 1964.
21. Zienkiewicz, 0. C. and Cheung, Y. K., 11 Finite Element Method of Analysis

of Arch Dam Shells and. Comparison with Finite Difference Procedures,"
International Symposium on the Theory of Arch Dams, Southampton
University, Pergamon Press, 1964.
-47-
TABLE 1. · SOLUTION OF SJMPLY SUPPORTED RECTANGULAR PLATE
t
I,
SS
a = 6.0"
PIN I
SS
h ::::;
v = 0,3
o. 03 11
LL ____ . ---'>--X
L~_J 2
x =0 y =0
Method
w M M
x y
4 2 2
qa /D qa qa
Model 3x3 .· 0.00563 0.0613 o. 0491

Model 6x6 0.00564 0.0623 0.0498
Ref. [18] 0.00564 0.0627 0.0501
-48-
· rrABLE 2. MAXIMUM POSITIVE AND NEGATIVE VALUES OF w, Ny , AND M

x
AT THE
MIDSECTION OF SIMPLY SUPPORTED CYLINDRICAL SHELL
2 2
E = 2x10 5 kg/cm Cl = O. 019 kg/cm
w cm N kg/cm M kg
' y ' x '
Pos. Neg. Pos. Neg. Pos. Neg.
Model 0.61 -0.22 No Pos. -67.8 153~0 -90.2

~
Value
Ref. [3] 0.60 Noti .
No Pos. -63.6 163.7 ~119.1
Given Value
~49-
TABLE 3. SOLUTION OF SQUARE PLATE, TWO OPPOSITE EDGES SIMPLY

SUPPORTED,, THE OTHER TWO EDGES FREE
1 SS
h =
v = 0.3
0.03"
11
mlC\l I Free
l i I
L~J~x
2
x = 0 ) y = 0 x = a/2 y =0
Method ·
w M M w M
x y y
4 qa 2 qa2 4 qa
2
qa /D qa /D
·-
. Model 5x5: 0.01320 0.0269 0.1225 0.01511 0.1309
Ref. [ 18] 0.01309 0.0271 . 0.1225 0.01509 0.1318

-50-
TABLE 4 .. MAXIMUM POSITIVE AND NEGATIVE VALUES OF w, NY' AND MxAT THE
MIDSECTION OF CYLINDRICAL SHELL, TWO OPPOSITE EDGES SIMPLY
SUPPORTED, THE OTHER TWO EDGES FREE
E = 2x10 5 kg/cm2 .q_ = 0.019 kg I cm2
w cm Ny ' kg/cm Mx kg
' '
Pos. Neg. . Pos. Neg. Pos. Neg.
Model 14.2 -2.3 858.7 --242.3 -- -498.2

Not
Ref. [3] 32.0 given 875.0 --225.0 -- -500.0
.,,
(l)
rg
~ ~
~ Q)
.-1
"'.'j
•l1 ~
•rl
a:
~0
~
().)
0
.
d
H
~
$'.:!
•rl
P-i
z
0
H
f;
u
~
0
0
t
H
0
I-;)
q
H
0
H
p::;
I'(\
0
.
H
µ.,
y
t Contribution Area
for Radial Loads
-----111- x
Concentration of Stiffness Contribution Area

at the Deformable Node for Tangential
Loads
FIG. 4 LOCATIONS OF DISPLACEMENTS AND LOADS

-55-
t ',I
,,......
'/
..
\I./
,,......
'~ -
\ /
'
,r-. 11'1. II'\ ,.I'\ , ..,.

j+2 ''-' './ \ l.J' '-1..1 J
,
(~
,.I'\ .~
j+l ""
\.1..1 '-IJ \1..1 'JJ ~D
,, 11'1. ,.,
j '"
\I/ ''-'
, r'\
\',I 'L-1 \V
j-1
( t'\
v ,. "'
\I/ "'"
\I.)
,. "
\I.I
, ....
\L.J \D
," ,
j-2 \I/
I I'\
\ 1..1
11'\
'IJ '"
'-!J \U
I
! I I
I I I '"
'I.)
,t-,
v (D
I
t I I
I I I , , ..,. ,.f,.
I ',I \I./ \l/
I I
I ,,
3 f '
\.I \',I - - - - l'I"\
\l.J' '"
-\:7
I")
/
,.,
2 ',I - - - - "'"
'I-' '"'
\l.I
,i-.
l/
, .... ,. ' ,,
1 ',I \.__,/ - - --- 11"\
....., -= ~
1 2 3 i-2 i i+l 1+2
FIG. 5 GRID POINT IDENTIFICATION

u.l - i J·
Cosa/2
FIG. 6 TRANSFORMATION OF u DISPLACEMENT

INTO TANGENTIAL PLANE
tan a/2
w. ~
1.u
L
sin a/2 = Lx/2 tan cx/2 =
x
H
:x
?R
x Cosc:d2
'
FIG. 7 EFFECT OF w DISPLACEMENT ON

E:<1"rENSIONAL STRAIN
-57-
FIG .. 8 EFPECT OF w DISPLACEMENT ON SHEAR

· STRAIN DUE TO THE TWIST OF THE ELEM:ENT
J ana
u.i+l"t.
2
wi+2j
displacement
Gosa/2
displacemen
exi+lJ. = Ll [2ui+lJ. tari a/2 +

x. .
Co~/2 ( w. +2 .-w .. )]
1 J lJ
FIG. 9 ROTATION OF RIGID BAR ij,i+2j
FIG. 10 EFFECT OF ROTATION OF RIGID BAR

ON FLEXURAL STRAIN
LL
xy
r-Area = -ir-
t
\..n.
\.0
(
· FIG. 11 SIMPLY SUPPORTED EDGE

_·owel Pin ··---
Frictionless Hinge
- - - - - Auxiliary Rigi.d Bars
;
CJ\
0
I
i
FIG. 12 LOCATION OF AUXILIARY RIGID BARS AT THE FREE EDGE

y~
/x
I
0\
t-1
&
FIG~ 13 FREE EDGE

-62-
a
2 u .. tan a/2
lJ+1
w.].- i·J+1
Cosa/2
FIG. 14a ROTATION OF BAR i-lj+l,ij+l
I
i
~ 71 Mxy. ij+2
l~I/
~I
I
FIG. 14b EQUILIBRIUM AT FREE EOOE

-63-
' / .
.....
..... ... ...
\,../
.... .....
'\
.d ..
't""\
,,,,,
..... ... .. ...

""
t"I'\ ,,, ~"\

\,/ '-V
'"'
FIG. 15 SEPARATION OF THE TWO NETWORKS

IN FLAT PLATES
Center Corner
o--.. . . . ~------------------~ ~~----------------------------~~:;m!lill.....o
0.001
0 .. 002 b....... ::;: L2

a
0.003
0.004
0,005
0.006
FIG. 16 DEFLECTIONS-ALONG DIAGONAL OF A UNIFORMLY

LOADED, STivll'LY SUPPORTED RECTANGULAR PLATE
-65-
-LO
q
..............
....:::t
ro -0 .. 5
01 y
....:::t
•0 0
i-1
x
....
~ 0.5
/o.642 z
LO
a = 1300 cm
b = 1800 cm
Rx = 1156 cm
-3 .. 0 h = 7 cm
-2.747 cp,_ = 68° 26'
.E-..
-2 .. 0 v = o.o
ro Uniforni Load
01
.... -LO
z~
LO
-LO
(\J /
a:l
011
(\J
I
0
r-1 0
x
1..0
FIG. 17 THE DISPLACEMENTS w, THE FORCES N , AND THE BENDING

MOMENTS M AT THE MIDSECTION OF AYCYLINDRICAL SHELL,
ALL EDG.J~SxSIMPLY SUPPORTED ,
-66-
-0.10
~
'- -0.05
..:::t
o:I x
01
..:::t
s
0 0
r-1
>< 0.,,041
....
~ 0.05
z
0.10
a = 1300 cm
b == 1713 cm
Rx = 1156 cm
-L5
Ry. = 1960 cm
-Ll42
h = 7 cm
-LO
CD = 68° 26 1
m x
0
·y = 51 50'
01
CD
"' -0.5
z» v ::: o.o
Unifo:rm Load
0
0 r:;
"' ~·
FIG. 18 'I'ID~ DISPLACEMENTS w, THE FORCES N.. , AND THE

BENDING MOMENTS Mx·AT THE MIDSECTtON OF AN ELLIPTIC
PABAEOLOID SHELL, ALL EDGES SIMPLY SUPPORTED
-67-
..
-50 y
t=l
~
...:::t
0
ro
a1
...:::t
I
0 50
r-1
z
><
.... 100
~
a :::: 1300 cm
150
b 1713 cm
cm
x = 1156'
R
-15.0
R =-1960 cm
y
h = 7 cm
-10.0
= 68° 26'
ro
01
°'x
eoy = 51° 50'
.... -5 .. 0
z» v = o.o
Uniform Load
0
5. 0
-10 .. 0
C\J ro -5.0
01
(\j
I
0 0
.-l
x
....
:;?::
x 5 .. 0
10."0
FIG. 19 THE DISPLACEMENTS w, THE FORCES Ny; AND THE

BENDING MOMENTS Mx AT THE MIDSECTION OF A HYPERBOLIC
PAHABOLOID SHELL, ALL EDGES SIMPLY SUPPORTED
y
Ref~ (5] ---

Model
0 z
..
s:l
•rl
....
0 .. 02 a = 180 in ..
~
b = 180 'in ..
0 .. 04 c = 36 in~
h i:: 2·cr5 in ..
E = 3x106 psi
-18C z = 50 psf
..
s:l v = 0.16
......_
-rl
.. -160
ti)
,0
r-1
....
-140
z~
-120
-20
_.
s:l"
.....•rl
0
c
-M
I
e
Ul
,0 20
r-1
....
>< 33,.2
~ 40
FIG. 20 THE DISPLACEMENTS w, THE FORCES N, , AND THE BENDING

MOMENTS Mx AT THE MIDSECTION OF AXSIMPLY SUPPORTED
HYPERBOLIC PARABOLOID SHELL BOUNDED BY CHARACTERISTICS
free
a = 25.88 ft ..
b = 50.00 ft.
R
x = 50 .. 00 ft ..
h = 0 .. 50 ft ..
Q) 30° 00'
x =
v ·- o· . 15
E = 4.32 x 105 ksf
1(
Z = Cos by
01
I
50
Cl
-6.o
+>
.
t"+-f
............
+> -4 . o
'+;
I
Cf)
p.
•rl
,.':I:! -2.0
"'
':t.><
0
FIG. 21 THE DISPLACEMENTS w, THE ·FORCES N , AND THE BENDING

MOMENTS M AT THE MIDSECTION OF AYCYLINDRICAL SHELL,
TWO EDGESxSIMPLY SUPPORTED AND THE REMAINING TWO EDGES
FREE
-70-
-l5o0
Roller
Htnged---
-10.0
q y
-=t~ -5$0
cd
01 -1 .. 89
; 0
0
r-1
x
.....
5. 0 free z
~
a = 1300 cm
10~0
......___11. 75 b = 1800 cm
15.0
Rx ~ 1156 cm
h = 7 cm
-20 .. 0 qJ
x.
::: 68° 26'
v = o.o
1(
-10 .. 0 Z ;;.:. q Cos by
10,0
20.0
-27a30
30.0
-1. 5
-1.218
(\j -LO
I
0
r-1
x -0.5
0.5
FIG. 22 THE DISPLACEMENTS w, THE FORCES N , AND THE BENDING
MOMENTS M. AT THE MIDSECTION OF AYCYLINDRICAL SHELL,
TWO EDGESXSIMPLY SUPPORTED AND THE ruMAINING TWO EOOES FREE
'-71-
Roller - - - -
Hinged---
-10 .. 0
a = 1300 cm
Edge Crown b = 1800 cm
Rx = 1156 cm
-35 .. 0 h = 7 cm
..,.___ -32 .11 c:p

x
= 68° 26'
-30.0 v = o. o
'\ I
Z = q Cos
rt
by
I
-25.0
'.\
\
I
-20.0
'\ I
\
-15.0
tU
a1
"'
'\
z~ \
-10.0 -
\
\
.
-5.0
0
\
\
\
\
\
Roller
,I>"'
A
/
/
..,,,.,.
---
\ /.
/
5~0
\ / I
\ ,_ ~ 8 .. 12
/
/
10 .. 0
FIG. 23 THE FORCES NX'l Ai.1\ffi N AT THE SUPPORTED EDGE

OF A CYLINDRltAL SHELL_, 'IWO EDGES SIMPLY
SUPPORTED AND THE REMAINING 1'-10 EDGES FREE
-72-
12 x6
6 x 6 ----
-50
~
..........._
...::r 0
Ctl
CJ' 43 . 9
...::r
B
0,......, 50
>cl
... a = 1300 cm
~
100
b = 1713 cm
Rx = 1156 cm
150
Ry = 1960 cm
h = 7 cm
-20
(f>x ::: 68° 26'
qJ,,. = 51° 50'
Ctl 0 .y
01 v = o.o
.... 1(
z» 20 z = q Cos b y
·:z;L7
40
56.8
60
-~·01
(\j
Ctl -6 .. 0
CJ'
(\J
I
0
,......, -4 .. o
><
....
>< -2 .. 0
~
FIG. 24 THE DISPLACEMENTS w, THE FORCES Ny, AND THE: BENDING

MOMENTS Mx AT THE MIDSECTION OF AN ELLIPTIC PARABOLOID
SHELL., TWO EDGES SIMPLY SUPPORTED AND THE REMAINING
TWO EDGES FREE .
-73-
Edge C own
-20
12 x6---
-12 . 8 _6x6---.
-10
0
l I
-
I
I
10
I
I
I
20
I
cd
01 I
"'
z:-,:..;
I
I
I
I
30
31.7
free
z
40
50
56. 8'
60
FIG. 25 THE FORCES Ny AT THE MIDSECTION OF AN ELLIPTIC

PARABOLOID SHELL, 'IWO EDGES SIMPLY SUPPORTED
AND THE REMAINING TWO EDGES FREE
80
-
70 I
~68 I
.................
.......
60
......... ,.... . - ._
....................
50
~
~r---_
ro r---_
01
"'
~»
40
- r-----__
~
30
20
10
0 0 0.004 0 .. 008 Oo0l2 0. 016 - 0.020 0.024
FIG. 26 THE PLOT OF THE FORCES N AT FREE EIX1E VS INVERSE OF THE .

SQUARE OF NUMBER OF SPACINGS FOR AN.ELLIPTIC PARABOLOID SHELL,
TWO EDGES .SIMPLY SUPPORTED AND THE REMAINING TWO EIGES FREE
-75-
A
.. 50
...........
...::t
ct1
01
0
...::t
I
0
r-1
>< 50
...
)
100
y
a = 1300 cm
b = 1713 cm
cm
x - 1156
R
R _ 1 nt:..A cm
= -.J...7uv
y
h = 7 cm
mx = 68° 26'
0
(j)y =·51 50'
v = o.o
i('
Z· =q Cos by
C\! 0
cd
01
(\J
I
0
r-1
5 .. 0
><
6.64
"'>< 10 .. 0
~
FIG. 27 THE DISPLACEMENTS w, THE FORCES ~r' AND

THE. BENDING MOMENT Mx AT THE MIDsECTION OF A
HYPERBOLIC PARABOLOID SHELL, 'IWO EDGES SIMPLY
SUPPORTED AND THE REMAINING TWO EOOES FREE
APPENDIX A
EQUILIBRIUM EQUATIONS
Alth011gh the eq_uilibriuill- eq_uations ·o;f the model are generated within
a digital computer, the eq_uilibrium eq_uations of typi_cal interior points are
presented here for referen.ce. ·
The equilibrium .eq_uation in the x direction at point i+lj is
LL
t
2 1
~ ( 1 +
2 [ 2 2 2 ) 2 (u.+3· .- 2 u. l" + ui-lJ.)
Cos a./2 4R Cos a./2 L l J ·. l+ J
x . x
2
1-v t 1 ·
+ - (1 + ) - (u - 2ul.+1J" + ui.+l·J·--2)
2 4R2 Cos2 a./2 L2, i+lj+2
x .Y
2
(1-v v l+v t
1
+ 2 + Cos a./2 Cos f3/2 + 2 4R R
.x y
Cos a./2 Cos f3/2) L L (vi+2j+l
x y
- v. 2. l - v .. 1 + v .. il - (_~ \ + R vI I ) L1 (w1·+2j-wiJ.)
i+ J- lJ+ 1J- _dos a./ 2 Y Cos a 2 Cos f3 2 x
x
2
v d (
1 - ) t 1
- Lr:-~ .wi+lJ"+l - wi+lJ"-1 + 2 3 (wi+4J" - 3~i+2J" + 3w. · - w. 2j)
X· y y 4R
x
Cos a/ 2 L
x
l J l- .
12
t (1-v · 1 l+v 1 1
+ 4Rx Cos a,/2 2.. Cos a/2 + 2 Cos f3/2) L L2 (wi+2j+2 -.2wi+2j + wi+2j-2
xy
2
- wl.J+
.. 2+. 2w l.J.. - w..
lJ -
2)] ~
1-v
Eh x 0
: (A-1)
The equilibrium equation in the y direction at point ,ij +l is
-76-
-77-
1 1 2
x y [ 1 t 1
-2- ( 2 + 2 2 ) 2 (v · ' 3 - 2 v · · 1 + v iJ' -1)
Cos ~/2 4R · Cos ~/2 L lJ+ lJ+
y y
2
(1-V V l+V t ) l
+ 2 + Cos a/2· Cos ~/2 + 2 4R R ·Cos a/2 Cos ~/2 . 1 1 (u.l+ l'J+ 2 - u.l- l'J+ 2
x y x y
2
1..,. v d t 1
- -L L ...--L ( w. 1 . 1 - w.. 1 . 1) + 2 -3 (w. · 4 - 3w · ' 2 + 3w · ·
l+ J+ i - J+ 4R Cos A/2 L lJ+ lJ+ lJ
xy x y f-" y
2
) t (1-V 1 l+V 1 ) 1 (
- wij-2 + 4R Cos ~/2 2 Cos ~/2 + 2 Cos a/2 2 wi+2j+2 - 2wij+2
y 1 x1 y
2
+ W.
. l- 2'J+ 2 - W. .. -wl..- J
l+ 2'J + 2wlJ
2.)] +~
1-V
y 0
(A-2)
The equilibrium equation in the z direction at point ij is
2 2
l - v d 1 "". v 2d t 1
- r:;r;- I:- (vi+lJ. - vi-lJ.) + r:;r;-L 1 wiJ. + 2 -3· (u. 3·
x y x x y x y 4.R Cos a/2 L i+ J
x x
-78-
2
t 1
- 3u. . + 3u. . ... u. .) + - . ( v. . - 3v. . 1 + 3v. . 1 - v. . )
i+ 1 J i- 1 J l- 3J 4R Cos 2 ~/2 L3 lJ+ 3 lJ+ lJ- lJ- 3
y y
+
t!2 v I
4Rx Cos a/2 IJ;os ~/2 +
1-V
2
1
(Cos a/2 + Cos
1
~/2)
J 1
L2 (ui+lj+2 - 2ui+lj
1
x y
+ ui+lj-2 - ui-lj+2 + 2ui-lj - · ui-lj ,...2) + 4R ·· Cos

y
t2
~/2 GCos va/2
+ l;V (co/a72 + co/1372)] L~L (vi+2j+l - 2 vij+l + vi-2j+l - vi+2j-l

x y
2
t l
+ 2v. .
1
,... v .. . ) +
2 1 2
=1+ (w. 4 . - 4w. 2 . + 6w. . - 4 w. 2 .
lJ- i- J- 4 Cos a/2 · L i+ J i+ J lJ i- J
x
2
t l
+ w. .) + . ~ ( w. . - 4w. . + 6w. . - 4 w ... 2 + w. . 4 )
l- 4 J 4 Cos 2 ~/2 L~ lJ+ 4 lJ+ 2 lJ lJ- lJ-
y
2 2]
t 1- 2V l-V l l .
+ 1+" i_Cos a/2 Cos ~/2 + 2 (Cos a/2 + Cos ~/2) _ 2 2 (wi+2j+2 - 2wi+2j
1 1
x y
· 2 + 4 w.lJ· - 2w.lJ-
+ wi·+2J·-2 - 2w.lJ+ · 2 + w.l- 2·J+ 2 - 2w.l- 2·J + w.l- 2·J- 2)}
2
l-V
- - -· z
Eh
0 (A-3)
APPENDIX B
A GENERAL DISCUSSION OF THE COMPUTER PROGRAM
To make greater use of the storage capacity of the computer,
the entire process of generating and solving the set of equilibrium equations
is accomplished by three sequential program.so
· In the first program the shell dimensiqns and properties are read
to the computero The equilibrium equations are then generated and stored
on an auxiliary tape o . The strain·-displacement relations corresponding to a
given set of boundary conditions are made available to the computer by means
of three FUNCTIONS. The loading is computed by a SUBROUTINE. By changing
the FUNCTIONS or the SUBROUTINE, the same program can be used for a variety
of boundary conditions or different loading combinationso In order to store
the equilibrium equations in the least numbers of locations, the grid points
are numbered in a one-dimensional triangular pattern. Furthermore, only the
coefficients of displacements between the first and the last non-zero columns
in each equation are storedo When a displacement component is zero, the
point is not given a virtual displacement, thus no equation is generated at
that pointo
In the second program, the equilibrium equations are read back from
the auxiliary tape to the computer one at a time. Gauss elimination method
is used to reduce the matrix of coefficients to an upper triangular matrix.
The solution is obtained by back substitution, starting with the last equation
in the matrix and the results are then stored on an auxiliary tape. The
maximum number of equations that can be solved depends on the storage capacity
-79-
-80-
of the computer and the maximum.band width (the band width is defined as the
number of terms between the first and the last non~zero columns in an
equation) of the equations"
In the third program the results are read back from the auxiliary
tape to the computer. The strains, the forces, and the moments are computed
and the desired values are printed out.
A general flow diagram of the computer program is shown in Fig. B-1.

Start
Read Shell
Dimensions
& Properties
Number the Do
Grid points Elimination
Process
,---1
1 Strain I Compute
Generate FUNCTIONS Strains, Forces,
-· - _ _J
nn No and Moments
All Equations?
Equation - ----,
I Loading I
L GUBl~OUTINE _J
I
CJ
Print !-'
Back Results
Substitute
out
Are All Equations

Generated ?
Call Second Call Third

Program Program
FIG. B-1 GENERAL FL0\"1 uIAGRAM OF THE COMPUTER PROGRl\M

Missouri University of Science and Technology
Scholars' Mine
Wei-Wen Yu Center for Cold-Formed Steel
Center for Cold-Formed Steel Structures Library
Structures
9-1-1971
The analysis and behavior of thin-steel hyperbolic

paraboloid shells
Peter Gergely
P. V. Banavalkar
J. E. Parker
Follow this and additional works at: http://scholarsmine.mst.edu/ccfss-library

Part of the Structural Engineering Commons
Recommended Citation
Gergely, Peter; Banavalkar, P. V.; and Parker, J. E., "The analysis and behavior of thin-steel hyperbolic paraboloid shells" (1971). Center
for Cold-Formed Steel Structures Library. Paper 19.
http://scholarsmine.mst.edu/ccfss-library/19
This Technical Report is brought to you for free and open access by the Wei-Wen Yu Center for Cold-Formed Steel Structures at Scholars' Mine. It has
been accepted for inclusion in Center for Cold-Formed Steel Structures Library by an authorized administrator of Scholars' Mine. For more
information, please contact weaverjr@mst.edu.
Department of StructuTal Engineering
School of Civil and Environmental Engineering
Cornell University
Report No. 338
THE ANALYSIS AND BEHAVIOR OF THIN-STEEL

by
Peter Gergely
P. V. Banavalkar
and
J. E. Parker
A Research Project Sponsored by the

American Iron and Steel Institute
Ithaca, New York September 1971

INDEX
Page
ABSTRACT i
NOMENCLATURE 1
I. INTRODUCTION 7
1. HYPAR ROOFS 7
2. LITERATURE REVIEW 9
3. SCOPE AND OBJECTIVES OF THE

PRESENT INVESTIGATION 16
4. ACKNOWLEDGMENTS 17a
II. DECK PROPERTIES 18
1. INTRODUCTION 18
2. ELASTIC PROPERTIES OF DECK 19
A. !·AEi'.1BRANE PROPERTIES 20
B. BENDING PROPERTIES 22
C. ELASTIC PROPERTIES FOR ARBITRARILY
ORIENTED CO-ORDINATE AXES 24
3. SHEAR RIGIDITY OF ORT~OTROPIC DECKS 25

4. DETERMINATION OF THE EFFECTIVE SHEAR
RIGIDITY FOR ORTHOTROPIC HYPARS 28
III. FINITE ELEMENT METHOD FOR STIFFNESS AND

STRESS ANALYSIS 31
1. INTRODUCTION 31
2. ELEMENT STIFFNESS 33
A. DISPLACEMENT FUNCTIONS 33
B. CURVED ELEirnNT (HETHOD "a") 36
C. FLAT ELEMENT (METHOD "b") 40
D. EDGE MEMBER 43
E. ECCENTRICALLY CONNECTED NON-

COMPATIBLE SUPPORTING EDGE MEMBER 45
F. ELASTIC SPRINGS 47
Page
3. lvIASTER STIFFNESS MATRIX 47
4. LOADING 53
6. SOLUTIONS OF EQUATIONS 56
7. STRESS ANALYSIS 57
A. DECK STRESSES 58
B. BEAM STRESSES 59
IV. GENERAL COI.1PARATIVE STUDY 61
1. INTRODUCTION 61
2. COivIPARISON OF METHOD 'a' WITH OTHER

NUMERICAL SOLUTIONS 62
3. COMPARISON OF METHOD 'a' AND METHOD 'b' 63
4. DISCUSSION OF METHOD 'a' 67
5. COMPARISON OF ANALYTICAL AND

EXPERIMENTAL WORK 71
A. EDGE-SUPPORTED HYPARS 72
B. CONCRETE UMBRELLA SHELLS 75
C. UMBRELLA SHELLS WITH STANDARD
CORRUGATED DECKS 79
FLEXIBLE EDGE t1E:·1BERS 81
STIFF EDGE lIBifBERS 86
6. SUMMARY 91
V. QUALITATIVE EFFECTS OF PRINCIPAL
VARIABLES ON BEHAVIOR OF HYPARS 92
\
1. INTRODUCTION 92
2. GEOMETRICAL SHAPE 92
3. DECK PROPERTIES 94
A. SHEAR RIGIDITY 94
B. THICKNESS OF THE CORRUGATED DECK 95
C. NUMBER OF DECKS 96
A. EDGE ME:MBER PROPERTIES 98

B. EDGE MEMBER AND DECK CONNECTION 99
5. LOADING 100
A. EDGE MEilBER WEIGHT 100
VI. INSTABILITY ANALYSIS OF HYPARS 104
1. INTRODUCTION 104
2. INCREMENTAL MATRIX FOR DECK AND EDGE HHMBERS 107
3. CHECKING OF THE INCREMENTAL MATRICES 108
4. INSTABILITY OF HYPARS Ill
S. DETERMINATION OF THE BUCKLING LOAD 112
6. NUMERICAL RESULTS AND THEIR DISCUSSION 114
7. DECK BUCKLING BY THE ENERGY HETHOD 120
8. STABILITY OF ISOLATED EDGE MEMBERS 123
VII. EXPERIMENTAL INVESTIGATION 126
1. INTRODUCTION 126
2. FLAT SHEAR TESTS 127
3. SADDLE-SHAPED HYPAR TESTS 131
4. NEDIUM-SCALE INVERTED UMBRELLAS WITH
VERY STIFF EDGE BEAMS 134
5. INVERTED UMBRELLAS WITH VERY FLEXIBLE
EDGE BEAHS 135
6. DISCUSSION OF EXPERUIENTAL RESULTS 137
VIII. SUMMARY, CONCLUSIONS AND RECOMMENDATIONS 145
BIBLIOGRAPHY 152
APPENDIX A. DECK PROPERTIES 158
Page
APPENDIX B. STIFFNESS MATRICES 166
APPENDIX c. DECK STRESSES 172
APPENDIX D. INCREMENTAL MATRIX FOR DECK 175

APPENDIX E. STIFFNESS AND INCREViENTAL MATRICES FOR
A FULLY cmIPATIBLE ECCENTRIC STIFFENER 179
ABSTRACT
An extensive experimental and analytical investigation
of thin-steel hyperbolic paraboloid (hypar) structures was car-
ried out to provide design information. As a result of this
work, empirical data is provided regarding the behavior of
such structures and computer programs are presented for the
analysis of thin steel hypar structures.
Hyperbolic paraboloid structures possess a unique combina-
tion of structural and architectural properties; some of them
are the following: 1) Due to the double curvature of the sur-
face the internal stresses in the deck are generally low and
the deflections are small. 2) Since a hypar surface can be
generated by straight lines, thin-steel or light-gage panels
may be used to form the shell; furthermore such panels are well
suited to carry the in-plane shear forces in hypar shells.
3) Basic hypar units can be combined in a large variety of ways
to produce attractive roofs (Fig. 1-2, page 212). 4) The dead
load to live load ratio is very low in the case of thin-steel
shell structures.
A hypar unit is a warped surface bounded by straight lines
(Fig. 1-1, page 211). The equation of the surface is z = Cxy/AB.
According to the simple membrane theory, a uniform load p pro-
duces pure shear forces Nxy = AB~/2C. This membrane shear
transmits uniform eccentric axial forces to the edge members.
The following are the major problems associated with the
design of thin-steel hypar structures: 1) The deflections,
i
stresses, and the stability of hypars depends greatly on the
shear rigidity of the thin-steel deck. This property must be
evaluated experimentally for each combination of decking, con-
nections to edge members, and seam connections. Furthermore,
in the case of hypars the deck is warped and thus the shear
rigidity may be different from that of an equivalent flat dia-
phragm. 2) The deck may buckle due to the shear stresses. and
the buckling load must be evaluated for highly orthotropic
shells. 3) The design of thin-steel hypar structures is gen·
erally governed by stiffness (deflections or buckling) require-

ments. The evaluation of the deflections is a very complex mat-
ter because it depends on the deck rigidity, the edge member
axial and bending stiffnesses, and on the eccentricity of the
deck-to-edge member connection. 4) If the curvature (or rise-
to-span ratio) of a hypar is small, the deflections may be
large and a considerable portion of the load is carried by bend-
ing rather than by membrane shear. 5) Partial or concentrated
loads may cause large local deflections, especially if single-

layer decks are used.
The present investigation studied all the above-mentioned
factors. The experimental and the analytical studies are sum-
marized briefly in the following paragraphs.
The ~~Eerimental investigatiQD consisted of four types of
tests: a) Four medium-scale (12 ft by 12 ft in plan) inverted
umbrella tests to study the stresses, deflections, and the
deck buckling; b) Test on a small-scale (2 ft by 2 ft) inverted
umbrella structure to study scaling effects and the overall
ii
buckling of hypars; c) Sixteen flat shear tests to determine
the shear rigidity of the decks used in the hypar tests; d)
Twelve saddle-s~aped hypar tests (5 ft by 5 ft in plan) with
various rise-to-span ratios to evaluate the effect of rise or
warping on the shear rigidity and to study other factors such
as partial loading and single versus double layered decks.
Photos of the various types of tests are shown in Figs. 7.1 to
7.5. The experimental program is described in detail in Chap-
ter VII.
Prior to the main test program, several small-scale (2 ft
by 2 ft) four-quadrant tests and medium-scale single-quadrant
tests were also conducted. These tests were however discon-
tinued because of the severe scaling effects in the case of
the small-scale models and the violation of the symmetry condi-
tions in the case of single-quadrant experiments where the
neighboring quadrants were missing. Nevertheless, these tests
produced useful qualitative information and experience with
manufacturing and testing thin-steel hypar structures.
The edge members of the umbrella-type specimens were made
of tuoular members since this afforded easy connection of the
warped surface to the straight edees. The decking consisted
of single or double layers of standard corrugated panels. One
layer was connected to the edge ~enbers with sheet metal screws
at various spacings. The seam connections between the panels
were also made by means of sheet metal screws. In the case of
shells with two layers, the top layer was connected to the bot-
tom layer in a similar manner.
iii
The medium-scale umbrella models were loaded using air bags
under each of the four quadrants. The saddle-shaped hypars
were loaded with sand, whereas the small-scale models were load-
ed through loading pads and suspended weights.
The following are the principal conclusions of the experi-
mental part of this investigation:
The effective shear stiffness of the cold-formed deck and
the rise (or curvature) of the structure are the most important
factors influencing the behavior of hypars. For low shear stiff-
nesses and for small rise-to-span ratios the deflections may be
large, the bending stresses tend to increase relative to the
membrane stresses, and the possibility of deck buckling increases.
As in the case of flat shear diaphragms, the shear stiffness
depends strongly on the seam and edge connections.
The increase in shear stiffness due to the addition of a
second layer of deck was found to be only about 1/3 if the
second layer was connected only to the first layer and not dir-
ectly to the edge members. Similarly, the deflections of a
double-layered shell are more than half of those of a correspond-
ing single shell. If the two layers are interconnected with
sheet metal screws (on an 8 in. grid in the present saddle-
shaped hypar tests), the deflections are further reduced by
about 10 to 20%, depending on the rise ratio.
A particular problem of certain types of hypar structures
is the deflection of unsupported outside corners (see Fig. 1-2,
page 212). The membrane shear cannot carry the load over such
flat corners and thus considerable bending and deflections may
develop. The tests showed that the bending stiffness of the
iv
edge members has a great effect on the corner deflections, in
fact, they indicate that the design of the edge members in hypars
with flat corners is usually governed by deflection limitations.
The measured bending strains in flat saddle shells (rise-
to-span ratio of 1.8). was much greater than the bending in hypars
with greater curvature (rise ratio of 1/3). The membrane theory
is insufficient for the design of flat hypar structures. How-
ever, the design of the connections (seam or edge) may be based
on the shear forces obtained from the simple membrane theory.
Several single and double layered saddle-shaped models
were tested under partial loading. Since such loads must be
carried mainly by bending of beam strips along the deformations
of single decks, relatively large deflections were noted. The
deflections under the 8 in. by 8 in. loaded area were about
three times greater in the single decks than on the double-
layered structures.
Since the effective shear rigidity of the deck is of para-
mount importance, the effect of curvature (warping) on it is
an important question. The effective shear rigidity of various
deck, edge member, and connection configurations are determined
by tests on flat diaphragms. The comparison of the measured
deflections for saddle hypars with various rise-span ratios and
the evaluation of the effective shear rigidities backwards from
the measured deflections indicated that the shear rigidity is
reduced by about 20% due to the warping effect.
The buckling of the deck is one of the design factors.
For small rise-span ratios and for low deck shear rigidities
the deck may buckle. As an example, a 12 ft by 12 ft model
v
having a single layer 24 gage corrugated sheet deck buckled
at a uniform load of 70 psf (see Fig. 6.14). This model had
relatively stiff edge members (3 in. dia. tubular sections).
The corner deflections remained linear with increasing load
beyond the buckling load.
The buckling load of double-layered structures is much
larger than that for single deck shells. A model, similar to
the above but with two layers of 28 gage standard corrugated
decks, did not buckle up to a load of 145 psf, when the test
was discontinued.
The major part of the ~~lytical investigation consisted
of two finite element approaches for the calculation of deflec-
tions, stresses, and instability. In addition, two simple
methods were developed for estimating the deck buckling load
and the buckling of the compression edge members, which would
suffice in preliminary designs.
Two types of finite elements were used: curved shallow
shell elements and flat elements. The details of the analysis
are described in Chapter III. Both approaches were verified
by comparisons with existing experimental and analytical results.
The stiffness of the eccentric edge members were properly
accounted for in the mathematical representation of the struc-
ture. The connection of the decks to the edge members may al-
low rotation about the axis of the edge members and movement
normal to the edges due to slip at the connections. These pos-
sibilities were also considered in the analysis.
The instability of the decks was studied with the help of
the incremental stiffness matrix approach. The effective stiff·
vi
ness of the system is reduced due to the in-plane forces in
the deck. The in-plane forces depend on the deflections of
the shell and to obtain the buckling load, the eigenvalues of
a large order system need to be evaluated. In the present
study the load incrementation method was used instead. The
effect of the in-plane forces was evaluated iteratively at
successive load increwents. The buckling load is obtained
from the nonlinear load-deflection curve, (Fig. 6-6, Page 276).
The comparison of the results of the flat element and
the curved element approaches reveals that both give good re-
sults for shells supported around the perimeter. However, the
flat element method gave better results in the neighborhood of
unsupported flat corners.
The analysis of the structures tested in this and in other
studies confirmed the conclusions of the experimental part of
the investigation. The stresses in most types of hypars are
low and the design is usually controlled by deflection or
buckling linitations.
The relative stiffness of the deck and the edge members
is an important factor. For stiff edge members the deck tends
to bend between opposite edges, whereas in the case of flexible
cantilevered edge members the shell partially supports the edge
members. Analysis of a structure including the weight of the
edge members indicated that this effect may have to be con-
sidered in the design of hypar structures.
The analysis of buckling of hypar decks showed that the
buckling load of double-layered shells is three to four times
vii
greater than that of single decks. The predicted buckling
loads compared well with experimental or previous analytical
evidence. The buckling load does not depend much on pre-buckling
deflections, however it depends on the axial stiffness of the
edge members.
The finite element analysis was also used to calculate
the deflection of an unsymmetrically loaded inverted umbrella
structure. The results, which compared well with experimental
data, showed that these deflections are about four times great-
er than those due to symmetric loading. This increase of de-
flections obviously depends on the type of structure; in this
case much of the flexibility was due to the bending of the
central column of the umbrella structure.
Since the instability analysis of hypars by the finite
element method involves considerable amount of computer capac-
ity and expense, approximate methods were developed for the
calculation of buckling loads. The buckling of the compression
edge members was studied by isolating them from the structure.
The instability of columns loaded by tangential axial forces
that remain parallel to the member during deflection was evalu-
ated. The results are tabulated in Fig. 6-13, page 284.
The buckling of hypar decks was also investigated by the
energy method (Section VI-7. The resulting equation has to be
minimized to get the critical load; this can easily be done
with the help of a computer. This approach is much simpler
than the finite element instability analysis and is preferable
in preliminary designs.
viii
A few buckling analyses of cold-formed hypar shells
showed that the critical load for double-layers is about three
to four times greater than a shell with a single deck.
The finite element analysis computer program will be made
available to designers by the American Iron and Steel Institute.
ix
-1-
GLOSSARY OF TERMS
Length of hypar quadrant in x-direction, inch.

Cross-sectional area of the beam, inch 2 .
~ Area of deck along x-direction, inch 2 •
Ay Area of deck along y-direction, inch 2 •
Ac Area enclosed by the cellular portion, inch 2 •

A. ,B. ,C.
1 1 1
Dimensions of the i-th quadrant.
a Length of the element in x-direction, inch.
B length of hypar quadrant in y-direction, inch.
b Length of the element in y-direction, inch.
c Rise of hypar quadrant, inch.
D Bending rigidity of an isotropic deck, lb-inch 2 /
inch.
Rigidity due to Poisson's effect, lb-inch 2 /inch.
Bending rigidity of the deck in x-direction,
lb-inch 2 /inch.
Dxy' Dyx Torsional rigidities of the deck, lb-inch 2 /inch.
Dy Bending rigidity of the deck in y-direction,

lb-inch 2 /inch.
E Young's Modulus, lb/inch 2 .
Modulus of elasticity in x-direction, lb/inch 2 .
Extensional rigidity of the deck along the x-
direction, lb/inch.
Inplane shear rigidity of the deck, lb/inch.
Modulus of elasticity in y-direction, lb/inch 2 .
Extensional rigidity of the deck along the y-
direction, lb/inch.
Modulus of elasticity due to Poisson's effect,
lb/inch2.
Inplane coupling rigidity of the deck accounting
for Poisson's effect, lb/inch.
-2-
G Shear modulus, lb/inch 2 •

G' Shear rigidity of deck (=Exyt), lb/in.
Effective shear modulus, lb/inch 2 .
Effective moment of inertia of a corrugated
d ecr.:,
1 inc
• h4; inc.1
· h..
t3
Moment of inertia of the deck section, 12 ,
inch4/inch.
Moment of inertia of a beam about the y-axis,
inch4.
Moment of inertia of a repetitive deck unit
about the x-axis, inch 4 /inch.
I'y Moment of inertia of a repetitive deck unit
about the li~e of connection parallel to the
x-axis, inch /inch.
Moment of inertia of a beam about the z-axis,
inch 4 .
St. Venant torsional constant, inch 4 .
Bending curvature in x-direction, inch- 1 .
Twisting curvaturet inch- 1 .
Bending curvature in y-direction~ inch- 1 .
[k] Stiffness matrix.
[k]bb Element stiffness matrix for plate bending.
[k]bbm Element stiffness matrix for w-displacement.

[k]bm Flexural and membrane coupling element stiff-
ness matrix.
[kl curved Element stiffness matrix for the curved element.

[k]eb Effective stiffness matrix of the supporting
edge member with respect to the eccentric axes.
[k]eff Effective stiffness of the deck after inclusion

of the instability effects.
[k] et Stiffness matrix of the plate element in the
global co-ordinate.
[k] flat Element stiffness matrix for the flat element.

-3-
k .. Element of the stiffness matrix in the i-th

1J row and the j-th column.
Element membrane stiffness matrix.
Length of a basic repetitive unit of the deck,
inch.
ix,ty,R.z Direction cosines of the local element axes.
M
x fioment about y-axis, inch-lb/inch.
Mxy Twisting moment about x and y axis, inch-lb/inch.

My Moment about x-axis, inch-lb/inch.
Direction cosines of the local element axes.
[N] Incremental matrix.
[NJe Incremental stiffness for the flexural displace-
ments (w) •
[N]global Incremental matrix for deck in the global co-

ordinate system.
N .. Element of the incremental matrix in the i-th
1J row and j-th column.
Force per unit length along the x-direction,
lb/inch.
Shear force, lb/inch.
Force per unit length along the y-direction,
lb/inch.
Direction cosines of the local element axes.
Generators of the hypar surf ace
ox,oy,oz Local orthogonal element axes.
oz' Normal to the hypar surface at point o.
[P] Load vector.
UniformlY distributed load, lb/inch 2 •
Reduction factor for the torsional rigidity of
a cellular deck.
s Developed 1ength of a basic repetitive unit of
the dec1', inch.
-4-
s.1 Stiffness of the i-th spring.

(T] Component transformation matrices for the plate
element.
Coefficient for torsional fixity between the
The coefficient for inplane fixity between the
Rotational transformation matrix for the beam
element.
Translational transformation matrix for a beam
element.
t Thickness of the deck, inch.
Thickness of the base plate in a cellular
deck, inch.
Thickness of the hat section in a cellular
deck, inch.
Strain energy, inch-lb.
Strain energy due to bending of deck, inch-lb.
Strain energy of a berun, inch-lb.
Average axial displacement of a beam section
measured at its centroid, inch.
Strain energy due to the membrane action of the
deck, inch-lb.
Strain energy due to the warping of the deck,
inch-lb.
u Displacement along x-axis, inch.
Displacements of the i-th node.
v Potential energy of the applied loads, inch-lb.
Potential energy of the in-plane forces, inch-lb.
y-axis, inch.
v Displacement along y-axis, inch.
Z-axis, inch.
-5-
·w w Differenciation of w-displacement with respect

x' y to x and y.
w Displacement along z-axis, inch.
x.1 x-coordinate of the local origin of a hypar
quadrant with respect to the global coordinate
system.
Cartesian coordinates of the center point o
of an element.
x,y,z Orthogonal cartesian global coordinates.
x',y',z' Orthogonal axes other than the global cartesian
axes.
Eccentricity of the centroid of the cross-
Y·1 y-coordinate of the local origin of a hypar
quadrant with respect to the elobal coordinate
system.
Eccentricity of the centroid of the cross-section
of the beam in the z-direction from the shell
surface, inch.
section of the beara in the z-direction from
Derivatives of the surface Z(x,y) with respect
to x and y.
a Ratio of shear rigidity of a corrugated deck to
that of a flat deck with the same thickness.
Warping constant of the deck, inch 6 /inch.
Warping constant of a beam, inch 6 •
Shearing strain in the x-y plane.
Generalized displacement vector.
Generalized displacement at the i-th node.
w-displacement at the point a.
-6-
op Load increment ~sed in load incrementation

method, lb/inch .
Increment in the displacements due to the
application of the incremental load op.
Strain in x-direction.
eX ' e :':v''
' Y }!JC Y
Strains measured along x' and y' axes
ey Strain in y-direction.
n Number of interconnected decks.
e Twist rotation of a beam, radian.
e' Rate of change of twist rotation,
radians/inch.
e1. Angle of twist of a beam at the i-th end, radian.
elX
! Rate of change of angle of twist at the i-th
end, radians/inch.
Rotation about x-axis, radian.
Rate of change of rotation, radians/inch.
Rotation about y-axis, radian.
Eigenvalue for the critical buckling load.
\) Poisson's ratio.
\) \)
xy' yx Poisson's ratios in x and y-directions respec-
tively for the equivalent orthotropic plate.
0
x
Stress in x-direction, lb/inch 2 .
o
x '' o
y ,' t
x 'y' Stresses measured along x' and y' axes.
Stress in y-direction, lb/inch 2 .
Shearing stress, lbs/inch 2 .
Total potential energy of a system, inch-lb.
CHAPTEP I
Ii·JTP.O DU CTI ON
I. 1. HYP P.R ROOFS

T:ie hyperbolic paraboloid shell rol)f, like any other
form of shell is one of the types of construction that makes
efficient use of materials by depending primarily upon the
form or shape for strength rather than on mass. The doubly
curved surface of a hypar shell is composed of straight lines
in two directions (Fig. 1.1). From the construction point of
view, this property is very attractive. It facilitates the
use of straight members for formwork and reinforcing steel in
the case of concrete hypars. This very feature also allows
the use of light gage steel deck panels, ,.~hich could be easily
warped to the required degree to form the hypar surface.
The hypar surface shown in Fi~. 1.1, can be pener-
ated in two ways 1 (1) The surface can he defined by moving
a convex parabola ODC in a direction parallel to itself, over

a concave parabola BDA. The parabola ODC lies in the plane
perpendicular to that of BDA. (2) The surface can also be
defined as a warped parallelogram. The surface can be gener-
ated (Fir. 1.1) by wovinr along y-axis, a strai~tt line that
reJ11ains parallel to the xz-plane at all tiTI1es but pivots while
sliding along the straight line AC. Physically the surface
can be visualized as a warped parallelo~ram OBC.I\, obtained by
depressing the corner H through a distance CH. By means of
-8-
similar triangles (Fig. 1.1); it can be easily shown that the

surface equation is,
CH
PQ = Z = Kxy where K = OA x OB
The sur.face is called hyperbolic paraboloid because

any plane parallel to the xy-plane, intersects it in a hyper-
bola; whereas a plane perpendicular to ODC intersects it in
a parabola.
For simplicity, the structural action of a hypar
can be visualized as a net of intersecting arches and cables.
The convex parabolas (arches) parallel to ODC carry compres-
sive stresses, whereas the concave parabolas (cables) parallel
to ADB carry tensile stresses. This implies· that the element
I is in a state of biaxial stress, compression parallel to
the arches and tension parallel to the cables. On the other
hand, the element II is in a state of pure shear. In the in-
terior, the membrane shear is carried by the shell. Along the
free edges, stiffening edge members are usually provided to
sustain the membrane action. These edge wembers themselves
carry gradually increasing tensile or compressive forces de-
pending upon the geometry of the structure.
By the combination of a basic hypar unit, such as
shown in Fig. 1.1, different elegant hypar structures can be
built. Four such configurations are shown in Fig. 1.2. The
edge members are provided to build up the membrane action and
also to stiffen the structure. Tension tie rods are commonly
employed to balance the horizontal reactions between the low
corners.
- ~-
Because of its architectural beauty, ease of con-
struction and ability to provide large column-free working
space, the hyperbolic paraboloid shell has been used for in-
dustrial plants, churches, assembly halls, etc. In 1'1oJTlbasa 2 ,
the hypar structure was used as a footing on low bearing capa-
city soil. A 225-feet double cantilever hypar roof is under
construction at Los Angeles for the AJ11erican Airlines jet
hangar. The roof uses a cellular form of deck. This struc-
ture may very well prove to be the forerunner of many more
similar structures.
I. 2. LITERATURE REVIEW
Like any other shell, hypars carry load hy both J11em-

brane and bending actions. The rneMbrane theory, as indicated
before, results in a state of pure shear. The li~itations of
this theory were realized by the niost investirators and the
necessity of probing into the bending behavior of the shells
became apparent.
The shallow shell theory of '!arguerre 3 and Vlassov 4
is often used to analyze the bendinp action. Though this theory
is approximate, it is considered fairly accurate for a shell
surface where the slopes of the tangents are very small co~pared
to unity. Two basic approaches were used to formulate the

In the first approach, two fourth-order coupled par-
tial differential equations in ter~s of normal displacement w
and Airy-stress function F are formed. ~eissner 5 using this
approach, determined the buckling load of a uniformly loaded
-IC-
isotropic hypar with moment-free rigid edges, with the edge mem-
bers of infinite axial rigidity but negligible bending rigidity
6
in planes tangential to the shell. Apeland and Popov reduced
these two equations to a single eight-order differential equa-
tion. Using Levy-type boundary conditions (with two opposite
edges knife edge supported) they tried to establish the effect
of edpe disturbances in the sa~e way as that for cylindrical
shells. Their important conclusion was that the effect of the
edge moment does not die off very rapidly in the case of hypar
shells.
The formulation in terMs of widdle-surface displace-
ment u-v-w, results in three coupled partial differential equa-
tions two second-order (u-v) and one fourth-order (w). Sal-
1
vadori and Bleich 7 using Vlassov's shallow shell equations fol-

lowed this approach. Assuming u=v=O all over the middle sur-
face, the fourth-order differential equation reduced to that
of a plate on an elastic foundation.
However, it must be emphasized that in order to obtain
the solutions to these mathematically complicated equations,
silTlplifying assumptions l'.'ere IT'adc. The choice of boundary con-
ditions was dictated by the possibility of ohtaining solutions
rather than simulating the exact boundary conditions in a phys-
ical model.
The shortcomings of the classical solutions resulted
in the realization of the importance of numerical approaches
for the solution of these differential equations.
Chetty and TottenhaM 8 applied a variational method
for the analysis of shallm" shell equations (l'J-F). However,
the choice of approximating functions limited its applicability
to the specific boundary conditions. Besides, non-classical
boundary conditions presented serious difficulties.
The numerical scheme of finite difference provided a
very useful tool in the solution of these equations. Das
Gupta 9 ~ ;.arza 10 , Russell and Gerstle 11 ~ 12 using the ~·r-F approach
applied this method to different hypar structures using mean-
ingful boundary conditions. The edge members were also incor-
porated in their analysis. Everybody use<l the classical beaw
theory. Mirza solved an umbrella shell. However, the magni-
tude of the corner deflections obtained for an umbrella shell
and the boundary conditions used along- the line of sy1l'lmetry,
raise serious doubts about the validity of the method 13 .
f::ussel an:d Gerstlel2 mainly analyzed two-corner and four-corner
supported hypars. .'.'Jon-dimensionalized design parameters were
provided. T~e main contribution of their work was to show the
importance of the line-load along the edge me1!'lbers. Croll and
Scrivener 14115 used the u-v-w formulation. The effect of the
eccentric connection of the beam to the shell is discussed. One
of the important features of their work is a complete discussion
of the convergence characteristic of the solution with relation
to the relative proportions of the shell and the edge member
stiffnesses. A comprehensive review of the above mentioned
method is presented by Brebbia 16 .
The finite clement rrtethod, which is nothing else but
the matrix formulation of Rayleiph-~itz method of variational
-L>
principle, was successfully employed by several workers. The

ease with Nhich this method can handle realistic boundary con-
ditions, made this method very suitable for hypars. It is
also believed that the variational principle used in the finite
element method will yield better results than the finite dif-
ference method because it involves integration rather than dif-
ferentiation as used in the latter method.
Connor and Brebbia 17 using shallow shell approxima-
tion, formulated the stiffness matrix for a thin shell curved
element, rectangular in plan. Similar formulations w~re worked
out by Deak 18 and Parker 19 . Pecknold and SchnobrichZO,Zl work-
ing alonr the same lines~ extended the work to the skewed
shallow shells. All these workers used linear displacement
field for u, v displacements. The fiajor difference was the
displacement field they used for the normal displacement w.
Deak 18 and Pecknold 20 both used the Birkhoff-Gara-
bedian interpolation formula. Connor 17 used a twelve-term poly-
nomial l·'hcreas Parker19 used the Lagrange interpolation formula.
Parker extended the '~ork to orthotropic light ga~e steel hypars
and compared the results with the experiments. None of the
Refs. 17, 18, 20, 21 mentioned any correlation with experi-
mental results but \·1ere content ·with the comparison to solu-
tions obtained by other numerical methods.
The buckling and nonlinear analysis of hypars are
also reported in the literature. Ralston 22 continued
Reissner's~ work by investigating the buckling of a hypar
under its o'l/m weight. Dayaratnam and Gerstle 23 presented a
- 1.5 -
solution to the buckling problem of hypars with edee bearns 9
simply supported at their corners. The in-plane displacements

u and v were assumed equal to zero. A double sine series de-
fleeted shape was assumed and the total potential energy of
the system was minimized using Ritz procedure. ~he critical
load was determined by equating the determinant of the result-
ing matrix to zero. It was concluded that for all positive
values of the ratio of bending rigidity of the edrre ~ember to
the deck, the deck buckling always preceeded the edge me~ber
buckling. The study in Ref. 19, indicates that for a very flex-
ible edee member, the possibility of overall buckling- nrior to
the deck buckling does exist. The erroneous conclusion of ref.
23 was reached because of two reasons:
(i) A. non-compatible displacement field between the shell
and the edge beams.

(ii) In determining the eigenvalue the off diagonal terms
were neglected.
These points are discussed at length in Ref. 19. ~'1uskat 24
studied the buckling of hypars with corrugated orthotropic
deck. A method for determining the buckling load of the deck,
considering pre-critical deflections of the entire structure
was developed~ using the energy approach.
Brebbia and Connor 25 presented a consistent finite-
element displace~ent formulation applicable to the shallow
shell elements using the Newton-Raphson iteration scheme, by
linearizing the incremental equation. The load deflection
curve for fixed hypar was presented.
-1i:i;~
~ large number of experimental studies have been re-

ported for hypar shells. ft major portion of the experiments
were carried out to measure deflections and stresses on small
to medium scale models. The results were correlated with the
approximate theories and by changing the design parameters,
certain predictions on the overall behavior of the hypar shells
were made. Rowe and r:irza 26 tested plastic Models with two
adjacent edges fixed and the other two free. The effect of
rise to span' ratio was studied on the free corner deflections
and axial strains. By changing the depth of edge beams along
the free edges, it was concluded that by increasing the depth
of the edge beam both the axial strains and the vertical de-
flections are decreased in the shell portion. Rowe 27 also
tested medium to small scale concrete models and used unsym-
metrical loading. In order to study the ultimate load carry-
ing capacity of the umbrella shell hypars, an experimental re-
search program is underway at the Cornell University, where
small scale concrete umbrella shells arc being tested to
f a1·1 ure 65 .
!!uskat 24 and Leet 28 tested small scale models to

determine the buckling characteristics. Leet tested plastic
models subjected to uniform normal load. The effect of imper-
fection was studied on the fixed shells. Edge beams were used
to study the effects of different edge conditions on stresses,
deflections, shell buckling and overall buckling. The effect
of different boundary conditions was not very pronounced on the
shell buckling. By studying the effect of different beam
-15-
sizes Leet concluded that the cross-sectional areas of the

edge beams have a significant effect on the deck buckling.
Most of the experiments were conducted with medium
scale models. Ref. 19 has listed most of them. Bertaro and
Choi 29 tested an 8' x 8' model. The model with edge beams and
two diagonally opposite corners supported was chemically pre-
stressed by using expansive cement. The model was tested in
the inverted position using air pressure loading. Deflection
profiles, crack patterns, and stresses were presented.
In the last 8 years or so, the use of light gage
steel decks as a hypar shell has gained some momentum.
~kDermott 30 tested two 8' x 8' orthotropic light gage steel,
saddle-shaped (Fig. l.2b) models. In the first case, the
steel deck was welded to the edge meri1bers whereas in the second
case the steel deck was glued. The rubberized canvas bags were
pressurized by water in the first case whereas air was used in
the second case. He also tested a large-scale model with a
single layer standard corrugated deck. The edge members con-
sisted of built-up sections. The loading was applied with
sand bags. Strains and deflections were measured.
Yu and Kriz 31 tested a concrete inverted umbrella
shell 24' x 24' in plan, in which upturned edge beams were
used. The symmetrical and unsymnetrical loading was simulated
by the discrete loads. The measured strains and deflections
were presented.
-16-
Three large scale tests were conducted for hypars

using orthotropic deck as a shell surface:
Nilson 32 tested a 15' x 15' hypar quadrant with simu-
lated boundary conditions of the adjacent quadrant. A single
layer of cellular deck was connected to the channels, which
were used as the edge members, by means of a warped plate. Uni-
form loading ·was applied by 25 jacks. The load deflection curve
and the measured membrane stresses were reported.
Two large scale hypar models 50' x 30' in plan, were
tested recently 33 . Two different cellular orthotropic single
layer decks were used in each case. 14 WF sections were used
as the edge members, which were allowed to move freely in the
plane of the hypar but were supported against the vertical move-
ment. The normal uniformly distributed load was applied by
creating a vacuum in the enclosed chamber.
1.3 SCOPE AND OBJECTIVES OF THE PRESENT INVESTIGATION
The main purpose of the present investigation was to
determine the deflections, stresses, buckling and collapse loads
of light gage steel hypar shell roofs and to provide design in-
formation. An analytical procedure was developed so that it
could be extended to include the analysis of hypars with various
support conditions as well as hypars subjected to partial loadings.
The finite eler.ient method was selected because of its
versatility. The entire project was approached from an engineer-
ing point of view. The validity of the r.iethod was established by
comparing the theoretical and experimental results for different
kinds of hypar structures.
Two separate computer programs were written, (1) For
-17-
stiffness analysis; (2) For instability analysis.

(1) Stiffness Analysis
A linear elastic analysis of the structure was per-
formed and the deflections, the edge member and deck stresses
were computed. The program can handle the following variables:
(a) Different types of orthotropic decks.
(b) Different configurations of the hypar structure.
(c) Realistic physical boundary conditions such as
eccentric connections of the edge members and
discontinuity between the deck and the support-
ing edge members.
(d) Different loading conditions such as uniform,
unsymmetrical, line loads, etc.
(2) Instability Analysis
A linear load incrementation method was used for the
instability analysis. The effect of the in-plane forces Nx,
Ny and Nxy was included.
The experimental part of the investigation included
the determination of the effective shear rigidities of standard
corrugated decks and the determination of deflections, stresses,
and buckling loads in inverted umbrellas and saddle-shaped
hypars constructed of corrugated steel decking and tubular edge
beams.
-17a-
1.4 ACKNOWLEDGMENTS
This investieation was sponsored by the American Iron
and Steel Institute. The financial assistance and the technical
guidance provided by the Institute is gratefully acknowledged.
We thank the Task Group on Shell Roof Structures, and especial-
ly its chairman, Dr. J. B. Scalzi, for their interest in this
project.
Special thanks are due to Dr. George Winter, principal
investigator of the project, for his great interest in the
investigation and for the many wise suggestions throughout the
work. His insight into structural behavior often steered the
study back from tangents and away from pitfalls.
The nonconservative stability analysis was carried
out by Dr. R. Huskat. His and Hr. A. Banerjee's help with the
experimental program was very valuable.
CI!f PTEP II
DECK PR0PE'RTIES
II.L INTRODUCTION
A material in which resistance to mechanical actions
is different in different directions, is called anisotropic.
Orthotropy is a special case of anisotropy, where the body
possesses elastic properties which are symmetric about three
mutually perpendicular planes.
Orthotropy can be further classified into two cate-
gories namely, natural and reometric. The natural orthotropy
is a result of the material property itself. A classical
example of natural orthotropy is timber, where the modulus of
elasticity, along the direction of its prain in tension, is
su1'stantially hieher th.an the corresponding modulus in the
direction perpendicular to it.
In the geometric type of orthotropy, the difference
in elastic properties in the perpendicular direction, as shown
in Fir. 2.1, is due to the reometrical confiruration of the
structural element even though it is made t~ of an isotropic
homogeneous ~aterial. Different types of decks belonging to
this category are shown in Fig. 2.2.
In both the cases, the definition of elastic con-
stants in two mutually perpendicular directions is required.
However, geometric orthotropy is of particular interest for
the hypar structure dealt Hi th in this 1•rork.
-18-
-19-
In the following analysis the decks, shown in Fig.

2.2, are replaced hy equivalent orthotropic plates, having the
same physical properties as the decks, such as extensional and
bending stiffnesses. It is impossible to achieve a complete
equivalence between the actual physical systetJl and the ideal-
ized orthotropic plate, in all respects such as strain energy,
deflections or moments at different points under different
criteria. The properties of the idealized system can be deter-
mined either by equivalence of stiffness or equivalence of
strain energy between the ide8lized and the physical systef.1 34 .
The equivalence of stiffness is established by equating only
the deformations between the actual and the idealized system.
The equivalence of strain energy is obtained by equating the
work done by tl .c internal forces in
1 bot~ the systems ,vhen sub-
jected to identical loading and boundary conditions.
The elastic constant for the idealized plate material
is assumed to be the sawe as that of the parent :material. The
properties calculated depend only upon the direction consid-
ered and not on the position of the corresponding point on the
actual deck. T:bc orthotropic plate theory is applicable to
the decks, shown in Fig. 2.2, provided the ratio of the dimen-
sions of the repetitive unit (~) and the overall span of the
deck, is very small, i.e., <<l.
II.2. ELASTIC PPOPERTIES OF DECK

In the case of a geometrical orthotropy, as present
in corrugated or closed formed decks, in order to calculate
elastic- properties in two mutually perpendicular directions~
-20-
it is necessary to separate the in-plane membrane and the

bending action.
A. MEMBRANE PROPERTIES
The principal directions of elasticity in an ortho-
tropic deck are those along which the extensional rigidities
are either minimum or maximum. When the axes x and y coincide
with its principal direction of elasticity, the equation of gen-
eralized Hooke's law for plane stress-strain can be written as
ax vxy
e:x = B-; - Ey 0y 2·la
a
£
y
a + ily 2-lb
2-lc
Solving the equations for the stresses we get,

... 0
ax Ex vxyEx
1
a
y = (1-vxy vyx)
v E
yx Y
E
y 0 e:
y
2 .. 2
TXY 0 0 (l-:-vxyvyx)Geff
The terms vxy and vyx represent the coupling effect o the
actions (stress or strain) applied in two perpendicular direc-
tions. From Fig. 2.2, it is obvious that these coefficients
(vxy and vyx) cannot be equal. For example, consider the
cellular deck shown in Fig. 2-2c, where a uniform stress
applied along the bottom plate of the deck in the x-direction
will produce a negligibly small strain in the y-direction, in
the hat portion. Whereas a uniform stress applied over the
entire cross-section along the y-direction will produce a
-21-
strain in the bottom plate proportional to the waterial con-

stant v, (Poisson's Patio). Though "xy 'f "yx' by ~1axwcll
Betti reciprocal theorem the following relationship holds:

2-3
As explained by t!:e physical behavior above, "xy is

equal to v and he;1ce 1
E
= E
_!. " 2-4
y
The value of Geff is given by,

Gef f = a.G 2-5
where ~ is called the relative shear rigidity factor, the

value of which depends upon a number of factors. A complete
discussion of a, together with a description of the experi-
mental method of determination of a, is given in Section II-3.
The methods for the determination of constants Ex? Ey, "xy
and "y x , for both closed and open form decks, are given in
Appendix A.
Since the decks are idealized as uniform orthotropic
plates of constant thiclmess, it is convenient to express the
meJTlbrane stiffness constants and the forces in terms of their
thicknesses. l~ul tip lying the first TOl.'! of Eq. 2-2,
Ext E1 t
axt = (l-vvyx) EX
+
(1-vv yx ) Ey 2-6
2-7
2-8
The other rows can be modified similarly (see Eq. 3-13).

Depending upon the ~ethod of connections the elastic
-22-
constants, calculated above purely on the basis of geometrical

configuration, thickness and material properties, need modifi-
cations.
The elastic const.:.mt Ex for an open deck is given
by, (Appendix A)
I E
0
Ex = y-t 2-9
y
The value of Iy' depends upon the line of application of the
load. A substantial reduction in the value of Ex results from
the eccentric connection. The cellular decks (Fiff. 2-2c) are

usually connected to the supporting rne~ber along the botto~
plate leaving the hat portions free. As shmm in Fip.-. 2-3,

the forces applied along the bottom plate will be partly re-
sisted by the vertical sides of the hat. Depending upon the
joint efficiency betNcen the hat and the botto:r: plate, the
effective area of the hat section in resisting the forces a y
will vary. In the absence of the test data, it will be con-
servative to consider the bottom plate only as being effective
in resisting the in-plane forces cry. In the computer programs,
a provision is made to include the effective width of the web
plates of tPe hat.
B. BENDING PROPERTIES
The bending rit:idities, -::::x, Dy, Dxy and D1 , for a
geometrically orthotropic plate cannot be obtained directly
from the directional elastic constants riven in Eq. 2-2. The
relationship between the bending riFidities and the moments,
is given in F.q. 3-16. In the case of a light-gage orthotropic
deck (corruffated or cellular) Dy >> SO.nx. This property can
-23-
be used to approxi~ate properties such as n1 and vyx· Since

the twisting rigidity Dxy is not equal to Jvx (see Appendix
,
A), the average of the directional twistine rigidities is
used 35 .
In the case of an open formed deck there is a
tendency for the·warping of the surface. The method of cal-
culatin~ the \varping constant r, based on the assumption that
an individual unit 35 of a plate twists about its center of
rot at ion is we 11 known 36 . However 9 considering t:he plate as
a whole unit, the plane of twisting changes, depending upon
the distance of a point from the support. The extent to which
this warping restraint alters the behavior of the deck is not
very clear.
The moment of inertia used for the COMputation of
DY is calculated at t}e centroid of the repetitive section.
The effect of eccentric location of the ribs or hats in the
case of decks, on the bending properties is discussed in great
detail by Mas sonnet and Bares 37 . Ho•·Jever its use in practical
problems is difficult. Due to the local buckled form for high
wi"<l t h_ to t h"ic kness ratio
. f or a compression
. fl ange 39 (F"ig. 2 . 4) ,
the effective moment of inertia for both deflections and stress
analysis is reduced. The reduction in the :rnmnent of inertia
is a function of u~e stress level. This factor may be of iJT1.-
portance in the cellular deck. Depending upon this reduction
of the effective section~ a second analysis of the hypar shells
~ay be necessary though in most cases the stresses are small.
The procedure of calculating I ea~ is discussed in ~efs. 38, 39.
-24-
The me ch an isr:- of rcduc t ion in bending rigidities due

to the connection between the different panels is not very
clear. Experimental results on the determination of these
rigidities for a continuous single panel have been reported 40 .
The bending risidities Dx and Dxy are particularly affected by
the dis continuities between connect ions of different panels.
Exawination of structural problems solved in this.study shows
that the magnitude of both these constants (Dx and Dxy) is so
small for an open formed deck that the reduction even of the
order of 1/10, does not affect the results significantly
(<<5% variation).
The properties for two or rnore decks placed perpen-
dicularly can be calculated by atl~ing the corresponding direc-
tional constants.
C. ELASTIC PJ">0PERTIES FOR ARBITPARILY ORIENTED
CO-OPDINATE P..>'ES
1
~hen the structural axes and the principal axes of
orthotropy coincide, the elastic properties of the deck cal-
culated on the basis of principal axes can be directly used in
the analysis without any modifications. As explained in the
Section I.l, the structural behavior of a hypar can be broken
up into mutually orthogonal arches and cables. W~1en the decks
are placed along the axes of arches and cables (Fig. 2.5), the
structural axes x and y, do not coincide with the principal
axes of orthotropy x 1
and y'. The elastic constants in terms
of the structural axes are expressed by the principle of work
equivalence 4 l,.~Z. The Membrane strain energy in two systems
-25-
of axes can be given by,
V' = l2 0
x
'e:: ' +
x
!2 0 y 1 e:: y ' + !2 T x ' y 'y x I f I 2~1ob
Expressing strains in terms of stresses in both e~~ SYsteThs

of axe~ and equating,
V' =v Z-1oc
one can obtain the equivalent elastic constants. lre eaUi-
valent bending constants can be determined by eq~~tJng the
bending strain energy along both axes (Eq. 3-17)' 1he elastic
properties for an orthotropic tnaterial in terrn.s 17~ Pl'ly 9rbi-
trarily oriented co-ordinate axes are given in 0 0\~il bi
Lekhnitskii 41 .
II. 3. SHEAR RIGIDITY OF OPTP.OTPOPIC DECKS
The shear stiffness of an orthotropi~ 9),~t~, flSsum-
ing Nxy = Nyx (Section III.2 B), is given by t~e \j9~e 5 9ion
Exyt = G' = a. G nt 2-11
where n is the nurnher of interconnected decks, 1;'. ~, 'the thick-
ness of each deck and G is the shear modulus O:f t~( l'natel'tal.
Cl is the relative shear rigidity coefficient P:ivi~; 'the l'atio
of the shear stiffness of the actual orthotropiC ~oci sfStem
and the isotropic plates of thickness nt. In t~~ ?a~e of an
isotropic plate, the vertical load on a hypar i~ b?rtly ~arried
by the membrane action in shear and therefore tP' 9hear stiff-

ness of the deck is of utmost importance in its ~tlQ~t~ra1 be-
havior. Experimental and theoretical determina~\ 0 f of a
was carried out by several workers. The fact~r \ J~peods
-26-
upon a number of factors listed below:

(a) The different shapes (corrugated sine form, N-type,
etc.) of diaphragms (Fig. 2.2), show different resistances to
the in-plane shear loads. The shearing leads produce bending
and twisting of the corrugations and also set up membrane
stresses and shearing strains. On the basis of the assumed
displacement field of the corrugations, energy stored due to
each of these above mentioned actions (bending, twisting, etc.)
is reported in Refs. 43, 44.
(b) The spacing of the connectors, between the deck and
the edge members, transverse to the corrugations have a pro-
nounced effect on the value of , whereas the spacing along
the directions of the currugations has a very minor effect.
Flat shear tests on a 26-G. standard corrugated deck, 6' x 6'
in plan, were conducted in this investigation (Chapter VII).
Two tests were performed with the connectors between the deck
and the edge members at each valley and one with the connectors
spaced over every third valley. The other factors, in the
above tests, being the same, the value of the shear stiffness
obtained for the former cases was reported nearly three times
as large as that of the latter.
(c) According to Luttre11 45 , the shear stiffness of a
panel also depends upon the length along the corrugations.
Accordion-like warping results due to the connection of the
diaphragm to the edge members. It was found 45 that the length
of penetration of this warping is independent of the overall
span of the diaphragm and this warping reduces the shear stiff-
-27-
ness, particularly for short spans.

(d) According to ~ef. 43~ the seam-slip between the ad-
joining deck panels and the connection of the deck to the edge
members contributes substantially to the shear flexibility.
(e) As found in the experiments (Chapter VII), for the
limited range of thickness of panels studied, the shear rigid-
ity increases linearly with the effective thickness as given by
Eq. 2-11. However, consideration should be given to the con-
nections between two or more decks and their connections to
the edge ~embers. The results are reported in Chapter VII.
The two layers of corrugated decks placed perpendicular to each
other were connected to the edge merabers through the connection
of the lower deck (Fig. 2.6). The additional flexibility
provided by the lower deck corrugations, reduces the effective-
ness of the upper deck. The position of the screws with respect
to the direction of the applied shear also affected the stiffness
of the shear panel. It was found that only 33% increase in Exyt
was noted for two decks connected as shown in Fig. 2.6, whereas
the value of nt doubled.
All the factors mentioned above are important for open
form decks as shown in Fig. 2.2a,b. In the case of cellular
decks or stiffened panels (Figs. 2.Zc,d) the flat plate of the
deck is directly connected to the edge members. In view of the
low shear carrying capacity of the out-of-plane hats, a major por-
tion of the shear is carried by the flat plate. Knowing the
seam-slip characteristics between the adjoining panels, the
shear stiffness for the cellular deck can be estimated conser-
-28-
vatively as that of the flat plate.

The dependence of the relative shear rigidity co-
efficient a on some other factor such as connecticn with the
intermediate purlins, etc., is discussed in detail in Pef. 43.
II. 4. DETER~~INATION OF THE EFFECTIVE SHEAR

PJGIDITY FOR ORTHOTROPIC HYPARS
The theoretical determination of shear rigidity co-
efficient a, was done by equating the work done by the applied
shearing force with the strain energy stored in the deck panels
due to deformations. In Ref. 43t an excellent correlation be-
tween the theoretical ~~d experimental results was reported.
Since a major contribution to the shear flexibility was from
the connection between the deck panels 43 and the connections of
decks to the edge members, a previous knowledge of the seam-
slip characteristic is required. Based on an experimental in-
vestigation, the method for the determination of the shear
rigidity from flat shear test is given in Pef. 45.
The next question arises as to whether the relative
shear rigidity deterrnined by the flat shear tnst can be di-
rectly used for hypar structures without any modification. To
correlate the shear rigidity coefficient a determined by the
flat shear tests and that of the warped deck in the actual
hypar surface, twelve saddle-shaped hypars~ uniformly loaded,
with tubular edge ti.embers of 3'~ diameter and !n thickness
(vertically supported all around) were tested (Chapter VII). The
1ower corners were connecte d b y Z
ln x ln tie
. b ars, in
. order to
2 4
restrict the horizontal spreading. The tests were conducted
-29-
for both single and double decks with varying rise to span
ratios.
It was found that for the lower values of a (<0.12),
the center deflection of all-supported hypars was proportional
to the value of a. Both single and double deck hypars were
tested for three different rise to span ratios (1/8, 1/5, 1/3).
The theory developed in Chapter III correctly predicts the
effect of a and rise independently on the central defl~ction
6 0 (Key Sketch, Table II.l, II.2). The value of a in the

actual hypar surface was interpolated from the structures
analyze ..: ·with different assumed a values for the constant rise
to span ratio.
The results obtained for the effective value of a
are compared with the results given by Parker 19 , who ar alyzed
the same shells on the basis of plate on an elastic foundation
(u=v=O) and did not include the effect of the tie bar. For
both single and double deck, a certain a~ount of increase in
the value of a was noted for the low rise to span ratio (1/8)
but with the increase of the ratio, the value of effective a
reduces. The results obtained by direct interpolation for
single deck hyµars appear quite reasonable (Table II.1). The
direct interpolation if applied to the double d6cks 9 results
in extremely low effective value of a partio1larly for the
ratio of 1/5 and 1/3 (Table II.2). The fallacy in the ~ethod
of interpolation can be explained as follows:

All the test results marked with an index (I) (82l(I))
were the tests where two transversely placed corrugated decks
- 30-
were inter-connected Hhereas in the rest of the tests, the top

deck was connected only along its periphery to the bottom decl~
which in turn was connected to the edge member. The center

deflections obtained for Case (I) is roughly 20-30% less than
that where the decks were not inter-connected. In the mathe-
matical model there is a complete coupling between the in-plane
displacements of u-v and the normal displacement w (see Chapter
I I I). Secondly when two decks are placed on top of each other
in the analysis> the in-plane stiffnesses of the top deck rxt'
Eyt5 Exyt and E1 t are assumed to be fully effective in re-
sisting the load.
In case of the decks only connected at the edges,
the deviation between the mathematical and physical model is
very drastic, and therefore the deflections given by the anal-
ysis are very lm·"· 1
~.1 ith these considerations in view, acer-
tain amount of discretion must be used in estimatine the values
of a. The results obtained by the analysis were compared with
the test results • The resultinz reduction in values of a
can be attributed to the warping of the surface and the deck
and the edge member connections. These factors are common to
both single and double decks (Figs. 2.7, 2.8).
1.'!hether the increase in the effective relative shear
rigidity for the lm·1 rise to span ratio of 1/8, is an inherent
behavior of the hypar or whether it is the shortcoming of the
theory in tPe region of transition between flat plate and warped
hypar surface is not very clear. It will be conservative to
neglect any increase in the value of a obtained from flat shear
test. The importance of the values of a and the recommended
reductions are further discussed in Chapters IV and V.
CliJ\PTEP. I I I
FIN I TE ELEI'ENT i T~TH0D FOR STIFFNESS AND ST!?_ESS AI~ALYS IS
III.I. INTRODUCTI0N
As discussed in Chapter I, the solution of the shal-

low shell equation for realistic boundary conditions is an
extre~ely co~plicated mathematical pro~osition. This necessi-
tates the use of numerical methods. The finite element method
was chosen because of its versatality in handling realistic
boundary conditions, different structural confirurations, ortho-
tropic deck materials and any forms of loading, with ease.
The method has also de~onstrated good convergence character-
istics.
The finite ele~ent method based on the stiffness
analysis uses the principle of rnini~um potential energy. The
total potential energy of an elastic system, for a geo~etri
cally admissible state can he represented by~
qi =u + v 3-1
where U is the strain energy store~ in the system and v is the

potential energy of the applied loads. Both C and V are ex-
pressed in terms of displacements at the joints of an idealized
structure. U is a quadratic function of the nodal displace-
Jllent. The principle of the minirmn1 potential energy states
that, ''The total potential energy is rriinimur.l; when an elastic
body is in equilibrium".
= _a_
()/j..
(P+V) = 0 3-2
1
-31-
-32-
-6V
P. = - - 3-3
1 aL\. 1
For the linear elastic analysis, these expressions

can be put in matrix form,
[P] = [K] [~] 3-4
The ele~cnts of the stiffness matrix can be obtained by the

second differential of the strain energy,
K' •. = 3-5
1)
In the case of a framewor}: composed of linear members such as

beams, struts, etc., the individual ele~ents are connected to
each other at their nodal points. 1/.1ell-defined boundary con-
ditions at these joints enables one to solve the physical
proble~ without any difficulty. In this case, there is one to
one correspondence between the matheJnatical and the physical
model. However, in the case of a two-dimensional structural
medium such as a plate or a shell surface, the discrete ele-
ment approach does not give a one to one correspondence between
the element used in the matrix analysis and the forces in the
actual surface. Here the entire structure is idealized into
discrete elements, curved or flat, connected to each other at
the nodal points. The displacements of the nodal points are
interpreted as those occurring at the correspondinr points in
the structure. The state of stress and strain inside an ele-
ment is defined completely in terms of its nodal point defer-
:ma tions. The success of determinini:r the elastic properties of
an idealized structural element lies in the equivalence estab-
lished between the actual w.odel and its equivalent discrete
-33-
model.
The linear elastic stiffness analysis consists of
four important steps:
(1) The formulation of the element stiffness matrix.
(2) The formulation of the master stiffness matrix for
the entire structure by assembling individual elements.
(3) The solution of Eq. 3-4 for the given boundary con-
ditions and loading.
(4) The interpretation of the deflected shape and the
computation of stresses and forces.
Two alternative stiffness formulation methods are
studied here:
Method 'a': The use of rectangular curved elements. based on
Method 'b': The actual curved shell surface is approximated by
the assemblage of flat rectangular elements.
III.2. ELEMENT STIFFNESS
The elements rectangular in plan are selected. These
elements are very simple to formulate and for the structure
under consideration, their limitation of application to the
rectilinear rectangular boundaries, is not considered to be of
any serious consequence.
As shown in Eq. 3-lc, the element stiffness matrix
can be derived from the strain energy U of an element, ex-
pressed in terms of an assumed displacement field.
A. DISPLACEMENT FUNCTIONS
The displacement fields assumed for the analysis are
-34-
as follows:
1
u = ab [(x-a)(y-b)u 1 - x(y-b)u 2 + xyu 3 - (x-a) yu 4 ] 3-6a
V = ao
1
((x-a)(y-h)vl - x(y-b)Vz + xyv 3 - (x-a) yv 4 J 3-6b
+ (3ax 2 -zx 3 )(3ay 2 -2y 3 )w 3+(a 3+2x 3 -3ax 2)(3by 2 -2y 3)w 4
- ax(x-a) 3 (b 3+2y 2 -3by 2 )ey 1 -a(x 3 -ax 2)(b 3+2y 3 -3by 2 )ey 2
- a(x 3 -ax 2)(3by 2 -zy 3)e y.3 -a(x-a) 2x(3by 2 -zy 3)e y 4
+ b(a 3+zx 3 -3ax 2)y(y-b) 2 ex 1+b(3ax 2 -zx 3)y(y-h) 2 ex 2
3
+ b(3ax 2 -2x 3)(y 3 -by 2 )ex 3+b(a 3+2x 3 -3ax 2)(y -by 2)ex 4
2 2 2 2
+ abxy(x-a) (y-b) exyl+abxy(x -ax)(y-b) exyZ
2 2 2 2
+ abxy(x -ax)(y -by)exy 3+abxy(x-a) (y -hy)exy 4 ] 3-6c
The same functions were used by Gallagher 50 and

Yang 51 , but in their studies the x,y cartesian co-ordinates
were replaced by the more general curvilinear co-ordinates al
and etz . As seen here, the displacement field inside an ele-
rr,cn t is directly expressed as the function of its nodal dis-
placements rather than in terms of undetermined parameters.
An element, as shm,m in Fig. 3.1, has six degrees of freedom
per nodal point and a total of 24 degrees per element. The
displacements u,v,w,ex, and ey have a physical meaning at each
aZw
node. The term exy represents the t'\•rist curvature axay· Using
the right cork-screw notation,
-35-
aw · e = - d'».T : e
ex = a-Y 3-7
• Y ai, xy
One may also note that the inplane rotation about the z-axis
is being omitted in this formulation.
The function for w, normal bending displacement, is
of the cubic order. The terMs corresponding to the degrees of
freedom w, ex and ey are obtained by the cross product of the
corresponding tcrws for the beam function in x and y-directions.
In order to represent the constant strain corresponding to the
. .
tw1st1ng a2w 1.
term a·xay' . e.' t h e terJTl ' xy ' ' t he a <ld.1 tiona.
. 1 d cgree
of freedom in the form of exy is added to the displacement.

The displacement functions are ~eometrically syMrnetri-
cal and include the constant strain and rigid bo~y modes for
the flat plate. It is obvious that the assumed displacement
fields for u, v and w are not of the same order. Whereas those
for u and v are linear,_ as stated_ before, w displacement is
cubic. If the displacement fields of u and v were of the same
order as that of w~ eac} node would have 12 degrees of freedom
thereby having a total of 48 deprces per element. Besides
this, the linear edge member elements would have to be given
the same order of stiffness Jl'atrix. The additional degrees of
freedom would involve more computational work and this effort
could only be justified if goo~ re~ults~ without sacrificing
the required degree of accuracy, could be attained with a fe1 11 er
number of elements.
Any combination of displacements which can be accomp-
lished ~ithout straining the structure are called rigid body
displacement modes. These displacement modes can be easily
-36-
recognized in the case of linear members or flat plates whereas

their role is not easily recognized in the displace~ent field
for the curved shells. The assumed displaceJnent field does not
include rigid body motion for the curved element. Cantin and
Clough 52 used the displacement field for a cylindrical shell
clement and added the appropriate rigid body motion terms in
the form of trignometric functions. It was shown that with the
inclusion of the rigid body motion terms, there was an improve-
ment in the rate of convergence. Haisler and Stricklin 53 and
Connor and Brebbia 17 , have shown that inspite of omitting the
rigid body motion terms with the refinement of the prid size,
the convergence is still insured.
Pecknold and Schnobrich 20 proposed the most logical
method for the inclusion of the rigid body motion terms. It
Pas suggested that these terms should satisfy the ho~ogeneous
part of the strain displacement relationship used for the

curved eleMent (Eq. 3-9,3-10). The inclusion of these terms
involves more computational effort but they seeM to have cer-
tain advantages, which are further discussed in Chapter IV.
B. cunVED ELE;~NT
Strain Displacement ~elationshi~
The strain displaceir.ent relationship:; used in the

curved element formulation are simplified according to the
shallo'\'r she 11 theory. The following assumptions are made:
(1) For a given surface defined by the equation, z=F(x,y),
the slopes of the surface z,x and z,y are considered negligible
in comparison with unity. In general, the shallow shell theory
-37-
will be quite accurate as long as z'x' z, y -·

< 1/8, and often
accurate enough for practical purposes as long as z,x, z,y

~ 1/2 5 ; though the second limit could be considered as too
liberal.
(2) The sides of a differential shell element, which are
orthogonal in the projected co-ordinate plane, are assumed to
be orthogonal in the plane of the middle surface of the shell.
In other words, the geometry of the surface is approximated by
that of its projection on the co-ordinate plane.
(3) If the equation of the middle surface is of the
second order as is the case for a hypar, the assumption (1),
leads to the approxiMation that the curvatures of the surface
are constant.
The errors resulting from these assumptions increase
as the depth of the shell increases. For the shallow hypar
surface defined by the equation,
Z -- C xy
Arr 3-8
the linear strain displacement relationships for the thin

shells assume the followinr form 20 , 17 :
The membrane strain displace,.,,ent relationships:
u x 3-9a
'
v,y 3-9b
2C
Yxy = u ,y + v
,x - -
AB
w 3-9c
The bending strain displace~ent relationships:

Kx = -w
,xx 3-lOa
-w 3-lOh
,yy
-38-
ZK xy = - 2w ,xy 3-lOc
Strain Eri.ergy
The strain energy for a typical element consists of
two parts: the mernbrane and the bending strain energy.
U = Ub + LT
m 3-11
The membrane strain energy for an element is given by 54 ,
3-12
The stress-strain relationship for an orthotropic material

(Fi~. 3.2) can be represented by
lJ r:: 0 ex
x ...,xt Elt
Ny = Elt
T"
£'yt 0 £y 3-13
N 0 0 Exyt Yxy
xy
The method of col'lputation for the above mentioned elastic con-
stants is discussed in Chapter II.
Using the strain displacement (Eq. 3-9, 3-10) and the
stress-strain relationship (Eq. 3-13), the membrane strain
energy (Fi~:'· 3.2) of an element can be expressed as follows:
1 2 3
b a 2
1
1J
m= 2 0J 0J {Extuz,x + Eyt ·~ 1y + 2E 1 tu,xv'y
4 s 6
+ Exyt[u2~Y + v2
'x
+ 2u,yv,x]
7 8
c
- 4Exyt(AB) w [u, y + V,X)
9
2
+ 4 Exyt(~B) w2 } dxdy 3-14
-39-
The same strain energy expression is also reported in r.ef. 19.

The bending strain energy is given by the expression 54
b a
1
ub = -
2
I !
0 0
[~1
x C-"r 'xx )+rry (-w 'yy·)+zr· xy (-111 'xy ) ]dxdy 3-15
The moment strain relationships (Fig. 3.3) are:

r··x D
x Dl 0 -i•r
, 'xx
1'"
·y = T''"l Dy 0 -w' yy 3-16
p 0 0 -Zw 'xy
xy Dxii
-
The bending strain energy is r;iven by 38 ,
This energy expression assumes that ;·xy = ~·'yx· However, as

pointed out in the Chapter II, for an open orthotropic dee¥,
l>'xy =I ilyx and therefore the value of r·xy eiven in F.q. 3-16, is
to be interpreted as the average value. The energy expression
3-17 neglects any ener2;y stored in the deck clue to the re•
.
straine d warping. As po1nte . l 35 , in an open f orm
d out \.oy .s m1t~
deck, the twisting moment i'l consists of t1:10 parts:

yx
p yx = -(2Dyx w 'xy - Erw,xyy) 3-18
The warping strain energy is given as,
u . 1
warp = -2 Er 3-19
The omission of this energy term is not considered to be of any

serious consequence. The stiffness matrix for the curved ele-
ment is obtained by usinp Eq. 3-5. The complete stiffness
matrix is given in Appendix B.
-40-
C. FLAT ELEf·.7.ENT
The stiffness matrix for the flat elements can be

obtained directly by putting c = 0 in the curved eleJ11en t stiff··
ness matrix. The strain displacement relationships for the
flat elements are the same as those for the curved elements,
except for the shear strain given by the Fq. 3-ac. For the
flat plate, the last term in Eq. 3-qc, due to the twist curva-
ture of the hypar surface disappears. The difference between
the curved and the flat element stiffness matrix can be shm~n
schematically as follows:
[ K]
m [K]~m u
v
[K]curved = ------- -------- 3-20
.,, [Y.]bb
[K]~,) Ill 1°•.T
+[K]bbm
u
[K]m 0
v
[K]flat = ------- -------- 3-21
0 [K]bb 1'.T
The meMbrane stiffness matrix [K] m, is obtained from terms

1 6 of the membrane energy expression Eq. 3-14. This
stiffness matrix is common to both curved and flat elements.
The coupling matrix terms [K]bm and [KJbbm are obtained from
terms 7 , 8 and 9 respectively, of Eq. 3-14. These terms
are zero in the case of the flat elements. The stiffness
matrix [K]bb' common to both. types of elements, is obtained
from the bending strain energy given by Eq. 3-17.
The stiffness ~atrix derived for the orthotropic flat
plate was checked term by term~ with. the stiffness matrix for
-41-
an isotropic plate reported in Ref. 51. To establish the

validity of the stiffness matrix further, the results for the
plates shown in Pip:. 3.4a,b,c were checked with available class-
ical solutions. For the isotropic plates shmm in Fig. 3. 4a
and b, the comparison between the classical solution and the
stiffness analysis for the center deflection ~B was excellent.
The error was less than 1% in hotL cases and the geometrical
symmetry in the nodal displacements was satisfied.
The third example (Fig. 3.4c) is of particular in-
terest. The 28-G standard corrm':ated
. '-·
deck can be considered
as an extreMe case of orthotropy. Pere the bencinr rigidity
Dy is 1678 times Dx. The bending rigidity constants were used
as given in Ref. 54. It is reported in Ref. 19 1 that the rec-
tangular elements proportioned in the ratio of their bending
riridities, in two directions, would give better results than
the square elements. A quadrant of the plate was analyzed by
usinp: square (6x6) elenionts and elongated rectangular (2 x 12)

elements. The results for (lx30) size elements were also corn-
puted tut are not reported since there is practically no dif-
ference betueen these results and those with a 2x12 ele111ent
grid. The deflection profile across the corrugation is plotted
in Fig. 3. 5. There is practically no differen,.·e (see Table
III-1) between the rectanpular and the square element solutions,
a dip in the deflection profile near the support is seen in
both the solutions. Similar deflection profiles are also re-
ported in Ref. 19. The solutions obtained by the stiffness
analysis are cowpared with the classical solutions given by
-42-
Timoshenko 54 in the form of a double sine seri1:s, for the uni-

formly loaded, simply supported orthotropic plate.
00 00
w = i: i:
m=l,3,S, ... n=l,3,5, •..
amn s in m; x. s in n JY 3- 2 2
where~
a mn 4
3-23
+ n 4DY)
b
where,
3-24
The first few significant terms in the sine series

were computed. The comparison of the results is shown in Table
III-1. Both the classical and the stiffness analysis solutions
are well within the limits of practical accuracy.
A strongly orthotropic plate such as the one under
consideration, priMarily behaves as a plate on an elastic
foundation. The stiffer beam strip near the support attracts
more load because of the presence of the plate action. In
other words, the deflected profile of the plate is the func-
tion of the assumed bending rigidities and does not appear to
be dependent upon the shape of the element (rectanpular or
square).
The non-monotonic convergence characteristic of this
problem, observed in the Ref. 19, is probably the function of
the de cl( properties. This analysis indicates that elel'lents
which are square in plan can be used for the single deck hypar
structure.
-43-
fl. EDGE J".Ei' "BEP

.
Th e conven t 1ona . 5 5 1s
1 b eam st1. ff ness matrix . pr1mar1
. . 1y
based on two assumptions:

(1) The shear center and the centroid coincide~ (ri~. 3.6a).
(2) The bending of the section takes place about the axix
of symmetry (Fig. 3.6a).
In the case of a syremetrical channel (Fig. 3.6b), the
shear center (S.C.) and the centroid (C.G.) of the section do
not coincide. ''Yhen the section is subjected to a. vertical load
P (Fir. 3.6c), acting at a distance e from the shear center, it
not only deflects in the plane y-z but also twists through an
angle a about the x-axis, passing through the shear center. In
other words, the bending displacements v and w need to be ex-
pressed at the shear center of the section. This observed fact
was elegantly expressed by Bleich 56 and Hoff 57 , in the strain
energy of a beam of arbitrary cross-section. The total strain
energy of a beam (Fie. 3.7) is piven as,
a
ubb = 1
2
f
0
[DI
y
·w~ xx +EI zv~. xx+GJe~ x +:Erhe~
·- . xx
+EA 1"u~ x ]dx 3-25
where, wand v are the displaceJTlents of the shear center (S.C.)

measured along the axes y and z parallel to the principal
ccntroidal axes of the section and u is the average lonpitudinal
displacement of the section along the axis x measured at the
centroid (C.G.) of the section. e is the angular rotation of
the section about the x-axis.
Neglecting the warping of the section, the conven-
tional beam stiffness matrix can be obtained by assuming the
-44-
following displacement fields:
3-26a
v = 13 [(a 3 +zx 3 -3ax 2)v 1 +(3ax 2 -zx 3 )v 2+ax(x-a) 2 ezl

a
+ a(x 3 -ax 2 )e 22 ] 3-26b
w = .!3 [(a 3+zx 3 -3ax 2 )w 1 +(3ax 2 -zx 3 )w -ax(x-a) 2 e

a 2 y1
- a (x 3 -ax 2)ey 2 l 3-26c
3-26d
Using Eqs. 3-25 and 3-5, one can obtain the conventional beam
stiffness matrix.
However 1 the warping restraint is of practical im-
portance, particularly for thin-walled open sections. To in-
clude the warpinr effect~ the displace~ent field for e is
assumed to be of the same form as that of v and w58 .
1
3 [(a 3 +zx -3ax 2 )el+(3ax -zx )e2+ax(x-a) e~1
3 2 3 2
e =
a
+ a(x -ax 2 )e~ ]
3
2 3-27a
0, = ae 3-27b
x ax
The stiffness matrix (14xl4) for this member can be

directly obtained from Table III-3 by substituting Y
c
= zc =
Y
s
= Zs = 0 · The additional degree of freedom (8 xy ) intro-
duced, does not present any problem. If the beam shown in Fig.
3.8 is attached to a shell surface or a plate along the line
passing through the shear center, one can easily find the cor-
relation between the degree of freedom for the plate and the
-45-
beam.
e = ex and e'x = exy =

a2w 3-28
axay
0mittine the rotation about the z-axis (e ), the
z
plate and the beam element can be connected to each other with
a one to one correspondence in the nodal <legree of freedom.
E. ECCENTRICALLY co1nmCTED NON-CmiPATIBLE
srPPORTING EDCE r~r~ER
The above mentioned matrices can be used only in the

case of non-eccentric loads or concentric connections of the
e<lee members to a shell or a plate surface. In practice, con-
centric connections between the supportine edge roe~bers and a
shell or a plate are seldom possible (Fig. 3.9). The eccen-
tric connections modify the effective stiffness of the support-
ing wemters. Tte modified stiffness for the non-compatible
edge f!1ember can be obtained by the use of simple linear con-
gruent transformations similar to the one suggested in Ref. 59.
The difference between the compatible and the non-compatible
eccentric ~embers is Jiscussed in Appendix E and the relevant
stiffness and incremental matrices for the co~patible element
are given in Table E-I and E-II. The linear transformation
assumes that the beam bends about its own neutral axis (Fig.
3. 10).
1 0 0 0 +Z c -Y c 0
-u
7
1 v
0
. ~
s 0 0 0
1 .. y 0 0 0 w
s
p.] s = 1 0 0 0 ex 3-29
1 0 0 ey
0 1 0 ez
1 A
-46-
= rP-] s 0 J 3-30
0 [A.] s
The effective stiffness of the edge ~ember can be given as,
[K] eh = [T] ! [K] [T] s 3-31
Similar stiffness matrices were independently de-

rived in Ref. 21. The stiffness ~atrices, with t~e linear
variation and the non-linear variation of the twist angle are
given in Table III-2 and III-5 respectively. The convergence
characteristics and the accuracy of the stiffness matrices, was
checked by solving three cantilever beams (Fi[Y. 3.11). Only
axial (Fig. 3.lla) and bending (Fig. 3.llb) loads were applied
eccentrically to the beam. All displacewents were computed
along the line PQ (Fig. 3.11). T~e deflections and the twist
angle at the free end Q, were compared with classical solutions.
As it can be seen, the convergence in all cases is insured and
the results for the case of six elements are within 0.2% of the
classical solutions. The loading II ~as applied to two cases:
Ila) Free torsion with a linear variation of twist. Ilb) In-
cluding the ~arping restraints at the fixed end. The classical
solutions for the case I lb, Iere obtained from the 11ef. 60.
l1
The convergence in case of the restrained varping is slightly

slower than in the case of free torsion. The influence of in-
eluding the warping degree of restraint can be seen from Fig.
3.13a. The free end deflection o0 in the case IIb, is about
63.5% of the deflection obtained in the case Ila.
-47-
F. ELASTIC SPRING8
In order to simplify the mathematical solution without
undue loss of gencrality 1 certain structural elements are ideal-
ized in the for~ of concentrated spring stiffnesses. For ex-
ample, the central column in the case of an umbrella shell, if
idealized as a physical roember with its end points, will not
only create an additional node point for the master stiffness
matrix but v ill also disrupt the regularly arranr:ed griil pattern
1
and will warrant a modification in the entire assembly routine.

In order to avoid this)> the stiffness of the column can be
idealized into six discrete springs accountin£ for its axial~
shear, bending and t'• isting stiffnesses (Fig. 3.14).

1
These
stiffnesses are given as,
-a -
3-32
12EI lZEiz
5sY v S S GJ
= a L ' SZ = a ; T = a
These spring constants are added alonr the main dia-

gonal elements of the master stiffness matrix. This idealiza-
tion is not alway•, satisfactory. In Fig. 4.1) a tension bar
connecting the lm·rer corners of the saddle sha:red hypars are
replaced by four springs in the u and v directions, two at each
corners f and b. This ic"iealization eliminates the interaction
between the nodes f and b. The validity of this approxirn.a tion
can only be assessed by engineering judgeMent.
III. 3 I:ASTER STIFFI'JESS !''.ATP-IY.
The stiffness formulation presented so far is for an

individual element. The total stiffness of the structure is
-48-
developed by the assemblage of these individual elements. The

two different methods ('a' and 'b'), mentioned in Section III.I,
differ in the formulation of the stiffness matrices.
a) Curved Element
The element stiffness is formulated by using an ele-
ment of the same shape as the shell middle surface (Refs. 17,
19, 20). H01·rnver, the fact that the shallow shell assumptions
are made in this formulation, should not be overlooked (a point
which is discussed at lenrth in Chapter IV). The assumed
shallowness of the shell does not warrant any form of co-ordin-
ate transformation. The strain displacement relationships
given in Eqs. 3-9 and 3-10 are based upon the displacements u,
v ~ and w \· hich are measured along the tangent and normal to
1
the surface.
b) Flat Element
The middle surface of the shell is approximated by a
series of flat plates. The geometrical approximation of the
actual surface needs three important steps:
(1) Definition of Surface -
As pointed out in Chapter I, different hypar struc-
tures can be built with various combinations of the basic units
(Fi~. 1. 2). It is necessary to express the equation of the
generated surface with reference to the chosen global axis.
The general equation of a structure using the hypar units can
be expressed as,
c.
Z
= A.B.
_1_ (x-i1.) (y-y1.) 3-33
l 1
Figs. 3.lSa and 3.1-Sb show two structures and also give the
-49-
values of the constant, defining the surface equatL:m for each

quadrant. In Fig. 3.lSb, points P,Q,R,S are the local origins
of the quadrant surface. x' and y', are the local co-ordinate
axes passing through the local origins.
(2) Element Size -
For shallow shells, the size of the element can be
approximated with the size of a rectangle projected on the co-
ordinate plane xy (Fig. 3.16). For example, the size of the
curved element PQRS is approximated by the projected element
P11 Q''R"S 11 •
A better approximation for the size of the element

PQRS can be made by calculating the actual lengths PQ, QR, etc.,
and using a rectangle P'Q'rt'S', of an equal area. For the low
rise to span ratio (~1/5):. the error introduced by using the
projected element is very small (2-3%). To take advantage of
·this fact, a provision is made in the computer program to
choose betwe~n the above mentioned approaches. The difference
in results when using these two :methods was about 10% for the
structure shown in Fig. 4.3. The computation of the exact
lengths gives different stiffness matrices for each element.
(3) Co-ordinate Transformation -
It is not possible to generate a smooth curved sur-
face by using flat elements with rectilinear boundaries. This
results in gaps and non-compatibilities between the adjacent
elements forming an idealized uneven surface (the picture of
the idealized surface is left to the imagination of the readers).
Such gaps and discontinuities occurring at the boundaries of
-so-
adjacent elements, have been known to produce undesirable and

. .
non-ex1st1ng . 61 , wh"1ch
no d a 1 ·f orces~ f ors h e 11 s o f revo 1ut1on
had significant effects on the solutions. However, no such
noticeable difficulty was encountered in the solution of shallow
hypar shells. The solutions obtained for these shallow shells
did not show any necessity of placing local tangential ortho-
gonal axes at each nodal point.
In writing the master stiffness matrix and the overall
equilibrium equations, local nodal axes can be chosen. Instead
of these, a simple ancl approximate approach is used. .A plane
tangent to the surface is dra~m at a point 0 (Fig. 3.17). The

most logical point for the tangential plane is the center of
the element. In the case of UJrlbrella shells, the flat portion
near the free corner shows a pronounced hending action. To
estimate this bending action conservatively, the tangent planes
were drawn along the horizontal lines PQ and PS instead of at
the center point 0 (Fig. 3.17a). When usiny the corner point
transformation, one has to exercise proper care to retain the
symmetry of the solution. For a large number of elements, both
methods should give about the same results but the corner point
transformation is more cumbersome and therefore it is not used
in the analysis .
. In Fig. 3.17a, the line OZ' is normal to the surface
whereas OX' and 0Y' are generators of the surface. The direc-
tion cosines for the lines OX', 0Y' and CZ' are riven as,
-51-
zx
OX' = 0
J1+Zf Rz
OY' = 0
1 y 3-34
-z
p;
-z
'R
oz I y 1
=
l+Z x2+z y2 l+Z x2+z y2 j+z2+z2
x y
where
zx = AB
£__ y
o
and x0 .
As discussed in connection ·w·ith the shallow shell

assumptions, the angle between the penerators OX.' and OY' is
not equal to 90°. Hence a new set of mutually orthogonal axes
O~, oi, and oi are obtained, where OZ coincides with OZ'. The
procedure for obtaining the direction cosines of ni, oi and o2
is a simple application of the three dimensional solid geometry
(Fig . 3 . 1 7b) .
The transformation matrix for each node can he repre-
sented as
p,] 1 0
[T]eli, = 3-35
0 p,] 2
6x6
li,x ty 1z R.x 1y 0
[A] 1 = x my
1'l mz [A] 2 = mx my 0 3-36
n x ny nz 0 0 nz
The in-plane rotation ez is omitted. From the matrix [A]z it

can be seen that exy the additional degree of freedom, is trans-
formed f roJT1 the local to the global axis in tPe same way as w
-52-
except for the assumption that there is no coupling betwe.,·n the

rotation exy and the rotation ex, ey. This transformation can
be viewed as an approximation.
The master stiffness matrix for the shell surface,
using method ·'b' (see Section III.l) is completed by transform-
ing each and every element stiffness from its local axes (OX,
OY~OZ) to the respective global axes (OX,OY,OZ). Depending
upon the direction cosines of the local axes of the individual
elements, every coefficient of the transformed element stiff-
ness matrix can have a non-zero value, i.e., there is a coupl-
ing between u, v, and w displacements, expressed in terms of
the global co-ordinate.
Beam Element
The co-ordinate transformation given by Weaver 55 , to
transform the stiffness of the beam element from the local to
the global axis is used. The transformation matrix with a minor
modification to suit the problem at hand is given in Table III-4.
After orienting the axis x of the member, it is also necessary
to define the orientation of the principal axes y and z, since
the stiffness of the beam element is expressed in reference to
its principal axes. The angle S defines the orientation of the
principal axes. The definition of the angle a is given in Ref.
63. (Fig. 3.18).
For the method 'a', using the curved element, the
.stiffness of the edge members is added without any co-ordinate
transformation. For method 'b', using the flat elemen.t;s, the
edge member stiffness is added with a proper co-ordinate ttans-
-53-
formation as given in the Table III-4.
II I. 4 LOADING
A uniformly distributed load acting on a rectangular

element can be replaced by statically equivalent loads of
equal intensity acting at each nodal point. This procedure is
acceptable if the size of the eleMent is small.
The alternative approach known as the Nork equivalent
load is based on the equivalence of energy. The nodal forces
are so assigned that during any virtual displaceMent the work
done by these forces is equal to the corresponding work done
by the actual distributed load.
The work equivalent nodal loads for the rectangular

element, with unit normal load, are given ~elow:
1·;
·1 ab/4 8 x1 - ~b 2 /24
ab/ 4 8
xz ab 2I 24
ab/4 8 -a 11 2I 2 4
x3
at/4 2
= 8 x4
= -ab /24 3-38
2
a b/24 8 xyl a 2 ~ 2 /144
2
a b/24 8 xy2 -a 2b 2/144
a 2b/24 8 xy3 a 2b 2/144
2
-a b/24 8 xy4 _
3 2~2/144
The nodal loads associated ~ith exy de~ree of freedom

do not have any physical significance. Ttc ~oment ~ is
associated with ey degree of freedom whereas the moment My is
associated with ex . For an interior point, the work equivalent
load reduces the static load for uniformly sizecl elements. The
effects of the nodal moments cancel out along the boundaries
-54-
whereas they add up in the direction normal to the boundaries.

All throughout this work, work equivalent lodds are used for
uniforIBly distributed loading.
In the case of a uniform load acting nor~al to the
surface, statically equivalent projected uniforin load is cal-
culated. The work equivalent nodal loads then calculated for
the ~edified load intensity are directly applied to the struc-
ture, in terms of the global co-ordinate axis without any
transforJTJ.ation. Tl1 is is again an approximation. /\more
accurate metho~ of determining the nodal load would involve a
co-ordinate transformation from the local to the plobal axis.
Besides the uniformly distributed load, a concen-
trated force or moment can be applied to the structure by
specifying the magnitude of the load at the corresponding de-
gree of freeco~ in the load vector.
III.S BOUNDARY CONDITIONS
The boundary conditions for a structure can be

broadly classified into two categories:
(1) Force boundary conditions.
(2) Displacer,ent houndary conditions.
In the conventional stiffness analysis, the latter can te
easily satisfied whereas the former can be satisfied only in
the variational sense. A detailed discussion of this point is
reported in Ref. 21.
The typical boundary conditions for the edge where
x is constant, are:
-SS-
Boundary
Conditions u v w ex xy
Hinge 0 0 0 0
Knife-edge 0 Q 0
Fixed 0 0 f) 0 (I
"'·'
Free
Symmetric 0 0 0
For the free edge, no displacement Poundary con-

ditions are specified. All the boundary conditions are applied
with respect to the global axes. The boundary conditions for
the ID\::rnber PQ (Fig. 3. 19), \·1hich was supported vertically but
allowed to slide along its length, should be specified in terms
of the local axes x, y and i but instead they : arc specified
in terms of the axes x, y and z. This is an approximation and
the error due to this will increase with the increase in depth
of the shell. The procedure to express the bo~ndary conditions
1n the local axes
-
x~
-
y
-
z is given in Ref. 62.
1
The connections of the edge members to the deck

present a problem in expressinr the correct boundary conditions.
Two non-corr:patible boundary conditions are shown in Fip. 3.2'1.
In Fig. 3.20a, the <leek bends freely without twisting the edge
member. This moment-free deck to edre ~ember connection is
quite common in practice. The op~n-form <leeks are discretely
connected to the edge members whereas the close forrn decks are
connected only along their bottom plates (Pip. 2.2). In both
cases there is no transfer of moments between the edge member
and deck.
The other type of discontinuity can result in the rela-
-56-
tive displacement betueen the deck and the edge member, normal
to the boundary (Fig. 3.20b). This type of a connection can
result because of an oversized hole, loosely connected screws,
or due to tearing of the deck. Depending upon the continuity
achieved between t11e deck and the edge members, different
values of fixity coefficients are used. Tr and T8 represent
the torsional and the horizontal fixity coefficients, respect-
ively.
In the case of the moment-free deck to edge member
connection, Tp ~ 0. The edre roei:'ber stiffness matrix is modi-
fied by multiplying the columns and rows corresponding to the
twisting degree of freedoms (ex and axy) by TF.
The problem is furtter complicated by the eccentric
connections. As sho1m in Fig. 3.2la, even with a discontinuity
of the rotational degree of freedom, twisting can still be
introduced in the edge member because of eccentrically trans-
ferred vertical or horizontal load. This problem is not solved
satisfactorily. By the ~ethod of fixity coefficients, the
twisting action introduced by these eccentric forces is elimin-
ated. There is no w-0ment transfer when t~o elements are inter-
connected by means of hinges and this results in the local re-
lease of the member forces. This for~ulation does not include
the effects of these releases. The details for the incorpor-
ation of these local member releases are given in Pefs. 62 and
6 3.
II I. 6 snu:TIONS OF EQl1ATTONS
The equation 3-4, relating the applied nodal loads and

-57-
the generalized nodal displacement can be solved. To obtain

the displacement vector,
[6] = [k]-l [P] 3-39
The inversion of the large matrix [k] not only re-
quires a very long time but also needs a large storarc space
in the computer. The structural matrices are usually well
banded about their main diagonals and are also symmetrical.
Banavalkar wrote a subroutine which stores only the half band
of the matrix in a vertical fashion (Fig. 3.22). The equations
are solved by the Gauss-elimination scheMe 64 • With the linli-
tation of the available core size and the computational cost,
a total of 486 equations with a maxiwum half band width of 66,
were solved for a normal problem (64 square elewents). The
rectangular matrix of 486x66 was formed and stored in the
computer.
However, there are problems where the structural con-
figuration destroys the close-handedness of the matrix. For
example, the tension-tie connecting the lower corners b and f
of the hypar (Fig. 4.1), creates sparse entries in the stiff-
ness matrix (Fig. 3.22c). In ;uch cases, instead of revising
the entire solution procedure, the structural element is ideal-
ized in t 11e form of discrete springs (see Section III. ZF).
III.7 STRESS Ai~ALYSIS
Since the main aim of the project is to establish the

behavior of the hypar shells, the physical interpretation of
the computer results is very important. The deflections, as
well as the stresses in the different structural components
-58-
such as deck, edge mewbers, etc., represent t~e physical be-

havior.
In the finite element analysis, the generalized nodal
forces are related to the stresses. But because of an error of
discretization and applied joint loads, the resulting nodal
forces for the adjoining elements show deviations. In order to
obtain some form of average stresses, the element forces are
calculated at the mid-point of an individual element.
A. DECK STRESSES
The deck stresses are calculated at the center point
of an individual element. Depending upon the choice of method
of analysis, method 'a' (curved elements) and method 'b' (flat
elements), corresponding strain displacement ·relationsh~ps are
used at the center point (see Eqs. 3-9, 3-10). The forces N
x'
Ny and Nxy and the moments Mx and My are calculated per unit
leneth. For the complete derivation of these forces, see
Appendix C. The major difference between the computation of
stresses for a curved and a flat element is that in the case of
the former, consistent with the shallow shell assumption (see
Section III.2B) displacements tangential and normal to the sur-
face can be used directly. But in the case of flat elements.
the displacements obtained in the global co-ordinates are trans-
formed into the local co-ordinate axes (see' Section III.3) and
the relevant displacements in the local co-ordinate axes are
used. The difference between the strain-displacement relation·
ships for the curved and the flat elements was already shown
in Section III.ZB.
It must be realized that the forces are calculated on
-59-
the basis of orthotropic plate theory which can be considered

as an approximate mathematical idealization. The forces cal-
culated per unit length are multiplied by the lengths of the
basic uni ts (Fig. 2. 2) . In the case of a uni f orroly loaded
structure, this wethod can be considered to be fairly accurate.
For the light gage sections witL high width to thickness ratio
of the in<lividual components, the effective led of the section
and the location of the neutral axis need modification in
accordance with the level of the load (Ref. 38).
The stresses calculated do not include the local
bending behavior. For example, the bottom deck plate AB bends
locally between the vertical web plates of the hat (Fig. CZ of
Appendix C). The problem of deviation between the mathematical
and the physical behavior of the orthotropic deck is dis-
cussed in detail in Refs. 46,47.
B. BEAM STRESSES
The nodal forces calculated in the loc~axis of the

beam can be directly used to calculate the beam stresses. The
method of calculation of stresses for the concentrically con-
nected beam nember is well knm, n. 1
The imaginary forces calculated alon? the line PQ

(Fip. 3.11). are to be transferred to the shear center and
centroid to calculate the relevant stresses. Ttc tea~ forces
[P]b can be calculated by,
[P]b = [T] ! [K] [T] r [fl] 3-40
The pre·multiplication of the ?lobal displacement

[6], by [T]~ (Table III-4), transforms the global nodal dis-
placements to the local axes whereas the pre-multiplication of
-60-
[K] [T]R [~] by [T]~. transforms the forces to the shear center
or the centroid of the beam.
Because of the mathematical idealization, certain
difficulties are encountered. A beam with an eccentricity in
the z-direction is shown in Fig. 3.23. The variation of tl:e
axial forces is shown in Fig. 3.23b and 3.23c. Since the forces
are balanced at point 0, the axial fotce also contributes to
the equilibrium of the moments at point 0. This results in the
inequality of the moments along the axis of the beam PQ. The
problem becomes particularly critical in the case of rapidly
changing axial force and a deck with strong bending rig~di ty
(e.g., concrete hypars). No suitable solution is found for
this problem as of this moment. In the absence of definite
guidelines, the deflected shape of the structure should be used
to decide the sign of the moment.
Experience shows that the bigger of the two reoments
Ct.\sp or MoQ) is always in conformity with the correct deflected
shape of the beam. The difficulty experienced in computing the
stresses of an eccentric edge mewber is one of the shortcomings
of using the nodal points only along the shell surface.
The results obtained by this stiffness analysis are
compared with experimental and the available solutions in the
literature in Chapter IV.
CtTAPTI'P IV
A GENERAL COf'PARATIVE STUDY
IV. 1 1;,JTRODUCT ION
As discussed in Chapter III, two ~ethods ~ere used

to analyze hypar shells: method 'a', uses rectanrular curved
elements based on the shallow shell theory; whereas method 'b'
approximates the actual shell surface by using a series of
flat rectangular elements. The solutions of selected proble1'1s
are presented here with three purposes:
(1) To substantiate the use of the finite element method,
hy method 'a' only, by comparinp the solutions for prohle~s
for which analytical or other numerical solutions are available

in the literature. The comparison for flat plates and linear
beams are already presented in Chapter III.
(2) To compare the solutions obtained by ~ethods 'a' and
'b', for typical hypar structures. The cowparison is done pri-
marily with a view of assessinr: tlteir suj_tability in applica-
tion to the practical problews and also t 1) find out their
shortcomings and limitations.
(3) To compare the analytical solutions with the experi-
mental results obtained by earlier workers. The details of
the structures analyzed are given in Table IV-1.
All the analytical results are further used to study=
to a limited degree, the effects of different structural nara-
meters on the behavior of hypar shells, such as relative shear
-61-
-62-
rigidity factor a~ rise to span ratio, etc. The effects of

these parameters are further discussed in length in Chapter V.
IV. 2 COrtPA.RISON OF METHOD fa WITH OTHEP Nm"EPICAL SOLUTIONS

. ------ I
Connor and Brebbia 17 presented the .centerline deflec-

tion profile for a saddle shaped hypar structure (Struc. 'l',
Table IV-1), with clawped boundaries all around. Their results
·were based on exactly the same procedure as used in this study.
The only difference is that they used a 12-term polynomial for
the normal displacement w, whereas this approach used the 16-term
displacement function as given in Eq. 3-6c. Fig. 4.6 shows
the results obtained by the author. For the grid size of 8x8,
the deflection profile along the line oa is similar with the
one reported by Connor and Brebbia 17 . The deflection·prof1le
along the diagonal ob is also plotted to check the sy~metry of
the solution.
The convergence characteristics of the solutions are
checked by refining the grid size for the above mentioned
structure. As seen in Fi~. 4.7, the convergence for the center
deflection at point o (Fig. ·4.1) is monotonic and rapid. Dy
refining the grid size from 6x6 to 8x8, an i~provement of only
2.3% is obtain~d in the result.
Pecknold and Schnobrich 20 presented the centerline de-
flection profile for the same type of a structure, with the
perimete,r knife:..edge supported (StTuc. '2 '). As pointed out
earlier, they used the Birkhoff and Garabedian interpolation
formula for w displaceMents. Besides that, the complete rigid
body modes of displacements were included by solving the homo-
-63-
geneous part of the strain displace~ent relationships piven in

Dq. 3-9. The inclusion of the rigi~ body motion terms des-
troyed the interelement compatibility and put additional re-
straints on the in-plane displacement fields of u and v.
u = F(ui) + F1 (w) 4-la
v = F(vi) + F 2 (w) 4-lb
The displacement functions F1 (w) and F2 (w) are the

results of the solution of the homogeneous part of the fq.
3-9c. Pecknold and Schnobrich compared their solution with a
Navier-type (double sine series) solution, for which SO ter!Tls
in
· eac h d 1rect1on
· · · 1 u d e a20 .
were inc The deflection profiles
along the center line oa and the diagonal ob, obtained in this
study (grid size 8x8), are shown. in Fig. 4. 8. The central
deflection obtained in this manner differs by +0.2% from that
obtained by the series solution (9.18xl0- 3 inch.); wtereas it
differs by approximately -1% from the finite element solution
of Pecknold and Schnobrich.
The solutions obtained by method 'a' for both Struc.
'l' and '2' mentioned above, are considered to be quite good.
IV. 3 COLP ft.RISON OF r;fETHOD ~ J\lH~ rtTHOD 'b'
Both structures solved hy method 'a', were solved

again by using method 'b'. The deflections obtained by the two
methods usinr the grid size of 8x8, are sho"rn in Tahle IV-3.
Methods 'a' and 'b' show similar deflection profiles along both
the center and the diagonal lines oa and ob respectively. The
central deflection obtained by method 'b' for both structures
is on the higher side, as compared with the one obtained by
-64-
method 'a'. For Struc. 'l', the difference in the central de-
flection is about 0. 8~; whereas for Struc. '2', the d; fference
is about 1. 309ii. The central deflection for Struc. '2' is only
0.5% on the higher side of tho deflecti~n obtained by Pecknold
and Schnobrich.
The correlation obtained by methods 'a' and 'b' 1s
excellent for these two structures. However, it wust be
pointed out that both of these struct~res~ taken from refs. 17
and 20, are supported all-around. 'From the practical point of
view, these structures are only of academic interest. The
boundary conditions such as free edgest encountered in an UJll-
brella shell (Fig. 4.2), provides a more critical test for the
comparison of the di ff er en t P1ethods.
It ·was not practical to compare tnethods 'a' and 'b'
for all the exampl~s, therefore only a selected number of struc-
tures were chosen for coJll.parison (Struc. '6' and '9' were
'
used). Struc. '6' is a swall scale concrete model. In this
structure, the stiffening edge members are located eccentri-
cally~ on top of the shell. The idealized edge members are
considered eccentric only in the z-direction (see Fies. 4.2
and 4.5).
Struc. '9' is also an umbrella shell hypar with 28-G
double layered standard corrugated decks placed perpendicular
to each other. Here the edge wember is connected eccentri-
cally to the deck with the deck on top. In the case of Struc.
'6' there is full fixity between the edge member and the shell,
whereas in the case of Struc. '9', the connection between the
.-6 5-
deck and the edge member is moment-free.

The comparison of the deflected profile obtained by
methods I a' and ! b I anJ. the corresponding deflected shapes are
presented in Fies. 4.13 and 4.23. Comparing the solutions ob-

tained by methods 'a' and 'b' for Struc. '9', it is obvious
that the method 'a' underestimates the free corner deflection
ob. Method 'a' gives a good correlation between the theory and
experiment for the deflection oa.
The deflection profile obtained by method 'b' for
the edge wember abt where a major portion of the load is car-
ried by the bendinrr action, is very good when compared with the
experimental results. The relative deflection between the
points a and b according to experiments is 1.2 inc~es, method
'b' giving a rela.tive deflection of 1. O inches; l·.rhereas that
predicted by the method 'a' is 0.73 inches.
A distortion in the deflected profile for the ~ember
ab and the underestimation of the relative deflection between

the points a and b, results in the underestimation of the bend-
ing and the total stresses at the point a. The bending stress
at point a by method 'a' is 8.44 ksi, whereas that by P.'ethod
'b' is 17.14 ksi. The total stress at the point a by method 'a'
is 12.50 ksi, whereas by method 'b' it is 19.90 ksi. The cor-
ner deflection ob and the stresses at a (Fig. 4.2) are of a
great practical significance for a designer, from both the choice
of edge P.'elTlher sizes and the overall behavior of the hypar
structure.
Another i~portant shortcoP.'ing noted of the method 'a'
is that the statical check for the total vertical load is not
-66-
satisfied at the column. Because a very flexible deck was used

for the shell surface, the deck could not transfe~ a substantial
. - - .; .
amount of load near the column and the resultant· axial component
and the v'ertical shear in the edge member oa should sum up to the
total applied vertical load; only 73% of the total vertical load
is accounted for by method 'a' whereas 98% of the applied load
.• .
is accounted fo~ by meth~d 'b'. T~i~ discrepancy of the statical
check was also noted when working with ;the computer program formu-
lated by Park~r 19 •
In the case of Struc. ''6', the deflection profiles obtained
by rriethods.'a' and 'b' along the compression member are reason-
ably close. However, these two nethods give entirely different
. I , .
deflected shapes along the tension member ab. According to

method ea', the point b (free corner) instead of deflecting
downwards relative to point a, it deflects upwards. The same
difficulty was also encountered regarding the corner deflec-
tion when using the computer program of Ref. 19. Because a
part of the load near the column is also carried by the con-
crete shell, the thickness' of which is quite comparable to the
depth of the edge members oa and oc, it is difficult to figure
out the statical check for the total vertical load.
The corner deflection of Struc. '8', which is identical
to Struc. '6' except·· for the fact that the edge members are
downturned, was found to be quite low when analyzed by the
''·
method of Ref. 19, as compared with the experimental results.

Briefly, the shortcomings of method 'a' can be summarized as:
-67-
(1) The method underestimates the deflection of the free

corner of an umbrella shell where the shell surface degenerates
almost to a flat plate.
(2) The prediction of the deflection profile along the
eccentrically connected tension members ab and be is not con-
sistent and leads to the underestimation of the bending and
total stresses in::rhe edge rnenbers, which are of practical
importance.
(3) A discrepancy for the statical check of total verti-
cal load is noted (Struc. '9').
Because of these shortcomings, it was decided to use
method 'b' in the analysis of all structures. It rnust 11e
emphasized that method 'b' does have certain shortcomings,

though none as serious as the ones associated with method 'a'.
~· 'ethod 1
b' is discussed later in Chapter V.
IV. 4 DISCllSSION OF t.1ET!IOD ' I
The umbrella shells with flexible edge members show

a pronounced bending action near the free corner b (Fig. 4.2).
This bending action was observed in tests conducted on con-
crete shells (Refs. 31,65) and Struc. '9' (using hynar with a
corrugated deck) tested at Cornell. The shell in this region
acts almost like a flat plate. In Ref. 1, this observed bend-
ing behavior of the shell was termed as secondary bending and,
based on the non-diinensionalized parameter of ~ii' the hendinp
moment coefficients for this region were f!iven. '~ethod 'a'
baseo on the use of the shallow shell theory fails to predict
-68-
this localized bending behavior at the free corner.

Before discussing this shortcoming of method 'a', it
is necessary to point out that in the formulation of the curved
element, the nodal displacements (u,v,w) are JTleasured along the
tangents and normal to t11e surface, rather than along the car-
tesian axes. In other words, the strain displacewent relation-
ships given in Eq. 3-9 are all expressed alone, the lines of
generators of the surface. The elel"lent stiffness matrix based
on these displacements eliminates the co-ordinate transforma-
tion. In the solution of the master stiffness matrix, method
'a' gives the displacements along the generators and normal to
the surface whereas the method vb' usinp; flat elements gives
these displace~ents along the global cartesian co-ordinates.
However, because of the shallOl.rness of the shells,
(see Section III.2B) the surface co-ordinates along the gener-
ators are approximated by the Cartesian co-ordinates defining
the surface. Because o~· this approximation, a constant shear-
ing strain term -;ric is add'ed to the shearing strain of a flat
plate (Eq. 3-9c). This term does not reduce to zero near the
flat corner b (Fig .. 4. 2). It is believed that this term· adds
extra stiffness to the free corner where the structure behaves
almost like a flat plate. This addition.is probably the cause
of the underestimation .of the corner deflection. The deficiency
of method .'a' ·in predicting the deflection of the free corner
needs further .investigation. The solution could possibly be
improved by refining the "grid. size or by the use of higher order
strain terms 11 ; But this will definitely entail additional
-69-
computational work.
The strain-displacement relationships for the curved
·element are dependent only on the twist curvature ~B irrespective
of the shape of the actual structure. To explain this further,
consider only the quadrant oabc of a structure of Type I, Fig.
4.1. One could build two cantilever hypars from this quadrant.
The first structure would have eclpes oa and oc fixed whereas
edges ab and be would be free. In the second structure, the
fixed ancl the free edges would he interchanged. If both these
structures are subjected to the same loadinr conditions, method
'a' would give identical deflections and absolute values of the
stresses.
The solutions bv, roethod 'a' for Strucs. 'l' and '2'
did not show any advantare of using a 16-term displacement
function for w-displacernent, ,.,hich ensures the slope compati-
bility normal to the boundaries of the adjoining element as
against the non-compatible 12-term polynomial used in °.ef. 19.
The solution obtained for Struc. '2 1 ~ith the inclu-
sion of complete rigid body modes 20 and that obtained in this
study , without the incl us ion, . did not show 111uch of a differ-
ence (Fir. 4.6). To study tte effects of inclusion of rigid
body modes further and also to evaluate the differences in the
solution usinr 16 or 12-tcrm polynomials for w displacement,
Struc. '15' was analyzed. The results are plotted in Figs. 4.37
and 4.38. It may be worthwhile to note that in this structure
the rise to span ratio is rather high for it to be considered
as a shallow shell (see Section V.2). The deflection profile
-70-
across the diagonal ob shows that there is practically no dif-
ference between the solution obtained by the use of a 12-term
polynomial for the normal displacement w, and the function used
by Banavalkar. The maximum difference of 2% is seen in the
corner deflection ob (0.090 inches by the present method and
0.092 inc~es in Ref. 19). Even the u-v displacements all over
the shell, obtained by the two methods were within 0.5% of each
other. The striking similarity in the results tends to confirm
the conclusion that both methods give the same results for the
uniformly loaded hypars. This view is also shared by Pecknold
and Schnobrich 21 . The comparison nay not be as accurate for
unsymmetrically loaded hypars where the 16-term displacement
function for the normal displacement w would possibly give bet-
ter results.
However, the comparison with results reported in Ref. 21
shows a difference both in the deflections and stresses (Figs.
4.37, 4.38). Though the deflection profile and the stress
variation are alike, the added flexibility of the curved ele-
ment with the inclusion of rigid body modes is apparent in Fig.
4.37, where the corner deflection is nearly 60% larger than the
one obtained in this study as well as by the method used in Ref.
19. Though the solution obtained in Ref. 21 used a 12 x 12 grid
size as against a 8 x 8 grid size used in this investigation, it
is not believed that the difference in results is due to refin-
ing of the grid size.
Analysis of the same structure by method 'b' using flat
elements, results in the corner deflection ob (0.123 inch)
being nearly 33% larger than that given by the method 'a'. As
-71-
pointed out el:;lrlier, the defiections by. wethod 'a' are given
normal to the surface whereas for the Pethod 'b' they are in
the global axes. However, this does not affect the corner de-
flection ob. !lore over' the deflection profiles along the com-
pression mewber oa and the tension member ab, are different for
the two methods. It is quite interesting to note that both of
these methods~ which give close results for edge-supported hypars
(see Table IV-3), could differ in the case of this structure
(Fig. 4.37). The inclusion of t~e rigid body modes in the
solution seerns to account for the correct behavior of the flat
corner but since no cowparative results - with experiments - are
presente d 21 , i.t
. . not poss1. b~ 1 e to comment on t h e va 1.1 d.1ty o f
1s
the method in Ref. 21.
IV. 5 THE COLPARISON 0F AfiALYTICAL AND EXPERU1E!JTAL wrnK
Because of the shortcowings encountered in rrethod

'a'~ the analysis reported hereafter is carried out by wethod
ib'. The experimental results used for the cowparison can be
basically cateeorized into three types:
(1) Hypars supported vertically alonr. the line of gener-
ators all around the perirreters. Strucs. '3', '4' and '5' come
under this catep;ory (see Table IV-1).
(2) Small scale concrete models of umhrella shells 65
(Strucs. '6 1 , '7' and 1
8').
(3) Umbrella shells havin rr standard corrui:;ated open decks
for the shell surf ace (Strucs. 1
9' - '13 1 ) .
All the above mentioned experimental tests were con-
ducted at Corne 11 1 except St rue. '5' ('Ref. 33) • The testing of
-72-
the concrete, hy:pars l~as .conducted as a part of a research pro-

: . . 65
ject c;:urren1:1Y ·in progress at the Cornell Un1vers1 ty .
: rh~ compariso~ between the analytical and experimental
results fpr the deflections, ~dge member stresses and the deck
stress es: .i·S given i~ Figs. 4. 9-4. 36. In all the analytical
solutions, the surface of the hypar is approximated by the tan-
.. gent planes drawn at the center of the element except in Strucs.
1
6' and '8' where these planes are drawn at points a.long the
free: boundaries ab an cl be (Fie. 4. 5). However, these two

structures were not reanalyzed because of minor differences
(<10%) in .the results of other siMilar cases.using both methods

of, transformation.
A. EDGE-SUPPORTED HYPAPS
The saddle shaped hypars (Strucs. '3 1 and '4 ') "rere
..
analyzed mainly to £ind the effect of rise on the value of
shear rigidity factor 'a'. The values of the central deflec-
tions are given in Tables II-1, II-2. In the experiments, only
the central deflection cS (Key sketch Table II-1, II-:2.) ·was
0
measured. The results obtained in the'analysis of .these
structures are used in thapter V, to study the effects of the
variation in the structural parameters.
Struc. '5' was a large scale model with the plan
dimensions of SO'x30 133 . A single layer of a cellular deck
(see Table IV-2) was welded to th~ edge Members usinp a warped
plate connection. ·The hat secti~n· ~as welded to the base plate
with spot welds l" o.c. The adjoining deck pant:?ls were butt
welded so as to develop the f~ll strerigth of the flat plate~
~ ;
-73-
The edge members were free to move in the plane tangential to

the shell boundaries but were supported vertically. A 3z111 dia-
meter, high-strength steel tie bar connected the points a and
b (Fig. 4.3). A uniform load normal to the surface was applied
by vacuuming the enclosed chamber. A predetermined tension
force was applied to the tie by weans of a 500 ton jack which
prevented the relative displacement between the points a and c.
However, t!1e details of the connections of members oa and oc
were such that there was no force on the rneMber at the ends a
and c. The members ba and be were free to move at the end b.
The stresses and the deflections were measured at various lo-
cations. The complete details of this test with the instrumen-
tation are given in Ref. 33.
In calculating the membrane constants for the dee~,
the stiffening effect of the hat is nerlected. The menbrane

stiffness calculated only on the basis of the properties of the
base plate~ is on the conservative side. Since no seam-slip
was noticed during the tests, t~e shear rigidity factor a is
taken equal to unity. The deck is highly orthotror>ic as is
apparent from the bending rigidities (Dy= 29,300 Dx)· The
bending constants calculated on the basis of the geo~etrical
shape are used in the analysis without modifications. As given

in Ref. 33, the equivalent projected loa<l is calculated on the
basis of equating the shear force at the point o on the actual
surface and that given by the membrane theory for an equivalent
projected load. The load intensity used in this analysis is
5% on the conservative side of the criteria given in Chapter
III, Section II1.4.
Th~ ei~e~i~erital defl~~iion profile alorig the ·1ines
~ -.,. ' 6
de and fg (Fig. 4.3) are corrected by subtracting the vertical

dispiacements at poinis d, e, £, and g. Th~ c~ntcr deflection:
c,ilculated analytically is 5% on the higher side of the exper-

imental d~flecti~n 2.30 inches (Fig. 4.9). As shown in Fir.
4.10, the axial stresses in the edge members are very close to
half the values given by the me:w.brane stresses. The reported 33
strain JT1easure:rnents on the edge members tend to confirJn this
observation. It is quite logical to expect the forces in the
edge members to be 101.·.rer than those given by the rrembrane theory
because a part of the load is carried by the flexural action of
the deck.
As shown in Fig. 4.11, the difference·'between the
analytical and experimental results for the shear stresses mea-
sured by the rosettes 1 and 2, is even less than 5%. The var-: ·
iation of the sLear force all over the hypar surface is shown·
in Fi:. 4.12. As expectec, the value of the shear force over
a major portion of the shell surface is less than that given
by the membrane theory. The increase in the shearing force
noted at the corners a and c is due to the restraint offered by
the tie; whereas the value of the shearing force in the fixed· · .
corner o is almost twice as that.given by the membrane theory.
The 'connection between the edge tTternbers and the deck
should be adequate· e~ough to car~y this high ~alue of shear.
The bend1ng stress~s- calcu.late'd -~i·e the· center of the ·span on
.,
the top of the hat, do not show good correlation.with the exper·
-75-
imental results. The bendin~ stresses.calculated usin~ the

effective inertia led (Chapter II, Section II.2.B) is 6.70 ksi,
whereas the measured total stress in the y-direction at the
same location is 12. SO ksi. One reason for this un<lerestima-
tion i~ that the measured stress is total whereas the calcu-
lated stress is only due to bending. Since in calculation of
the membrane stiffness only the flat plate was considered, it
is not l:nown as to what ex.tent the hat portion participated at
the center of the deck in resisting the membrane stresses.
The reduction in the moment of inertia calculated on
the basis of the full cross-section) is not affected by the
calculated compression stress in the top hat plate (the vari-
ation is less tl~an 5%). The change in the bending rigidity DY
does not warrant a new analysis.
B. CONCRETE lH'BRELLA SHELLS
The concrete hypars differ from the hypars using
corrugated orthotropic decks mainly in two aspects. For the
loads used in the elastic analysis of this study, the shell can
be considered as made of an isotropic material. Secondly the
bending and axial stiffness of the shell is quite comparable
with that of the edge me~ber.
The experimental work on Strucs. '6', '7' and 8' was 1
conducted at Corne11 65 . Strucs. 1 6' and '7' were identical

except for different eccentricity of the edge roe~bers (Fig.
4. 5). In Struc. '6' the beayns were located on the top of the
shell surface whereas they were located below the deck in Struc.
'7'. These structures were loaded uniformly using concentrated
-76-
loads applied discretely over the surfa~e, lvhereas only half

t~e
1
of structure was loaded in .the case of Struc. 81 •
The elastic properties of the concrete used in the

model were determined experimentally. In calculating the prop-
erties of the shell only the concrete section is considered.
The classical ceam theory which assuroes ~he linear variation of
the angle of twist is used in the. analysis. The beams are con-
sidered eccentric only in the z-direction. The bea~ prooertics
calculated are based only ?n the ribs projecting above the deck.
For Strucs. '6 1 and 1
7', the comparison betNeen the
experimental and the analytical results are shown in Figs. 4.13-
4.19. As shown in Fig. 4.13, ior Struc., _'6' the corrtpression rib
deflection o is about 10% sw.aller than.the experimental re-
• . a ..
sults whereas the free corner deflection ~b is about 5% larger
than the experimental value. For Struc. '7' (Fir. 4.17), the
deflection ob is about 60% and Sa is about 80% of the experi-
mental values. Though percentage-wise the error in .6a' in
Struc. '7 1 is about 40%, the maenitudes of the deflections are
very small. Except for t~e deflection profile along the dia-
gonal ob near the coluwn support, the general shapes of the
profile agree fairly well with the experimental values. The
deflection profiles of ~he tension members ab in both the
structures) where the bending action in the shell doIBinates
• • ,1
over the membrane action, is very good and al~ost parallel to

the one observed experimentally.
TQ verify the idealization of the edge member, Struc.
'6' was rea~alyzed, but a certain portion of the deck was in-
-77-
eluded as the effective width in recalculating the beaf1 prop-

erties. The modified eccentricity of the beam with respect to
the deck and properties were recalculated. It is obvious that
in doing so a certain portion of the deck is duplicated, with
the result that the properties of the edge members are over-
estimated. For the same structure it was found that the free
corner deflection ob remained alniost unaltered (0.022 instead
of 0.023) ·whereas the deflection oa reduced from 0.016 to 0.012
inches. This observation shmvs that important deflections are
insensitive to the edge member properties for this particular
structure. However, there is a redistribution of the bending
and axial stresses in the shell, which are of a relatively
sJllall magnitude. The upturned bearis used in Struc. '6 1 seem
to have a pronounced effect in reducing the corner deflection
ob as seen from the analysis as well as experi:::ients. The free
corner deflection ob for Struc. '6' is nearly half that of
Struc. '7' whereas the compression rib deflection cSa for Struc.
'6' is larger than that for St rue. '7'. These points are
further discussed in Chapter V.
Because of the varying size of the edge members, the
axial forces are plotted instead of axial stresses. The ratio
of the calculated axial forces to that given by the membrane
theory is 70- 80% for the compress ior. meml-:ers oa and oc and
50-60% for the tension rne~bers ab and be. The analytical and
experimental values of the stresses for the tension member are
in close agreel'!lent, 1vhereas the analytically calculated results
for the cor.pression. members are on the conservative side.
-78-
Even though part of the vertical load near the column is car-
ried by th~ concrete shell, in order to satisfy the static
equilibrium·for the vertical load it appears that the experi-
mentally measured forces in the compression rib are quite low.
The axial and the bending stresses are measured along
the diagonal ob at an angle of 45° with the x and y axes (Fig.
4~15). The measured axial stresses show excellent agreement
i..ti th the analytically calculated value of 72 psi. An important
point :to note is that the calculated and the experimental, values
are about 34'% higher than those given by the rnembrane theory at
a load of 4.0. 9 psi, the reasons for whicl:i. are not readily
apparent. The values of bending stresses are very low and arc
not compared here. The variation of the shearing force is
plotted all over the shell for both the structures. Though
there are minor differences in the shape of variation of shear-
ing forces, two important observations can be mad~. The values
of the shearing forces over a substantial portion of the shell,
are larger than those given by the membrane theory. The shear-
ing force near the colu~n is nearly twice as large as that
given by the membrane theory. Tlds sudden increase in the
. shearing force clearly indicates that the shell participates
in tr:nsrnitting a certain portion of the vertical load. The
same behavior is also noted in Struc. '5'.
Struc. '8' is .the same as Struc. 1
6' but it is sub·
jected to an unsymmetrical load (Fig. 4.20), where half .of the
structure is loaded uniformly. Only half the structu;re along
the line ·cf. €Fip. 4. 20) is analyzed, using 1.6 elements in each
-79-
quadrant. The statically equivalent load is used in one

quadrant. The central column is idealized by means of con-
centrated elastic springs as given in Chapter III, Section

III.2.F. The comparison between the theoretical and the ex-
perimental results is shown in Fig. 4.20. The deflection pro-
files appear to be quite reasonable though the magnitudes of
the deflections ob and oe are 30-40% on the lower side of the
values obtained experimentally. A static chect for the ver-
tical load is satisfied at the center column though a dis-
crepancy in the overturninp woment is noted.
A highly irregular pattern of axial forces and moments
is obtained which unfortunately could not he verified properly
because of the difficulties encountered during the experiment.
A better solution can be obtained by using a finer grid (64
elements in a quadrant) and also by using work equivalent loads.
It was not possible to check the imp1.ovement in the solution
because of a limited comT'uter core capacity. The example how-
ever, clearly showed that the theory can solve unsy~metrical
loadinr conditions such as wind load, etc., and can satisfac-

torily predict tr.e overall behavior of the shell.
The corner deflection <\ in Struc. 1
8' is nearly
three times as large as that ·obtained for the uniformly loaded
Struc. '6' . The increase in deflect ions in the l~aded quadrant
is mainly due to the twistinf of the shell about the line ah.
C. UMBRELLA SHELLS WITH STA1'!DARD CO~_RUGATr.D DECKS
Four medium scale urnLrella shell models (Strucs. '9',

'11',. '12' and '13'; Table IV-1), 12x12' in plan
- and with a
-80-
rise of 14.4 inches, were tested at Cornell. Struc. '10'

is a hypothetical structure analyzed to study the effects of
change of shape in Struc. '9' due to the excessive deformations.
Self-tapping screws were used to connect the adjoin-
ing deck panels and also to connect the deck panels to the edge
members. The main supporting edge member frame consisted of
circular pipes (for sizes see Table IV-1) connected eccentri-
cally below the deck.
For the structure havinp, two decks placed in a mu-
tually perpendicular manner, the decks were not only connected
along the peripheral edges but were also connected intermit-
tently all over the surface. In the case of the two deck sys-
te~~ the bottom deck was directly connected to the edge member
whereas the top deck was connected to the bottom deck (Fig.
2.6). All structures were supported at the center column and
a uniform load was applied using pressurized canvas rubber bags
with. one bag placed under each quadrant (see Chapter VII).
The properties of the decks used in the analyses are
given in Table IV-1. The gaee thickness of the deck was
checked by the microweter screw and the properties correspond-
ing to the uncoated decks are usec1 in the analyses. To account
for the effect of rise, the shear rigidity factors used in the
analyses are modified from the values obtained by the flat
shear tests (Figs. 2.7 and 2.8). The reduction in these values
of a is roughly 25% for.the single ~eek whereas it is about 15%
for the double decks. Zero torsional fixity between the deck
and the edge members is assuwed for all the structures analyzed.
- fl 1-
In order to have a bett:~:r und:::rstandin:.;r of the he-

..
havi.or of these· structures, t}:ey ;:ire classified into two cate-
gories. This clJssification is based on the ratio of the rel-
ative stiffnesses of the deck and the supporting edge J1'1embers.
Strucs·. '9 1 and '10 1 are considered to have flexible edge f!'1em-
bers whereas Strucs. '11', '12' and '13' are consiclered to have
very stiff edge rne~bers. The edge ~embers used in Strucs. '11',
'12' and '13' arc 4.37 ti~es stiffer axially an<l 236 times
stiffer flexurally as compared with the edge members used in
Strucs. '9 1
and '10'. This large difference particularly in
the bending stiffness alters the behavior of the umbrella
shells.
C. 1. INVBf!.TED U~ 1 BRELL1' SHELL lflTP FLEXIBLE
FI'GE ; "EPBERS
Strucs. '9' and 'H'' \·.rere analyzed us inp- the boundary

condition u (Table IV-2) ,,1J1ich assumes full horizontal fixity
between the edge Members and the deck. The convergence char-
acteristics for the corner deflection ob for Struc .. '9' are
sho~m in Fip. 4.22. Fy refining the grid size, the free cor-
ner deflection increases. This is because of the effect of

eccentrically connected edge members (Fig. 3-12). The difference
in the corner deflection 6b bet¥een 6x6 grid size and that of
8x8 grid size is less than 2%.
The deflections and the edge ~ember stresses obtained
for Struc. '9' are compared witr the experimental results in
Figs. 4.24-4.28. The analysis underestimates the deflection 6a
by 32% whereas the deflection ,5e is overestiMated by 40%. The
difference between the analytical and experimental results for
-82-
the corner deflection cb is 10~6. Comparin,? the' relative magni-

tudes of thes8 deflections (oa, ob and oe)' it is apparent that
in the case of a flexible edge member the free corner deflec-
tion ob is of utmost importance. The shape of the deflected
profile for the member ab and the relative deflections between
points a and b, by theory and experiments are in close agree-
ment (error~ 1%). T'1e reasons for the underestimation of the
compression rib deflections are discussed later in this section.
The corner deflection oh.., is greater than 10% of the
rise of the hypar shell, \'Jhich is 14. 4 inches. In other '11ords,
the change in the shape of the structure is quite important.
To estimate the effect of the change of shape, a very approx-
imate method was used whereby the same structure (Struc. '9')
was reanalyzed by only modifyinp its rise from 14.4 inches to
13. 8 inc.hes. The reduction of 0. 6 inches in the rise was cal-
culated by taking half the difference betveen the relative de-
flections of the points a and b. The analysis of Struc. '10'
using the.modified rise, shows an increase in deflections. The
error in the deflection 6b in particular is reduced further to
4%.
_t\ coinparison between the experimental and the analy-
tical results for the axial and bending stresses, and the abso-
lute value of the total stresses for the edge me~ters is given
in the Figs. 4.25 and 4.26. The bending and the total stresses
show very good correlation with a maximum error of -15% for
the tension members. Comparing the analytical and experimental
results for the axial stresses, it is noted that the calculated
-83-
compression: s~resses for members oa and oc are on the high side

whereas in case of tl:e tension P"•2'mbers ab and be are on the lo-w
side. The measured axial stresses are only about }th in ~agni
tude of the total stresses and therefore the devi~tion (-55% for
,the member ab) between the theory and experiment is not consid-
ered to be a s6rious handicap.
To examine the validity of the solution and also to
help to understand the behavior of hypars, 'the variation of
th~ bending moment My and the in-plane shearinp force Nxy are
plotted over the shell surface (Figs. 4.27 and 4.28). Along
the colurm line 1 (Fig. 4. 2 7), the deck bends with the tens ion
member like a cantilever (negative moment) whereas in the in-
terior of the span' it acts as a simply supported span between
the opposite edge Members. Along the column line 8, 'the deck
has a repion of negative bending moments near the supporting
column. The variation in the shearing force (Fig. 4.28) is
similar to that indicated for the concrete hypars (Strucs. '6'
and '7'). Near the center of the quadrant, the shearing force
Nxy is larger (by 10%) than the values given by the membrane
stresses. Hm,1 ever ~ one major difference noted between the con -
crete and corrugated deck hy~ars is that near the column the
deck does riot carry a substantial portion of the vertical load
as is seen in1the case of Figs. 4.16 and 4.19. The comparison
of the axial, bending and total stresses calculated by theory
and measured experi~entally at point e is given in Table IV-4.
The, calculated stresses are compared with the aver-
age measured values obtained for the top anrl bottom deck. Be-
-r.4-
cause; pf !he very small :magnitude of the stresses, the varia-

tion.in
. '
their measured value was extreme. The variation in
the measured axial stress ranges from 210 psi to 1780 psi
whereas that in the bending stresses ranges from 140 psi to 2840
psi. Though the calculated values appear to be in the vicinity
of .these measured values, a direct comparison would not be
fruitful.
·. ··.
In the analytical solution of Struc. '9', it-is noted
that the deflection oa is underestimated. Fir. 4.29 shows a
. .
typical connection between the tension Member ab and the com-
pression Member oa. Because of the eccentric connection be-
tween the deck and edge members, all the node points are along
the.top of the edge members ab and oa. The in-plane forces on
the ~ember a~ are transferred eccentrically to the member oa
at the noc,~ a, resulting i:i its up1..1arcl deflect ion as shown in
Fig. 4.29. In order to illustrate the effect of this eccentric
: . ·.·
. transfer of the in-plane forces 9 Strucs. '9' and '11' are
analyzed for the two boundary conditions V and VI (see Table
Ill - 2) •
For Struc. '9', inspite of certain redistribution of

forces due to the change in the boundary conditions, there is
practically no change in the ~~flections oe and ob. The re -
.lease of the in-plane shear of 328 lbs. ac·ting eccentrically
at point
0.17 inches (Fig. 4.30). This shear, if resisted entirely by
.,
the compression member oa acting as a cantilever supported at
. . .
point o, produces a deflection of 0.22 inches. Except for the
-85-
bending stresses in the tension ~emhcrs ab and he, the changes

in the stresses for both the edge meJT!bers and shells are in-
significant. As sho 1 vn in Fig. 4.31, the decrease in the verti-
cal shearing force due to the release of the in-plane forces
results in the reduction of bending stress at point a in mew-
ber ab.
From the consideration of the ~arnitude of the in-
plane shear and its eccentric transfer, Struc. '11' represents
an extreme case. As shown in Fig. 4.32, the deflection pro-
file along the diagonal ob rewains practically unalter~d for
both boundary conditions for St rue. '11'. Because of the very
high in-plane rigidity of the 311 diameter pipe, the value· of
the in-plane shear developed at the junction a (Fir. 4.29) is
quite large (742 lbs.). Though s~all in ~agnitude, the in-
crease in the compression rib deflection oc and oa is alTJlost
200%. The increase in deflection exceeded that which ~·rnuld
have been obtained by considering the edge Members oa and oc as

cantilevers, acted upon by the eccentric shears at points a
and c respectively. A srrall increase in the deflection ob is
noted and it JTlust be pointed out that the transfer of the
eccentric force also exists at the junction of the tension mem-
bers but it is of minor i~portance.
In the case of concrete hypars l·.rhere full fixity be-

tween the edge meTJlber and the shell exists, this transfer of
ec~·.entric forces in two wutually perpendicular cHrections does
not present a problem. To get an exact solution for the dis-
continuities between the edge member and the steel deck, equa-
tions of compatibility Hill have to be satisfied at the addi-
tional nodal points thereby incr~asing "the complexity and the
,,
stotage requirement for tlic computer program. Both cases pre-
sented here', particularly Struc. ,· 11' j
'
represents an extreme
···oi. ·pt6ble ms. which ~il'l be hardly en'counte;ed in. practice.
1
·class
.. ~ . ' '
Besides the eccentric connection, the in-plane stiffness of

circular pipes is equal to the vertical bending rigidity. On
the assumption of full horizontal fixity between the edge mem-
beis and the deck, the horizontal stiffness attracts high in-
. plane shears, the ·T!lagnitudes of 'vhich raise the question of
its validity.
In practice, the rolled sections such as channels
and I-sections have very s~all in-plane stiffnesses as compared
with their bending stiffnesses. Secondly, these members will
be usually connected along their shear centers by weans of
warped plate connections (Fig. 3.10). One way to correct the
deflection oa is by applying the moments, equal in magnitudes
but opposite in directions, to those produced by the eccentric
shears at the junction of two eccentric mewbers (Fig. 4.29)
and recalculate the deflections of only the supporting frame.
Since the exact amount of horizontal fixity is not
h"Jlown, the other al tern a ti ve is to rf analyze the structure with
a complete release of the in-plane forces (boundary condition
VI) ancl use the conservative results for the design.

C.2. U!'fEPTED m~BPELLA SflELL 1 \TITH
STIFF EDGE t •Ei''BERS
Strucs. '11' and '12' used single corrugated decks

1·~hereas Struc. '13' used two perpendicularly placed in tercon-
-87-
nccted decks. The structures are analyzed using the boundary

condition V~. This boundary condition is on the conservative
side as far as the computation of deflections are concerned.
Fig. 4.33 shows the comparison of tlie experiIPental
and analytical deflection profiles along the diagonal ob, for
all the three structures. Besides this, the comparison between
the measured and calculated deflections at points a, b, c and
e is given in Table IV-5. During the experi~ent5~ difficulty
was encountered in obtaining tl'.e symmetry of deflections. The
unequal rate of leakage froni each canvas bag, placed under the
quadrant resulted in an unequal pressure loading being applied
to different quadrants . In order to show this resulting un-
symmetry in the solution, Table IV~S shows the average, waxi-
mum and minimum measured values for the deflections. P. cow-
parison between the results is based on the averaee value. In

general, the shape of the deflection profile along the diagonal
ob shows a reasonably pood correlation between theory and ex-
periment. The deflection 6e at the center of the quadrant for
a sinple deck hypar (Strucs. '11' and '.12') is ovcresti111ated
by the theory whereas the deflection for a double deck (Struc.
'13') shows a difference of only 10% from the measured value.
Except for the minor scatter of the deflections 6 8 , 6b and 6c'
the analytical results are within 15% of the average experi-
mental values.
The axial and bending stresses are measured at five
locations (Table IV-6). For the bcndin~ stresses greater than
2000 psi, the experimental and analytical values show a devia-
-88-
tion -of less than 20~. For very small 1J'lapnitudes of stresses
(such as less th:m, 2000 psi) the calcula.tion of the ~rror hased
'
on the measured stresses 1•.•ill ·be misleading.· T~e acquacy of
. .,. " ' .' ~
measurer.ent for the small magnitudes of stresses is. al\,rays

less. The. measured total stresses also show .a fair amount of
agreement with the analytical solutions.
The major discrepancy arises in· the comr~rison
, .. be-
tween the meas.ured and the calculated axial stresses.. Based
'. '! .
. '
purely on the Tl'ernbrane theory? the maxiwum axia.l .stress should

·be 1570 psi; as against this, the me~stir~d value of ~tresses
reaches as high as. 2440 psi (Struc. '11 ') which is. nearly 55%
· . ·,1
larger than that given by the J11ernb rane · theory. This appears
,,, I,
,.
inconsistent i~i th the expected behavior, since a part of the
load is also carried by the b~nding k~tion.
In order to unders'ta.ncl' t!1e ·difference. in behavior be- . I{'
· tween the single .layer and doubl'e layer d~cks, Fi).Js ... 4. 34 and
4. 35 show. the variation of the axial stresses and the ,verti-
cal shearing forces carried by the edpe mepber,. For Struc.
· '11'' (which has a 28-G single layer deck), bot1' the...~ompression
member oa and. the tension member be placed across . th.~ corruga-
tions (along the weak axis), carry hith axial loads as compared
·to the me~bers ab and oc, placed perpendicular.to the direc-
tion of the corrugations. This trend is also observed exneri-
,
mentally. Because of very 1011 in-plane stiffness across the
corrugations, the effective area of the deck resisting the in-
p 1 ane s h.. ear a 1 ong w1•t.1
t t.:e
l e d re mem,Jers
1-, oa .lb.
anL~ c is very srial l
· , , and therefore the entire shearinlS forces are resisted by the
-89-
edge members alone. As against this, in a direction along the

corrugations a part cf the deck shares the in-plane shear and
subsequently results in the reduction of the axial stresses
in the edge members.
Fie. 4.35 shows the transfer of the vertical load
to the edge members. ~ith the strong axis of bending placed
parallel to the lines oa and be (Fie. 4.2) the deck basically
bends between the supporting lines oa and be. \•Tith
. Dy = 18. 4 5
Dx, practically no load is transferred directly to the edge
meMbers oc and ab. However, from the conditiors of compatibility
at points b and c, the member be is supported at its end by
members ab and oc. The nepative shearine force at the point
b on the wember be and the constant shearing forces along the
members oc and ab confirm this expected behavior~ This manner
of transfer of load for a sin~le deck is also reflected in the
bending stresses at points a an<l e (Table IV-6) '\'!hich are
higher than tf>.ose for double decks (St rue. '13 ').
The measured axial and bending stresses at the center
of the quadrant were hir;hly erratic and did not show any con-
sistent behavior. The minimum measured bending stress was
half the value of the maximum measured value at the same lo-
cation. This wide range of scatter is due to two reasons, first
the magnitudes of stresses.are too srall to be ~easured re-
liably and secondly there was an unsymmetry due to unequal
pressure loading. For completeness, the comparisons between
analytical and experimental values for the decl: stresses are
given in Table IV-4. The bending stresses for Strucs. '11' and
-90-
'12' are overestimated by the analytical method whereas they

are underestimated for Struc. '13'.
Fig. 4.36 shows the variation of the in-plane shear
force Nxy over the entire shell surface for Struc. '11'. i\n
almost identical variation in the in-plane shearing force is

also obtained in Struc. '12' (24-G sinr;le deck) which has 61%
j ·~ '
larger shear and bending rigidities than those of St rue. : '11'

(28-G single deck). For Struc. '13' with d 28~G double layered
deck, the shear force distribution is very similar to that.ob-
tained for Struc. '9' with 1 1
diameter flexible edge mewbers.
However, the maximuw values of the shear force a.re about 5-10%
lower for Struc. '13'. The only noticable difference for the
variation of the shear force for single and double deck struc-
..
tures is that, in the case of the former structure, the maxi-
mum value of the shearing force does not exceed the shearing
force given by the membrane theory whereas it exceeds the mem-
brane shear force in the latter case. It may be of interest
to note that the results for the deflections of the deck are
quite close to those reported in Chapter VII. With the stiff mem-
bers, as those used in Strucs. '11'~ '12' and '13', the deflec-
tions along the free boundaries are small and therefore the
behavior of the shell is quite close 'to that of an edge-sup-
ported hypar for which; as pointed out earlier, wethods 'a' and
'b' give the same results.
The salient features differentiatine the beliavior of
the hypRr with very stiff edge memhers (Sfrucs. '11', '12' . and
'13') and the behavior of t~e·hypar~ with very flexible edge
-91-
members (Strucs. '9' and '10') are further discussed in detail

in Chapter V. The effect 6£ the edge member weight on the
behavior of hypars is also discussed in Chapter V.
IV.6 SUMMARY
The validity and the accuracy of the finite element
methods were assessed. Both approaches were found to converge
sati~factorily. A grid of 6 by 6 gave essentially the same
results as a grid of 8 by 8.
For hypars with fully supported edges, both the flat-
element and the curved-element methods yielded deflected shapes
that are identical with those given in the literature. Satis-
factory agreement was also found with experimental results
even when the effects of eccentric edge members were included.
However, the deflections of flat corners, such as those at the
outside corners of umbrella-type hypars, are underestimated by
I
the curved-element method. The flat element approach predicts
the experimental deflections and stresses of various types of
hypar structures with satisfactory accuracy.
CHAPTER V
QUALITATIVE EFFECTS OF PFJ:~CIPAL VAPIABLES

· ON BEHAVIOR GF HYPARS
V.l INTRODUCTION
. '
Based on the analysis of some selected structures
(Table IV-1) it is possible to show qualitatively the effects
of different parameters on the behavior of a ·hypar shefl.
Since. the number
'
of parameters affecting the behavior of the
shell is quite large and iheir interaction is very complex, _
attempts to show their effects on the structural behavior by
means Of forlfl.Ulae would involve extensive COJTlputational work.
During the following discussion so~e of the parameters which
were not investigated are ~entioned.
The structural variables affecting the behavior of

the shell can be broadly classified into four categories:
(1) Geometric shape of the hypar shell.
(2) Properties of the deck used as a hypar surface.
(3) Boundary conditions.
( 4) Loading.
V.2 GEOMETRICAL SHAPE
All hypar surfaces have a constant twist curvature

~~ The effect of rise to span ratio on the central deflec-
tions of the saddle shaped hypars (Strucs. '3', '4', Table IV.l)
is illustrated by plotting the deflections against the non-
.
d 1rnens .
1ona :z
1 parameter o f AB ( F1g.
. 5 . 1) . • t.h
l.\T1 t h e increase of
c
-92-
-9 3-
rise to span ratio, the curvature of the surface increases.

This increase in curvature reduces the bending action of the
shell whereas the membrane action is increased and this even-
tually leads to the decrease in the central deflection.
The cf fect of the rise can be shown by comparing the
central deflections of a siinply supported 28-G square plate
(60;'x60" in plan) with those of a hypar having a rise of 7. S
inches (rise to span ratio= 1/8) (Fig. 5.1). The deflections
in the latter case are nearly 40% of those obtained in the
former case.
The sensitivity of the structural behavior to the
change of rise is well dewonstrated by comparing the deflec-
tions and the stresses for Strucs. '9' and 1
10' (Table V.l)
where the rise of Struc. '10' is only 4. 3 % smaller than that
of Struc. '9'. The increase in the bending action with the
reduction in rise is evidenced by the increase in the deflec-
tions oa, ob, oc, and oe and also in tpe hending stresses.
The bending stress in the center of the deck increases
from 1870 psi t~ 2130 psi. Accordinp to the ~embrane theory,
the in-plane shear force is inversely proportional to the rise
to span ratio (N xy = q 2CAB) . The same trend is also observed
in the increase of the in-plane shear and the axial edge me111-
ber stresses (Table V-1).
For larger values of AB CcA < 1
8)' the membrane action
c2
is reduced to a minimum and the entire load is practically
carried by bendin~ action. The calculation of the in-plane
shear on the basis of the membrane theory) as C approaches zero,
-94-
is meaningless. The theory given here is primarily.good for

rise to span ratio of ~ } (AB/C 2 : 15) but it can be used for
·greater rise with loss of accuracy. From the construction
point of view, the choite of rise to span ratio will be also
governed by the warping of the deck.
V.3 DECK PROPERTIES
In the case of an open form deck, the membrane elastic
constants Ext' Elt and the bending constants Dx, D1 and Dxy
(Fig. 2.1) are very small and their influence on the behavior
of the shell is insignificant (for the stiffness coefficients
see Appendix B). However, in the case of the closed cellular
decks, though 'the magnitudes of DX ~rid D1 are small and insignjf-
·icant, Ext' Elf and Dxy are compar~bl~ in magnitudes to the
properties EY~' Py and Exyt and there£6re their influence on
the structural behavior cannot be overlooked. Since onli one
structure was analyzed for the celltilar deck (Struc. 'S'), the
discussion given below primarily concerns the open form (stan-
dard sinusoidal) ·decks~
A. SHEAR RIGIDITY
According to the membrane theory, the normal 16ads
on the hypars are carried by the in-plane shearing force xy • N
In reality, though a part of the load is carried by bending,

the magnitude of the in-plane shear Nxy is quite comparable to
that given by the membrane theory (Figs. 4.12, 4.16, 4.19, 4.28,
and 4.36) and even exceeds it in certain regions of the shell.
Therefore the in-plan~ shear resistance Exyt • Geff•nt, is very
important in the behavior of hypars~ As discussed in Chapter
-9 5-
II, the effective shear modulus (Geff) is obtained by reducing

the shear modulus of the material by th~ factor a. In the
s:.1ddle shaped hypars (Strucs. 1 3' and '4 1 ) for a rise to span
ratio of 1/5 (AB/C 2 = 25) (Fig. 5.1), the reduction in the
shear rigidity a. from 0.06 to 0.04, shows an increase of nearly
30% in the central deflection. The behavior of the hypar shell
is very sensitive to the values of a< 0.10.
To illustrate the effects of a. on the behavior of
the shell, the results for Strucs. 1
13 1 and '13a' are compared
in Table V-1. With the increase in the value of a., the deflec-
tions (oe' ob) and the edge memter and deck bending stresses
are reduced whereas the axial stresses in the edge members and
the in-plane shear force Nxy are increased. Except for the
axial forces in the edge meribers and the in-plane shear Nxy'
the response of the structure to the variation in Cl is similar
to that of the variation in the rise to span ratio. The
ootimum value of a. in orthotropic hypar structures is a = 0.1
since larger a. does not improve the behavior much. Factors
which improve the value of a "rnre already discussed in Chapter
II.
B. THICKNESS OF THF CORRUGATED DECK
In the case of an open deck the important memPrane
properties such as Eyt' Exyt and the bending riridity DY are
directly proportional to the thickness of the deck. However,
it must be pointed out that the bendinp rigidity of the deck
is small compared with the membrane stiffness. norcover it is
the change in the shear stiffness that influences the behavior
-96-
of the hypar shells and therefore the effect of increasing theI I ' '
·,.
thickness is analogous to that of increasing the. value o,f a..
To substantiate this observation th.e comparison between the de-
flection and stresses for Struc. '11' (28-G single deck) and
Struc. '12' (24-G single deck) is given in Table V-1.
The variation of the in-plane shear rigidity, which
is directly proportional to the thicYness and the shear
rigidity factor a, also affects the manner in which the verti-
cal load is transferred to the supports by the membrane action.
Because of the high shear rigidities for the concrete struc-
tures ('6' and '7') and Struc. 'S' using the cellular deck with
the full effectiveness of the bottom plate, .the values of the
in-plane shearing forces show a substantial increase near the
supports (Figs. 4.12, 4.16, 4.19). The increase in the shear-
ing force indicates the participation of the deck in_ carrying
a part of the vertical load. As against thisJ Strucs. '9'-'13'
with low shear rigidity do not show any incr~.ase in the in-plane
shearing force (Nx.y) near the supporting columns . (Figs. 4. 28,
4.36). In other words, in these structures the entire vertical
load is primarily carried by the edge members.
C. HUMBER OF DECKS
As far as deflections and stresses in a hypar are con-
cerned increasing .the number of. decks has. the same effect on
the behavior of the shells as that of increasing the shear
rigidity factor a. and the thiclmess>. How~ve.r, this observation
does not app.ly for buckling (see .Ghapter VI)..
I • . .
As discussed in
Chapter II, the effectiveness of the deck in resisting the
-97-
loads depends upon the JTJanner in which two or more decks are
interconnected and connected to the supportinr edge merrhers.
However, it must be pointed out that.in order to avoid chatter
and get a better structural performance, it is desirable to
interconnect the decks all over the surface of the shell.
\'Then two decks are used, they are placed in a mut -
tially perpendicular manner and. this gives an equal bending
rigidity to the structure in both directions, thereby distri-
buting the applied loads more evenly to the supporting edge
wembers. The comparison. of the results for Struc. '11' using
a single deck·(28-G) and those for Struc. '13a' using the
double dccl~, all other constants being the same, shows that the
uniformity of the stiffness in Struc. '13a' has more even
distribution in the edge meinber_axial stresses (Table V-1).
Though the corner deflection shows practically no change, the
center deflection 5e for Struc.: '13a' is nearly half that of
Struc. '11'. The change in the bending stresses of the edge
member is very small but because of the increased membrane
action the bending stresses i~ the center of the quadrant are
reduced by nearly three times.
In practice, the use of a double deck with two decks
placed mutually perpendicular is more desirable than a single
orthotropic deck.
v. 4. BOUf'DARY CONDITIO~\rs
From the practical point of view, boundaries such as

siw.ply supported, knife-edge supported or fixed all around, are
not real is tic. Boundary conditions '~hi ch consider the proper-
-9 8-
ties of edge members· and the manner in which they are connected
~b th~ deck are r~alistic from the practical point of view.
A. E'DGE MEMBER PROPERTIES.
The edge mei:iber properties Ab, Iy, .I.z, J and r b. for
available rolled sections show varia.tions over a wide range.
A sufficient nm111:-er of analyses could not be carried out to
formulate any definite rules by which the effect of the varia-
tion of· these· individual properties on the behavior of the
·shell can be assessed·. Except for the. concrete hypars, the
analysis was carried out for zero torsional fixity and there-
fore the influence of the torsional constants .J and r is not
clearly known:.
To get the general idea of the effect of the stiff-
ness of the edge members, one can compare the results of Struc.
'13' with very stiff edr,e members and Struc. '9' with very
flexibl~ edge members. The difference in the behavior of these
two extreme structures is obvious from the deflection profile
along the diagonal ob (Figs. 4.24 and 4.33). In the case of
Struc. '13' because of very high bending rigidity of the edge
members, the deflections along the periphery are quite small
and the deck bends freely between .the opposite supporting
edges. The simply supported plate bendin~ action is quite
dominant in ttis case. Because of the small bendinp, rigidities
of the edge members in Struc. '9' , it appears 'from the deflec-
tion profile along the diagonal ob that it is the deck that
supports the edge wembers near the free corner and therefore
. ..
the deck stiffnesses (both bending and mewhrane}'are quite
-99-
important for this structure.

The fact that the corner deflection ob for Struc.
'13' is not very different than that of Strucs. '11' and '12',
where single layer decks with different shear ripidities and
thicknesses are used, 6i~arly indicates that the deflections
along the periphery of these structures priBarily depend upon
the properties of the edge members. In order to optimize the
interaction between the deck and the edge members to give a
satisfactory structural performance, the ratio of the bending
rigidities of the deck and the edge members would have an
optimum value between the tlrn extreme cases (Struc. '9' and
Strucs. 'll'-'13').
B. EDGE :MEMBER AND DECK CONNECTION
As sho.Nn for Strucs. '6' anc '7' (Table IV-1), the

eccentric location of the edge members affects the deflection
of the structures (Figs. 4.13 and 4.17). The difference in
behavior of the edge members is shown in Fi!!. 5.2. For umbrella
shells to reduce the vertical deflection for tl:e compres-
sion member? it is beneficial to connect tr:e deck on top of the
edge member whereas in the case of the tension JT1embers~ it is
beneficial to connect the edge member on the top of the deck.
The experimental as well as the analytical results for Strucs.
'6' and '7' seem to confirm this conclusion.
No comparative results are presented for the zero
and full torsional fixity, though results are· presented for
the full and zero in-plane fixity (TH) between the edge ~embers
and the deck (Figs~ 4.30-4.32)• Though the results are very
-1'10-
limited, it is believed that providing fixity along the periph-

eral edge' ~embers tends to attract ~ore vertical load on the
~dg~ members (Fi~. 4.31).
In the case of saddle shaped hypars, th~ increase in
area of the ten~i6n tie bar connectinp the lower corners (Key
. .
sketch 5 Table II-1) of the shell and the in-plane bending
benefic·i~'i
T
rigidity of the peripheral. eJge members have effects

.· . . • '. 10 , : .,
in rcducing;th~ bending action of the shell "' The effects
.'
of these v~riablcs May need further irivesiigation.
V. 5. LOADI:~G
All the conclusions given above on the behavior· of

the hypars are bas~d on the analysis for the uniformly distri-
buted verti~al loading. In reality the structures are also
subjected to unsymmetrical loads such as wind or drifting
snolr. The stren~th of' the· structurp under these· -kinds of .loads
is tested more severely than under the conditions of uniform.
loads. The unsymrnetrical.ly· loaded Struc. '8' shows the: cor-
ner deflect ion ob nearly -three .times as large as tha_t obtained
for the uniform loading condition~
A. EDGE MEMB.ER .l'1EIGHT
In case. of some shells, such as.umbrella shells, the

edge me~ber weight is distributed alone the periphery of the
s:he.11. The customary procedure of sI"ea.ring this load uniforir.ly
over the whole surface and analyzing the structure can lead to
a, gross. underestimation of hoth the deflections an<l the
stresses. To de~onstrate t11is, a!'. umbrella shell with each

quadrant of 20'x20' in plan having a rise of 4', (Table IV-1)
- l':Jl -
is an:,ilyzed. The edge Tl'ernber sizes and the deflection tolcr-

ances used for this structure represent the values ¥hich are
encountered in practice. The deflection profiles and the bend-
inp stresses for the edge rnenbers, wit~ and ~ithout the inclu-
sion of edge rner.ibcr 'mights, are shown in Figs. 5.3 and 5.4.
The weight of the edge ~eMher is 20%.of the total uniform load
of 40 psf over the whole surface. A simple frame analysis
consi~erinr only the edge meFher weight and edge mernhers, would
have given an increase of 0.74 inc~es in the deflection of
point a (as against 0.33 inches) and 0.20? inches in the def1ec-
tion of ~oint b relative to point a (as against 0.15 inch).
This shmrs the effectiveness of the shell in carrying the
weight of the edg:e meJY\bers. The cable and t};e arch action along
the diagonals ac and ob is evident in Firs. 5.3 and 5.4. The
increase in the deflections 8 a and 8 c produces an upward <le-
flection at the center of the span.
The axial stresses for both the tension and the com-
press ion members shm.; an increase of nearly 20%. This is equal
to the increase in the total load of the structure by the in-
clusion of the edge member weight. The bending stresses for
both the tension ancl the compression edge members show an in-
crease of nearly 50% in the rnaxivum stresses (Points o and a,
Fi7. 5.4). The increase in the bending stresses clearly in-
dicatcs the unconservativB assurnption of Srr'earing the edge
meIPber weight over the whole surface.
The effectiveness of the shell in carrying the weig~ts
of the edge members raises an important question as to the

method of construction. The ~ituation is analogo~s to that
encountered· in 'a composite construction using steel''beams and
concrete ,slab.· There.are three alternatives for the ;construc-
tion . . Depending upon the size and··shape of the shell, 1~i th the
decks in p<;>.si tiort, a hypar can be built on the ·ground and erected
. .
in position; or it can be built in place ·by using- an adequate
shoring for the edge members. In these methods of construction,

the effectiveness of the shell in carrying the weight of the
edge member wi11 be utilized. The third way of construction
will eliminate the shoring and d~pend entirely upon the strengths
of the ·edge members. The economics ~ill obviously decide the
method of construction.
From the analysis of- the different structures, it is
found that the axial stresses ih the edge wembers derived on
the basis of the membrane theory are ~lways overestimated
(Table V-2). Because of the relativ~ly small magnitude of the
edge meMber stresses in comparison with the bending stresses
and with the uncertainty in the calculation o:f the exact bend-
ing stresses, the design of th6 edge membet~ for the; axial
stresses based ori the rrernhrane theory cannot be consiclered to
be on a very conservative side~-
From the analysis of different structures it appears

. . . 1 parame
, a t . th e non- d unens1ona
th . aCn t prov1
. t er ~ .. d.es a goo d index
.
for th·e behavior of the shells. The higher the value of this
constant, the more dominant· is the membrane action. As dis-
-103-
cussed earlier, the beneficial membrane action reduces both

the bending stresses and the deflections of the shell. A
second good non-dimensional parameter would be the relative
stiffness of the deck and the edge ~embers. This however
would need further study.
CHAPTEr VI
INSTABILITY ANALYSIS ()f HYPAT~S
VI. 1. INTRODUCTION
The linear stiffness analysis given in Chapter III does

not include the effects of middle-surface forces Nx' Ny and
Nxy on the behavior of hypar shells. The omission of these
effects precludes the possibility of the analysis of insta-
bility of the individual finite elements. The accumulation of
the instabilities of the individual elements eventually leads
to the general instability of the structure.
In the case of a hypar ·ri th light gage steel deck
1
used as a shell surface, the effr:ct of the in-plane forces is

manifested in three different types of instabilities:
(1) Local Butkling - In the case of a !!-type open deck
or a cellular deck (Fig. 2.2b and c) the deck is composed of
flat plates. These individual plate components, depending
upon the thickness to width ratio and the boundary restraint
offered by the adjoining plates, may buckle locally when sub-
jected to in-plane compres5ive and shearing forces. In spite
of the uncertainty in the degree of restraint offered by the
adjoining plates, this local buckling can be approximated on
the basis of the stress level in each component plate 36 . The
effect of the local buckling on the behavior of light gage
beam section is discussed in detail in Pefs. 38, 39.
The local buckling of the individual plate components
-104-
-105-
results in the redistribution of the total stiffness of the

shell. The theory used in this chapter does not account for
the local buckling and therefore the effect of local buckling
on the shell stiffness cannot be predicted. The local bucklinp
can be prevented by choosing proper thickness to width ratio
for each individual plate element.
(2) Deck Bucklin. i:r - In this mode of buckling~ the edge
beams rewain stable whereas the deck, used as a shell, buckles
as a unit. To understand the deck buckling, consider the
umbrella shell in Fig. 4.2. The deck acts primarily as a com-
pression arch betNcen the points o and b, and therefore it can
buckle along the diagonal ob; but the shell edge members re-
main stable.
(3) Overall Buckling - The shell and the edge meribers
buckle as a whole unit. One can iITlagine an umbrella shell,
foldinr ·dol"m as an umbrella turned inside out. Overall
buckling could occur either simulataneously with the deck
buckling or it can happen after the decl• has buckled.
According to a si1nplified an3lysis by Parker 19 , the
possibility of overall buckling for the practical size of edge
members is very remote. Very high values of deflections and
stresses for both edge merr hers and the deck will indicate the
1
possibility of overall buckling. The conclusion that overall

buckling is very unlitely is further verified by Struc. '9'
(Table IV-1) tested at Cornell , 11•here l;' cliameter standard
pipes were used as eclge me~bers. The resulting structure was
too flexible to be used in practice. In spite of excessive
-106-
deformations (nearly one half the rise of .14. 4n), the struc-
ture did not show any tendency of overall buckling though the
deck buckled.
The present study was primaTily ~oncerncd WlLh d"rk
buckling. However, the overall instability due to the buckling

of the edge members can also be predicted from the load deflec-
tion curve. The assumptions used during the analysis and the
liwitations of the theory are as follows:
(1) A linearized stability analysis was carrie.d out to
predict the bifurcation point of buckling 51 , 66 . The prebuck-
ling deformations were within the limits of small deflection
theory.
(2) No attempt was Made to predict nost-buckling behavior
or the post-buckling streneth. To be able to predict the
post-buckling behavior, one needs to retain the higher order
strain terms in the strain displacement relationships and have
. l 67
h 1g1er or d er ma t rices
. . It is extremely difficult to formu-
late these rratrices explicitly and one has to resort to
numerical integration. The non-linear equations can be solved
by the use of methods such as Newton-Raphson scheme 25 , energy
I
searc~
.
tee h nique 67 , c t c.
(3) The possibility of local buckling was totally

neglected.
(4) The material was assumed to be linearly elastic.
(S) Buckling was assumed to be conservative 36 •
Both curved and flat elerent approaches were used.
VI. 2. INCREHENTAL MATRIX FOR THE DECK AND EDGE ~fE~BERS
In order to represent the instability effect in the

finite element analysis, the change in the potential energy
due to the middle-surface forces Nx, NY and Nxy' which occurs
durinr the flexural action is to be included 66 . . For the con-
stant values
' of N x , Ny and tJ x y at any prescribed
' load level,
the potential energy due to in-plane forces assumes the form of,
b a 2 2
f f (N (~!!) +N (~l·J) +ZN
xy (~oX' )(~oy·' )]dxdy·
11 1 1
V,.
l·J
= -12 · X oX y oy 6-1
0 0
With the inclusion of the work done by the in-nlane

forces, the total potential energy can be \'rri tten as,
¢ = [~) [K] {i\} + 4-1 [N] {6.} - [6.] {P} 6-2
For stable equilibrium, the first variation of the total' po-

tential energy is zero.
{P} = [[K] + [rJ]]{6.} 6-3
{P} = [K]eff {A} 6-4
The matrix [N] is called the incremental matrix and.it is ob-

tained by the second differential of the potential energy (Eq.
6-1) with resocct to nodal displacements.
a2,,l'·J
[I'J] . .
l)
= atJ.. a6. 6- )
1 J
The coefficients of ti1e increJT1ental ·matrix [N], depend only
upon the geometrical parameters of an element, such as its
length. The incremental matrix is identical for both ortho-
tropic and isotropic cases.
For constant values of Nx' l';y and ?lxy the incremental
-108-: .
matrix for' a .sffal'low sh'ell h)tpa.r· elemf?nt and that for· flat
plate· e;I.ements. ar,e: identical. ~he OJ?.l.Y diffe.rence is the manner
in which ?the Jn-plane forces are determined (see Chapter I I I) .
'' ' :
.The incremental matrix [N] for the deck is given in Appendix

D, Tables D-I to D-IV .
. Due to the presence of the , axial force 'i·l x , the
. effec-
. tive stiffness of a beam element is also modified. Neglecting
the torsional mode of buckling, the potential energy due to the
axial force Nx can be obtained by putting the values· of NY an<~
N
xy .equal .to .zer.o_,. in I.q. 6-1. The procedure for determining
the increwental matrix for the beam !s identical ~,ri th that for
the deck. The incremental matrix for a beam element is given
in Appendix E. The incremental matrix for the whole structure
is obtained by the same procedure as described for the formula-
tion of the master stiff!less matrix in Cr.apter III.
VI.3. CHECKING OF THE. INCPEt~NTAL MATRICES
Before analyzing hypar structures, it is necessary

to establish the validity of the increrwntal matrix given in
Appendix D. The determination of the in-plane lJuckling loads
(Nx, NY, Nxy) for flat plates provides a ~ood ~;eek. At a

critical load, absolute magnitudes. of the deformations are
indeterwinate and the determinant of the effective stiffness
matrix [K]eff must vanish;

l[(K] + >. [N]]I =·o 6-6
1\lhere A. is the eigenvalue which 'depends upon the applied state
of membrane stress e.g. for an uniaxial1y·conipressed plate,
along· the x-direction it ·11ii1 giv~ eirenvalues corresponding
-109-
to the in-plane force Nx (see Table VI-1). From the structural

point of view~ one is only interested in· 'the Y'1inimum critical
load. To achieve this, it is necessary to rearrange Eq. 6-6
to get the first eigenvalue corresponding to the critical load 68 .
1 r} c11 + CK1- 1 CN111 = o 6-7
The negative reciprocal of the first eigenvalue of the matrix

[Q]: where,
[Q] = [[K]-l[N]] 6-8
\"Jill give the critical value of the IBembrane force. A sub-
routine named '·Nn.OOT!i available in the IBV system/360 Scientific

69
Su~routine Package , calculates eigenvalues and eigenvectors
of a real, square, non-symmetric matrix given in Fq. 6-8,
where both [K] and [NJ are real symmetric matrices and [K] is
real positive definite. In order to avoid underflow and over-
flow in the computer program, it is necessary to divide' both
[N] and [K] matrices by some large number, like 1000.
The first three problems solved were uniaxially com-

11ressed simply-supported ?lates with or without stiffrners (Table
VI-1). In general, the bucklinp in-plane force 'f'Jx (lbs/inch)
is given by,
6-9
where K is a constant depending upon the aspect ratio a/b and

also on the relative stiffness of the plate and the stiffener.
The ratio of the stiffnesses of the plate and the stiffener are
given by non-dimensional parameters y and 8 36 .
y = EI
-D-b 8 = A
bt 6-10
-110-
1
1 is the moment of inertia of "the eccentrically connected
·.:stiffener calculated about the junction of the stiffener and

the plate. For these problems, the torsional mode of buckling
of the stiffener is neglected. The error for the values of K
for these three cases is less than 0.2% as compared to the
classical solutions. One of the interesting observatiohs for
these problems was that the plate with the aspect ratio of
a/b = 2, buckles in a double sine wave with zero deflection at
the center line. However, with the attached stiffener it
buckles in a single sine wave. \11hile analyzing only a quad-
rant of a plate, proper boundary conditions are-to be applied
to account for this behavior.
Since the shearing action is of primary importance
in the case of hypars, the shear buckling loads for a square
isotropic and for a 24-G standard corrugated flat deck were
also calculated. The value obtained for the critical shear-
ing force !Jxy in the i~otropic square is compared with Timo-
shenko36 and that obtained for the corrugated deck is com-
pared with r·•cfarland 48 . The error between the classical solu-
tions and that obtained in this study for the· shear buckling
(Nxy) is more than that for the uniaxial compression (Nx).
One reason for the greater error is that the assumed displace-
1rent field for the. displacement w (He-rinitian Interpolation)
closely approximates .. the buckled surface for an axial compres-
sion. To approximate the bµckled wave form due ·tQ shearing·
load, a greater ~µmber of elements is: required to -achieve equal
accuracy. The error for the critical shearing force for the
-111-
orthotropic deck is 7. 46% 1 on the high side cowpared with a

· simplified formula 48 • ,. However, according to the authors 48
·; their formula underestimates the critical load· by as much as
by·S%. Therefore the actual error may be considerably less
than 7.46%. ·The error for the isotropic plate with only 6
elements is 4.65% (Tabl~ VI-1).
The correlatioh between the classical ~olutions·and
the solutions obtained here is considerec adequate to sub-

stantiate the incremental matrices for the beam and the plate.
A further check uill be presente~ in the Section VI-6, for
the ca.se of an isotropic hypar for which a classical solution
is available 5 •
VI.4. INSTABILITY OF HYPARS
The fncreJl)ental matrix [N] is a function of the in-

plane forces Mx, Ny and Nxy· In the case of plate buckling
problems there is a complete uncoupling between the flexural
and membrane action. This enables one to formulate the (N]
matrix from a ?iven distribution of the in-plane: forces which
are predetermin~d, independent of the flexural action. In the
case of a hypar, or for tl:at matter al!y curved shell surface,
the values of Nx , I~
and. Nxy arc dependent on th~ deflect ions.
y
With the change in the applied load.inr:, the rna.gni tudes of the
in -plane forces also chanre. In other words, . there is a
coupling bett·1een the wernbrane and flexural behavior. T!1e in-
crease in the in-plane forces resulting from the correspondinr
increase in the load causes some of the ele~ents to undergo a
marked decrease in the effective stiffness. This reduction in
-112-
the effective
.,, stiffness of an element will adjust the incre•
mental force dis.tribution. The accurnu la tion of :·these local··
? ; '
'
elemen,t instabilities will eventually lead to 'Duckling. The :
discuss.ion of this membrane and flexural behavior .for the shell
structures is given in detail in Ref~; SI, 66.
The incremental [N] matrix used for both curved and
'
flat elements is identical and can be represented by,
6-11
The only difference is. that u, v, and w £or the flat

elements are measured along local axos whereas those in the case
of the curved elell"ents are measured along the tangent and nor-
mal to the surface. As pointed out in Chapter III, the tr'ansfor-
mation from the local to the global system for the flat ele-
ments can result in non-zero entries in all elements of the
[N] matrix and the watrix assumes a generai form of,
-1111 Tl11
[ll ]global = ---------· 6-12
11 21 11 22
whereas its basic form remains unaltered in the case 0£ the

curved element.
VI. 5. DETEf<J~INATION OF THE BCCKLING LOAD
The linear eigenvalue formulation for the determin-

ation of the e~genvalue and thereby the lowest buckllng load
is well documented.in the Refs. Slj 66 and therefore i~ is not
repeated ,here ... Because of tf..e lack of a. reliable eigenv.alue
subroutine for the large-order systems and sufficient computer
-113-
storage, Banavalkar could not use the direct eigenvalue approach.

Instead, a linearized load incrementation !T!ethod had to be used.
In the load incrementation method~ as used by the
author, the assumption is rnade that the in-plane forces nx ,
and Nxy are constant during an increynental step and are equal
in magnitude to the value at the end of each step. The proce-
dure of the solution can be demonstrated by the use of Fig. 6.1.
In the incremental step I, only linear analysis is
carried out by solving the linear part of the equation assum-
inr [N] as a null matrix.
[~] = [K]-l [P] 6-13
From the known values of displacement vector [ti.], corresponding
in-plane forces Nx, ?ly and nxy are calculated and the incre-
mental matrix [N] is forI'"'ed. The effective stiffness matrix
[K]eff is used in iterative cycle II.
The iterative cycles are continued till convergence
is obtained for the nodal displacements and consequently the
incremental matrix [N] is consistent with the defor~ation~.
It is found that for small incremental loads, convergence of

displacements is obtained within three or four cycles. An
incremental load 8p is applied on the modified effective stiff-
ness matrix and the increase in the displace~ents o and ti. is
calculated by finding new values of the in-plane forces at the

end of the step II by the iteration as ~escribed before. The
analysis is continued by applying the incre~ent of the load op
on the previously determined effective stiffness ~atrix.
Any sudden change in the lo.ad deflection curve be-

-114-
tween any two load .ieyels indicates the occurrence of buc~ling.
In this method it is. possible to predict only the range within

which the buckling occurs ... Since the solution near the un-
stable configuration is very sensitive, it is not possible to
determine the e.xact point of buckling.
VL.6. NUMERICAL RESULTS At\JD THEIR DISCUSSION

. . 5 .
Reissner analyzed the case of a simply supported
isotropic hypar with edge members having infinite rigidity
along their axes but having zero stiffness in a plane tangent
to the shell surface. The saddle-shaped hyparj Struc. '~'
(Table IV-1) was analyzed for these boundary conditions. The

deflection profiles at three points along the compression dia-
gonal bf are plotted for the load level of 0.20-0.50 kgms/cm 2
increased by the interval of 0.1 kgm/cm 2 . The deflection pro-
files along the tension and the compression diagonals are sh0WTI.
in Figs. 6.2, 6.3. The sudden change of deflection profiles
(Firs. 6.2, 6.3) between the loads 0.40 and n.so kgms/cm 2 .
clearly indicates that the bucklinr occurs between these two
limits of loads and moreover close to 0.40 kgrns/cm 2 . The
analysis based on the curved element for the sa:ine structure,
also predicts the load between the same range though some dif-
ferences in the defle~tion profiles are note~.
The load thus predicted is slightly higher than tl1at

given by Reissner (0.38 kgrns/cm 2) but this is because of the
fact that Reissner used linearized membrane analysis. To ver-
. .
ify this fact, analysis was carried out where.predetermined
membrane shearing force Nxy =~ was used to establish the
-115-
incremental matrix. From the deflection profiles not shoh~
herein, the buckling of the shell occurred between uniform

2
loads of 0.35-0.4 kgms/crn which is in the range of the loads
predicted by Reissner. One of the interesting points is that
for the linear elastic analysjs. the normal deflections w were
symmetrical about the 1 in es bf, dh, ce and a g (Fig. 4. 1).
Similar observations were also made by Deak 18 . However, ar~~T
the inclusion of the instability effects the symmetry of the

normal displacement is still retained about the diaronals bf
and dh but there is no symmetry about the lines ce and ag. This
is because of the readjustment of the effective stiffness due
to the in-plane tension and compressicn forces.
In order to assess the effect of edge deflection and
the stiffness of the edge members on the buckling of the hypar
d ec k , an urn b re 11 a s h_e 11 w1. c ratio
1 AB
ti . equ1va
. 1 ent to t h at o f
Struc. 'l' (Table IV-1) was analyzed for tuo different sizes
of edge members. For all edge members of size 6x3 ems. (18
sq.cm. cross-section area), it appears (Fig. 6.4) that bucklinp
2
occurred between the loads 0.20 to 0.30 kgms/c~ , which is
nearly half that of the all-supported case discussed before.
~ne
According to Leet 28 , the buckling of the hypar shell prirnarilv

depends upon the axial stiffness of the edge me~bers and not
on the edge deflections. The boundary conditions used in
Reissner's solution correspond to the infinite axial stiffness
of the edge member. To verify Leet's conclusions, the same
hypar was analyzed with the edge bea~s having the sa~e proper-
ties as those of the 6x3 size beam except for the cross-sec-
-116-
tional. ·area, which was increased frofl' 18 sq.cm. to 108 sq.cm.

Though the deflection profiles did not show the buckling very
clearly up to a.so kgms/cm 2 , there was a small deviation in
the deflection profile. at about 0.40 kgms/cm 2 (Fig. 6.5). It
appears that the buckling depends upon the cross-sectional area
of the edge member and not so much on the edge deflection. It
is believed that for the stiff edge members used in Strucs. '11'-
'13' the deck buckling load can be predicted by calculating the
bucklincr of a single quadrant of the umbrella shell with all-
supported edres.
T1110 hypars ~ Struc. '13' ui th C.oub le 2 8-G corrugated
decks and Struc. '12' with a single 24-G corrugated deck were
analyzed to determine the buckling load. In the case of a
double deck structure, the deck buckling load is between the
range 200 to 243 psf. (Fig. 6.6). Experimentally the struc-
ture was tested up·to 145 psf and no deck buckling was obser-
ved, though deck tearing _along the lines of connection was
noticed.
Struc. '12' with a 24-G single deck was analyzed
usinr both curved and flat elements. The experimentally ob-
served deck buckling load for this structure was in the vicinity
of 75 psf, but from Figs. 6.7-6.9 hoth curved and flat ele-
ments analyses predict .a lower buckling load. One of the pos-
sible reasons for this is that the pressurized canvas bags
used as loading devices, tend to offer some restraint to the
buckling of the shell. It is known that even a small external
restraint to the shell surface is adequate enough to raise the
-117-
buckling load substantially. Comparine the deflection pro-

files in Figs. 6.7-6.9 it appears that the flat ele~ents pre-
dict the load in the vicinity of about 46 rsf whereas the curve<l
elements predict the load in the vicinity of 60 psf. fts
pointed out in Chapter IV, the curved and flat elements, for
hypars with all edges supported give very close results for
the linear elastic analysis. It appears that the flat ele-
ments give very conserv;itive results for the bucklinr of single
decks. As pointed out in connection with Eq. 6-12 for the flat
elements, the transformation of the incremental matrix from
local to the global axes results in the modification of the
flexural as well as Jllembrane stiffnesses. Since the in-plane
membrane stiffness for a corrugated deck is very small in the
weak direction, a premature buckling could possibly be trig-
gered by the reduction in the rnerbrane stiffness. As against
this, in the curved element formulation because of the
assumptions used in the analysis, the mer ·tbrane stiffness is not
1
modified.
For a double deck or an isotropic deck, because of
high in-plane rigidity in both directions, hoth ~ethods pre-
dict the buckling loads in the same range. As pointed out in
Chapter IV, the bending action is very dominant in t!'le flat
portion of the shell. But in the buckling analysis, the curva-
ture of the hypar in the center of the quadrant may be more
critical. This fact occurs even more in the case of a single
deck and therefore the curved elenient, which accounts better
for this curvature effect, probably predicts a hi~her load.
-118-
Reissner 5 in his analysis of buckling of isot~opic

hypar shell, has indicated that except for a difference in
numerical coefficients, the critical in-plane shearing force
for the shell differs from that of a flat plate by the fact
that a thickness . square factor is replaced by the product. of
shell thickness and shell rise. However, in the case of an
orthotropic shell, the increase in the critical shearing force
does not appear to be as high as in the case of an isotropic
shell. Because of equal bending rigidities inboth directions,
the buckling load for a double deck hypar shell appears to be
three or four times larger than that of a single deck. This
is also observed in the case of the critical shearing stress
for a flat plate.
Struc. '12' was also analyzed using the identical
boundary conditions as the structure analyzed in Fig. 6.7, but
allowing the edges to deflect. As pointed out previously,
thougt the exact point of deck buckling is not known, the deck
buckling occurs between the loads of 43.0-50.0 psf. The in-
teresting point to note here is that even though the deck
buckled (Pips. 6.10, 6.11), the deflection of the edge wernbers
was still quite linear. The same trend ''·'as also observed dur-
ing the experiwent. The bucl·ling also showed a similar trend
as ·Jbserved in the case of an all-supported hypar. The deflec-
tion profile along the tension diagonal did not.s~ow any
huckling (Fig. 6.10).
It may not be convenient to analyze every structure
in practice by this load incrementation wethod. According to
-119-
Reissner 5 , the buckling load for an isotropic hypar shell is

given as,
qcr = 4(C/AB) 2 ~D Et 6-14
where D is the bending rigidity and Et is the membrane stiff-
ness. In the case of a double deck, the bending rigidities of
the shell are equal in both the directions, however, the mem-
brane stiffness is affected by the shear rigidity factor a. If
the equivalent thickness in the formula 6-14 is replaced by
at, the estimated buckling load will be very conservative be-
cause of the fact that it will also underestimate the in-plane
axial stiffness. In order to eliminate this underestimation
of the axial stiffness the shear rigidity factor a is arbi-
trarily multiplied by 2(l+v) (the ratio of E/G for the parent
material).
qcr Double Deck = 4 (~-) 2 )2 Dy E(l+v)at 6-15
This formula is a conservative approximation of the buckling

load of hypars with double decks.
-120-
VI.7.· DECK BUCKLING BY THE ENERGY METHOD

The buckling of isotropic hyperbolic paraboloid shells Ii
" '<'.·
was investigated in a classical paper by Reissner 5 . He devel-

oped the.general shallow shell theory and reduced it to two dif-
ferentl~l equations in the displacement w and a stress function
F. From these equations he studied the buckling of simply:-
suppo,rte_d isotropic hypars under uniform loading. The assumed
displacement was a double sine series and the form for the
stress function F was assumed to b~ a double sine series with
an addi t_ional term containing xy. The substitution of these
series into the two equations result in homogeneous.algebraic
equations because the sine terms drop out. The condition of
non-trivial solution yields the buckling load:
2 2
p = 2E . h c
cr ·----··· :!::-2'
V3 (l-v 2) AB
.This solution was possible because each of the two dif-

ferential equations contained only one elastic constant: E
in one equation and D in the other. Thus the sine terns could
drop out.
In the case of orthotropic shells these equations contain
several elastic constants and therefore the sine terms would
not drop out. This means that one cannot use these shallow
shell equations to obtain reasonable buckling load. It may be
possible to find very simple assumed functions for w and F but
the accuracy of such a solution would probably be very poor.
Another alternative approach uses energy principles. The
potential energy of an orthotropic hypar shell is 19 :
-121-
b a
v = 21 ( ( [D Xw""X 2
·i •·
+ 2D
1
w 2
xx wyy + Dy wyy + 4Dxy wxy
2
0 0 "'
2
w2 +
+ 4 Geffh (C/AB) 2Nxy wx wy] dxdy 6-17
The selection of an assumed buckling shape requires very

careful attention. A double sine series is very difficult to
use because of the complexity of the resulting arithmetic.
Furthermore, the direction of the buckles is at an angle with
the coordinate axes in the case of orthotropic shells and the
deflection function must contain a factor which accounts for
this fact. Several types of deflected shapes were tried. By
far the best results were obtained using the following shape:
w = sin !f sin [n~ (x-sy)] 6-18
where the factor s represents the tangent of the angle of the

buckles measured from the y axis and n is the number of buckled
waves. This function was used by Timoshenko and Gere 36 and by
Easley and McFarland 48 for the buckling of shear diaphragms.
This deflection assumption corresponds to a simply sup-
ported shell (or diaphragm) where B (and the y axis) is measured
along the deformations (corrugations). Actually this expression
does not satisfy w =0
along they axis, but comparisons with
more exact analyses for the buckling of diaphragms showed 48
that this discrepancy is not serious.
Substitution of the assumed w into the potential energy
expression and integration yields:
-122-
where Q1 = Dxa 4 + 2D 1a 2b 2 + 2D 1 a 4 s 2 + Dyb 4 + Dya 4 s 4
2 2 2 4 2
+ 2Dya b s + 4Dxya s + C'
o.2 = 4D y a 2b 2s 2 + 4D xya 2b 2
a = nrr/A, b = rr/B,
Since V represents the change of potential energy under

a deflection w, the condition V = 0 corresponds to the buckling
load. Thus
From the membrane theory Nxy = pAB/2C, thus
Per=~ Ci) 2 [Dxu 2 + 2D 1 (l+u 2s 2) + Dy(l/u 2 + u 2s 4 + 6s 2)
+ 4D
xy
(l+s
2
u1 + c/u 2 ] 6-19
where
u = a/b = nB/A and c = C1 b 4 = 4Geffh(BC/A) 2/rr 4
This expression has to be minimized with respect to the

direction of the buckles (s) and the· number of buckles (n).
This can easily be done by trial and error, using a computer.
The analysis of the single-layer 24-Gage structure (No.
12) gave 50 psf for the buckling load as compared with the
experimental value of about 75 psf and the finite element analy-
sis of about 60 psf using curved elements. Minimum occurs for
n = 8 and the buckles form at an angle of 8.0 degrees with
-123-
they axis (Fig. 6.12). The buckling load was not very sensi-
tive to n and therefore the prediction of the number of buckles
may not be accurate.
For structure No. 13 (double-layers, 28 Gage) the above
analysis resulted in a buckling load of 192 psf at n =2 and
at an angle of 35.S degrees. The finite element analysis gave
about 200 psf and t~e test structure did not buckle up to a
load of 145 psf when loading was discontinued.
The main advantage of this energy analysis of the buckling
load is that it is very fast as compared with the finite element
analysis. Simple supports are assumed and therefore the actual
buckling load may be somewhat higher if some bending or in-plane
fixity is present. The deflection of the edge members is not
considered, but the finite element analysis showed (Section VI.6)
that edge member deflections tlo not affect the deck buckline
load appreciably. The deck buckling load is influenced by the
area of the edge members. If the edge members are very slender
then deflections will control; on the other hand, the above
analysis should give conservative results for structures with
heavy edge members.
VI.7 STABILITY OF ISOLATED EDGE MEMBERS
The edge members of a hypar structure receive uniform axial
loads along their lengths from the deck. Some or all the edge
members are in compression. The loading remains axial during
the deformation of the edge menbers and therefore it constitutes
a non-conservative force field. To obtain an approximate value
of the buckling of the compression erl.ge membe-rs, the membrane
-124-
shear force was applied to isolated edge members with various

idealized boundary conditions. The buckling loads determined
in this manner are conservative since the stiffening effect of
the deck is neglected.
The marginal member was analyzed both as a fixed free and
a fixed pinned member. The equation used to solve the non-
conservative force field problem was:
EI IV + S(L-x)y'' +
y
My = 0 6-20
where S = the shear force, L = the column length, M = mass per

unit length, y = deflection, and (') means differentiation
with respect to time.
The solution of the differential equation was obtained by
using Galerkin's Method for solving differential equations with
non-constant coefficients. The numerical solution for the fixed
free case was found in an article by V. H. Leipholz; "Die. Knick-
last des Einseitig Eingespannten Stabes mit Gleichmassig Ver-
tiel ter, Tangentialer Llingsbelastung", published in ZAMP, 13, 6,
1962.
The solution for the fixed pinned case was determined using
the same method as described above, but a computer program was
written to facilitate the trigonometric integrations and the
mathematical solution.
The numerical solutions are:
2
(ql)cr = 40.7 EI/1 for the fixed free case
2 for the fixed pinned case
(ql)cr = 122.6 EI/1
where q = shear per unit length transferred from shell to
marginal member. These values and some comparison values are
given in Fig. 6.13.
-12S-
The results show tlaat the buckling load of members loaded

by tangential shear forces is very much greater than that of
members under gravity load of constant direction parallel to
the original, undeflected axis of the member.
Thus, if case IV is compared with V, with loading and end
conditions the same except for the direction of the load, it
is seen that in the elastic range the edge member of a hypar
will buckle at a load (1.12/0.49) 2 = 5.2 times that which the
same member would carry under uniform gravity load. Similarly,
co~paring cases VI and VII, it is seen that the corresponding
ratio is (0.436/0.284) 2 = 2.4.
If the total load which will make the edge member of a
hypar buckle, is compared with that which the same member,
loaded as a hinged-hinged Euler column (basic case) would carry,
it is seen that the edge member in the fixed-free condition
(case V) will carry (l/0.49) 2 = 4.1 times the buckling load of
the basic Euler columnj or, for the fixed-hinged condition
(case VII) (1/0.284) 2 = 12.4 times the basic Euler load.
The above results do not represent a complete analysis of
the problem, which is really one of buckling interaction between
the edge member and the shell. However, if the shell action is
close to that of a membrane, as in the case of light-gage steel
hypars, the approximation should be reasonably good. This ap-
proximation shows that, while a buckling possibility exists,
these edge members are very much more stable than they would
be if they were part of ordinary fr~ming subject to gravity
loads.
-126-
CHAPTER VII
EXPERIMENTAL INVESTIGATION
VII.l INTRODUCTION
An extensive experimental investigation was carried out.
The purpose of the testing program was twofold: (1) to study
the behavior of light gage steel hypar roofs subjected to verti-
cal loads and (2) to provide a comparison with analytical ap-
proaches.
The following tests were carried out:
a) Sixteen flat shear tests - Properties and dimensions of
all the specimens are presented in Table VII-1 and a picture of
one is shown in Fig. 7.1. It was necessary to carry out these
tests to determine the shear rigidity G' of the decking used
for the hypar models. Twelve specimens were 6' x 6' in plan,
three were 5' x 5' in plan (all. referred to as "medium scale
tests:;), and one was l' x l' in plan ("small scale test").
b) Twelve saddle shaped hrpar specimens - All of them were
5 1 x 5' in plan with various rises (Table VII-2-VII-3). A
picture of one is shown in Fig. 7.2. The specimens are desig-
nated by three numbers: the first one indicating the rise/span
ratio, the second the number of layers of decking, and the
third whether it is an original or duplicate. For example,
for test no. 512' 1151!
indicates a 1/5 rise/span ratio, 11111
indicates one layer of decking, and "Z' indicates that it is
I
a duplicate test (each specimen was duplicated).

c) Three inverted umbrellas with verr stiff ed8e beams -
Numbers 11, 12, 13 in Table IV-1. All were 12' x 12' in.plan
-127-
with a 14.4 11 rise. A picture of one is shown in Figure 7.3.

They are referred to as "medium scale inverted umbrellas with
very stiff edge beams".
d) One small-scale inverted umbrella with very flexible
edge beams. The model was 2' by 2' in plan with a 3" rise.
The edge members were made of 3/16" O.D. and 0.014 11 thick brass
tubes, two were used for the interior compression beams. Two
layers of corrugated decking of 2 mil thickness formed the shell.
The deck was soldered at every valley to the edge members. A
picture of the model is shown in Figure 7.4.
e) One medium scale inverted umbrella with very flexible
edge beams - The model was 12' x 12' in plan with a 14. 4'· rise,
structure number 9, Table IV-1. A picture of it is shown in
Figure 7.5.
The tests are described in the following sections.
VII.2 FLAT SHEAR TESTS

1. Introduction
A series of flat shear tests were conducted in order to
determine the shear rigidity G' of corrugated steel decking.
Luttre11 45 investigated the shear behavior of light gage steel
diaphragms. Based on the results of numerous tests he concluded
that the primary variables influencing the shear rigidity are
length of diaphragm parallel to the corrugations and spacing of
connectors to the edge members in the direction perpendicular
to the corrugations. It was also determined by Luttrell that
the only reliable means of determining the shear rigidity of a
given diaphragm with framw was by experiment.
-128-
The formula for G' developed by Luttrell acc.ounts for the

deflection due to shear alone by subtracting from the measured
deflection the bending deflection of the cantilevered structure
due to axial deformation of the edge members. The value of G'
obtained from a given test can be applied to the following
case: Any set-up with exactly the same spacing of diaphragm to
edge member connections perpendicular to the corrugations, and
the same diaphragm. The size of the edge members should have
little or no effect.
It may be applied with simple modifications to the follow-
ing cases: (1) everything the same except diaphragm has differ-
ent thickness - G' is (approximately) directly proportional to
the thickness; (2) everything the same except length of frame
parallel to the corrugations is different - new G' may be ob-
tained from Figures 4-23 and 4-24 in Reference 45 for box-rib
and standard corrugated diaphragms, respectively.
2. Edge Member Frame
For all tests except two the edge members were light gage
channels. Two different size channels were used. One, desig-
nated as "heavy frame", consisted of 6 11 x 1 1/2" x .1046 11 chan-
nels and the other, designated ''light frame", consisted of 6" x
3/4" x .1046" channels. For the other two tests, the frame was
made up of tubing.
3. iJecking
All the medium scale tests employed standard corrugated
decking; including 24, 26, and 28 gage. Either one layer or
two layers, with the second layer placed with the corrugations
-129-
perpendicular to those of the first layer, were used.

The small scale test had 2 mil corrugated steel foil.
This material was produced by United States Steel Corp. as a
reinforcing center for cardboard cartons. Two layers, running
transverse to each other, were used.
4. Connections
For the nedium scale tests, #8 x 1/2 11
self-tapping screws
were used to connect adjacent sheets to each other along the
seams. #14 self-tapping screws with aluminum-backed neoprene
washers were used to connect the decking to the edge members.
For double layered decking, the second layer was attached to
the first layer around the perimeter only with #8 x 1/2 11
self-
tapping screws. The spacing of connections for each test is
given in Table VII.l.
In test no. 14, the effect of the relative positions of
the screws connecting the top layer to the bottom layer and
the screws connecting the bottom layer to the edge members was
noted. Referring to Figure 7.6, one possibility is sho~m in
(a) where the shear flow causes a region of compression between
screws 1 and 2 and the bottom layer distorts considerably be-
tween screws 2 and 3. The other possibility is shown in (b)
where the shear flow now causes a region of tension between
screws 1 and 2 and very little distortion of the bottom layer
occurs. However, in this case, because of the tendency of the
tensile region between screws 1 and 2 to flatten out, eventually
screw #1 tears out of the bottom layer.
For the small scale test, the top layer was soldered to
the bottom layer at every point of contact around the perimeter
-130-
and the bottom layer was soldered to the edge members at every
point of contact. One full panel was used for each layer.
5. Loading Apparatus
The mediuru scale frames were placed in a horizontal plane
and load was applied by means of a 50 ton hydraulic jack.
Vertical support along the loaded edge was provided by rollers
on beams. Reactions were provided by a fixed wall beam to
which the frame was pinned. Steel bricks were placed on the
loaded edge to prevent out-of-plane warping.
The small scale frame was attached to a wooden frame with
metal guides to prevent the specimen from warping out-of-plane.
The entire set-up was placed in the vertical position in a
Tinius-Olsen hydraulic testing machine and load was applied by
the machine.
6. Deflection Measurement
For each test, deflections were measured by .001" dial
gages .. At first, .0001" gages were tried but it was discovered
that they are too sensitive because the readings obtained from
then were erratic. Sufficient accuracy was obtained with the
. 001 · gages. If possible the stem of the gage was placed in
contact with the specimen, otherwise a thread was used to at-
tach the stem of the gage to the specimen.
7. Determination of Shear Rigidity G'
G' was determined by the use of the expression given by
Luttre11 45 . The initial linear portion of the load-deflection
curve was used to obtain the slope. The customary testing pro-
. cedure was to initially load the specimen to a predetermined
value so as to seat all connections and then ~o conduct as many
-131-
loading cycles as necessary to get good agreement between the

load-deflection curves of successive cycles. Usually, only
two or three cycles were required. The dimensions, properties,
and the values of G' and of a= G'/Gnt for each test are pre-
sented in Table VII.l. The load-deflection curve for test #14
is shown in Fig. 7.7.
VII.3 SADDLE SHAPED HYPAR TESTS
1. Introduction
A series of light gage steel hypar models, 5' x 5' in plan,
with various curvatures were tested in an effort to determine
if the shear rigidity as determined by a flat shear test is
valid for the plate on elastic foundation approach. If an
exact analysis of light gage steel hypars were possible, then
it would appear that the experimental shear rigidity could be
used with validity in the analy~is. However, the assumptions
which were introduced into the approximate plate on elastic
foundation approach may lead to errors in the results if the
shear rigidity as determined by a flat shear test is used in
the analysis. Better results may be obtained if a "fictitious"
shear rigidity were calculated backwards from experimental de-
flections. Thus, it might be possible to determine the ficti-
tious shear rigidity as a function of the curvature.
For the above reasons, the only variable in this series
of tests was the rise, i.e. curvature. Three different rise/
spa~ ratios were employed, 1/8, 1/5, and 1/3; one set for
single layer decking and the other for double layer. Two tests,
one a duplicate of the other, were carried out for each rise/
span ratio and number of layers of decking. The tests were
-132-
designated according .. to th~ rise/span ratio, number of layers,

and first t:est or duplicate as exvlained in Section VII-1
·2. Details of Test SEecimens. {
The edge beam frames .with supports were the same for all
the tests. They consi$ted pf 3 11 O.D. x 1/4" wall thickness
cold rolled steel tubular members welded together to give the
correct rise/span ratio for each specimen. The diagonally op-
posite corners, two low and two high, along with the midpoints
of each side were rigidly supported in the vertical direction
by steel bricks (Fig. 7.2). rhis support system together with
the fact thgt the tuLes have a large bending rigidity was con-
sidered to offer continuous rigid support in the vertical
direction.
A tie bar, 2 1/2 11 x 1/ 4" in cross-section, was used to
limit the spreading of the tw~ low corners due to the vertical
load.
The decking was made up of 28_gage standard corrugated
steel sheets with 2' cover. Three sheets were used for each
layer, with the middle sheet cut to fit the 5' width. Single
layered decking was connected to the edge beams by #14 self-
tapping screws with aluminum-backed neoprene washers at 8"
spacing while the two seams were fastened together with #8 x
1/2" self-tapping screws at 2 2(3" spacing. For the double
layered decks, the top layer was fastened to the bottom layer,
around the perimeter only, with #8 x 1/2" self-tapping screws
at 8 11 spacing. The two seams of the top layer were fastened
together exactly the same as those of the bottom layer.
-133-
Uniform vertf~al loading was achieved with sand held in

place by wooden sides and screeded to a uniform depth for each
400 lb. (about 15 psf) load increment.
For each one of the models, subsequent to the uniform load
test, a partial load covering an 8" x 12" area was applied at
the center of the same decking. These tests were designated
with a "C". In addition, for each one of the two duplicate
models with double layered decking and subsequent to the tests
with unconnected decks, the two layers were fastened together
with #8 x 1/2" self-tapping screws on an 8" square grid and
subjected to uniform loading first and then the partial load-
ing on an 8 11 x 12" area. These tests were designated with an
"I".
Deflections were measured with .001" dial gages. Strains
were measured at the center of the decking with SR-4 strain
gages with l" gage length. The strain gages were placed in a
valley on the top side and on the adjacent crest on the bottom
side, and in the direction parallel to the corrugations • They
were also located at mid-length of the tie bar, top and bottom,
on some of the models. The purpose of the gages on the deck-
ing was to determine the axial and bending stresses at the
center of the deck and on the tie bar was to determine the
axial force in it.
The experimental deflections and stresses at 40 psf for
the saddle shaped hypar tests are presented in Tables VII-2
and VII-3. The method of determin"ing the shear rigidity versus
cti~vature (6r rise/spari ratio) by usirig the experimental
-134-
deflections in the plate on elastic foundation approximation

I j" _:. : ) :1 ~. ~ . ,
is presented in Section VII.6. :

! :: ··~. i, " • "(' ~: f;
VII .4 MEDIUM"SCALE INVERTED UMBRELLAS l';lTH.--. VERY STIFF EDGE BEAMS

r .~ ' '
1. Introduction
'l "'~
Three ~eJi~m scaie models (Struct;ires 11, 12, 13, Table

·, .
IV-1) were designed and'~tested with;.1vhe main purpose being to
check the theory for the buckling of;(~rthotropic hypar shells.
J
•·. . .
Therefore, edge beams with' a large b¢nding . f·igidi ty were chosen

so as to remain stable· when the d.ec•}d;Jlg buckled. Vertical de-
flections as well as strains were measured.for the decking and
the edge beams.
" .',, "'1 J.·
The same edge beam frame was µ:Sed.,for all three tests. It
. ~ :· J..
consisted of 311 0. D. x 1/4 ., wall: thickness cold rolled steel
tubular members welded together. The dimensions of the full
inverted umbrella were 12' x ·1'2lvin plan, centerline to center-
line of the edge beams, ·with a 14.4" .rise (1/5 rise/span ratio).
Each interior edge beam·Xeonsistee! of. t:iro tuh,ular members side
by side (spot· welded toi~~t-he-r· 1 :trt~:l i)...-//l.' dnJel"vals) so that deck-
ing could b~''i~stened to: ciri.tl -Of· 1 :th~:·(t\e;mbers in a given quadrant
and decking \ri. :;fr{ adj acernt '.:quaidr.a:n1r-~.ould be fastened to the
adjacent member.
-' Therefore;"the: plan dimensions of each quad-
rant were 70. 5" x 70. S'' ~ d.mterlin~ t.o. centerline of the edge
'..!
beams.
2. Details of Test SEecimens and Procedure
Two of the modJis had one layer of standard corrugated
steel decking, 28 g•~e for·,one model and 24 gag~ for the other.
The third model had two layers of 28 gage st.andard corrugated
steel decking.
-135-
For both of the models with one layer of decking., #14 self-
tapping screws at 8" spacing fastened the decking to the edge
beams and #8 x S/8 11 self-tapping screws at 2 2/3.i spacing fas-
tened the seams together. Three panels, each with 2' cover,
were used per quadrant. For the model with two layers of deck-
ing, the bottom layer was attached to the edge beams exactly as
described above. The top layer was fastened to the bottom
layer, around the perimeter only, with #8 x 1/2 11 self-tapping
screws at 8" spacing. The seams for the top layer were fas-
tened together exactly the same as those in the bottom layer
as d9scribed above.
For all three models, uniform normal air pressure was ap-
plied to the inverted umbrella shape by four rubberized canvas
bags placed between the floor and the hypar in the inverted
position. A water mano~eter was used to measure the pressure.
Vertical deflections were measured by level sightings on
meter sticks held at each location. Strains in the deckinp were
measured with SR-4 gages with ln gage length and those on the
edge beams with SR-4 gages with 1/2" gage length. The vertical
deflections, experinental axial stresses, and experimental
bending stresses at 40 psf normal pressure are presented in
Figs. 7.8-7.10.
VII.S INVERTED UMBRELLAS WITH VERY FLEXIBLE EDGE BEAMS
1. Introduction
Two models were tested with extremely flexible edge beams.
The purpose of these tests was to determine the mode of failure
as well as to check how closely the failure could be predicted
-136-
by theory.
2. Small Scale Model
One of the inverted umbrellas with very flexible edge
beams (Fig. 7.4) had overall plan dimensions of 2' x 2', center-
line to centerline of the perimeter edge beams, with a 3" rise
(1/4 rise/span ratio).
The edge beam frame was made up of 3/16'" O.D. x . 014 11
wall
thickness brass tubes brazed together. The decking consisted
of two layers of 2 mil corrugated steel sheets soldered together
at each point of contact around the perimeter. The decking was
soldered to the brass tubes at each point of contact.
Uniform vertical loading was simulated by 64 discrete
weights (16 per quadrant) hung from strings passing through very
small holes in the decking and attached to 1 11
square pads.
Vertical deflections were measured with .001" dial gages.
The dial gages were positioned such that the stems did not
touch the shell. To take a deflection reading, the stem of
the gage was pressed until contact with the deck was made;
avoiding the application of any force by the stem as much as
possible.
Experimental deflections at 40 psf load are presented in
Fig. 7.11.
Collapse of this model occurred at 73 psf. It was ini-
tiated by splitting of the decking along a row of holes through
which the weights were suspended. This splitting caused the
brass tubes to fail in bending very close to the central sup-
porting column.
-137-
3. Medium Scale Model

The test set-up and procedure for this model were almost
the same as for the 12' x 12' inverted umbrella with two layers
of 28 gage decking described in Section VII.4 except for the
size of the edge beams.
The edge beams were 1 11 nominal diameter standard weight
black steel pipe with a 50 ksi yield point.
The vertical deflections were measured with .001 11
dial
gages. The stems of the gages were attached to the shell by
means of thread which was glued to the structure.
Experimental deflections and stresses at 40 psf load are
presented in Fig. 7.12.
VII.6 DISCUSSION OF EXPERIMENTAL RESULTS
1. Shear Rigidity G' of Standard Corrugated Decks
Comparison of the results for the flat shear tests, which
are presented in Table VII-1, reveals that the variable having
the largest effect on G' is the spacing of the connections to
the edge members. In particular, the spacing transverse to the
corrugations has the nredominant effect whereas the spacing
along the corrugations has little or no effect on G'. For
example, screws at every third valley as in test no. 3 produce
a G' approximately one-third that for screws at every valley
(one-third the spacing) as in tests no. 1 and 2.
According ~o Luttre11 45 , another variable which affects
the shear rigidity is the dimension along the corrugations; the
larger this dimension the larger the shear rigidity. From the
results presented in Table 3.1, it is not possible to verify
Luttrell's finding conclusively. For test no. 8, with plan
-138-
dimensions of 5' x 5', G' is less than that for 6' x 6' test
no. 5 but slightly more than that for 6' x 6 1 test no. 6. Thus,
it appears that scatter in the test results masks any effect of
the small variation in size.
The spacing of seam connections and number of seams have no
discernable effect on G' based on the results presented in Table
VII-1. Also, it appears, at least for a limited range of thick-
nesses of panels, that the shear rigidity increases linearly with
the thickness. For example, compare the results for tests no.
5 and 7.
Tests numbered 10 - 16 had two layers of decking running
transverse to each other. In each case the top layer was con-
nected to the bottom layer along the edges, rather than being
.· .
connected directly to-the edge members. The effe~t of this ar-
rangement of connections is to make the top layer less effec-
tive in shear than the bottom layer, which is attached directly
to the edge members. The reason for this is that any shear
deformation which occurs in the bottom layer along the edge
members occurs also in connections for the top layer which in
turn reduces the effective shear stiffness of the top layer
below that obtained by connecting the top layer directly to the
edge beams. From Table VII-1, tests no. 10 and 11 versus 1 and
2; 12 and 13 versus 4, 5, and 6; and 15 versus 8 and 9 show that
the shear rigidity for the double layer tests is less than twice
that for the corresponding single layer tests. It appears that
connecting the second layer in this manner results in only a 33%
increase in G' over that of the singie layer, although as men-
tioned previously the relative positions of the screws appear to
-139-
affect the shear rigidity. If all the connections between the

two layers were such as to create the condition shown in Fig.
7.6(b) then the shear rigidity should be larger than for the
case shown in Fig. 7.6(a).
Fig. 7.7 shows the load-deflection curve for test no. 14.
The linear portion extends to about 1800 lbs. load which corre-
sponds to a shear Nxy = 25 lbs. per inch. Beyond this point,
the curve indicates that the shear rigidity decreases with in-
creasing load.
Figs. 7.8 and 7.9 indicate that the shear rigidity of
standard corrugated steel decked ~ypars with a rise/span ratio
~ 1/5 is somewhat (20 - 30%) less than that of the same flat
deck. However, as mentioned briefly in Section VII.2, this in-
formation may be misleading because of the approximate plate on
elastic foundation approach used in conjunction with the experi-
mental deflections to obtain these curves. Two factors were
omitted in the theory used to obtain Figs 7.8 and 7.9, the in-
clusion of which indicate that the shear rigidity obtained from
a flat shear test may be valid for the hypar shell. One of the
factors omitted was the middle-surface deformations u and v.
The other was the spreading of the low corners of the saddle
shaped models. Even though a tie bar connected the low corners,
tension in the tie bar caused by the shear forces resulted in
its elongation. Duplicate tests no. 521 and 522 were chosen to
illustrate the effect of these two factors. If the u, v dis-
placements are included then the resulting a is .046 instead of
.042 as shown in Fig. 7.9 for the 12'' rise. If, in addition,
-140-
the spreading of the low corners is included by introducing

the stiffness of the tie bar into the analysis then the result-
ing a. is approximately .OS which is the same as that obtained
from the flat shear test. These results for only one example
are not meant to be conclusive but the indication is that the
shear rigidity in a warped element is not much different from
that determined from a flat shear test.
2. Hypar Deflections
a) Saddle Shape SupEorted All Around
Table VII-2 presents the maximum deflections at 40 psf for
the models tested in this investigation. The results reveal
the decrease in maximum deflection with increasing rise as well
as the fact that the maximum deflection of a given model with
two layers of decking is more than half that of the correspond-
ing model with one layer of decking; the reason being, as men-
tioried previously, that since the top layer is connected only
to the bottom layer it is not as effective in shear as the bot-
tom layer and thus the rigidity of the single layered deck is
more than half that of the double layered deck.
Table Vll-2(b) shows the effect of interconnecting two
layers all over and not just around the perimeter. Test numbers
with an "In indicate interconnected layers. The reduction in
maximum deflection due to interconnecting is seen to be only
10 - 20%. However, it appears that interconnecting would be
necessary in practice to prevent chatter.
Table VII-3 presents the maximum deflection due to a partial
load covering an 8 11
x 12" area in the center. A comparison of
the maximum deflection due to a 100 lb. load on a single layered
-141-
deck with that due to a 200 lb. load on a double layered deck
reveals that a double layered deck is more than twice as rigid
as a single layered deck for carrying a concentrated load; the
reason being that a concentrated load on a single layer is car-
ried mainly by a few beam strips whereas on the double layer it
is spread out and thus carried by more beam strips in each layer.
For all the models tested, the maximum deflection due to a 100
lb. load on a double layered deck is approximately one-third
that for a single layered deck.
b) Inverted Umbrellas With Very Stiff Edge Beams
Figs. 7.lO(a)~ 7.ll(a), and 7.12(a) show the measured verti-
cal deflection at 40 psf normal pressure for three inverted um-
brellas tested at Cornell. It is seen that difficulty in obtain-
ing ~ymmetry was experienced. The lack of symmetry was probably
due to unequal air pressures in the rubberized canvas bags. The
bags were not entirely air tight so that unequal rates of leak-
age from the bags could have caused relatively large percentage
variations in the pressures.
c) Inverted Umbrellas with Very Flexible Edge Beams
Figs. 7.13 and 7.14(a) show the measured deflections at
40 psf for the 2 ft by 2 ft small-scale test and test No. 9,
respectively. Fig. 7.13 reveals that rotation about one diagonal
apparently occurred during testing of the small scale model. Fig.
7.14(a) shows that a reasonably good condition of symmetry was
obtained for the medium scale model. For both models, it is
apparent that the deflection at the free corners is much larger
than that at the center of each quadrant.
-142-
3. Hypar Stresses
a) Saddle Shape Supported all Around
The experimental stresses presented in Table VII-2 reveal
the difference in structural action between a very flat hypar·
(rise/ span = 1. 8) and one with a much larger curvature. (rise/
span= 1/3). ·The bending stresses in the models with a 1/8 rise/
span ratio are much greater than those in the models with a 1/3
rise/span ratio. The decrease in bending with incr~asing rise/
span rati.:o .is accompanied by an increase in membrane action as
seen in·the last column of Table VII-2. The experimental force
in the tie bar for the models with a 1/8 rise/span ratio is much
less than ·that given by membrane theory whereas the force in the
tie bar for the models with a 1/3 rise/span ratio is almost as
large as that given by the raembrane theory. Thus, it. is demon-
strated experimentally that for very flat hypars, shear stresses
are much smaller than predicted by the membrane theory and that
bending constitutes the major part of the structural action
whereas for deep hypars (rise/span ~ 1/3) membrane stresses pre-
dominate with values close to those given by membrane theory and
bending stresses are insignificant. The axial stresses in the
deck, for all models, are seen to(be insignificant.
Table VII-3 shows that a concentrated load produces signif-
ic·ant bending stresses, especially in the flat hypars, but very
little membrane action.
b) Inverted Umbrellas '1ith Very Stiff Edge Beams
Figs. 7.lO(b), 7.ll(b), and 7.12(b) show the axial and bend-
ing stresses obtained from strain measurements at 40 psf load
-143-
for the three models tested at Cornell. The axial stresses in

the strong direction of the corrugated decking are seen to be
quite erratic and do not indicate any consistent behavior. It
appears that their magnitudes are too small to give reliable
results.
The bending stresses in the decking at the quadrant centers
vary from 6200 to 10200 psi for the 28 gage single layer, from
4800 to 5700 psi for the 24 gage single layer, and from 5100 to
10200 psi in the bottom layer of the 28 gage double layer decking.
The wide range of these values for each model indicates a con-
siderable departure from symmetry which was probably caused by
unequal pressures applied to each quadrant.
Experimental edge beam axial and bending stresses as well
as edge beam axial stresses from the membrane theory are also
shown in Figs. 7.lO(b), 7.ll(b), and 7.12(b). It is difficult
to observe consistent trends in the experimental axial stresses
in the edge beams. In some locations they are very close to
those values given by the membrane theory while at other loca-
tions the difference is quite large. It is to be expected that
the experimental axial stresses would be less than those given
by the membrane theory because of the fact that part of the load
is carried by bending. However, the results for the three models
do not give any indication as to whether or not this is the
case.
Again, referring to Figs. 7.lO(b), 7.ll(b), and 7.12(b),
the bending stresses at the mid-lengths of the perimeter beams
for the two models with one layer of decking are larger in the
-144-
beams parallel to' the weak direction of the decking.

' .
This is
to be expected because the bending which occurs in the decking
transmits a vertical shear loading to these edge beams. The
bending stresses in the interior edge beans for all three models
are small although the strains in these beams at points of maxi-
mum stress near the column were not measured.
-145-
CHAPTER VI I I
SUMMARY, CONCLUSIONS AND RECOMMENDATIONS
The behavior of thin steel hypars was studied experimental-

ly and analytically to determine the stresses, deflections, and
buckling loads of such structures under various conditions.
The experimental program consisted of tests on: a) Five
inverted umbrella type hypar structures, b) Sixteen flat shear
tests which were made to determine the shear rigidity of cor-
rugated steel decks used on the hypar models, c) Twelve saddle-
shaped hypars with different rises and with rigid edge supports
.
which were tested in order to evaluate the effect of rise or
warping on the shear rigidity of decks.
Several small-scale and single-quadrant exploratory tests
preceded the above tests. The results are not included in this
report because of experimental difficulties or uncertainties
experienced with those models.
Two different approaches based on the finite element method
were used in the analysis of hypars. These approaches basically
differ in the stiffness formulation for individual elements. In
the first, a curved element rectangular in plan was developed
on the basis of shallow shell theory. The displacements u, v,
and w used in the formulation of this element are measured along
the tangents and normal to the surface. In the second approach,
the actual shell surface was approximated by a series of flat
plates assembled in the global coordinate system. The stiff-
ness matrices for the eccentrically connected edge members were
developed. The effect of a tension tie rod and a column support
-146-
was incorporated in the analysis by replacing the physical struc-

tural members by idealized equivalent springs (see Section III-2F).
Computer programs were developed for the analysis of thin-steel
hypar structures.
The linear elastic analysis is applicable for various types
of hypar structures (e.g. umbrella shell, saddle-shape, etc.)
and also for different loading conditions such as uniformly dis-
tributed load, unsymmetrical load and the load due to edge member
weight. Deflections and stresses for both deck and edge members
were calculated.
T~e linear elastic analysis was further extended to include
the instability effect introduced due to the in-plane forces Nx,
Ny and Nxy· The load incrementation method was used to predict
the deck buckling and the overall (edge member) buckling of the
structure.
The solutions available in the literature for both the linear
elastic analysis (Refs. 17, 19, 20, 21) and for the instability
analysis 5 were co~pared with the solutions obtained in this study
(Figs. 4.6, 4.8 and 6.2). The analytical results were further
compared with the experimental work conducted at Cornell and
also with test results available in the literature elsewhere 33 , 65 .
Based on the finite element analysis of several structures
and the comparison of analyses with available experimental and
analytical results, the following observations evolved:
In the case ,of .hypars with all edges fully supported, and
for hypars with very stiff edge members (Figs. 4.6, 4.8 and 4.33)
the results of curved and flat elements show excellent correla-
tion. However, in the case of hypars with flexible edges (Fig.
-147-
4.23), the curved element formulation tends to underestimate

the bending in the flat portion of the shell (e.g. free corner
of an umbrella shell). In the case of a uniformly loaded hypar,
the fully compatible 16-term Hermitian polynomial used for the
normal displacement w, does not show any improvement· over a 12-
term non-compatible polynomial displacement field.
The relative stiffness of the edge members to the decks is
quite important from the point of view of behavior of shells.
In the case of moment-free connections between the deck and the
edge members, the type primarily investigated in this work, for
very stiff edge members the deck tends to bend freely between
the opposite sides. In the case of umbrella shells with flexible
edge members, the deck supports the edge member at the free cor-
ner. The change in behavior of shells with stiff and flexible
edge members is quite noticeable from the deflection profiles
for Strucs. '9' and '13' (Table IV-1).
The in-plane shearing rigidity of a hypar shell is of ut-
most importance in its behavior. The effective value of the
shear modulus of the corrugated decks is given by Geff = aG.
Depending upon the rise to span ratio of the hypar surface, the
value of a~ determined by the flat shear test needs modification.
It is recommended that a as obtained fron a flat test be reduced
by 25% for single deck structures whereas it should be reduced
by 20% for a double deck (Figs. 2.7-2.8). Since the deflections
and the stresses of the hypar shell primarily depend upon the
value of a, the conservative estimation of its value will provide
a significant factor of safety for the structure.
-148-
Since a part of the vertical load is carried by bending

action, the value of the membrane shearing force given by the
membrane theory is on the conservative side. However, the
analysis of several structures (Chapter IV) tends to indicate
that over a major portion of the shell the value of the in-plane
shear is very close to and at places somewhat higher than the
membrane shear (Figs. 4.19, 4.28). Connections between the ad-
joining panels designed on the basis of the membrane shear, are
considered adequate.
The axial forces calculated in the edge member on the basis
of the membrane theory are on the conservative side (see Table
V-2). The axial stresses form a relatively small portion of the
total stresses (axial + bending) in the edge members. Therefore
the computation of the axial stresses in the edge member on the
basis of the membrane theory neglecting the bending stresses,
cannot be considered as conservative.
The eccentric connections of the edge members to the deck
have a pronounced effect on the deflections of the shell (Figs.
4.13, 4.17). With a proper choice of the eccentricity, the in-
fluence of the eccentric connection can be used to advantage
(see Chapter V) to reduce deflections.
As indicated in the analysis of Struc. '8' (Table IV-1)
unsymmetrical loading in an umbrella roof produces considerably
larger deflections and stresses (Fig. 4.20) than uniform loading.
The unsymmetrical load due to wind or drifting of snow should be
given due consideration in selection of decks and edge member
sizes.
-149-
T!1e weight of the edge members is partly carried by the

shell action. To average it over the whole surface of the shell
will underestimate the bending stresses in the edge members (Figs.
5.3, 5.4). The effect of the edge member weight on the behavior
of the shell will depend upon the construction procedure.
The effect of the amount of restraint against spreading of
the low corners of saddle-shaped hypars on the deflection of the
center of the shell was studied by means of numerical examples.
It was found that a tie bar connecting the points of support is
very effective in restraining the outward movement of the supports
and thus in reducing the center deflections. Edge members with
large bending rigidity in a horizontal plane also restrain the
spreading of the supports effectively.
Partial loading on hypars was studied to a very limited
extent experimentally. Loads were applied on an 8 in. by 12 in.
area in the center of the saddle-shaped models with rigid supports
around the perimeter. It was found that such a loading on a
single layer of corrugated decking produces a maximum deflection
three times that for two layers of corrugated decking. There-
fore, two layers of deckinp are recommended for hypars which will
be subjected to significant loads on small areas.
The linear elastic analysis adequately represents the be-
havior of the shell for low levels of loads. However, as the
load level increases it is necessary to incorporate the effects
of instability in the analysis.
The_ buckling in hypar shells using a light gage corrugated
open deck is manifested in three different forms. The individual
plate elements composing the deck may show local buckling when
-150-
subjected to in-plane shearing or axial forces. This however,

can be prevented by a proper choice of thickness to width ratio
for each plate element. Deck buckling takes place when the deck
acting as a unit buckles along the compression arch (Chapter VI)
while the edge members are still stable. Overall buckling of
.the shell is defined as the one when the edge members along with
the deck buckle simultaneously. For practical sizes of edge
members there is only a very remote possibility of overall
·1·Ing 19 •
b UC.iC
Though the magnitude of the deck stresses is quite low,
the possibility of the deck buckling must not be overlooked.
Two decks placed perpendicular to each other for the same geomet-
ric configuration of the shell will increase the deck buckling
load roughly three to four times compared with that for a single
deck. From the analysis it appears that the axial stiffness of
the edge members is more important than the deflections of the
edges as far as deck buckling is concerned. It also appears
that the deck buckling load is roughly proportional to the area
of the edge member~.
The menbrane action in the shell increases and the bending

action decreases with increase in the rise to span ratio, thick-
ness t, and shear rigidity factor a. A non-dimensionalized param-
eter ~ can be used as a good index for the behavior of the
shell. A higher value of this non-dimensionalized parameter
indicates reduction in the vertical deflection as well as in the
bending stresses for both the deck and the edge members. Further-
more, it will increase the buckling load.
-151-
Though this investigation resulted in acceptable and re-

liable methods of analysis, the following recommendations can
be made for future analytical work:
The computer program using flat elements for the stiffness
analysis gave consistent results. It is not clearly established
as to whether the deficiency of the curved elements in predict-
ing deflections and bending stresses near the flat free corner
of an umbrella shell is due to element stiffness formulation or
whether its due to the use of stiffening eccentric edge members.
As for the response of the structure, for the variations of dif-
ferent parameters (rise to span ratio, a, etc.) both curved and
flat elements in general give the same pattern.
The assumption of shallowr.ess of the shell surface is used
in applying boundary conditions. This limitation can be elimi-
nated by a suitable choice of local coordinate axes along the
boundaries. This will need additional computer programming.
The computer program can incorporate the beams built along
the lines of generators. However, their influence on improving
the shear rigidity and reducing the deflection of the shell is
not studied. The local release of forces such as hinge connec-
tion (moment-free) was not incorporated in the solution.
-152-
BIBLIOCPAPh'Y
1. ·'non., ':Elementary Analysis of Pyperbolic Paraboloid

Shells, 11 Portland Cement ...Association, 1960.
2. Sondhi, J. R., and Patel, i·T. N., ,,Design for Hyperbolic
Paraboloid Shell Footings for a Building in r-~ombasa,
Kenya," Indian Concrete Journal, June 1961.
3. rarguerre, K., ''Zur Theorie der ·gekrilmmten Platte grosser
Formanderung, '· Proceedinp.:s, Fifth Int::rnational Congress
en Applied Hechanics, 1938.
4. Vlassov, V. Z., "General Theory of Shells and its Appli-
cations in Engineering' Ir nASA Ted1nical Translation F-99.
s. Peissner, E., "On Some Aspects of the Theory of Thin
Elastic Shells, 1 Journal of the Boston Society of Civil
Engineers, Vol. 42, No. 2, 1955.
6. Apeland, K., and Popov, E., "Analysis of Bending Stresses
in Trans la ti on al Shells, n Procer!dings of the Colloquium on
Simplified Calculation r~ethods, Brussels, September 4-6,
1961.
7. Bleich, H. H., and Salvadori, h. G., "Bendinp: r~oments on
Shell Boundaries," Journal of the Structural Division, ASCE,
0 ct ob er 19 59 .
8. Chetty, S.r1.K., and Tottenham, H., "An Investigation into

the Bending Analysis of Hyperbolic Paraboloid Shells, i i
Indian Concrete Journal, July 1964.
9. Das Gupta, N. C., "Edge :':lis turbances in a Hyperbolic
Paraboloid, 11 Ci vi 1 Engineering and Pub lie 1''orks Pcview,
February 1963.
10. r·~irza, J. F., ''Stresses and Deformations in Umbrella
Shells 1 ' ' Journal of Structural Division, ASCE, April 1967.
11. Pussel, :<.. rr., and Gerstle 1 K. H., "Bending of Hyperbolic
Paraboloid Structures,'' Journal of the Structural Di vi-
s ion, tSCE, June 1967.
12. Russel, R. IL, and Gerstle, K. H., "Hyperbolic Para-

boloid Structures on Four Su!'ports, 1 ' Journal of the
Structural Division, ftSCEr April 1968.
-153 ...
13. Schnobri'ch, iv. C., and Gustafson, W. C., "Discussion on

Reference 10," Journal of Structural f'ivision, ASCE,
January 1968 •
. 14. Croll, James G. A., and Scrivener, J.C., ' 1 Pdge Effects
in Hyperbolic Paraboloid Shells, 11 Journal of the Struc-
tural Division, ASCE; ~arch 1969.
15. Croll, James G. A., and Scrivener, J.C., "Convergence of

Hypar Finite-Difference Solutions 1 " Journal of the Struc-
tural Di vis ion, ASCE, r{ay 19 69.
16. Brebbia, Carlos A.~ nA Critical Review of Bending Solu-

tions in Hyperbolic Paraboloid Shells," University of
Southampton Report (England).
17. Connor, J. , and Breb bia, Carlos A. , 1 'St if fnes s ~·~a tr ix for
Shallm·1 flectangular Shell Element l>" Journal of Engineer-
ing: ~ echanics Di vis ion 1 ASC[ / 0ctober 1967.
7
18. Deak, Alexander L., "Analysis of ShalloN Shells by the

Dis p laceJ1lent ~iethod, n neport 6 7-A" ~~ni vers i ty of Washing-
ton, College of Engineering, Department of Aeronautics
and Astronautics.
19. Parker, John E., "Behavior of Light Gage Steel Hyperbolic

Paraboloid Shells, 11 Ph.D. Dissertation, Cornell rniver-
sity~ 1969.
20. Pecknold, David A.~ and Schnobrich, W. C., "Finite-Ele-

ment Analysis of Skewed Shallow S~tells," Journal of the
21. Pecknold, r. A., and Schnobrich, t11 • C., 11 Finite Element

Analysis of Skewed Shallow Shells,u A Technical Report
to t11c Office of naval nesearch, Department of the Navy.
22. Ralston, l\., "On the Preble~ of Duckling of a Hyperbolic

Paraboloid Shell Loaded hy Its o,,rn 1'.:eight, I' Journal of
i'-atheinatical Physics, 1956.
2 3. Dayara tnarn, P. ? and Gerst le, K. H. , "Buckling of Hyper-

bolic Paraboloids, n 1Vorld Conference on Shell Structures,
October 1-4, 1962.
24. Puskatl> ~ich.ard, '·Buckling of Ligr.t Gage Steel Eyperbolic

Paraboloid Poofs 1. Ph.D. l'is s ertat ion, Corne 11 Pni ver s i ty,
j
1968.
25. Brebbia, Carlos A., and Connor, J., ''GeoJT1etrically Non-

linear Finite-Element Analysis]' Enrineering ~· echanics
1
Division of ASCE, April 1969.

-154-
26. Rowe,P.. E., andr~irza, J. F., '?-.nExperimental Investi-

gation of Small Scale Hyperbolic Paraboloids,'' Proceedings
of World Conference on Shell Structures, San Francisco,
October 1962.
27. Pm·re, R. E., "Tests on Four Types of Hyperbolic Shells,r'
Proce~dings of the Symposium on Shell Research, Delft,
August 30-September 2, 1961.
28. Leet, Kenneth F., 1.study of Stability in the Hyperbolic
Paraboloid, 11 Journal of the Engineering Vechanics Divi-
sion, ASCE, February 1966.
29. Bertero, V., and Choi, John, ·~cheJTJically Prestressed Con-
crete I!yperbolic Paraboloid Shell ~1 iodel, 1 ' Proceedings of
October 1962.
30. !"cDcrmot t John F. , ''Single -Layer Corrugated-Steel-Sheet
$
Hypars, 11 Journal of the Structural Division, ASCE, June

1968.
31. Yu, C. ~·J. ~ and Kriz, Ladislav B., nTests of a Hyperbolic
Paraboloid neinforced Concrete Shell, i ' Proceedings of
Octorer 1962.
32. Nilson, A. H., '·Testing a Light Gage Steel Hyperbolic
Paraboloid Shell, 11 Journal of the Structural Di vis ion,
ASCE, October 1962.
33. 1
'1:.P,ry0:--::..s of Shear Proprietary Tests of Hypar Hetal Deck-
in<_., '' ~.Private Communication).
34. E::·:, :....1·~t011, N. J., Jr.,
Theoretical I'etermination of
11
r .i.:?_;_c~~ :~r
nroperties of Orthogonally Stiffened Plates' I'
Jcur.11d.l 0£ Applied Pechanics, I"arch 1956.
35. Smi.-t:h, Geoffrey E., 1 ·Elastic Buckli'.1£ in Shear 0£ In-

fin.i tely Lrng Corrugated Plates with Clamped Pa1·allel
Edges," A Thesis presented to Cornell University, for a
Paster of A1)ronautical Engineering 1957.
36. Tirnoshenko,. S. P. , and Gere, J. r'. , ' Theory of Elastic

Stability," McGraw Hill Second Edition 1961.
37. B~res, nichard and i'assonnet, Charles, "Analysis of Beam

Grids and Orthotropic Plates, 11 Crosby Lockwood and Sons,
Ltd.,·SNTL-Publishers·of Technical Literature, Prague, 1968.
38. Anon., 'Light Gage Cold-Formed Steel Design Manual,"
Edition 1961, AISI.
-155-
39. Winter, G. 1 "Commentary on 1968 Edition, Light Gage

Cold-Formed Steel Design I.'anual, ;i AISI.
40. Kinloch, H. :1 and Harvey, J. !T. , "Analysis and Design of

Anisotropic Plate Structures with Particular Peference to
Decking Systems,~? An /\rticle from "Thin-T•falled Structures, 11
Edited by Chilver, A. H. ~ John ~\Tiley and Sons, Inc.~ N.Y.
41. Lekhnitskii, S. G., 1 ;Anisotrcpic Plates,': Translated from

Second Fuss ian Edition by S. 111. Tsai and T. Cheron,
Gordon and Breach Science.
42. Zienkiewicz, 0. C., and Cheunc-l' Y. K., 11 The Finite Ele-

ment nethod in Structural and Continuu111 :cechanics, '' ~·cGraw
Hill Publication 1968.
'.43. Bryan, E. R. jand Fl-Dakhakhni, 1'.1 • rl., "Shear Flexibility

and Strength of Corrugated reeks," Journal of Structural
Division, ASCE 1 November 1968.
44. Falkenberg, J.C., "Discussion on "Q_efcrence 43, 1 , Journal

of the Structural Division, ASCE, June 1969.
45. Luttrell, L. D. , 1 'Structural Performance of Light Gage

Steel Diaphragl!ls,'' Ph.D. Dissertation, Cornell Univer-
sity, 1965.
46 .· Troitsky r.1. S., northotropic Bridges Theory and Design, r;

J
The James F. Lincoln Arc ~elding Foundation.
4 7. Anon. , nDesign i1anual for Orthotropic Steel Pl ate Deck

Bridges 1 •; AISC Publication 1963.
48. Easky, J. T., and Y'cFarlnnd, D. E. , "Buckling of Liph t

Gage Corrugated Petal Sl~ear i1iaphragl1's, ,- ASCE Journal of
the Structural Division, July 1969.
49. Wright,. R. 1'1., et. al. 1 ''BEF ./\nalogy for Analysis of Box
Girders," Journal of Structural !:livision, ASCE 1 July 1968.
50. Gallagher, P.. H., 11 The DevelopTI'.ent and Evaluation of

~~a tr ix tretl:ods for Thin She 11 Structural 1'.na lysi s, 11 Ph.D.
Dissertation, State University of ~ew York, Buffalo,
June 1966.
51. Yang, Henry T. Y., "/I. Finite Ele1n.ent Fonnulation for

Stability Analysis of Doubly Curved Thin Shell Struc-
turesi-'1 Ph.D. Dissertation. Cornell University, 1968.
52. Cantin;!'.• , and Clough, P., "A Curved, Cylindrical Shell

Finite Element, 11 AIAA Journal, Vol. 6, June 1~68.
-156-
53. H.a,is ler, l'J •. , and Stricklin, J. , "~igid-Body Displacements

of Curved Elements in the Analysis of .Shells by t·'atrix
. Displacernen t r,'ethod, 11 AIAA Journal, Vo 1. 5, No. 8,
August 1967.
54. Timos.henko, S. P., and Woinowsky-Kricge+, ''Theory of Plates
and Shells, 11 VicGraw-Hill Book Company, Second Edition 1959.
SS. Weaver, 1~7 il liam, Jr. , "Computer Programs for Structural
Analysis,'· D. Van Nostrand Company, Inc.
56. Bleich' Friedrich' t!Buckling Strength of r.fetal Structures' II
. McGraw-Hi 11 Book Company, 19 52.
57. ·Hoff, N. J., 11 The Analysis of Structures," .,John ;•riley
and Sons, Inc. , 1964.
58. Gallagher, IL and Barsoum, IL, 11 Finite Elewent Analysis of
Torsional and Lateral Stability Probelms, 1 Int. Journal
for. Num. I:eth. in Engrg., Vol. 2, No. 2, April-June 1970.
' '
59. Hall, A. s. and Woodhead, P~ 1r., ''Frame Analy~is,

1 11
John
Wiley and Sons, Inc., 1961. ·
60. Anon., 11 Torsion Analysis of Rolled Steel Sec.tion, 11 Steel
Design File Design Data from Bethlehem Steel, AIA, File
No. -13A-1.
61. Jones, R. E. an<l Strome, P. P., •:A Survey of Analysis of
Shells by the Displacement 11 1ethod," 11atrix r'ethods in
Structural ~·echanics Proceedings of the First ~Vright
Patterson Airfield Conference, Octobc! 1965.
62. Gere, J.P. and Weaver, l'J., "Analysis of Framed Struc-
tures," D. Van Nostrand Company, Inc., 1966.
63. Anon. , Stre ss, A C:s er' s !ianual," !~IT Press.
11
64. Brooks, D. F. and Brotton, D. ~·.,''Formulation of a

General Space Frame Program for a Large Computer," The
Structural Engineer, Vol. 44 s J\To. 11, November 1966.
1
65. 'Behavior of Hypar Shells,'' National S.cience Foundation
Grant GK-1282, A Project under progress at Cornell Uni-
versity.
66. Gallagher, P. H .. , et. al., "Discrete Element ·Procedure

for .Thin Shell Instability Analysis, AIAA Journal,
January 196 7.
6 7. Hallett, 'Z. H. and r1arcal, P. V. , ';Finite Element Analysis
of Hon-Linear Structures, 0 Journal of Structural Divi-
sion, ASCE, Septewber 1968.
-157-
68. Kapur; K. and Harz, B. J. ~ "Stability of Plates using the

. .
F 1n1te r1 ef'lenl."· "j e th
L ..:s 1 ' J ourna 1 o f_ E
,_.oc, . .
-:Tg1neer1ng ''···ec h an1cs
.
Division, ASCE, Vol. 92, April 1966.
69. Anon., nsystcm/360 Scientific Subroutine Packape
(360A-CH-03X) Version II, Programer's ~·anual rm: Applica-
tions Program flanual HZ0-0205-2, 1967.
70. Gustafson, W. C. and Wright, R. N., "P.nalysis of Skewed
Composite Girder Bridges, 11 Journal of Structural Division:.
ASCE, April 1968.
71. Argyris, J. H. and Scharpf, D. ;·r., "A Further Sequel tc
Technical I,Jotc 14 Puga Family of Elements for Folded Plate
Structures," The Aeronautical Journal of the Royal Pero-
nautical Society, Vol. 73, July 1969.
72. Oj al vo, I. V. and Ncntman, n. , ''I'Iatural Vibrations of a
Stiffened Pressurized Cylinder· with an Attached ~rass,"
AIAA Journal, June 196 7.
73. Scniggs, ?. 1·., 1 'An Analytical and Experimental Study of

the Vibration of Orthogonally Stiffened Cylindrical
Shell, 11 AIAA/ ASJ'RE 9th Structural Dynawics and Paterials
Conference, Palm Springs, California, April 1-3, 1968.
APPENDIX A
DECK PFOPEPTIFS
The elastic properties along principal directions of

"'
orthotropy are eiven for both closed and open decks. A detailed
computation of the elastic constants for an orthotropic deck
are reported in Refs. 34, 35, 36, 37, 40, a1, 46, 47, 48.
tl.-f CORP.UGATED OP.EN DECK

: '
Fig. A1 (a) shows an arbitrary cross-section of unit

length, in y-direction, for an open deck.
A- Ia rr.embrane Constants
Ex = The extensional elastic constant for the equi-
valent orthotropic plate, is obtained by equating the extension
Ax between the physical and the idealized system.
From Fig. A1 (a) ,
For constant thickness t of the dee¥.,
P. R. s z2t
l.l = l ds A-2
X I:t! 0
0
-Y-
s z2t
I T ds = Iy' A-3
0
where Iy' is the moment of inertia of the cross-section, per

unit length, about the line of action of load P, which coin-
cides with the x-axis.
For the equivalent flat plate, (Fig. A1 (b)),
-159-
P9..
= A-4
Ext
Equating equations A-2, A-3 and A-4, we get
IO
E
x = r' I:.: 1 A-5
y
The other membrane constants are given in Table A-I.
A-Ib Bending Rigidities
Bendinp rigidities for the x and y-directions are
given in Table A-I. The method of determining the constants is
given in f. 35. For the orthotropic plate, the twi~ting con-
stants Dxy and Dyx are not equal and hence the average values
o f tlese
1 . t.e
constants are use d 1n . 35 .
h ana 1.ys1s ~··
r inor d ev1a-
.
tions in the properties of rx and Dxy are noted in Pefs. 48,54,

however, the small deviations in these relatively uniwportant
properties are considered of no consequence at all.
A- I I CELLULAR CLOSED DECK
The Fig. A 3 shows a typical c~llular deck where a

trapezoidal hat section is connected to the base plate.
A-IIa i"embrane Constants
The principle for the computation of membrane con-
stants for the closed decks is the same as that for the open
de cl'. However, because of the continuous base plate, certain
modifications in the constants, such as Exy and vyx' are re-
quired. Ex - Assuming full fixity between the base plate and
the hat (Fir. A 3 ), for BC,
A-6
From Eq. A-5~ [xh' the elastic extensional constant, for the
hat only,
-160-
Elot2
Exh = Iy-,- A-7
1' • is the worncnt of inertia of the hat portion about the base
y
plate. By compatibility,
E t
p
2
= p1 ( 'x'h) c-1)
E t
A-8
1
The equivalent orthotropic plate is assumed to be of the thick-
ness t 1 •
E
x
= p,-9
( a + ~\
L t NJ
R,(l+( ~h)(~))
E t
1
For all practical purposes, the deno~inator of the above equa-
tion (A-9) is equal to unity.
A-10
EY and vyx - If the full cross-section in Fir. A
3
is effective
in resistinr the axial force along the y-axis then,
A-11
where Ay is the cross-sectional area of the section sh:wn in

Fig. A3 . ny the ;:axwell-Betti reciprocal theorem,
A-12
A
vyx = v2
to. A-13
''y
The shear rigidity for the closed deck can be considered equal
to that of the base plate which is usually cirectly connected
to the edge me~bers. Even a thin plate interconnecting the
free hat portions will increase the effective shear rigidity of
-161-
the section. However, the cost of construction will have to be

taken into consideration in this approach.
A-lib Bending Properties
These properties are given in the Table A-I.
A-Ile Torsional Rigidity
Here again the values of Dxy and Dyx are not equal.
As shown in Fig. A5 , the value of Dxy is negligibly small as
compared to that of Dyx· The value of Dyx can be obtained by
using Brendt's formula 46 •
a . r1 A-14
~
where Ac is equal to the area enclosed by the cellular hat
portion.
However, the magnitude of D x tacitly assumes that
. y
the shape of the deck does not deform (dotted lines in Fig. A4) 47 •
The in~plane shear forces set up due to the twisting, produce
secondary bending moments in the individual plates. Besides,
the bending of the overhanging plate, outside the cellular
portion, further adds flexibility to the cellular section.
In order to account for this reduction in torsional
stiffness, an effective torsional rigidity of the cellular deck
is defined as the torsional rigidity of the ideal system (shown
by the dotted lines in Fig. A4 ), free from individual plate
flexural deformations. The work of the deformations due to the
torsion of an idealized system is equated with the work due
to torsion and secondary flexure. The modified twisting
rigidity can be given as,
-162-
4A 2
.D
yx
= { (15+2h f) c + A-15
tz
The reduction factors given in Ref. 47 were checked. The

method of calculating the reduction factors is given in the
.
Re f . 46 . As pointe d out in . 49 sti"ff ness
. C'i.1apter II , t h e warping
for both closed and open form decks is neglected in the analysis.
.. 163·
-x
'
(a)
p - Fig. Az
(b)
Fig. A Equivalent Orthotropic Plate

1
p z
Fig. A3 Cellular Deck

-164-
Assumed llinges
~Midway Between
/ Ribs
\
Rotation in
an 1dealized
System z
/\ctrrnl
l<otat ion
Fi~. ,i\ (1-~er. 47)

4
Twisting of a Cellular Section

-165-
TABLE A-I
PROPERTY: DECKS
CPEN CELLULAR ISOTPOPIC
Membrane
Io
Ex E E p
.._.
Iy '
A
s _r_
F.
y F r F
P..x
E
\) \) \) \)
xy
I Q, A
x
0
\)
yx Vy-1 5- \)
Ay
\)
y
1
El vE I0 ' vE vF.
y
!~-
1- \) yx \) xy " 1. 0 1-v 2 x-xy 1-v 2
E
xy
aG aG ~
l.J
Bending
~ 3
3
iE t 3 .t. tl Et
D
x l2S 12(1-v 2 ) 12c1-v 2 )
n
·'~·
y f:Iyd EI vrl
I_. Ft
3
2
V[t 3 £ 12(1-v )
'I
VE t 3
()
'-·1 12S
12(1-v 2 )
3
Et
'- -
xy GI 0 0
24(1+v)
GI
0
s 4Ac 2G Et
3
D -9.- ds 24(1+v)
yx
Rrcd Ut) £
-166-
APPENDIX B
ST! FFNESS r ,ATRI CES .
The stiffness of the curved e1e~ent is given in three

parts: I: I!embrane Stiffness (u,v, displacements); II: Bend-
ing Stiffness ('t':, ex, ey, exy); III: The coupled terms of
u, v and w.
I: ~.~EJ".BJ?A~!E STIFFNESS VATRIX [K]m
TABLE B-I
Terre Ext Elt Eyt Exyt

' b a
4>1 3a Th
1 1
¢2 4 4
-------- . - -
6 6a __
31)
_..,
a ;.,
4>7 Th 3i
<Pg
a
60 - b3a
a b
<Pg 61) 6a
<Pio a b
3fi . 6a
a
e.g. 4>1 = b3a Ext + - E
3b xyt
-167-
II: BENDING STIFFNESS ~,ATPIX [K]
All terms multiplied by c 1 are resulted because

shear strain depends upon the ·w terws as well (see Eq. 3-~c).
These terms follow the same pattern as the bendine stiffness
terms derived from the Eq. 3-17.
TABLE B-II
Coeff.
676 ab
2
11 b 2 11 a 1 11
-7, -c. 7a- 35 b:z- 25 so
156 b 54 a 144 1 72 1
-35 ~a 35 ;3
t)
2S an -ab 234 ab
27 a 12 1 36 1
35 b2 25 a 25 a
12 1 6 1
25b - 2.Sb"
lrb 2 1 3
352
a 25 25
540 54 a 144 1 72 1
25 81 ab
35 a3 35 b3 ab 25 ab
13 b 2 27 a 12 1 6 1 211.. ab2
35 a3. 35 ~2 25 a 25 a 6
---·--
27 b 12 1 6 1
35'""7
a
2511 250
Continued
-168-
Con'tinued
Coeff. D D D
x y xy
1 1 169 2b2
25 so ~a
54 b 156 a 144 1 72 1 234 ab

35 a3 - 35 b3 - 25 ab - 25 ab
13 b 2
78 ~ 12 1 6 1
- _.3_s_a_3_ _ _~_ 1
s b 2~·_ _,_2_s____
6
_ _1_4____ a_ 25 a
27 b 22 a 2 12 1 36 1
- 35 ~2 35 h3 25 -b IT b
13 b 2 11 a 2 1 3 143 2b2
- 70 a2 35 :lJ l 25 25 18 a
~---- ~-------------~--~---~
4 b 52 a 16 b 8 b
35~ 35b zsa Ea
2
11 h " 1 61
- 35 z,
a.
, 25 so
2 b ''
3 22 a 2 4
'35 2 105 b 75 b
a
4 b 18 a 16 b 8 b
- 35 ~ 35 b 25 a 25 a
--.-2
2 b 13 a 4 _z.
b
35 ~ - 155 b 75 b 75
3 b3 9 a _4_b____2 b Q. ab3
35 a 3 35 b '-2 5 a 25 a 2
3 b3 ---13-T -:-· b-..- - - - b_ _ _ _ _
1_3_2_b_3_
- 70 2 - 210 r,-·. - 7s - 150 IT a
a
?
11 a- b b 11 2 3
105 ~ 75 TI 6 a b
52 · ·b· 4 a3 16 a 8 a 52
6 26 - a 3b
35,' a 35 b3 25 b IT b 3
2
22 b 2 a 3· 4 4 2 2 a 3b 2
~27.___-_1_0_~.~a___-_3_s_.~~2~----7-;_,_a~----2~s_a_______--~9:__~--
6
Continued
-169-
Continued
Coeff. Dy
3
26 b 3 a 4 a 2 a
35
~~-~~-~~-;;.--~-
a 3s b3 25 b 25 b
11 b 2 3 a3 a a
--10-5 -a 70 b2 75 25
9 b 4 a 2 a 9 3
ab
35 a 25 b 25 b 2
a a 13 31' 2
75 150 IT a -
16 a 8 a
25 b 25 b
4 2
75 Cl 75 a
16 ab 8 b
225 225 a
4 ab 2
225 225 ab
1 1
225 ab 450
ab
4 2
225 ab 225 ab
III: COUPLING rrATRIX [K]b

·--- m
These terms are formed due to the coupling of u, v
and w displacements (see Eq. 3-9c)
C - F ( C )
2 - -'xyt AAxB13
The stiffness matrix for the curved element can be

reduced to that of a flat plate by putting c=o; whereby c 1=o
and c 2=o. The schematic representation of these two matrices
is shown in Eqs. 3-20 and 3-21.
-170-
TABLE B-III
Coeff. c2 Coeff. Cz Coeff. C2 Coeff. c2
7 3 7 __!_ b 2
el 20 a es 20 a eg 20 b el3 30
-
1 1 2
e2 ITO7 ab a6 40 ab 8 10 _!_ b2
20 el4 lflO ab
1 2 1 2 6 11 1 2
03 -IO a 87 30 a 120 ab
1 . 2 1 2 3
e4 120 a .b es -180 a b el2 20 b
!91'.t e 1;6x I
u.,
SYMMETRIC
....
t
..........
I
TABLE B-IV
CURVED ELEMENT STIFFNESS MATRIX [K]
ti -t_l- -GL_~ ...~~-e?>.. G.,, ~4
4' Yo;
APPEIJDIX C
DECK STP.E
A typical deck element 1-2-3-4 rectangular in plan is

shoim in Fif. c1 . The deck forces and the deck moments with
notations sh.mm in Figs. 3. 2 and 3. 3 are calculated at the center

point o. The deck in-plane forces and moments are calculated
for a curved clement using the strain displacement relationships
2ive.n in Eqs. 3-9 and 3-10. The forces for the flat ele~ents
can be deduced frorn the expressions given beloN. The two major
diffe~ences are: (a) c=o; (b) The local nodal displace~ents
in the local axes for each element are to be used.
C-1
C-2
1l xy = (u, y + v 'x - 2 /~--

E w) Exyt
= Exyt{(-u 1 -u 2+u 3+u 4)/2b+(-v 1 +v 2+v 3 -v 4 )/2a
- ~ [ (t': 1+w 2+w3 +~.r 4 ) / 2+ ( ex + eXz - ex - ax ) b8

J.B
1 3 4
+ c-e +a -e -e ) -a + (e -e +e -e ) ab 11 c 3
Y1 Yz Y3 Y4 8 xyl XYz XY3 XY4' -
w , xx = [ Cay 1 - 8y 2 - a y 3 +8y 4) I 2a+ C- 8xy1 + e xy2 - 8xy 3+ 8 xy 4) h·] C- 4
w,YY + [C- 0 x1- 0 x2+&x3+ex4)/ 2 b+(-exyl+~xy2-exy3+exy4)~b] C-5

-173-
The moments per unit length can be obtained

;'!
x
= - [ D ,IT
x 'xx
+ n1w'yy ] C-6
71
y
= - [D 11v, xx + D N.
y 'YY
] C-7
Fig. c2 shows the bending stresses lvhich could be

intrcduced due to the local bencHng of each individual plate
elements. These stresses are not calculated here.
-174-
J I 2
---'1-x
I
b
INx ,Ny
2 ,Nxy
-·----· -10.IM_____
x ' My
-
h
2 I
I
hg. c1 Deck Forces at Point 'o'.
- p
Fig. C2 Local Bending of Plate AB

-175-
APPENDIX D
INCPJWENTAL l'<ATRIX FOR THE DECK
For si~plicity, the ~atrix is split up into two

parts. [N] 1 l·r!1 ich cont a ins the terms corresponding to the in-
pl ane forces ;.Jx and FY whereas the matrix [:N] 2 contains the
terms corresponding to the shearing force '·' xy l"r
[N] = [N] l + (N] 2 n-1
TABLE D-I
Coeff. Nx Ny Coeff. Nx Ny
78 b ·7s~a 27 b 78 a
 1 1 75 a
1 75 b
- - - - - - - - = -2 - - - -
 13 1 75 a - 175 b
11 b 13 13 b 2 13
<1>2 175 a
350 a
_ _ _ _ _ _ _1_3_ _ _ _ 11 a 2
<1>14 - 350 a
9 b
350
11
a
a2
) - 350 b - 17 5 b- <1>15 - 700 . 17 5 b
------- -------+--
11 b2 11 2 13 2 11 a2
2100 2100 a <1>16 - 4200 b 2100
- - - - - - - - - - - - - - - - - - - -3= - - - - - - -
78 b 27 a ~ 2 h 26 ab
------- 115 a 5 ns
'1'11 115 s25
~-----,t------------------
a
11 b 2 9 11 2 11 a2
<1>6 - m· a 700 a <l>1s - 211)0 b - 21QO
·~------=------------------
13 13 a 2 1 3 11 2
<1>7 - 350 b 350 o ¢19 1050 b 1575 a b
11 2 13 2 2 b3 3
<l>s 2100 b - 4200 a <l>zo - I7S a 175 ab
---------~2·=1~b~----~z-=1-a·~--~ 1 _3____1_3_a_2_b_
<l>g - 175 a - 175 b ¢21 1oso b - 3150
------ --+--------·---- ------
2
13 b 9 3 b3 3
<1>10 350- a 700 a <1>22 350 a - 700 ab
--------------
9 13 a2 1 3 13
<1>11 - 700 b - 350 5-- <1>23 - 1400 b 12600
------------~--~---t-----------=--------
13 2 13 2 3 ~' 13
cf>12 - 4200 b - 4200 a <1>24 - 350 tl·_:_ - 10 so· ab
Continued
-176-
Continued
Coeff. N
x
r~y Coeff. Nx N
y
1:-3 il-:-2 . 2 a
~) ... :
(,,,.,,.,,,. 1400 b 6300 a b 26 525 ab I75o <f>33 3150 ah - lOSIJ a
11 2 \ 2 b3 2 3
<P21 - 1S7S ab - 1 a3
<P34 1575 a 1575 a b
13 3 a 1 . 3 1 3
4>zs lOSO ab 350 b ¢35 3150 ab 1050 a b
11 2 1 3 1 3 1 3
<Pzg 6300 ab 140') a <1>36 4200 ab 4200 a b
3 3 a3 1 3 1 3
<P30 700 ab 3SO o 4>37 - 1050 ab - 3150 a b
13 2 I 3
<P31 12600 ab 1400 a
TABLE D-11
Coeff. N Coeff. N Coeff. N

xy xy xy
1 .a a2
el 2· 64 10 e7 60
e2
ab a 2b b2
so es 300 es 60
b 1-,2. a2b2
au
03 IO e6 300 e9 1800
N ab·
e.g. N(l,4) -.· xy
so
SYI1'2.TPIC
TABLE D-I I I
[~.!] VATRIX
1
-177-
·o · o o . e2 · .o e5 eY3
o e2 o ~~
5 o exy 3
SYt1lliTP.I C
0 . ·~
xy4
TABLE D-IV
(N] f"ATlUX
2
-178-.
-179-
APPENDIX E
STIFFNESS AND INCREVENTAL f'ATRICES FOR A

FULLY COJ\7PATIBLE ECCENTRIC STIFFEHEr.
INTRODUCTION
In the case of light ga?e hypars the orthotropic deck
is connected at discrete points to the supportinp edge merrber.
Because of this discrete connection, there is a certain amount
of non-compatibility between the edge members and the deck. The
stiffness matrix used in Chapter III was developed by the use
of direct co-ordinate transforMation. The co-ordinate trans-
forJilation assumes that the neutral plane of bendinr for the
original beam cross-section re~ains unaltered. ~ith the type
of connections used for the light-gage hypars, the assumption
made above represents the true behavior.
However, in the case of stiffeners which are rigidly
connected to a plate or a shell or a monolithically cast con-
crete beam, there is full corrpatibility between the strains at
the junction of th.e deck and th~ hearn. In this cornpati ble cas c.
a part of the deck also acts along with the stiffener (effective
width concept). The interaction of the deck and the stiffener
results in the adjustment of the neutral axis of the section.
To account for this change of neutral cixis, the bendinp nroµerty
of the stiffener will have to be rrodified hy arbitrarily assum-
ing the effective width of the deck actin~ along with the be~m
(see discussion on Struc. '6', Section IV-SP). ;Jencnding upon

-180-
the relative stiffness of ~he plate, this kind of arbitrary
adjustment in the stiffener property will present a problem.
This arbitrariness is removed by formulating the compatible
stiffener element.
STIFfNESS rrAT!UX
A typical eccentric member is shown in Fig. 3.10.
It is assumed that the mer.iber is uniform in size and its local
axes i, ~and i, througt the shear center and the centroid, are
parallel to the global axes x, y and z. Let Uc be the average
.
axial deformation of the section measured at the centroid of
the section (C. G.) and Vs and 1
\:
s be the bending deformations
measured at the shear center (S.C.).
Assuming a rigid connection between the stiffener and
the reference structural node points, the displacements at the
shear center and the centroid can be expressed as follows:
uc . = u - z
c
(aw)
ax - Y.c (av)
ax E-la
= v - ZS e E-lb
"s
ws = w + y SS E-lc
where e= !;. The total strain energy of the beam ele:rnent can
be given as
a au c 2 a a2vr 2
ub =
EAb
-z- l (--) dx + E~y l ( . s) dx
0
ax 0 ~
Elz a a2vs . 2 1 . a a
-z- 0~dx· + !.2 Er ·a·' 2dx E-2
cf C-z) 2·
+ cl.x + G.J ! I
ax 0 0
All the member properties are expressed with refer-

ence to their local principal axes. The displacement fields
-181-
for u, v, w and a are given by Equations 3-26~ 3-27. The

stiffness matrix (neglecting the in-plane rotation a I is given
z
in Table E-I. Though not tested on the hypar structure, the
element when checked for simple cases such as simple and canti-
lever beams, converged to the correct results. The element can
be used for a rigidly connected eccentric me!'1ber and correctly
locates its mm neutral axis under different types of loading.
It may be worthwhile to note that because of the
coupling of the u, v and w displacements, the resulting displace-
rnents for u are no lonrrer linear. The use of a Hermitian
polynomial of t!"'ce same order as used for the v and w displace-
l!lents will give more rapid convergence characteristic than with
a linear u displacement field. It was not possible to use u as
non-linear because of the lack of a correspondine degree of
freedom in the formulation for a plate or a shell element.
Similar stiffness matrices are reporteC: in Refs. 70 ~ 71, whereas
the theory of the element is developed in detail in ~efs. 72,
73.
INCREl~E~TAL MATDIX F'_iR UNIFOill'LY ccr:PPESSED

ECCENTRIC STIFFINER
The load acting on a fiber whose cross-sectional

area is dA, is cr(;Jl. and the change in the distance between the
end point of a fiber is given by ~u. The increase in the po-
tential energy due to the change of fiber is ziven by
8V =- f oln1dA E-3
/,
For the section which shm•s both TI'odes of buclrling (flexural

-182-
as well as twisting), thci change in the length of a fiber is

given by 57 , (Fig. E1 )
a 2 2
tm = 1 J {(aw+ ye') + (av - ze') }dx E-4
2 0 ax ax
For the uniformly loaded section, o is the constant

quantity and oA = P. Therefore the change in the potential
energy
a 2 2
oV l p 1 '{(()l..r) + Cav) 2Yc Caw) ,
=
2
0
ax ax +
ax 8
-2Z c Cav)e'
ax + e 12 p 2 }dx E-5
where Zc and Ye are the distances of the centroid of the sec-

tion (Fig. 3.10) and p ii the radiui of gyration for the polar
moment of inertia of the cross section about the point o.
2
p = -1 1 r.-6
A /\
The incremental matrix can be obtained by integrating

the expression E-5.and then obtaining the second derivative of
the total change in' the potential energy. The incremental
matrix (8x8) is given in Table ~-II.
The stiffnes~ and the incremental matrices for the

eccentric stiffeners can be used to ~redict the buc~ling load
for the stiffened p~ate. Becal:lse ,of. the non-availability of a
large-order eigenvalue solution routine, the formulation could
not be checked with.other available solutions. However a
simple case of .twist-bucl<l'j.ng of colulllJl section about the
forced shear center for a T-section and I-section ~.,,as compared
with the classical solution for hinged end condition given by
-183-
Bleich 56 .
2
cr _Q_ a J] E-7
c ·rr 2E I pc
I pc = Polar moment of inertia about the forced shear center 0.

The only difference for both the cases is that the T-section
has rb = o whereas rb for the I-section is 1590 inch 6 , In
both the cases, for simply supported ends, the twist bucl<ling
load was within <<1% in error, from the values given by the
classical solutions. A negligible error is observed in one
element solution. The buc~ling stress predicted with
2 and 4 number of elements showed an 3lmost insipnificant i~-
provement. ·
-184-
z
tJy = -ze
tJZ = Ye
z
y
Fi~. 1: 1 Displacement Due to Twisting of a Beam
0 0
\
a = 200"
,,
\\
\\
\\........ /\,.
/
/
\
\_ ST 12 I 39-95 \_ 12 WF 40
Fig. E2 Twist Buckling

·185-
'/(I
\} ' 1,.., t e It I 0 (f I 8 ><.~I
Ab£ AbE l.c.

·---- () 0 (' 0
0.. I a...
SYMMETRIC
_____.,. ..
(; 0 0 0
. ~ --
12.f [ 0 l 12.EAbYcl -12 £ l z ls i'EAi,Ycl~ 'EI• "l. 5
A.~ o.3 0. 3 a...2. o..•
t
···r- - --
12El0y 12.E.I'ijYs !'Elcy:-c.EliYs
- - - - - .---1 -
o.~ a~ tA.i-: o.2..
12-E l;r Y.s/~~ ·- . i-(,, El;/f~7i
TABLE E-I
r Eli/:1a_~'
11
re' EL ii y,-'i~" fli-c;I I&-'
(
STIFFNESS MATRIX FOR A COMPATIBLE
+~c,:r/o...; o.~~-GJ/to
ECCF.NTRIC STIFFENER
+ 12£rt. /a.:!i: _, E
1
rt/a.'.1.
Li E I 0 y !- 4El'lf Y6
I
a_ CL
4 E. I<J Ys -; a. 1
+'1 El 1l~/ o..
+fg GJ~
+4EI'i::,/a_
-186-
wl e xl eyl exyl w2 e x2 ey2 8xy2
6 6 a a 6 6 a a
5 5 Gl IO 10 Gl s S Gl 11) IO GI wl
a a a a
6
Gz - IO Gl IO Gz - S6 Gl - S6 G2 - IO Gl -
10C 2 9 xl
2 2 2
- rs 2r a a 1 2 1 2
rs a a ~)1
ro IO Gl - 30 a ~a G1 ey1
2 2 a a
- 10 1 2G
rs· a Gz - 10 Gl Gz 30 Gl - 30a 2 9xyl
6 6 a
SYMMETP.IC 5 S Gl 10 - a10 Gl \•T
'2
Gl = yc 6 a a
I S Gz IO Gl IO Gz 9x2
G2 =~
A 2 2 2 a2G
p IS a l~ 1 ey2
Note Common Factor a ---
r:ATRIX FOR 2 2c::
TABLE E-11
-
lNCP,m~mITAL rsa 12 e xy2
cm~PATIBLE ECCEHTJUC ST! FFET«TE P
--· ----
-187-
APPETDI X F
COJ'-,;Pl"TEr· PROGRJ'.}'S
Computer programs were written for linear stiffness

and instability analysis usinp both curved and flat ele~ents
(:!ethod 'a' and ~'.ethod 'b'). Since the input data for both
curved and flat eleBents is nearly the same, a general des-
cription of both stiffness and instability programs is given
here. All informatj.on given in this appendix, is pertinent to
Im~ 360/65 model availahle at the Cornell Universit~ nouble
precision is used in all programs.
STI FF:·rnss A:JI~AYS Is
Hypars have mostly heen analyzecl for a uniform grid

size 8x8 (64 elements). ~ependin9 upon the nvaila~le core
storage and required accuracy of the solution, the e-rid size
can be varied. For the flat plate elerents 1 the progr8M re-
quires a core size 365K and has a compilin~ tirne roughly of 55
secs. The computation ti~e for o~e proble~ is rour,hly 70
secs. The ti~e requirement for t~e curved element pro~rarn 1s
SoPle~·!hat less.
The input information can te categorize~ as follows:
1. Proper!ies of the ')eek
Properties of dee Ls shm:n in Fig. 2. 2a, b and c can
be calculated by a subroutine 'PI'EC" in which the geometrical
properties such as tl•ickness and other physical paraJneters can
be specified. For decks other than those ~entioncd above,
-188-
elastic constants Ex, Ey, ... etc., are to be calculated and

read in as input data. Besides the type of deck, the orien-
tation and number of decks also need to be specified. The
geometrical properties calculated on the basis of shape can
be further modified by the use of coefficients determined ex-
perimentally.
2. Geometry of the §urface
The eeometry of the surface is defined by specifying
the number of hypar quadrants and the property of each quadrant
in terms of its rise, spans, local origi.n (A 1. , B.,
1
C.,
1
x.1 and
~i' Chapter III). The groupinp of the elements in each quadrant
also needs to be specified. Structural shapes other than
umbre Ila she 11, can be handled v:i th ease.
3·. ~ring I'a ta
As pointed out in Chapter III) in order to retain
the close-handedness of the stiffness matrix, members such as
a supporting column and tension tie rod are idealized by
springs h11ving equivalent stiffnesses. The number of such
stiffness constants has to be specified.
4. Bearr Data
Be~ms ca~ be added along any lin~s of generators.
The beam data is given in the usual way with its location
specified by the start and end points and their co-ordinates.
Beam properties such as area, moment of inertia, w~rping can-
st.ant, ~eccentricities and distance· of extreme fibers for the
calculation of stresses are required.
-189-
5. Loading
The solution can be obtained for three different
loading cases or any combination of the three. The three
cases of loading are: a. uniformly distributed load over the
whole surface; b. weight of the edge rnewber; c. discretely
applied forces at any nodal points.
OUTPUT
The output is given in the following order:

(i) Six components of displacements at each node point
(u, v~ w, ex, ey' exy).
(ii) Axial, bending and total stresses for beam.
(iii) Deck forces per unit length (Nx, Ny) j~xy' !~x' and
~1y) .
(iv) Reaction forces in each idealized spring.
INSTABILITY ANALYSIS
Instability analysis is done only for uniformly dis-

tributed loading for the grid size of 6x6 with 36 elements
though results can be obtained for any general type of loading.
As pointed out in the stiffness analysis, the r-rid size can be
varied to meet the requirenents. The compiling til"e is roughly
50 secs, whereas the time for individual iterative cycle
ranees a1ong 20 to SO secs.
The input stream is basically the same as that for
the stiffness analysis, the additional infor~ation neeaed here
is the startinp: load point (First increment, see Chapter VI)
subsequent incremental loads and the num.her of iteration
-190-
cycles required at each step.

The output gives only the load level and the corres-
ponding displacements at all node points.
The program using the flat plates can he easily
modified to solve any shallow shell problems by defining the
equation of the shell surface in the global co-ordinates.
-191-
TABLE 11-1
SINGLE DECK
EXPT. AVERAGE THEORETICAL

TEST •0 INCH o0 INCH 0 INCH
No. 0 0 a.interpolated
811 1.14 1.14 0.07 1. 23

812 1.14 0.05 1. 45 0.077
511 0.80 0.06 0.68

512 0.77 0.78 0.04 0.87 0.050
311 0.33 0.32 0.06 0.28 0.052

312 0.32 0.04 0.38
TABLE 11-2
DOUBLE DECK
2 1/2" x
1/4"
TEST EXPT. AVERAGE THEORETICAL

No. o0 INCH o0 INCH 0 o INCH a.interpolated
821 0.86 0.75 0.07 0.71 0.080

822 0.65 0.06 0.77
0. 0 5 0.84
821(1) 0.65 0.04 0.93
521 0.59 0.58 0.06 0.38 0.040
522 0.58 0.04 0.48
521 (I) ·o.48
321 0.24 0.26 0.06 0.14
322 0.32 0.040 0.20 0.037
321(1) 0.21
Test No. 521(1) means:

5 = The curvature of the hypar, same as that of a
quadrant having rise/span ratio of 1/5.
2 = Two decks.
1 =Test No.1
(I) = Both top and bottom decks were interconnected.
'•.'
-192-
TABLE III-1
cm~PARISOl~ OF CLASSICAL Nm FINITE

ELm1ENT SOLUTIONS F0P. sr"PLY-SUPPOPTED
AND UNIFOI!MLY LOADED 28-G
n:r.THOTROPI C PLATE
FINITE ELEr-1ENT SOL UT ION

DOUBLE SINE
LOCATION GP.ID SIZE SERIES SOLUTION
6x6 12x·' 2
c INCH 8 INCH 0 INCH
A 0 0 0
1/6 6.117 6.106 6.31
1/3 7.397 7.397 7.36
1/2 7.137 7 .139 7.09
2/3 6.961 6.962 6.96
5/6 6.944 6.944 f.. 9 8
B 6.952 6.952 6.97
See Firs. 3-4c and 3-5.

~-r----~--
u, vl tv 1 e 1L1 elf i e-e 1. 1..l~ w'l- e?(.2- fJ(JL e-c.a.
_
\/ 1-
---·--
~ 0 0 0
AbE:G, , _AbEl'c. ..... A.bE
0 0 0
_ AbE!c: Ab£1c,.
·~
0.. a.. 0.... i a... a.,
.. - .- - --.--~--- -
-l2E I'll. ~S ' £Ii. - t.lfl:1: )12.E.I&~S G.E.li!
0 0
o~
D 0 :z...
Q,1; o..'J. a..o a..?> 0...
·- - ~.
··-•<, ---· C---·-
12. EI ;if 12..EI~ l& I -ft( I':I .. ,,.EI3 -12EI.a "fs - 'E l/f
'l.
0 0 0 0
o..~ o..~ Q,. a.,3 o.. a ~
.. .,. . - 2- - '-·-----·--····-- .. -· -···-- .. ·---·- ----- -t'l.EI,.?:YJ
1u11 -io/'J
,.. -iaEI;r1;-'Elt~ t'lEl~~ -11.E..t11 -0 £Ii:,l.S
:z. """' El;f't'!i
it I 'l. El '1~ 0
11 o.?,
t-12-El ll )Id '}...
2. d- 0. 0...2. (),,
- ~
0.
+6¥a.
... ~ ....1: • >'
~----~ ----- t-····· -·----- ~~°".- ...
~AbE. .:Zc.
"'"E 'l:c.
c1.
'-AJ-"l'c.lc -Ai,l'l-c ro£I4 l'£I~Ys a. IA bE ""( -i, 1
I
0
I+ 4E l;i 0.... <3. o..l- i 0.1... ~.it:rff a.. I
... 0.
- ·1
IAbE l'c:;.
·-- -. --· -----·-r·-·----·---
I
a.,.
-A"ET:rj
I:> c.
I
~ IAbE ">'c:;. -,EI l I1GEI.ls Alf.Zc."'4 .()\..

!
It u.££ 1
~
a.. - 0: 0
I
I
o.l- Ck-
LL~£ I~
0..
I
!
i
!
SYMMETRIC
AbE
-- 0 0 0 Ab£i!e -A11EYG I
()., a.
°'" -·-- )
J
12.E 1 1 - 2.&:I~~s - <;, E.I!

TABLE III-2 a.3
0
o..3
a 0..2...
STIFFN¥.SS MATRIX FOR NON~

~-------· -~----~
----
12Ell 12.E I(} Ys ,_er i1. 0
COMPATIBLE ECCENTRIC BEAM a.3 Ct~ 0. 2.
!
:LEMENT (LlNfiAR VARIATION OF 12..EI'! l.% 3
I
i
i
ANGLJ: OF TWIST e) .:OE IJ1'& ~ tizlg.
fl-12Eii"l'0 I
2- I
0.,1..
a. I
+ GJ"/ex \ I
I
AbE t.Z- I
I
a.. '-Ai:,E'fc.'l,i
ti- 4eI~ j a..
I
I
().., i
...,. I
··r:4-(J I 'U. l- 'V,_ wi. ex :.z... e<f z. e.,, ,. e-..~,.
AbE A&,E le -AbEY'-
'
---
a.
0 0 0
-
0..
...
d.
---·---·~ .
0
··-
- "be:
Oi.
0 0 0 '-AbE l:, AbE 'IC,
~ 0.
0
12£ 1 l. '-12.t I~ ls -11£1:; 12.Eii. i!s ~ E.Iz

- a,A 0
0.. '1
0 'fl&
Ao.
0 0
a.a
0
0. '3
0
(),,'l.
0
-- -
'-' u,
---~---·--·--
!
~-f.[1 UEI11s 1-12. S IJt -t:z. £!1 Ys -'-E.Il
0 0 0 0 0 0 I
&.l ~lo a.• ~3 ~J.
s " -. I-··. ,, .. ~
~---~---
.._~
Yo/'.J.'o.'iI .:.·11u 1 ~~
IZ£La
~ISU 1 l~'-vEl,y~ _,£11~

-GJ
ro
0 11n,\ -1.u1,'fj -.1i.£I~Y~
I to
GJ
,Er4> -"EI~'fs-'£I«1 ~ t0£P1

+ tc..YC1 (l. ,J. o,2.
l'I"'
a.,2. 0. 3 Q.3 ~ - GYa. a.;jl. o.'- 0:-

+- 1.z El'._/a ... -:_12 Efth.~
.. -.. -- - -~ ··~---- ·-·----~
4U~ ,.,..A.t,El.,"'(. 2.£Iih

~bE.lc;. 4'.EI~ bEI 1 'fs A!fl'c.~
...... a.. 0 0 ... A'-E1t 0
... 79- (L
-·
~
_.. _______ a,J. o.:l- 0...
... -----·-
lf.Ei lict Aa.E Y<. -'1 E Il: ~ E.. i. .s AbE:!,'fc
...A,,E'i"%
. ,.,.EY~ 0
a.. cl. J-
0
...
a. :z.. 0.
irUI~
0
-~-----
~GJa -GJ/10 ioGla.

16
~4E~
0 0 0 0 i
0
-GErbfa.1 tf-2.Er~
AbE AbEic -Ai,EYc.

- a.. 0 0 I! 0
0., ()..,
0
SYMMETRIC -----!---
12.£ It -12.E. r a 'l-s i-"E Ii
TABLE III-3 0 0 0
<:1_3 oc.. 3 o.i
··-·---~ ---·-
STl fFNESS MATRIX FOR 12.f.I'I Ys (,£14
12.£Il 0 0
NON-COMPATl BLE ECCENTRIC a. '3 OI. 3 o.."'
l\EJ\M l~LEMENT (CUBIC •2,El~lsfd ,_ G-.J
1. 10
VI\ Hl/\TION OP ANGLE OF TWIST, 0) lf. 12.£ l./'/d
6fI:i'fs 'Ell~
+. §.4-Va -G£rb
5 l\. i.. o.l al-
+ 12Eil/a.l
-~·
4-£f:f/o. -Alf.Ye.~
1tAfl~ 0
0. i
4H1;/o.
0
M:'-la
!.
16
GJC
~~
-195-
TABLE II!-4
ROTATIONAL TP..AJJSFOP 1"ATION t~ATPIX
[A] =
CZ
-cyczcosS+CxsinS
p,]1= J x
C2 + CZ
z
Cy Cz sinS+C x cosS
-jc!+c~sinS '
j c~ + c;
0
r::z:-::2cos 0. ., 0
J1...x,.1.;z
[:l.]2=
0 0
cx L
y - y
See Fip. 3-18.

96
TABLE IV-1
STRfTCTURES 1'NALYZED
·--~-
BOUNDARY LOA.;
'RUCTURE GEOI 'ETRY DECK EDGE ~.~EMBEPS
CONDITION~ P!1
-
I • TYPE
A
Ir~CI!.
I
INCH
--
c
-- INCH
~NO. OF~TH~CKNESS
TYP DECKS ·-INCH
~-- . ---
I d
INCE4 /INCH (Y,
1:10BS/INCH 2
6
\) Bl B2 TYPE PSF
x
I xso xso xlO I 1 0.80 - - 1.0 28.! 0.40 - - I ioJ
--- --·-
'" 6.46 6.46 1. 304 1, I
0. 2.5 - " r..5 0.39 - - II 144
----- ·----·--··
~ 30 30 var- II 1 28-G 0.00047 var- 29.5 0. 30 3 1'0. D. 3"0.D. III 40
ies 0.0149 ies !:;' ~T~1 k ~II Thk
pipe pipe
-------·-
. L ii !) II
2 l< n Tl n ll
"
,. I! n
---···"' ,,
• IV 360 600 100 . IV 1 See 1. 532 1. 0 t:
14 WF 14 NF " 80
69 Fig. 43 68
4.3 .,
- -
• III 24 24 6 I 1 0.25 - f
2.77 0.15 +See Fig.
4.5 :+
IV 40
9
--- ·----~---
fi
r '' ' I
'· I fl
" 1:
' " f'.'" I 11
"
·-·--
t
•• r. I 1: ,.. 7• fl It i; ti ~ Ii ti *28
' VI II E
---- - ------ -·
f II 72 72 14.4 II 2 28G 0.00047 0.04 29. 5 0. 30 1" <fJ 2-1' ¢> v 4(

0.0149 Std Std VI
pipe pipe
- 1; ,, ,. I· ti 11 1;
II ii I
13. 11
" fl
"
80
fABLE IV-1 C0NTINUED
BOUNDARY LOA
STRUCTURE CEOMETRY DECK EDGE FEr,'BERS CONDITION~ HJ'
-
~o. TYPE AM
Ii~Qi I~CH
c NO. OF,THICKNESS
INCH TYPE DECKS INCH
I .
yd
INCH,. I INCH
;
a
, Exi0 6
LBS/INCH 2 v Bl B2 TYPE PSF
11 II 72 72 14. II
·.
1 28G 0.00047 0.06" 29.5 0.3 3 ''C. D. 2-31!0.D. v 40
.. 40 0.0149 ~tlThk Jii"Thk VI
pipe pipe
-- ... r;
..
.
12 . ' 1; '
·•.
11 11 II
24G 0.000753 1; II II
"
ll
VI II
,, 0.0239 .
..
13 - II
.
""
1i
..
,. II
II 2 28G 0.00047 0.05
. ,,
" I!
" II II
~' ' .. 0.0149

.~
- ..
•· !! I! ,, "
13. II ~-
·- ' ·' " II

" !I
0~06·
fl
" " "
-·· ~-
,,
14 ~ : ..
..
240 240 48 '" 16G 0.00192 I! •·: II
,12[40 2-12 v • 17
0.060 ,. [40
15 ;.
12. 12. 5.216 I 1 0.25 - 1.0 a.so 0.39 +NO BEAHS+ IV 14 ..
92 92
x Dimensions are
·-·
in ems
-
+ gm/ cm 2 ..
only
..
1: alf. the structure was loaded
-
• Two lc·ading conditions are analyzed:
1 With 40 psf only 2 40 psf + weight of the edge JTlernbers.
For types of structures see Figs. 4.1 - 4.5.
-197-
-198-
TABLE ·1v-2
DECKS
= Isotropic deck of uniform thickness
I
II = Corrugated sine-form. Fig. 2-2a.
III = Corrugated trapezoidal. Fig. 2-2b.
IV = Cellular trapezoidal deck. Fig. 2-2c.
BOUNDARY CONDITIONS
= Edges
I x = + A and = + B are fixed.
y
II = Edges x = + A and y = + B Knife-edge supported.
III = TF = O for edge members along x = +A and y --+ B.
(i) w =o along x =+ A and y =+ B
(ii) a tension bar connects low corners. (f and b -

Structure I and a and c =Structure IV).
IV = TF = 1 and TH = 1. For all edge members. For the
eccentric locations of edge member with respect to the
deck, see Fig. 4-5.
(i) x = y = o Lines of symmetry.
(ii) x = + A and y = + B Free edges.
(iii) x = y = o rigid support.
V = TF = 0 TH = 1 for all edge members. For the eccentric
connection of edge members see Fig. 4-2. For
(i) x = 0 u = o.
(ii) y = 0 v = o.
(iii) 8 ya = 6xya = 6yh = 6xyh = 6xc • 6xyc s exf =-
_exyf = o.
(iv) x =y • o rigid support."·
-199-:
VI = Boundary conditions i<lentica1 \oli th v except TH = n

v
for all edge mernbe'rs.

VII = TF. = 0 TH ·= 1 for all edge members.·· Edge members are
·· connected con~entrically. Boundary conditions (i),
·{ii.)·.-and .(iii) are same as in the case V.
VIII = ·Boundary and fixity conditions. Same as given in IV.
The line cf (y=o, Fig. 4.2) is the only line of symmetry.

Also.see Fig. 4~20.
-200-
TABLE IV-3
. ' ., . '
cm.1PARISON OF DElJLE·CTIONS BY
J'IETHODS 'a 1 and 1 b 1 ; ; FOR 5-TPUCTURES. 1 and 2
STRUCTtJRE 1 STRUCTURE 2
LOCATION Method 'a' Method 'b' · r.'~ethod 'a 1 Method 'b'
-2
10 ;.. 2 cms 10 ems .10 -3..H~ch · -3.
.10 · ~nch~
. .
s. s·o ·
, ,.
c 1.22'8 1. 235 5.552

2.352 2.384 8 .<S98 8.466
2.574 2.602 9.149 9.262
0 2.531 Z.551 9.196 9.322
6
0.742 0.750 3.271 3.251
2.218 2.248 7.526 7.561
2.613 2.647 9.081 9.184
0 2.531 2.551 9.196 9. 322
NOTE: See Figs. 4-6 and 4-8.

..
-201-
TABLE IV-4
DECK STRESSES IN PSI AT THE CENTER

OF A QUADRANT (Point e in Fig. 4.2)
AXIAL STRESSES BENDING TOTAL STRESSES

STRESSES
STRUCTURE
NO.
EXPERI-,CALCU-
DESCRIPTION MENTAL LATED EXPT I LATED
CALCU~ EXPT CALCU-
LATED
9 28-G
Double Deck 832 848 1460 1870 2292 2718
11 28-G
Single Deck 890 62 6820 10700 7710 10762
12 . . 24-G
Single Deck 69 5600 6200 6269
13 28-G
Double Deck 2780 22 6505 4510 9385 4532
For the Structure numbers, refer to Table IV-1.

-202-
DEFLECTIONS IN INCHES OF INVE!(TED u~rnRELLA

SHELL WITH STIFF EDG~ '~MBERS
', J• • • , ,· EXPERir'ENTAL
~· ., . '
LOC.ATION ANALYTICAL AVEPAGE ~··AXIPtJ!·' lr,fJNnmr~ I

STRUCTUBE 11
0
·a 0~133 0.20 0.26 0.15
'. . ~
'«5b 0.29 0.29 0.,48 O.lR

oc 0.16 0.15 0.15 0.15
t5'
e 0.91 0.70 0.92 0.59
STPUCTURE 12
" 6a 0,. 12 0.11 0. 180 ' 0.04

0
',.' .b
0.26 0.23 0.26 0.32
oC 0.13 0.095 0.150 0.040
oe 0.57 0.42 0.62 0.26
STRUCTURE 13
0
a
0. 18 .. o.is 0.26 0.040
ob 0.31 0.31 0.55 0.15
oc 0.18 0.15 0.26 0.040
oc 0.56 0.54 0.92 0.26
See Fig. 4.2.

-203-
TABLE IV-6
EDGE BEM' STRESSES FOR INVEPTED u~·IBPELLA SHFLL
- . WITH STIFF EDGE ~"TINBERS (q = 40 PSF)
AXIAL BENDING TOTAL (Absolute)

LOCATION EXPT. ANALYTICAL EXPT. ANALYTICAL EXPT. ANALYTICAL
1 STRUCTURE 11 28-G Single Deck a=0.06

a 2440 1070 4850 4481 7290 5551
b 1400 456 3120 3990 4520 4446
c - 580 - 586 1660 1167 2240 1753
d - 900 2320 2496 3396
e 1620 856 3020 3050 4640 3906'
2 STRUCTURE 12 24-G Single Deck a=O. 06

a 1740 1093 4140 4385 5880 5478
b 530 420 3500 3801 4030 4221
c -1060 - 609 540 955 1600 1564
d -1070 -1075 1100 2165 2170 3240
e 1100 874 2300 2668 3400 3542
3 STRUCTURE 13 28-G Doul) le Deck a=0.05

a 2250 435 4110 4228 6360 4663
b 930 435 5260 4228 6190 4663
c .- 900 - 518 2040 1650 2940 2168
d - 560 - 734 2020 2106 2580 2840
e 1110 357 2320 2561 3430 2918
All stresses are in psi
-
d-i----·-'-e
1--~-----
c ·• ~-------1
TABLE V-1
SELECTED EXAMPLES SHOWING THE EFFECTS OF CHANGE OF STRUCTURAL PARAMETERS
DEFLECTIONS EDGE MEMBER STRESSES PSI

VARIABLE IN INCHES MAX. AXIAL
•structure
Type Value oa oc ab be
Rise 'C' 14.40" 0.38 1.57 0.38 0.69 -5329 -5329 2747 2747
9 • 10 13.so" o.41 1.69 o.41 o.75 -5518 -ss18 21s2 2752
Shear 0.05 0.18 0.31 0.18 0.56 - 856 - 856 4~; 43~
13 • 13a Rigidity -11:110·- -o,,,...... rs----0-.-2. .s...--o-.
. . . 1. ....s--o-.,. .s---
. --.s. .s,.......1--.
. . a-s
. .1----.4
.. ....s-1--..4. .s. . .1-
..
£ actor a
11 • 12
Thick- 0.0149 0.13 0.29 0.16 0.91 -1129 - 802 456 1070
ness
0.0239 0.116 0.26 0.13 0.57 -1150 - 870 420 1093
No.
Single
11 a 13a of Deck 0.13 0.29 0.16 0.91 -1129 - 802 456 1070
Double
Deck Deck 0.15 0.28 0.15 0.48 - 881 - 881 481 481
EDGE MEMBER STRESSES PSI DECK
MAX. BENDING BENDING STRESS Shear force at
*Structure --~--------------~----~------------------------~---------
PS I Center for Nxy
oa oc ab be lbs/inch
18,357 18,357 17,144 17,144 1,870 52.75
9 8 10 19,077 19,077 18,179 18,179 2,130 54.96
•
6,648 6,648 4,228 ~,228 4,510 50.0
13 6 13a
5,225 4,634 3,990 4,480 10,700 36.96
__1_2___________4_._7_2_3~___3_._9_7_4____3......_8_01________4_,_3_8_5____~--6~2_0_0___________3_7_.7_5___~
__1_1__&
5,225 4,634 3,990 4,480 10,700 36.96
11 & 13a 5,899 5,899 1.095 4,095 3,780 Sl.28
For locations of deflections and stresses see Fig. 4-2.
* For the type of structure according to number see Table IV-1.
204
TABLE V-2
THE COPPJ\RISON OF AXIAL STRESSES IN THE EDGE r·m1BER AS GIVEN
BY f''E!'BPPNE THEORY AND BY ANALYSIS
~~EMBRA'l\IE
ANALYTICAL PSI % OF MEMBRANE THEORY
THEfi'P.Y
STRUCTURE STRESS
NO. a PSI oa oc ab be oa oc ab be
SINGLE DECK
.
5 1. 0 *19660 10700 9070 ~440 11040 54.4 48.7 50.6 56.1
----·.. ~·--
,_ ____
18640
---- ·-
11 0.06 1570 1129 802 456 1070 71. 8 51. 5 29. 68.2
---·- ------ ...
----------~ ~ -- ..
12 0.06 1570 1150 870 420 109 3 73.2 55.5 26.7 69.6
-- ·-
DOPBLE DECKS
- ·-
9 0.04 7100 5329 5329 2747 2747 75.Z 75.Z 38.8 38.8
13 0.05
---54.4 27.7
1570 856 856 435 435 54.40 27.7
~------ ---~----
13a 0.06 1570 881 881 481 481 56.2 56.2 31. 30 31. 3
-·
14 0. 06' 3420 1673 1673 1575 1575 49.0 49.0 46.0 46.0
--~ ....
* Viember sizes are different, see Table !V-1
-205-
-206-
TABLE VI-1
STIFFENED PLATE :BUCKLING PROBLEMS
DitvtENSIONS
LOADING a-inch b- inch t-inch y ·6 K % 'Error
-::
::
J
b 108 108 5/8 - - 4.0

..
..
....
-
...
.:s /? -
'
2
108 l08
... ..
5/8 5· 0.10 :Li. :to
·-·'
....-
... 10-8" 108 5/8 - - ,.
. 4. 0
b <0.2
1 2 108 108 5/8 5 0.10 20 .·5
a/2
..
..
j 108 54 5/8 - - 4.0
~
b
2
~ 108 54 5/8 5 0.10 7.69
..
a/2
'---a~
t i 16 16 0.10 - - 9.34 4.65

1 - - + ___.,
~b
... a
1
- 1 24-G
70.5 ·. 70. 5 - - 0'•·670 . 7.46
~b
l --- - -
0.0239
TABLE VII·l Fropert1es ~~d ~i~enslo~s ~: flat Shear Tests
Teat l>h»nstons Steel Edge No. of Seam Deel< ... id~e G'
-
No.
l,
(ft)
6x6
Dockins
26G s.c.•
He:::bors
6''xl~"x.1046"
Pa.~els
l
Connections Cor:.~cctio:is
screws @ 8" scret.Ts @ every

c
.163
(l'.::.1,,l '.I 'J
33100
1 layor channels valloy and @ 4"
2, 6 x 6 26G S.C. 6"x3/4'tX.'t046" 3 screws @ 8" scret:s @ Gvcry .159 32300

1 layer channels valley and@ 4"
3;... 6 x 6 26G s.c. 6"x3/4"x.1046 11 3 screws @ 811 screws @ every .056 11400 '
N
1 layer channels 3rd valley and @8 11 •• ....
c:>
•
4 6x6 2SG s.c. 6 1'xl\ 11x.1046" 4 screws@ 8" screws @ every .011 12000
1 layer channels 3rd valley and @8°
5 6 x 6 28G s.c. 6 1'xl\ 1'x.1046" 3 screws screws @ every .077 13000

1 layer channels @ 2-2/3 11 3rd valley and @8 11
6 6 x 6 28G s.c. 6 11x1%''x.1046" 3 a crews screws @ every .066 11200

1 layer channels @ 2-2/l" 3rd valley and @8"
1 6x6 24G s.c. 6"xl\''x.1046" 3 screws. screws @ every .078 21100

1 layer channels @ 2-2/l" 3rd valley and @8"
8 5 x' • 28G s.c. 6"xl\''x.1046" 3 screws screws @ evary .068 11500

1 lnyer channols (I 2-2/3 11 3rd valley and @8"
* Standard Corrugated
TABLE VI I -1 Cont im1ed
Test Dincnsions Steel ::dga ~o. o! Sca.':1 Jeck - E:dge G'

:.;o. (ft) Decking :·:c~bors Panels Cor..neet:ic:'ls Conncctio:;s ,.,... (lb/i:i)
9 5x 5 28G s.c. 6':x13i 11x.1046" 3 screws screws @ every .068 11500

1 lnyar ch;:m. . . ~ls @ 2-2/3 11 3rd vnl ley and @8 11
10 6 x 6 26G S.C. 6nx1~ 11 x.1046 11 3 screws screws @ every .098 39800

2 layers ch~i.nnels @ 8" valley and @ 4 11
11 6 x 6 26G s.c. 6 "x3 /4 11x.1046" 3 screws scrcYs @ every .114 46300

2 layers chnn.i.-:.e ls @ 8" valley and @ 4"
12 6 x 6 2BG s.c. 6"x1.!z 11x.1046 11 4 screws screws @ every .056 18900

2 layers channels @ so 3rd valley and @S" I
N
0
13 6 x 6 28G s .c. 6 11xl!i•::<..1046 11 3 screws screws @ every . .040 13500 co
2 layers channels @ 2-2/3" 3rd valley and @8 11 '
J4 6 x 6 28G s .c. l '' std weight 3 screws screws @ every .045 15200
I
2 layers pipe @ 2-2/3 11 3rd valley and @8"
15 5 x 5 28G s.c. 6''xll:i"x.1046" 3 screws screws @ every .050 15400

2 layers channe~s @ 2-2/3 11 3rd valley and @8 11
16 1 x l
2 layers. brass tubes
·1;-
2 mil ce>rr. 3/16 lfonx:. 014"t 1

- soldered @ every
valley and cont.
.030 1360
"1''4,
-209·
TABLE Vll·2 t:xporimont;il Results for Sc'ldclle Shaped Hypc'lrs

Supportod All Around (q ~ 40 psf)
(c'l) One Layer of Decking
Test wmclx in St1·ess at Canter in Strong Dir. Axic'll Force in

No. inches Bending (psi) Axi.al (psi) Tie Bar (lbs)
811 1.1'. 20800 ·640 1820 (5660)"'*
812 1,14 19100 ·1700 1630
Sll o.ao 11.100: .
512 o. 77 15900 -
1260
311 o.33 2500
312 o.32 3560 810 1710 (2120)
(b) Tuo J,Clyers of Deck l ng
Test ln
Wlil.lX Stress at Center ln Strom~ Dir. Axial Force
!\o. inches Bending (psl) Axial (psi) in Tie J3ar
___ ~· Lc'lyer Top Lay;er Bot. !.ayer Top Layer _.;..(_lb_s.>_ _
321 0.86 12400 10000 -290 1810 1560
822 o.6S 13000 13000 ·820 740 1310
821I o.65 10300 7600 ·680 1750 1250
521 o.s9 11300 7700 1660 650

522 0.53 11000 7800 nao 360
522! o.48 10000 5200 80 2060
321 o.24 4000 4500 570 920 1720

322 o.32 5560 2310 ·560 1220 1570
3221 0. 21 3990 640 -:uo 1670 1570
* ll.'.-1.scd on one str:lin. gago at extreme fiber, axlnl stress

assuned to be zero.
-ltk (
Villuc ln ) ls calculated from membrane theory.
-210-
TABLE Vll-3 !·:xp•!rir.1cnlal Rc.'.1ults fot· S.'ld<llc Shnp<!d Hyp.'lrs

~upportcd A 11 Aro1111<l (8 11 x 12" area londcd)
(il) One Layer of Deckinr, 1 Lond ""' 100 lb.
w,. 1 ;i~~
in St_LcSS_!tt_ Center in Stron& Dir. Axial Force in
i nc:lCs !knclin.~ (psi) Axial (psi) Tie Bar (1 bs)
:a1c 0.(,J ----·--
17000 4300 150
fll2C 0.67 17400 2300 130
311C o.39 1990

3J.2C o.37 2050 0
(b) Two Layers of Dcc!<ing, Loa<l.~ 200 lb.
'fest "'1:tn:< in Stress at Center in Stron~ Dir. Axial Force

No. inches ;?,end.in.~~ (psi) Axi;il (psi) in Tie Bar
- · - - Dot •__ £-ayer Top_!._~_yer Bot. Layer Top J.ayer (lbs)
i..i21C a.so l)l.OO 20000 -160 -1380 470
822C 0.42 11300 17000 .. 400 210 530
:->2 lIC o.39 13600 10900 -1650 -1300 390
521C o.t.4 14000 16500 1060. -1090

522C o.41 15300 11300 460 360
522IC o.32 13000 .9300 860 ·350
321C 0.24 15700 10100 1220 750 140

322C o.2a 10200 9900 10 -40 180
322IC 0.17 6800 7700 1060 -340 110

-211-
0
-212-
'A
'/
/
)
·;
(I I
, /
1 1/ ,_..._
'rj Vl
y '-' s::
0
·~
,/' +.)
m
'{ I-<
~
.-i' O{J
1 I .,.;
/1 ! ··I 4-4
s::
/,.I u
0
/'. I-<
re
p..
I >-
::c
µ
s::
11'
$-<
11'
'+-!
'+-!
·~
c.
N
i
rl
00
•rl
µ...
-213-
0 x 0 x E D
x' x
s
.& ..
-
s
) El,.Exy
Dl,Dxy
.... ;-
y y ,Ey, Dy
Fig. 2-1 Typical Orthotropic Deck
a) Sinusoidal
Corrugated
b) Trapezoidal
I- l ·-~
f Q j \ j \ j \f c) Cellular
~--~--·-~
d) Stiffened
Panel
I I
[ ]
I
'
e) Box-type
I 11 11 I
l---!--.,~
I
'
Fig. 2-Z Section S-S Showing Different Types of Decks

-214-
Effective Width
Varies
z
Fig. 2-3 Effective Cross-Sectional Area
of a Hat for Axial Force
e/2 e/2
Effective Width
1
-- t
I• .. l
Fig. 2-4 Effective Width of Compression
Flange in Bending
-215-
Principal Directions
Fig. 2-S An Arbitrarily Oriented Orthotropic Deck
r- Connectors U per Deck

...,__ _ _ Lower
Deck
Fig. 2-6 Edge Member and Decks Coanection

0.10
'\
\ Present Study
\
\ /:., Ref. 19
0.08
,...
0
.....
u
aS
u.. Flat
Shear
>.. Test
..... I
•r-1
"O
•r-1
DC,
G
A
....C\
N
..... &. I
c:::
J-4
0.04
m
C)
..c:
[/)
.. s l -0 II
·~
"'
0.02
c in inches
:~c
4 8 12 16 20 24 28
0
Fig. 2-7 Shear Rigidity Factor (X t Vs Hypar Curvature

f
(One Layer of Decking)

0.10
.\
'\ Present Study
0.08 P_ef. 19
....
N
....
I
S'-0"
.. r-
..."
0.02 S'-0"
C in inches
4 8 12 16 20 24
0
Fig. 2-8 Shear Rigidity Factor 'a.' Vs Hypar Curvature
(Double Layer of Decking)
-218-
e ex
1
xl
"" 2 n
ul Uz
wl exyl
lw 2 8xy2
e e
~
X3
ey4 C
U4
ey 3c/
3
"'- U3
W4
1e
xy4
w3 8 xy3
V3
Fig. 3-1 Nodal Displacements
o~_Y_L
M
y
M
xyc
0
x x
1
-
Mc )Mx
x
, Nxy
J
-
U--'M
Nxy
! y
M
y
! '
xy
Fig. 3-2 Jn-plane Forces Fig. 3-3 Moments

-219-
l ---
!s-s
a) a • b • 80 inches
1
I
I t = 0.05 inch
E • 29.Sx10 6 lbs/inch 2
h/2 s-s \) • 0.30
IB q = 1 psi
s-s No. of Elements = 64
------+--- x (Full Plate)
0
0.00406qa ..
h/2 B = D
i.·- - - '-------+---.
s-s
...
-~L~- J
'
I
I- a/2 -·----..-- ...
1
--
1
y
b) a = 96 inches
b = 144 inches
t = s.o inch
E • 3xl0 1 lbs/ ·
b/2 F.E. inch 2
s-s 'V = 0 .. 30.
q = 200 lbs/;ft2-
~o. of Elements
• 18 (Half
Plate)
. ..
b/2 ~ '
0.0064a ..
s-s oB • D • •
a 2 a 2
l y
I
s-s
I c) a • b = 70.50 inches
t • 0.0149 inch
s--s 28G Std. Corrugated
I
lo
b/2 I> r-, •>
- D.e.ck ·
E = Z9 .sx10• lbs/ inch 2
·- ~I A v • O. llJ
- - ~
-x q = 0 • 3·0.. psi
I'(o .. ·of El~ments
b/2 •Varies _(Quadrant)
; ; 5B s !k ~t~-
a/2 a/2
Fig. 3-4 Plate Bending Problems

Fig. 3-5 Deflection Profile Across Corrugations
0
(see Fig. 3-4C) for Uniformly Loaded
simply Supported 28G Standard
Corrugated Steel Deck
q = 0.30 psi
2 !J)
(l)
.c:
u
s:::
......
....i::: I
4 s::: N
0
....
.µ
N
0
I
u
t>
.--4
'M
t)
i:l Deflection of
6 Beam Strip
B
= 6.96"
A
8 Present Study
A Series Solution
0 1/6 1/3 1/2 2/3 5/6 1

I
-221-
p
e
0 C.G S.C.
Y- IJt- - __,,__,_ __
z z
(a) (b) (c)
Fig. 3-6 Typical Cross-Sections of Beams
- - - - - - - - - - s. c.
y,v
-1
x
z ,w
Fig. 3-7 An Arbitrary Cross-Section of a Beam
Fig. 3 .. s Nodal Bending Displacements

of a Typical Beam Element
-222-
z 'w
z
y
8
z
y,v
----b-----···--·- ·---------~ x ,u
ax
Fig. 3-9 Typical Eccentric Edge Member
-Plate e
. .__.
C.G., C.G. s.c.
s.c.
~Bracket
(a) (b)
Fig. 3-10 Eccentric Connections

,;-- 0. 2SK/"
{a) Case 1 {b) Case Ila and IIb
Fig. 3-11 Loading on Eccentric Edge Members
•o 4
.....s::
- 2
0
No. of Elements
Fig. 3-12 Convergence Characteristics for Vertical Deflection

·~Q' for Case I (Fig. 3-lla)
-224-
9.0 ....
~ --· ~~ ~~8 . 6 4,-,- w-,~~-=~-=:•r-e_e_T_.:.._r_s_i_o_n_(_I_I_a_)__c
..c:
u
~
=
•l"'i
"--- ~Warping Restrained (Ilb)
'<I
6.0
--~ -5.:~ -- . Q
4.0 2 4 6
No. of Elements
Fig. 3- l 3a Effect of Restrained Warping on
Vertical Deflection 'oqt
8.0
(IIb)
6.0
(!Ia)
~
0
....
....
1.1-l
4.0
2.0
6
0
No. of Elements
Fig. 3.13b Convergence Characteristics for
t 0
Qt and 'a Q'
x,u
al y,v
I
I
Column
_j__
Fig. 3.14a A Cantilever Column
Fig. 3.14b Idealized Spring System

·226-
A =A =A =A =A
1 2 3 4
B =B =B =B =B
1 2 3 4
c1=c 2=C 3=c 4=C
x- 1 =x- 2=x 3=x- 4=c
- - - -
Y1=y2=y3=Y4=o
y
(a)
A A
A =A =A =A =A
1 2 3 4
B 1 =B =B =B =B
2 3 4
c1=C 3=C
C2=C 4=-C
x1=x 4 =A
.__,._._ _-+-_ _ z_ __, _...,.. x ' -2 -
x =x =-A
3
/R(A,B,O)
y• . Y1=Yz=B
y
y =y =-B
(b) 3 4
Fig . .3,.15 Geometrical Definition of a Hypar Surface

-227-
pt Q'
s• R'
P'Q' = R'S' = PQZRS

P'S' = Q'R' = PSZQR
y
(a) (b)
Fig. 3-16 Element Size
Pr--_ _ _ _Q"'t----- x 0
ct
-
x
-----..... x'
x'
s
I -
y
!,Z'
I (a)
y'
(b)
Fig. 3-17 Co-ordinate Transformation

-228-
y
-
x
Fig. 3-18 Co-ordinate Transformation for

a Beam Element
y \
z z
Fig. 3-19 Boundary Conditions in Global Co-ordinates

·229·
-- Deck - Deck
1
--~Edge
Member
(a) Moment-Free (b) Sliding-Connection
Connection
rig. 3-ZO Deck and Edge Member Connections
Shear ;Deck
-~ I p
'. \
I
u"' ~·I ·f-+llo.'-....--...M'--i
1··~
I <{,-----.....
'J
(a) (b)
Fig. 3-21 Twisting of an Eccentric Edge Member for Tp=O

-Z30-
1
llnlf Band Width
1_~ _ r
' !'I
><
"
.... ""
"' ''
"' " "'
"' '
'
' ''
'
'
x ''
'
(a) Stiffness Matrix [K] (b) Vertically (c) Stiffness Matrix

Stored Half- with Sparse entries
Band
Fig. 3-22 Solutions of Equations
0 ~ 1
x
-- •· ··---. - ..... y
z x z
(a) 1
M-p
p
xl -[ __ '" --·-- .-_]-)---?-p x 1
p x 2 ···-{--l.___ _.I- ---p x 2 l:M
0
= 0
MoQ,
(b) MoP + (Pxz-Pxl)Zs
oQ = 0 - M-
M-oP - MoQ = CPxl-Pxz) Zs

~xz
(c) Mc;q 'I Mc;q
Fig. 3-23 Computation of Forces in Eccentric Edge Members

-231-
A
--A ---1
h
I ·-
I
s/
b
Fig. 4-1 Structure Type I
g
1
<
Fig. 4-2 Structure Type II

-232~
B/4 B/4 B/2
y - -·
r; T---·-·-r:i ·-·· . ,b
Bl
Deck
.... > . 07 7" r-.
8. 871"
0
A.onfl}.16"
c [1 Rz
2
LYi·il
Typ. Deck
Section
Fig. 4-3 Structure Type IV

Note: 1 & 2 Refer to the Locations of Rosettes
1 2 3 4 s 6 7 8
h
I I I I I I I I a
B -
-
c ·-
,__. . -- -- -~~ -
-
e
E - ·-
F ---
G ·-
II -
c 0
Fig. 4-4 Quadrant of a Hypar

-233-
J! .. _ _ _ _ _ _ _ 1...._ 2 _....
h___
tl"(TYP
- ~Varies
1. . 1
i c
, -1 / 2 ,, 0 Strucs.
f
i '6' 6 '8'
B 1/4" Thk
Shell
<l A a b
A
Plan
Strucs.
'6' & '8'
2-1/2" square
column
Struc. '7 1
Section P-P 2.
Fig. 4-5 Structure Type III

-234-
lo a
1. 0 .
I Symmetry
~/
2.0
**2.55
3.0
Scale
1" = 10- 2 cm
I .~-Symmetry
I // 1.0
r 2.0
I
1+2.46
3.0
* Method ta•
*Ill Method 'b'
+ Ref. 17
(Structure '1')
0.035
E
v
....c:
0
'O
c: Scale
....0
~
1" = ().0025 cm
u •
Q)
,..... 0.030 N
'+-!
C>
Q
"""'•
,...
QS
...e
0
z
0.025
2x2 4x4 6x6 8x8
Grid Size
'~
Fig. 4-7 Convergeace Characteristics for Deflection 0 ' (Structure '1')
-2l6-
I r Symmetry
r-·/
I·9.20*
9.32** 10
Scale
l"=Sxl0- 3 "
Jo b
T·----.--------_,,_-_,..,
I Symmetry
l ..
oxro 3
r
. 9. 18+
9.28+ 10
* Method 'a'
**
+
Method 'b'
Ref. 20
(St rue tu re '2.')
-237-
d e
z.41 Scale
l"•l"
f g
-·lilr·- Experimental
Analysis

(Structure '5 ')
-238-
9.32
7.50
7.50
Scale
l ":r.7. SO ksi
1/2 The Membrane
Stress~..-..-___.~-~ 9.32
7.50
9.83
7.50
Fig. 4-10 Axial .Stresses in Edge Members

{Struc~ure 'S') ·
100
Rosett-e ~o. 1
-Max~ Reading
.
't-4
Gage BrO"ke When
Local Buckling
!ll
I=!. 80 of Plate
....= 7150 Occurred
Rosette
No.
•
NI
e,,e
~
•
e Anaj.ysis
-·--- Experimental
2000 4000 6000 8000 10,000 12,000 14,000

0
Shear Stress in psi Along Plate to Plate Welding
Fig. 4-11 Load vs Shear Stress (Structure 'S')
-240-
I A I B c I D I E I F I G I H
1- h -- - ·- .. __________,. _____ -- - - - c
r- - ----- -- ------:i
2 --
L - =::J
~------:-~
3 . ---
4---
L_ -- .:sJ Scale
1···=1000
v-------:--~ lbs/"
5-
~- -------. --_---1
6-
c
~-----------]
.· :=:J
7- L:--------~~---------- . ==
_____ _j ______ _
r Membrane Shear
= 653 lbs/inch
8- a o
(Structure '5')
-241·
·:'
0.016
0.02 0.018
a b
o.o
Scale
1"•0.02"
0 b
0.02
- ·-A-- Experimental
-·-e·- Method 'a'
~1ethod 'b'
Fig. 4~13 Deflection Profiles

.(Structure '6')
-242-
324
300
"' '
''
''
Vi
.D
r-i
200 /Membrane Theory
.....s::
Q)
u
"' ' 'l
""
0
1-1..
"'
100
+ "'
'''
a b
Tension Member
Scale l" • 100 lbs
324.0
300
''
.D
!/)
" "" ,../ Membrane
Theory
- · - 4 · - Experimental
Analysis
r-i 2 00 e
.....s::
GJ
u
""
''
0
µ .. . -A......
100
~~
0
Compression Member a

(Structure '6 ')
--6- Experimental
-
0
c
Analysis
80 /.
Axial Stress ~
4
.
t+-1
aJ
#
II) 60
Po
#
..""'
s:::
·~ I
N
~
aS
0 I
14
40 M•mbrane
Theory
Stress in psi at Point 'e'
Fig. 4-15 Load vs Stress at Point 'e' (Strucutre '6')

-244-
I A I B I c I D I E I F I G IH I
[ -- - -- ---
1-~ b
~ c
•
,-~
~---·-
::sJ
2--
p=:
3--·- ._
_____________
--.---~
____:J
_____
4_r:--- ----~:']
5-r:
Scale
:J 1"=25 lbs/"
6-L ---:i
7 -- _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___.
Membrane Shear
= 13.48 lbs/" - ,
--· ----- - - - --~-- -
a o
(Structure '6')
-245-
0 a
-i!r. ---
. -- ---· --·-.o.
.
0.009
015
0.025
a b
o.oso
0 0
Scale
l"=0.025"
0. 02 5
o.oso
-·-A-·-· Experimental
e Analysis
Fig. 4 .. 17 Deflection Profiles
(Structure 1 7'1
-246-
324
300
t/j
..0
.......
zoo
.....s:!
160
Q)
u
J..<
0
µ..
100
+
a Tension Member b
Scale
1"=100 lbs.
300
Ill
..0
......
~
zoo
•..-!
(I,)
u
'""
0
µ...
100
0
Compression Member a

(Structure 1 7 1 )
-247·
IA ja le ID IE IF IG ltt I
-------]
=::: ~
--- -- -
_k:::
-
v:--::
z-~
1 -~
- --j
-
3-~-·---~
4-c- -::sJ
s-r::: ---- ::1 Scale
ltt•2S
lbs/V
6_p=:---. ------1
7_
-- =- -=-----]-
k--==-====
MembTane
r3·
- ___ ___ _/
Shear
48lb-/" .
_.,......_
a o
Fig. 4·19 Shear Force Nxy lbs/inch
(Structure • 7 •)
-248-
0.10
d +
"""" ........... -~
.
0.068
-~0.098
0.10
0
0.10
0.10
0
--itr·- Experimental
Analysis
Fig. 4-20 Deflection ~rofil•s

(Structure '8')
so '
' .,,,,,..,,- ~
~ .. ~"
'
..__ob
~od __ . ~·.-::::
---- ---·.~· ~
.;,_...-·
./· /.>\.
40 - : _...,,....... - · ..ei- · - Experimental
,.,,... . '
1..- 0d Analysis
I .
. ./ ob
I
.I
I " I
....
N·
fO
'
I
I
.I
0.10 0.20 0.30 0.40 0.50 0.60 0.70
Deflections in inches
Fig. 4-21 Load vs Deflections (Structure 1
81)
-250-
Grid Size
4x 4 6x6 8x8
1. 60
l/l
CIJ 1. 574
..c:
u
i::
·~
i:::: 1. so
·~
..0
'<'
i::
0
·~
+-'
u
(!)
rl
l.H l.'!>72
Cl)
Cl
1. 30
Fig. 4-22 Convergence Characteristics for Corner

Deflection '6b' (Structure '9')
-251-
0 • 3B <1--.__...._
0. 51 Scale
0.56 l "•O. 50 11
1. 0
1. so 1. 57
-·A-· - Experimental
• Method 'b'
o.s - -a- -- Method 'a'
1. 0
1. s
Fig. 4~23 Deflection Profiles
(Struc. •9 '}
-252-
6
o.s
Scale
1"=0.50"
1. 0
1. 5
- ·-A·- Experimental
Structure '9'
0. s ---&--- Structure '10'
1. 0
]. so
Fig. 4·-24 Deflection Profiles
(Strucs. '9' & '10)
6
-253-
3.62 Axial
. ·--A- S.tress
s.~3 1t'•5 ks i
5.52
Bending
Stres.s
1 u•lO ksi
19.0
0
0
Total
Stress
10.0 1"==10 ksi
-·-A·-·- Experimental
2 O' Structure '9'

- - ...g.. - - - Structure '10'
a
Fig. 4-25 Stresses in Compression Member
2.1slt;---~~-e----'il____..~-..---::~,,_.,.::::~;:::r-
2.1s
s.o
s. 2 -- ---· --- Axial Stress
1"•5 ksi
a
a--------..------r-----..,...----r-b
Bending
10.0 Stress
1"==10 ksi
17.3 17. 14
18.20
(1
Total Stress
10.0 1"111:10 ksi
-·-Ii.· - Experimental
0
Structure '9'
--.;a.--. Structure '10'
(J
Fig. 4-26 Stresses in Tension Member

-2SS-
I A I B Ic I D I E I F I G IH I
1~-------~~~----~~~
b ~------------~c
z-.~
3-L~ ~
4 L ::: ::::.; ~ 9 Scale
1"-10
lbs-inch/
inch
s C_75--- ~
6-·~~
7~-~
8----~~~------~~~~~..;;;a..--~-.
a o
Fig. 4-27 Bending Moments My (Structure '9')

-256-
I A I B Ic ID IE I F I G IH I
,~-------------1
1- [ - l
b c
r=- - -:...:;::----------=:..::-::...:- - -1
z_L. ~
3_ p:___~~- -w -j Scale
1"=100 lbs/"
4-r::= ,-:~-
[-- -----
s -~
[--- ----™
6_ t::::=
----]
7-~ :J
8_ [
Membrane Shear= 49.50 lbs/"
______ ------ ---j
a o

(Structure '9')
-257-
Deck
(a)
M=P II e
(b)
l_ __-;;::__.,,,,ar- - - - --
t.. f - ·-- ··--
- ·'-·-==--
-.
0
--- ---- ..__ ..........
2
M9.
2E!
(c)
Fig. 4-29 Effect of Eccentric Transfer of Force at

a Junction of Tension and Compression
Edge Members
-258-
0 0.38
0.50
Scale
l"=0.50 11
1. 0
1. 50
0 b
-·-A-· - Experimental
0.50 B.C. V
---e--- B. C. VI
1. 0
1. so Fig. 4-30 Comparison of Deflectio~

Profiles for Boundary ·\
Condition, With (V) and
Without (,VI.) in:.. plane fixitt'
(Structure tgt)
-259-
a b
10.0
B~ding Stress
13.4 1.,•10 ksi
Bending Stress
a b
Shear Loads
1"•25 lbs
25.0 -·~· B.C. VI
/
/ B.C. v
50.0
75.0
Shear Loads
~ig. 4-31 Effect of in-plane Fixity on Bending Stress and
Vertical Shearing Loads on the Tension Member
(Structure '9')
-260-
()
--.--- r·-
-----a---e--a--- -a 0 . 0 4 3
---:r--
8
0.133
0.20
6
0 c
. I I•
_.,_ ..........,__-fJ--w..._...a0.053
0.20
6 Scale
l"=0.20"
0.29
6
0 b
Scale
o.so
1. 0
6
--- _--
_.....__
Fig. 4-32 D flection Profiles

(;tructure 'llt)
-261-
0.29
0.29
o.so
1. 0 0.91 Scale
l"=0.50"
Structure 1
11'
0
.--r ---...----r----r----.--.......--....---...,·- b
-...........
~
~ __ .-40.23
· '-.~o 42 0.26
• •·-&-·-'A
A-· _r,;;..__..·
0.50
() Structure '12'
0. 31
0.31
0.50
0.56 - · - A · - Experimental
e Analysis
Structure '1:5'

(Structures 1 11','12','13')
-262-
667
+
c b
Scale
a 1"=667 psi
667
0 c
0 Max. Membrane Stress
= 1570 psi
667 Analysis
0 a

(Structure '11")
-263-
1s.o~~~-_+__.._ _
a
Q_ _• __ ._I
• ___
150.0
c b
+
o----- I I I
c
30 0. 0 .
Scale
l"slSO lbs
150.0
+
0
a
Fig. 4-35 Vertical Shear Force Diagrams

(Structure '11')
-Z64-
IA IB Ic ID IE IF IG IH I
[------ - - -------]
l-~
h
.. ~
[-- - ---- ····-- ----]
c
2-v::
------- ·- ---··
~
I ---------------J
3-v
[---
~
----------------
4-v ·- ---····--------------1
~
Scale
1 11 = 100
lbs
5-~ ~
6 - ~- - --- - - - -- --s;J
7-v----------SJ
c---~----
8-~ .
------J
. _-Membrane Shear = 48 lbs/"
:'>J
.,
C· 0

(Structure '11')
-265-
·-.A.-·-..a...... .......... 0.029

.........0 .046
o.os
a--~~-----------------...-.-----------------------.--~
o.os
~.
~ 0.092"
0.10 0.090
0.123
0 .... b
o.os
--4- Method I a'
Method I bf
0.10
·-- --- -- Ref. 21
'* Ref. 19
I 0.123
6
j 0.144

(Structure '15')
-266-
Section y = -zb
10-0
Scale
1 11 = 10
lbs/"
20-0
(a) Shear Force Nxy lbs/inch

. . . . . -.-\.41
/ 2. OS ,
\
Scale
11! = 1.0
Section lb· -incftl'
a
x = 2
+
e Method 'a'
-·-·-Ref. 21
(b) Bending Moment My lbs.•inch/inch

Fig. 4-38 Shear Force and Moments
(Structure '15')
for c = o
" = 3.38'' Single Deck
0
1. 67
~
0
~ 1.69 Double Deck
1. 33
Single Deck
Double Deck
1. 0
N
•
0\
.....
•
0.67
AB
2
20 30 40 50 60 79
2 (Structures 3 t and '4')

Fig. 5-1 Central Deflection '6 0 t vs AB/C t
-268-
~1
0 a
----·r-= 4 .. .. ---..:.._
- ,
·-
• - - • - - ·
---........._.._, a
(a)
Deck
-·
--- ---
-. -
.4fo_ __....,......_== - -
. -·--
. -·
- ·-
-··- . \a
--
0
(b)
Fig. 5-2 Effect of Eccentric Connections

Between Deck and Edge Member
-269-
a
............
-..
--~ 'A.....
'"'lL.
a.so "'It. 0. 4 8
0.12
Scale
1"•0.50"
0.4 ·er-. -6--._.._
·-&. ._...,
' -. ·--....._ ... o. 72
o. s1r-----..___
1. 20
0..--~~~-.-~~~~...,.--~~~.....,~~~~-,----b
e With Edge Member wt
-·&-·- Without Ed1e Member wt

a.so
.f
I. 0
I
I
I
I. 50
Fig. S-3 Effect of Edge Member Weight

on Deflections {Structure '14')
-270-
I
I 0.48
Ir Sym.
0.81
~/ Scale
1. 25 l"=O. SO"
h.'33 ._.._ ---

Deflection Along Tension Diagonal
a b
Scale
3.27
l"=S ksi
5.06
Tension Member Bending Stresses

0 a
/
s.o / --·-&:·--Without Edge Member wt
6.7 e With Edge Member wt
Compression Member Bending Stresses

Fig. S-4 Effect of Edge Member Weight on
Deflection and Bending Stresses
-271-
Linear
Elastic
Analysis
cSp II
Value of De-
formation at
the end of p I
First Step
Fig. 6-1 Load Incrementation ~ethod

o.so /
/
/ I \ Buckling
/ I \ Load
/ I I
0.40
q.J.L~gm/ c;:m
N
·-·-· ·-· -- g h j
e
u
...........
ee.o
.:.i:
0.30
Reissner7 3
i::::
......
1 I
"Cj f c N
C'd 0
.....,
0 N
....:I I
d a b
Point 'l' Point '2' Point '3' Key Sketch
1. 0 2.0 1.0 1. 0
I tS in cm.
Fig. 6-2 Buckling of an Isotropic Hypar (Structure 'l' Table IV-1)

-273-
g b
0.20 kgm/cm 2
I
I
/
1. 0
/
/
\ I
\ , 1 0. so kgll/cm 2 Scale
2.0 \ I 1''•1 cm
v
Compression Diagonal 'gb'
2.0 Buckled
Tension Diagonal 'dJ'
Fig. 6-3 Buckling of Structure 'l' (Table IV-1)

-274-
0 . --
2.0
4.0
Scale
Compression Diagonal 'ob' 1 11 =2 cm
a ,. c
I
I
/
/
/
/
./
2.0
- - ---Buckled
Scale
Tension Diagonal 'ac' l"=l cm
Fig. 6-4 Buckling of an Isotropic Umbrella Shell

(Edge Member Sizes 6cmx3cm)
-275-
2
kgm/cm
1. 0
2. 0
Compression Diagonal 'obt Scale
a c
0.20 Scale
1 ~'=I. Sent
a.so 0. 30

Fi~. 6-4 Buckling of an Isotropic Umbrella Shell
(Edge Member Areas • 108 ca 2 1
243
/
I I \ Buckling
I I I Load
I I
20 2 I I
200
b
4a
~
. 150
Cfl
p.
s::
•!""!
1 I
N
...a
"tJ
Cd
0 100
2 °'
I
....:I
3
-(do-
c
0
Point '1 t Point t 2' Point ' 3 t Ker Sketch
2.0
o in inches
Fig. 6-6 Buckling of a 28G Double Layer All-Supported Hypar (Structure '13')
-277-
\
.\ 23 psf
\
. 0\\ /"-·- 46 psf_ _ _ _ _ . /
\ I
/ ---
1. 0
'" Compression Diagonal 'ob'
' '' 23 psf
0. 50
' ' 46 psf
1. 0
' \
\
v
,,,.....--- --- __
69 psf
.,,,.,,.,,
Scale
1 ''•0. SO"
Fig. 6-7 Buckling t>f a 24G Sin11le Deck All-Supported

Note: Results With Flat Elements
23 psf
0.50
1. 0
Compression Diagonal 'obt
23 psf
0.50
46 psf Scale
'·--- --- _.,,.,

62 psf l"=0.50"
1. 0
Fig. 6-8 Buckling of a 24G Single Deck All-Supported

Note: Results with Curved Elements
,,/iJ
/ /
~ .... -~
I / /1"
\·
60 \
'I
. • I e Flat Elements
4-4 I
Ill
'
p..
45 I Curved Elements
.....=
""C:J
t1S
Note: For Key Sketch
0
...:l
see Fig. 6-6
I
N
......
'°I
Point 'l' Point '2 1 Point '3'
1. 0 1.0 1.0
tS in inches
Fig. 6-9 Buckling of a 24G Single Deck All-Supported Hypar (Structure '12')
-280-
o - · - · ·····T------.------y-----,..----,.-----..- h
0.25 ~
\
\
\
o.so \
\
\.,.....
....-- -,
____
'-.._ ............ SQ. 5 psf . /
/
0.75
Compression Diagonal 'ob'
0
0.25
36 psf Scale
0.50 l"=0.25"
0.75
Fig. 6-10 Deflection Profiles Along Tension and

Compression Diagonals (Structure '12')
·281-
60
so
.
'+-< 40
fJl
p,
i::::
·~
'"CJ
~
0
~ 30 Scale
l"•0.10" Deflection
1"•10 psf Load
20
0.10 0.20 0.30 0.40

o in inches
Fig. 6-11 Load vs Free Corner Deflection '0 b f

(Structure '12 ')
-282-
Fig. 6-12 Deck Buckling from Energy Metltod Analy1l•

-283-
Figure 6.14 Photo of buckled hypar deck (Test • 11).

- 284 -
"P ip:? .,,

.. .
\ •.
. ~ ~
~
~ ~
\'
\ I
~
"Ii. { \I
,Ii!~
~
• l t
I I
'I I
I
. {
.~
,\
J, ,~
~
,~ }..
l
I
I I •
;:I'
4 t
I
,•' ~ I
f
I I
I~ ~ ~
I
,'
I
I
•
~
~
' f • I~
'Ii
•
I
I
'
JI IIf
~
nz y
' I
YH
Case Eguation ~
I p = tfEI/(k1) 2 1.0
er
II p • 11 2EI/(kl) 2 2.0
er .;
Ill Per • El/(kl) 2 0.699
IV 2 2
(q 1) s T El/ (kl) 1.122
er
v (ql) • ~El/(kl) 2 0.492
er
VI (ql) 2
s ir EI/(kl)
2
0.436
er 2 2
Vll (ql)
er
= 1T EI /(kl) 0.284
Fig. 6-13 Follower Force Buckling of lselated Edge Members

-zss-
Figure 7.1 Flat shear test
1 •. 2 Saddle hypar test .. 821

-286-
Figure 7.3(a) Umbrella Shell No. 11
Figure 7.3(b) Test No. 11 in inverted position

-287-
Figure 7.4 Small-scale test
Figure 7.5 Test No. 9 in inverted position

-288-
Direction of Shear
r-------------'""-:--------------------1Top
Layer
..._~..+or-o-.._.-____....,..,.toro---:.-~r""'r""~~---'."P"'ll:~...,...~....,,.rBottom
~~--.-&..-..-..;--~r----11111&---11~---......-..-.._----~~Layer
Edge Beam
(a) Local Compression - Laroe Distortion
Direction of Shear
1---~-------------~---------------~Top
...,_----:i.,.._________..,..i---__ ir--__,._~tor--,_.-~--.
Layer
Bottom
l-.l!K----::1.._......'l...._---i...,.._.;...,._~..__..;Jte;..___.....__ _ _ __...,"'-t Layer
Edge Beam
( b) Local Tension - Tearino
,
/
Figure 7.6 Effect of Position of Screws

-289-
MH)O
J:.iOO
-)000
)'jOQ
1I
- I iOO
Il I
Figure 7. 7 1
iI
Lo.-1d versus Ocflcction
j
:flat SJ1:-ar T"st l\o. 14
1,;:io /'o
·l
j
l°
v --- ---- - ~ --
0.2
26 in inches
o.3 o.4 o.s
-- L ··-·· --·--- -'----~•--------
-0. :o
.~0.08 13.S2C· Figure 7. 8 Sr.car ::!igidity G' ... cent

rlat Shear Tests Versus Hypar Curvature for
No. ::, , 9 Cne Layer of Decking
•> Eypar Tests No. I
N
Sll, 812 \0
0
f-o. 06
\
101.40 I
E~,p~r Tes ts No. Hypar Tests No.

a G' 511, 512 3U, 312 5I
I
'
l....0.04 6760 ! j
-1
~ \~ '
I
I
~ I
'
i
I: ~
tc I
~
I ..---1.
r-0002 3380
'I
I' !
I
! c in inches I
I
24 28
l~ 3 12 16 20
!
L.i::.10
1
33.;.00
I
I
I
Figure 7.9 Sh~.:ir ~i~idity G' • ~Gnt

Vcrs:Js ;-:yp.:ir Curvature: for
Two Layers of Decking
27040
Ey?:::- 'fcsts ~o.
021, 022 '
G' '
0.06
I N
IQ
~
I
- - - - - F l . ' . l t Shear Test
10.04 1352G
No. 15 No.
Hypar Tests No.
1
~
321, 322
5'
I
!
i
I
I
\;
f
I
Iro.02 5760
! c in inches
I 4 s 12 16 20 24
! I I I I I
-Z9Z·
.26
+ .66 ' ::t. ~ 88 ' : '

+ .73
• ll +.59
'.+.44~
+ .22 .
• 15 • l l • 01 • 01• • 01• 0
' :·:-.~.'E=-=-
1. ls .55 .66
.1 5-t-
.48 ..
+ + +·
-. ()7 .59 • 6()
+ +
--
.n •.-,2 ._io .17 '
+- +
•r:r. .ss
• l ') + +
• 73 14!)6 • 77
.18 + ·I-
+.33
• 22
-----·-· !------- ~---
.lG .22 .15
-------1---
.11
I
.1s
----·----------.......1.18
Figure 7 .10 (a) Experimental Vertical Deflections

in inches at 40 psf, Umbrella
Test No. 11
-293-
-1060 (+1550)
---·--·-
+2860
--· - - - - - --- - - - - - - - - - - - - t,.:_:__ _ _ _ _ _ _ _ _ _ _ _ - - ,
.,'• 1 2 0 ' (-1160)-·

. - -··. -....
---- - .. - ... ~-~·::··. - -+-16?.~1160)
!li~r030) +2320 I · - -· -·· · - -
+302
+30l2i_
·:·637u• .
+:I" 2500 _._ .
- .. --··
+4030
-2990 +
+7650_
-l·lj~g l (~7_7_0) ·--· . +-
-230 _- __ --·--
. . ·- . -- .. - ·-· +7070. - .. -----
+](>0\520)
+JO'--
l__
· _ _-· . --:strong dir..
:> S '.O( 520)

' 9'l 0 ( l 030) l- : -: ~-~WR:-( 1:3: ~--- . ---- -
1
.:!. +1400(1550) ...!.

+3370
Figure 7.lO(b) Experimental Stresses in psi at

40 psf Load for Umbrella Test
No. 11
Axial stresses are directly above

bending. Values in ( ) are axial
stresses from membrane theory.
-294-
• 26
.13 j • 07
• '..iS
+-
.33
+
.07
.22 29 • L8 .22
+ ·+ + +
• 11
.l'.) .29
+
.T3 .13
+ +
• :? <) .MJ .26
+ + +
rl~.15
2') l ..
·-·- .11 .04
Figure 7.ll(a) Experimental Vertical Deflections

in inches at 40 psf, Umbrella
Test No. 12
-295-
·:· 53 0 ( l 5 s0)
--- ----- ------ ----- - - - -+3500
--.r.=-----------------
- 1 ll~( (-770) -1030

+125< ++5660
·i-1270 -1110
-;-l~l) 50 + - +5080 +
:: ~~~;.~K~_:;1 o),_::;:~-~o~-~~~?:&==~- 1_0 7 (~~-~,1.::__c~c...:;o~_-__-______- 71~-==o+~(-=7=7=0=)==:i;~r sso)
~·
; ;1r010)
l 770
-1010
+1100
(-1160)
-880
+1100(1160)
·:-230
t +350 ++2580
+3860
-10 - t0:~c (-770)
+ + 1150
-;-t,:.;oo t :-1 l3r
+5620
-1 o:;o( s20)
-:-in
-'J40 ~ (-380)
+70
strong dlr.
+000(5~0) +~30(1010) +700(1030)

·--------- - -I - ---- -- -- 1-- - -- ------ ------!------ - --------.---
Figure 7.ll(b) Experimental stresses in psi at

40 psf Load for Uabrella Test No. 12
Axial stresses are directly above
bending stresses·. Values in ( ) are
axial stresses from membrane theory.
-296-
. -·-· ·--····.40
·t ····---~--,..----~----·.26--- --- --------~.....-----'-0
.37
,.,, . 33 . '
• ) :.>
I '. • T :.
i . . ' 1
• aI 59 . ' ..
t- ~- ~·i- :
!
.17 - • 07 •l8 I:
.29
+
f ' ! '
• I 1 .1') .11 .1)7 • O/~

I i ·1 . i· ... ···+ ..
- • 04
•ll . +
')?
,_ ,_ • ll •ot~ 0 .11
+ + + +
.15 ' '
'
I
'
• la • 29 .04 ' '
+ ·'· . +. ':' ' '
• l :i
·r~ ·~
I] •ot~
.l-
'.~ () r .l l
.• 113
• 15 • :U1 • 22 - .04
+ + +
• 27.
1 u
I .01
i . '
. ' ' . ' '
\I . . . . '
I · ···I · ·· I· - -1-- -+----·--!1- --.---·------~~-- -- _: ------~---.J
• I u'·' 0 .1.5 .U7 .07 • 07 • Ol~ • 15
Figure 7.12(a) Experimental Vertical Deflections in

inches at 40 psf, Umbrella Test No. 13.
-297-
Axi<1l stresses arc directly above bending stresses.

V;1lucs in ( ) arc i!:<lal stresses from membrane theory.
r·- ··- --t, ;I •!· ·

I l !
; ! .
t ' t 1
' '
' l ' '
i .
-1320 ,,(-770) .' +5960
+ +z9oonl . . ++10150
; . I
+-1120.
+!~3 70 + i ! .
' i
' . I
i
.
t
' l ' 1
-700(-776) . +225C (1550)

---r~-=~. ~-==::.:_:-_-_ -=:;::::-_-
- ' . . +411
·:-(.;.j()
+2200· r( 1. 030) • 1·3330
' l
'1·2970
·:·20')0
·•·'.Jo,;o+ + +t-560
+6l;10
+ '.' 'j!l('.)20) .
·•·')(;t
l
I
I
'
Figure 7.l~(b) Experiaental Stresses in psi at 40

psf Load for Uabrella Test No. 13.
-298·
.300
----.:.....135
I •
I
.026 -.067
+ +
' .
'
I
I
.
' · .o:rn . 1
.029 .01s
• 063
+ + .032 .
+ !- .057. +
fr •<W2
II .
-~-·-----: taz · ·---~~1t1-- ·------·::lie-~---'·· .315
Figure 7 .13 Experimental Vertical Deflections in

inches at 40 psf, 2 ft x 2 ft x 0.25 ft,
Umbrella Test (Section VII.Id).
-299·
I. 73 1.01 .48 .33 1.56
- • • - - .. ;· -- --·· • ·- ; I , •. '
.32 : ~ ..' ....

• 92
+ • I
I
'
I • .
·-i~ ' ' ·fi
• "· I
: ... .,' .
.t~ 7
+
.14 .:.· . . ·+2 . : : •i~
l' - ..
I
' - -i -
I I '
' '
.76
.60
J ..
' ' I !
'
.os .. i .79
r
I
. I' I '
Figure 7.14(a) Experimental Vertical Deflections in

inches at 40 psf, Ullbrella Test No. 9.
,\;.:i.1l :.;tt·e:;:;es ;1rc directly nbovc bending stresses,
Deck :>trc~;~<.·:• in bot ton lnyor .1ro to the lcf t of those
in top l.1ycr.
+ :;320 +6t.so +3550

+ 16000 -i· l 6000 +5970
l-·--·----1-------...,......-..
-f-
' . '
. \
+~lO +128~ +17SO · ·· 1-·

+920+ +57 . +2.060 : .. ' ; I
l
·: t' •
I ·I • .. _.. i • l; i
: I ;
•· ••I
'
I ; ! ..
.I
• I
' ?
, ! : •
; : ~ !
I . I .
'' I' I ' :
I
i ' ''
'. ·Hit~o +s10 !· .. ; I '
·i-2C4o-!-+2270 · :
; ! ~
...
! ' j
':
' ..
.. ' :
L: .
I ! '
: _, I
l
I
I J "
'. .'
I
I '
·--- ---------l-- -----1---:__~~1----J_~_;_._·~-~----··. ·. .-

+40~0 +4900 +6450
+5470 ~3580 +17200
Figure 7.14(b) Experimental Stresses in psi at 40

psf, Umbrella Test No. 9.
archi DOCT The e-journal for the
dissemination of doctoral
research in architecture.
Supported by the ENHSA Network | Fueled by the ENHSA Observatory
9
July2017
www.enhsa.net/archidoct
ISSN 2309-0103
PROTOTYPING STRUCTURES
IN ARCHITECTURE
European Observatory
enhsa
enhsa
european
european
networknetwork
of Doctoral Research
of heads
ofof
heads
schools
of schools
of architecture
of architecture
in Architecture
ISSN 2309-0103
77 www.enhsa.net/archidoct
Vol. 5 (1) / July 2017
//
Shell structure: Analysis of hyperbolic paraboloid in paper
Ana Laura Rocha Peña
Shell structure: Analysis of hyperbolic

paraboloid in paper
Ana Laura Rocha Peña // Polytechnic University of Catalonia-Barcelona Tech
Abstract
In contrast to the opinion of many people, who think that technology will eliminate paper, tons
of paper are used daily worldwide.The annual average paper consumption is 48 kg per person,
which is equivalent to approximately 347, 035, 970.30 tons. In the last few years the architect
Shigeru Ban has motivated research in the field of paper as building material. In this paper, we
evaluate the implementation of paper as building material in shell structures. At first, paper
does not seem to be strong enough to be used as structural material, however we propose
shaping it, in order to improve its structural behavior.We have simulated a hyperbolic paraboloid
(HYPAR) of paper in Abaqus software based on the finite element method (FEM), and analyzed
its structural behavior.The analysis results demonstrate the feasibility of using paper as building
material.
Keywords
Shell structure, Paper as building material, Hyperbolic paraboloid, Form optimization.
Note
This essay is the result of an extensive research conducted for the author’s Master Thesis
at the Polytechnic University of Catalonia-Barcelona Tech (UPC), completed in October
2016. Due to the complexity of the calculations of the HYPAR, we first calculated a plate
by using the Navier method. Then, we compared these results with those obtained from
a simulation model of the same plate in Abaqus. Likewise, we use this model as reference
to compare it with a HYPAR structure and made a comparative analysis between paper
and concrete, since concrete is a material well known and by now, widely studied.
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Introduction
Shell structures are constructed systems described by three-dimensional curved surfaces, in which
one dimension is significantly smaller compared to the other two. They are form-passive and resist
external loads, predominantly through membrane stresses. A shell transfers external loads to its
supports predominantly through forces acting in the plane of the shell surface, which are called
membrane stresses. The shell surface is normally stressed in compression, or in combined com-
pression and tension. A ‘thin’ shell has to be sufficiently ‘thick’ to carry these compressive stresses
without buckling. Shell structures can be constructed as a continuous surface or from discrete ele-
ments following that surface. Geometric shapes such as sections of spheres, hyperbolic paraboloids
and regular polyhedrals are also commonly used.
The challenge is to find geometries that can work entirely in compression under gravity loading.
These geometries are not limited only to masonry, but are often built of any material. However, for
traditional masonry structures, the dominant load is often due to the self-weight of the structure,
and the applied live loads, with smaller effect, due to wind or snow (Adriaenssens, et al., 2014).
The antecedents of shell structures are founded in Gothic architecture evolving from the heavy
brick vaults to the slender and ribbed vaulted alloys.The most direct precedent in time is the barrel
vaults. These consisted of several layers of fine brick, the first of which is placed with the help of
small wooden guides and is overlaid with plaster paste, constituting a collaborative formwork. This
sheet reproduces the shape of the inner curve of the vault, the following layers are superimposed.
One of the most important developments in shell structures was the application of reinforced con-
crete. At the end of the Second World War (1945), due to lack of steel, reinforced concrete favored
the development of molds with its ability to work in compression and traction, as well as to provide
monolithic construction.
The construction of reinforced concrete shells became a process of study, experimentation and
innovation, whereas respective contributions made by Eugène Freyssinet, Eduardo Torroja and Felix
Candela were decisive (Carceles Garralón, 2007) (Figure 1).
Paper as Building Material
The first cultures to use paper as building material were the Chinese and Ancient Egyptian ones. At
the beginning, it was used in form of papier-màché, followed by the development of papyrus by the
2nd century B.C. By the 9th century A.D., the Japanese culture started to use paper elements in the
construction of sliding doors and walls, called shoji-fusuma.That was the first time, when paper was
utilized as an interior building component. France was the first to use paper in furniture production
in the 19th century, and later as wall covering, introducing for the first time in history paper as a
decorative element.
Paper products were being used in the production of aircraft and tank components in World War I.
When realized that aluminum had problems with expansion and shrinking, the substitute for alumi-
num sheeting on aircraft wings came in form of plaster-made molds for shaping and cellulose rein-
forced sheets of paper combined with starch or similar adhesives. By 1920’s, paper and cardboard
started to be used as electrical insulation in the United States. In the same period impregnation
experiments began with the introduction of cellulose fiber laminates in industry. First, phenolic-res-
in was used, until the development of melamine resins led to an increased popularity of paper and
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
cardboard as building material.
In following decades, several architects began to experiment with paper as structural material. The
first building principally constructed out of cardboard was ‘The 1944 House’ that was followed
by a period of slow development in the field. Several architects have influenced the progress of
cardboard applications in architecture, with the two most influential figures, Buckminster Fuller in
1950’s and Japanese architect Shigeru Ban, most recently (Sekulic, 2013) (Figure 2).
Geometry Definition
Among different shell structural configurations, a hyperbolic paraboloid was selected for the sim-
ulation. This is an anticlastic surface, whereas the center of curvature is located on opposite sides
of the surface. In this type of structure, normal loads are transferred to its surface by tangential
stresses (compression in the convex curve and traction in the concave curve).
The hyperbolic paraboloid equation can be written as:
z=kxysin(ω)
Where k is a constant representing the unitary warpage of the paraboloid; k = AA ‘/ (OB • OH •

sin ω). Then x and y are axes that will only be perpendicular in the case when a = b and in this case,
ω shall be equal to 90°.
Another way to understand this surface is to consider it as generated by a main parabola P1 moving
parallel to itself along another main parabola P2. Thus the surface has two systems of paraboloid
generatrices (Oliva Quecedo, et al., 2011).
Paper Mechanical Properties
There are three significant factors, which determine the mechanical properties of paper: The prop-
erties of fibers, the interfiber bonding and the geometrical disposition of the fiber (Figure 3). It’s im-
portant to know how these factors influence the properties of paper. In laboratories, when sheets
are produced for experiments, even if these have the same composition as machine produced ones,
they do not have the same properties, because paper properties do not depend only on the com-
position, but also on the production process. Laboratory prepared paper is different than machine
prepared paper (Sekulic, 2013).
Stiffness Values of Paper
Table 1 contains a collection of directly measured values of the elastic stiffness parameters for a
few paper grades. Many values are missing, because of measurement difficulties caused by the small
thickness of paper.Various estimation schemes have been developed to avoid direct measurements.
The table demonstrates that the ZD (Z-direction) stiffness of paper is generally low compared to
the in-plane values. The negative value of the Poisson ratio Vxz for the paperboard shows that uni-
axial tensile loading in MD (machine direction) increases with thickness. In compression, at least, the
elastic moduli, perhaps even the Poisson ratios, are usually equal to the corresponding tensile values.
In general, the density of paper is between 300 and 900 kg/m³.The elastic modulus usually increases
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Figure 1.
Los Manantiales restaurant, Félix Candela
Figure 2.
Japan Pavilion, Expo 2000 Hannover, Germany by Shigeru Ban
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
with density and ranges from 1000–9000 MPa, when the effect of anisotropy is removed by
averaging over MD and CD (Cross-machine direction) (Niskanen, 2011).
Simulation Process
Analysis Parameters
Following the geometry exploration process, a hyperbolic paraboloid was chosen for the
prototype design. For the simulation, an initial model was used starting from a quadrilateral
of 6000 x 6000 mm, with height of 5000 mm, and thickness varying from 60 to 10 mm.
We have simulated the hyperbolic paraboloid of paper in Abaqus software based on FEM
(Abaqus, 2004), in order to investigate its mechanical behavior (Figure 4). To this model,
two general analyses (linear and nonlinear) have been conducted. A static linear analysis
provides a first approximation of the structure’s behavior, by considering equilibrium of the
system without deformation. For obtaining precise results with regard to the response of
the structure, a second-order analysis is required that incorporates the effects of material
and geometry nonlinearity. The nonlinear analysis considers the properties of the material,
the surface loads and the boundary conditions. All elements are assigned with an elastic
modulus of 5420 MPa and a Poisson’s ratio of 0.38, selected from the table of measured
values of elastic stiffness (Niskanen, 2011). A uniform vertical load of 1 kN (0.001 N/m²) is
applied to the surface. As to the boundary conditions, two of the four nodes are pinned with
zero displacement.
Development of the Analysis
The development of the simulation model provides information about the physical behavior
of the Hypar structure in paper. In all cases, the maximum deformation and the maximum
stress by Von Mises in the center of the Hypar have been registered. Probably the most
important properties of structural materials are their strength and stiffness. The limit stress
refers to the maximum strength value of the material. The type of paper chosen for the
analysis has a maximum strength value of 5.00 MPa. Currently no standard regulations exist
with regard to paper as building material. We have chosen a maximum deformation limit of
L/100, i.e. 8485/100= 85 mm.
To calculate the longest distance of the system, the following equation applies:
Lmax=L.√2

where L is a side of the quadrilateral that makes the Hypar.
Only maximum values of the stress and deformation in the center of the Hypar have been
considered in the analyses, and any respective results superior to the limits set were dis-
missed. For better comprehension, the analysis process is divided in three stages. The first
stage, static linear analysis, considers the equilibrium of the structure without deformation.
Then a second stage, static nonlinear analysis, takes into account effects of the deformed
geometry. And finally, a third stage, plastic nonlinear analysis, considers effects of nonlinearity
of the geometry and the material behavior.
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Figure 3.
Wood fibers, interfiber bonding and geometrical disposition of the fiber
Table 1.
Measured values of elastic stiffness parameters in tensile loading for some machine made papers (Niskanen, 2011)
Figure 4.
Simulation model, visualization of the Von Mises stress and deformation
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Stage 1 – Static Linear Analysis
The initial model was ideal for obtaining a first approximation of the Hypar behavior. As
shown in Figure 5, the highest stresses develop at the ground supports areas and the
center of the Hypar on both sides (superior and inferior). This is due to the fact that
the shell transfers the external loads to its supports through forces acting in plane of its
surface. For the specific analysis, the resulting stress values in the center of the Hypar
have been registered.
Figure 6 summarizes the Von Mises stresses in the center of the Hypar obtained from the
static linear analysis. Based on the analysis conducted, the maximum stress increases with
thickness reduction, even above the material strength limit.
The specific Hypar structure works along one axis as an arc and along the other axis as
a suspended arc. While compression stresses tend to deform the membrane along one
axis, traction stresses along the other axis, tend to counter this deformation. For this rea-
son, the maximum deformation develops in the center of the Hypar as shown in Figure 7.
Figure 8 summarizes the maximum deformations obtained from the static linear analysis.
Two systems have higher deformations than the respective limit set.
Stage 2 – Static Nonlinear Analysis
The difference between linear and nonlinear analysis is the system’s stiffness. When a
structure deforms under an external load, it is the stiffness that changes due to the ge-
ometry or the material properties. This second stage of nonlinear analysis comprises a
static nonlinear analysis that considers the system’s deformation effects. In Figure 9, the
results of the simulation model are presented. The last two systems do not converge.
Figure 10 summarizes the Von Mises stresses in the center of the Hypar obtained from
the static nonlinear analysis. The last two models with 20 and 10 mm thickness do not
converge.
As shown in Figure 11 the maximum deformations develop in the center of the Hypar.
Two of the six systems do not converge.
Figure 12 summarizes the deformations obtained from the static nonlinear analysis. The
last two systems with 20 and 10 mm thickness do no converge. The other four systems
develop favorable results within the respective allowable limits set.
Stage 3 – Plastic Nonlinear Analysis
Finally, for obtaining most accurate results, mostly similar to reality, with regard to the
Hypar structural behavior, a plastic nonlinear analysis has been conducted. The analysis
at this stage considers both, the effects of nonlinearity with regard to the geometry and
the material. Figure 13 shows the results of the simulation models.
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Figure 5.
Results of the simulation
model - Von Mises stress in
static linear analysis
Figure 6.
Structural response -Von
Mises stress in the center of
the Hypar based on static
linear analysis
Figure 7.
model - Deformation in
static linear analysis
Figure 8.
Structural response - Defor-
mation in the center of the
Hypar based on static linear
analysis
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Figure 9.
static nonlinear analysis
Figure 10.
Structural response - Von
nonlinear analysis
Figure 11.
static nonlinear analysis
Figure 12.
Structural response - De-
formation in the center of
nonlinear analysis
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Figure 13.
plastic nonlinear analysis
Figure 14.
Structural response - Von
the Hypar based on plastic
nonlinear analysis
Figure 15.
plastic nonlinear analysis
Figure 16.
Structural response - De-
formation in the center of
the Hypar based on plastic
nonlinear analysis
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
Figure 14 summarizes the Von Mises stresses in the center of the Hypar obtained from the plastic
nonlinear analysis. In this particular stage, all simulation systems show favorable results within the
allowable limits set.
The maximum deformation of the systems develops in the center of the Hypar, as shown in Figure
15.
Figure 16 summarizes the deformations obtained from the plastic nonlinear analysis. None of the
systems exceeds the allowable limits set.
Conclusions
Research activities in the field of paper as building material have increased in the last years. Paper is
an excellent material with regard to providing innovative, new ways of application in construction.
This material can be used as structural material for construction and as formwork of complex
structures, in both cases, offering opportunities for sustainably and economically sensitive designs.
Following analyses of a hyperbolic paraboloid based on FEM, the following can be concluded: In a
first stage of a static linear analysis, as well as in a second stage of a static nonlinear analysis con-
sidering deformation effects, the response values obtained are almost in the range of the allowable
limits set in the analysis, except for the models with 20 and 10 mm thickness that have passed these
limits. Finally, in a plastic nonlinear analysis, most similarly to reality, the results obtained are all fa-
vorable and within the allowable limits set. The developed stresses of the models obtained values
within the range of 0.56 to 4.79 MPa, all below the respective maximum strength of the material
of 5.00 MPa. All system maximum deformations are less than L/100, starting from 3 mm for 60 mm
surface thickness and reaching 73 mm for 10 mm thickness. Therefore, the results obtained from
these three stages of analyses, favor implementation of paper as building material in shell structures.
Future research will investigate the mechanical behavior of the different types of paper, ways for in-
crease of the strength of paper in combination with glued composites, of reduction of the humidity
of paper, and among others, the fiber of paper as material for the 3D-printer.
ISSN 2309-0103
Vol. 5 (1) / July 2017
//
References
ABAQUS, 2004. ABAQUS keywords: Reference Manual: Versión 6.5. West Lafayette:
ABAQUS Inc.
Adriaenssens, S., Block, P.,Veenendaal, D. and Williams, C., 2014. Shell Structures for
Architecture : Form-finding and Optimization. London : Routledge.
Carceles Garralón, F., 2007. El paraboloide hiperbólico comogenerador inagotable en las
estructuras laminares. E.U. de Arquitectura Técnica (UPM).
Niskanen, K., 2011. Mechanics of Paper Products. Berlin: De Gruyter.
Oliva Quecedo, J., Antolin Sanchez, P., Cámara Casado, A. and Goicolea Ruigómez,
J.M., 2011. Análisis estructural de algunas obras de Félix Candela mediante modelos de
Elementos Finitos. Hormigón y Acero. E.T.S.I. Caminos, Canales y Puertos (UPM).
Sekulic, B., 2013. Structural Cardboard: Feasibility Study of Cardboard as a Long-term
Structural Material in Architecture. Universitat Politècnica de Catalunya.
Introduction to Shell Structures
Assoc. Prof. Adrian Dogariu

Introduction to Design of Shell Structures
General
• Metalic shells
General
• Natural shells
General
• Natural shells
General
• Definition:
• A shell is a thin structure composed of curved sheets of material,
so that the curvature plays an important role in the structural
behavior, realizing a spatial form
• Motivation:
• A shell is the most efficient way of using the material, and can be
very useful in case o storage of fluids and solids (uniform loads)
Difficulties
• The curved form may lead to different failure modes and often
unexpected behavior occurs
• The analytical formulas are very complex and complicated in
comparison with all the other structural forms
• Shell structures are very attractive light weight structures which are
especially suited to building as well as industrial applications.
Range of application
• The shell structure is typically found

• in nature
• as well as in classical architecture.
• There are two principal uses of shells in civil engineering:

• industrial structures:
– silos, tanks, cooling towers, reactor vessels etc.
• aesthetic and architectural special structures
Main documents
• Eurocode on strength and stability of Steel Shell

Structures – EN1993 Part 1.6 (2007)
• Generic normative standard on shells for
chimneys, towers, masts, silos, tanks, pipelines
• Buckling of Steel Shells European Design
Recommendations 5th Edition (ECCS – 2008)
8
General
• Built structural shells
General
• Built structural shells
Reinforced concrete
Steel
Aluminium alloys
Plastics
Glass
Timber
Structural typologies
Eliptic paraboloid Hyperbolic paraboloid
Circular cylinder/cone
Structural typologies
Shells are the most difficult form of structure to analyse and the form with
the most complex behaviour. As a result, all but the simplest conditions
must be analysed using computers.
Examples – Steel reticulated dome
US pavilion Expo 67 Montreal
Architect: Buckminster Fuller & Shoji Sadao
The 250ft diameter by 200ft high dome roughly

presents a three-quarter sphere, while geodesic
domes before 1967 were hemispherical. The dome
consists of steel pipes and 1,900 acrylic panels. To
keep the indoor temperature acceptable, the design
included mobile triangular panels that would move
over the inner surface following the sun. Although
brilliant on paper, this feature was too advanced for
its time and never worked.
Examples – Aluminium alloy reticulated dome
Spruce Goose dome, Long Beach, USA
Architect: R. Duell and Associates
Engineer/builder: Temcor
A - Aluminum cover plate with silicone seal

B - Aluminum gusset plates, bolted to struts
C - Aluminum batten secure silicone gaskets
D - Triangular aluminum panels
E - Wide-flange aluminum struts
F - Stainless steel bolts
Examples – Timber-steel free form grid shell
Multi-hall, Mainz, Germany
Architect: Mutschler, Frei Otto, consultant
Engineer: Ove Arup
The multi-purpose dome for the 1975 garden

show spans max. 60m with 50x50mm twin
wood slats of 50cm squares that deformed into
rhomboids.
1 - Form-finding model
2 - Interior
3 - Mesh detail (steel bands resist shear)
Examples – Timber-steel free form grid shell
Architect: Thomas Herzog
Engineer: Julius Natterer
Wood grid shell with PTFE membrane

The theme pavilion advanced the philosophy:
• Wood is the only renewable material
• Requires the least energy for production
• Use of wood maintains healthy forests
Shell Analysis
Curved
shapes Bending stress state
Continuous
Membrane stress state
Plated
Shell structures
Reticulated (bar structures)
• Continuous (or reticulated) shells

• Linear behaviour
• Non-linear behaviour
• Elastic
• Elastic-plastic
Shell Design
• Resistance
• Stability
• Highly sensitive to imperfections
• Buckling is a process by which a structure cannot withstand loads with its original
shape, so that it changes this shape in order to find a new equilibrium
configuration. This is an undesired process (from the point of view of the
engineer), and occurs for a well–defined value of the load.
• The consequences of buckling are basically geometric:
• The are large displacements in the structure
• There may also be consequences for the material, in the sense that
deflections may induce plasticity in the walls of the structure
Local buckling Global buckling

of a tank of a wind
turbine tower
Steel Shell Design: Codification
Conceptual design
Design for strength and buckling
Detailing
Behavioural phenomenology of shells
• Behavior of a given structure (slender!) can be controlled by design if
the three characteristic ranges of load-deformation curve are correctly
defined
• Pre-critical range P
• Critical point (or range)
• Post-critical range
Pcr Post-critical
Critical point range
Pre-critical range
P(0, Pcr]  Structural stability

D
P > Pcr  Structural instability
elastic
Buckling: plastic
dynamic
• Instability phenomenon e.g. bifurcation instability of cylinders
N xsup
,cr
N xinf
,cr
L
L
• Instability phenomenon – Jump of Equilibrium or Snap Through Instability
• Affects shallow arches and shells, reticulated shells
EREN Exhibition hall,

Bucharest, 1963
Critical and post-critical behaviour of elastic structures
P
u length
w Perfect bar (unloaded) P
(unloaded)
w
length
P
u Perfect
length w
cylinder Perfect
(unloaded) P plate
u
P
P Perfect plate
P Perfect P
bar Perfect cylindrical
Pcr Pcr shell
Pcr
A
Imperfect
Imperfect plate
bar Imperfect cylindrical
shell
w0
w w0
w
w0 w
Columns Cylinders Plates

indifferent post-critical path unstable post-critical path stable post-critical path
Favourable and unfavourable effects of spatiality
• Curvature effect in axial compression
 2E  t 
2
 cr , p   
31   2   b 
 2E  t 
2 2
E b
 cr ,c  2  
 2 
31     b  4  r 
Stable Unstable
component component
increase in sensitivity to
increase in critical load
geometrical imperfections
Favourable and unfavourable effects of spatiality
• Curvature effect in axi-symetrical compression
increase in sensitivity to
increase in critical load
geometrical imperfections
Coupled instabilities for plate and shell elements
W – weak interaction
M – moderate interaction
S – strong interaction
VS – very strong interaction
Coupled instabilities for plate and shell elements
• Erosion of Theoretical Critical Buckling Load
Instability phenomena: Influence of imperfections
• Agreement of theoretical and experimental values
bars
shells
Instability phenomena: basic types and models
• Dynamic propagation of instability or progressive instability
• Domino effect
(double layer grids)
• Instability propagation
(single layer reticulated shells)
Models and Methods of Analysis
• Pre-Critical, Critical and Post-Critical Analysis
• Generic classification of structures in terms of characteristic instability

types and sensitivity to imperfections
• Linear, nonlinear, elastic, plastic models
• Linear buckling analysis (eigen-buckling) – LBA
• Geometrical nonlinear imperfection analysis – GNIA
• Geometrical material nonlinear imperfection analysis – GMNIA
• Pre-critical solver methods (Newton – Raphson) or
• Post-critical solver methods (Arc-length); Designed load checking or
load-deformation curve
Models and Methods of Analysis
• Design flowchart for the Design of Shells according to EN 1993-1-6
Methods of Analysis
• Methods of Analysis – Global Frame Analysis
• Finite Elements Methods for Analysis and Design
Load-displacement curves found using different

analyses of the same structure (Rotter, 2011)
Methods of Analysis
• Basic modes for behaviour
a) Membrane
b) Bending shell
c) Shell like a member
a)
c)
Methods of Analysis
• Basic Equations
• Simplified Linear Shell Theory
• The Love-Kirchhoff assumptions (simplified model)
• The shell thickness is negligibly small in comparison with the least radius of
curvature of the shell middle surface (shell is thin)
• Strains and displacements that arise within the shell are small (products of
deformations quantities that occur in the derivation of the theory may be
neglected, ensuring that the system is described by a set of geometrically
linear equations)
• Straight lines that are normal to the middle surface prior to deformation
remain straight and normal to the middle surface during deformation and
experience no change in length (analogue to hypothesis for beams – plane
sections before bending remain plane after bending)
• The direct stress actin in the direction normal to the shell middle surface is
negligible (not valid in the vicinity of concentrated transverse loads)
Methods of Analysis
• Model of an axi-symmetrical Loaded Shell
Model of an axisymetrically
loaded shell
Geometrical parameter of the

spherical shell
Methods of Analysis
• General Rotation Shell
• Membrane Theory: Equilibrium Equations for Unsymmetrical Actions
 N


N r0 

r1  N r1 cos   Yr1r0  0
 N

 
r0 N   r1  N r1 cos  Xr0r1  0

N N
 z0
r1 r2
Methods of Analysis
• Cylindrical Shells
• Bending Theory Axisymmetric Loading
dN x
a  dx  d  0
dx
dQx
a  dx  d  N dx  d  Z  a  dx  d  0
dx
dM x
a  dx  d  Qx  a  dx  d  0
dx
d 4w Eh Eh3
 wZ D
;
 
D
dx 4
a 2 12 1   3
Methods of Analysis
• Shells
• General Bending Theory
N x N x
a  0
x 
N N x
a  Q  0
 x
Q N x N x
Q a  0
a x  N  q  a  0 x 
x 
N N x M x 1 M
M x M  a   0
a   aQ  0  x x a 
x 
 2 M x  2M x  2M x
1  M
2
M x M x N  a    qa  0
a  aQx  0 x x2 x a  2
 x
Methods of Analysis
• Basic Equations
Aspect Equation Unknown
1. Equilibrium equations 5 8
(static)
2. Deformability 9 12
compatibility (geometric)
3. Physical aspect 6 ---
TOTAL 20 20
Methods of Analysis
• Buckling of Cylindrical Shells in Compression
• General Case
Equilibrium equations for elastic buckling:
N x N yz
a  0
x 
x, u
N y N x  2 v M xy M x
l a  aN x 2   0
z, w  x x x a
2w  2 M x  M yx  M y  M xy
 2 2 2
aN x 2  N y  a    0
y, v
x x 2 x a 2 x
with solutions:
2a mx
u  A sin n cos
l
mx
v  B cos n sin
l
mx
w  C sin n sin
l
v0
n0  u, w  f x  axial  symetrical buckling
Methods of Analysis
• Cylindrical Shells
• Membrane Theory Application for Wind Action
q 2
N  cos
r0
N  2q sin 
N  qr0 cos
Methods of Analysis
• Simplified Design Formulae (Cylindrical shells)
• Two possible approaches
• Overall column buckling if l/r ratio is large
• Shell buckling which involves the cross section deformation and can
be, in general, either:
• Axisymmetric, when the displacement are constant around
circumferential section
• Asymmetric (chessboard shape), when waves are formed in
both axial circumferential directions
r=a
l
Methods of Analysis
• Simplified Design Formulae
• Axial-symmetric buckling of cylindrical shell in compression
d 4v d 2w w
D 4
 Nx 2
 Eh 2
0
dx dx a
Eh3
D

12 1   2 
Radial displacement:
m x
w   A sin
l
N
Elastic critical axial stress (  cr  cr )
h
 cr 
Eh m Eh
; 4 2
a 3 1  2  l a D
l
For =0.3   1.72 ah
m
• In case of axial-symmetrical buckling, the critical shear does not depend of cylinder length!
• If one of the cylinder ends is free (w0), cr drops to 38% compared to simple supported case.
• Cylinder is highly sensitive to tangential displacements at the boundaries. If v  0, critical
stress drops to 50%!
Methods of Analysis

 Post-elastic critical buckling
h EEt
 cr 
fy E  tg  0 
a 3 1  2 
fp
Et  tg  l Et
 1.72 ah
m E
0

Methods of Analysis
• Post-critical buckling: stable and unstable components
Ncr  m2 2 E l2 
 cr   D  2 
h  hl 2 2 2

2 a D m 
 m 2  Eh  h 
2
Ncr  D     
 l  a2  m 
 
Stable component Unstable component

(inextensional bending (extensionally circumferential
deformation – x direction) deformation – y direction
Methods of Analysis
• Post-critical buckling equation
2
N 1 3 Eh  l  2
 1  m2  m
Ncr 8 8 Ncr a 2  m 
Stable component Unstable component
• The effect of circumferential extensional deformations increases the value of

critical load, but change the type of instability from stable to unstable!
Methods of Analysis
• Examples: medium length cylinder
• (i,j) i = no. of longitudinal half-length waves;
j = no. of circumferential half-length waves
Methods of Analysis
• Examples: long cylinder
Methods of Analysis
• Examples: short cylinder
Methods of Analysis
• Principle of ECCS approach (ECCS Recommendations 1998)
• Real cylinders are highly sensitive to imperfections
• “Knock-down” factor  is introduced to account for
imperfections and for plastic effects
 d   cr
•  depends on:
• Shell geometry Test data and design
• Loading conditions curve (typical) for
cylinders subject to
• Initial imperfections
axial compression
• Material properties
Methods of Analysis
• Principle of ECCS approach ECCS design formulae for
• For unstiffened cylinders, unstiffened cylinders
is similar to the one for
column in axial compression
 2   cr  0.5 f y
n  1  1
 2 
f y     M1
Graphical presentation of
 2   cr  0.5 f y ECCS “knock-down” factor
n
 1  0.41231.2
fy
Methods of Analysis
• Buckling of cylindrical shells under external pressure
• Membrane (hoop) stress in practical range
external
pa
y  ; x  0 uniform
h pressure
pa
y  ;  x  0.5 
h
• Von Misses formula for critical pressure, cr
  
Eh  1 h2  2 2n 2  1   “hydrostatic”
 cr    n 1 
 
a  n2  1 32 12a 2  3 
  type pressure
2
a  n
1  ; 3  1    Simplified formulae for long cylinders
 1 
 
l
1 Eh3 n 2  1
h2 2 2   1  1 
2
Eh  1 pcr 
12a 1   
pcr   2 2  n 3  1     3 2

a  n 3 12a 2  2  2  
 
n  min pcr
Methods of Analysis
• Principle of ECCS approach
pu 1
0   1 
py 1   2
pu 
 1  2 ;   0.5
py 
 py 
   
 cr 
p
h
Pcr  E    min
a
1.5 ECCS design strength for
0.855 a  h 
 min  unstiffened cylinder
for l/a0.5 
   
1  2 l  a  under uniform pressure
Methods of Analysis
• Principle of ECCS approach
• Wind action is more complex than simply an external pressure
• It is needed to check the cylinder stability separately for:
• Wind radial pressure
• Wind axial effects
• Wind tangent effects
• Interaction of the three
• Approximately, wind critical pressure can be taken as 1.6 times critical
external pressure (Maderspach, Gaunt, Sword )
• ECCS Design Recommendations (No. 125/2008) offers also a solution
Methods of Analysis
• Buckling of cylindrical shells under compression and external pressure
x p

 x ,cr pcr
Interaction curve
Methods of Analysis
• Buckling of cylindrical shells in bending
• (Flügge) 
 cr , x   
M Nx
 1.33  cr , x
• Long cylinders (Brazier) 
0.99
M cr  Eh 2a
1  
2
Methods of Analysis
• Buckling of cylindrical shells in torsion
• (Swerin and Flügge)  long cylinders
32
E h
 cr  34a

3 2 1  2   
• (Donnel)  short and medium long cylindrical shells

• Fixed end
2 3 2 
12
E h   
 2 l  

 cr       
  
  
4.6 7.8 1.67 1
1  2  l     2ah   
 

• Simple Supported End

  3 2 
12
h   2 
 
2
 cr 
E
      2 l

l    
1  
2.8 2.6 1.4 1
2     
  2 ah   
 
Methods of Analysis
• Cylindrical shells under interactive buckling
• Bending + torsion • Bending + external pressure
2 2
    M p     p 
2
    1   1 or  x      1
 cr   cr    x,cr 
M cr pcr    0.9 pcr 
• Compression + torsion • Axial compression + external

2 pressure + torsion
    2
    1  p   
 x,cr   cr       1
 x,cr pcr   cr 
• External pressure + torsion • Axial compression + bending +
2 torsion
p    x x   
2
   1     1
pcr   cr 


 x,cr  x,cr   cr 
N M
Design of Steel Structures: Strength and Stability of Shells
• Basis of design and modelling
• Shells shall be designed in acc. with EN1990 and, in particular, to satisfy the following
requirements:
• Overall equilibrium
• Equilibrium between actions and internal forces and moments
• Limitation of cracks due to cyclic plastification
• Limitation of cracks due to fatigue
• Types of analysis:
• Global analysis
• Membrane theory analysis
• Linear elastic shell analysis
• Linear elastic bifurcation analysis
• Geometrically nonlinear elastic analysis
• Materially nonlinear analysis
• Geometrically and materially nonlinear analysis
• Geometrically nonlinear elastic analysis with imperfections included
• Geometrically and materially nonlinear analysis with imperfections
requirements:
• Types of analysis: approximate treatments of certain parts of

• Global analysis the structure
requirements:
• Types of analysis: Conditions of use:

• Global analysis - the boundary conditions are appropriate for
• Membrane theory analysis transfer of the stresses in the shell into support
reactions without causing bending effects;
- the shell geometry varies smoothly in shape
(without discontinuities);
- the loads have a smooth distribution (without
• Materially nonlinear analysis locally concentrated or point loads).
requirements:
• Global analysis
-linear elastic material law
- linear small deflection theory (undeformed
geometry)
requirements:
• Global analysis
• Membrane theory analysis LBA
• Linear elastic shell analysis - linear elastic material law
• Linear elastic bifurcation analysis - linear small deflection theory
• Geometrically nonlinear elastic analysis- imperfections of all kinds are ignored
- the basis of the critical buckling resistance
evaluation
requirements:
• Global analysis
GNA
• Linear elastic shell analysis -change in the geometry of the structure
• Linear elastic bifurcation analysis - the elastic buckling load of the perfect
• Geometrically nonlinear elastic analysis structure
requirements:
• Global analysis
• Linear elastic bifurcation analysis MNA
• Geometrically nonlinear elastic analysis- gives the plastic limit load and the plastic
• Materially nonlinear analysis strain increment Δε
requirements:
• Global analysis
GMNA
• Materially nonlinear analysis - gives the geometrically nonlinear plastic limit
• Geometrically and materially nonlinearload
analysis
and the plastic strain increment
requirements:
• Global analysis
• Materially nonlinear analysis GNIA
- where compression or shear stresses
• Geometrically nonlinear elastic analysis with imperfections
dominate in the shell included
- elastic with imperfections
buckling loads of the "real" imperfect
structure
requirements:
• Global analysis
• Geometrically and materially nonlinearGMNIA
analysis
GNIA
• Geometrically nonlinear elastic analysis- gives
withthe
imperfections
elasto-plasticincluded
buckling loads for the
- where
• Geometrically and materially nonlinear"real" compression
analysis with or shear stresses
imperfections
imperfect structure
dominate in the shell
- elastic buckling loads of the "real" imperfect
structure
• Shells shall be designed in acc. with EN1990 and, in particular,
Material to satisfy the following
Shell
Type of analysis Shell theory
requirements: law geometry
• Overall
Membrane equilibrium
theory of shells
membrane not
perfect
equilibrium applicable
Linear elastic shell linear bending
• Limitation of
analysis (LA) cracks due to cyclic
and plastification
stretching
linear perfect
• Limitation
Linear of cracks due to fatigue
elastic bifurcation linear bending
linear perfect
analysis (LBA) and stretching
• Geometrically
Types of analysis:
non-linear
non-linearapproximatelinear treatments of perfect
certain parts of
• Global
elastic analysis (GNA)
analysis
the structure
Conditions of use:
•Materially
Membranenon-linear
theory analysis linear non-linear perfect
analysis (MNA) --linear
the boundary
elastic conditions
material laware appropriate for
Geometrically and materially transfer
- linear
LBA smallof thedeflection
stresses in theory
the shell(undeformed
into support
• Linear elastic bifurcation
non-linear analysis (GMNA)
non-linear
analysis reactions
non-linear
without causinglaw
perfect
bending effects;
• Geometrically nonlinear elastic analysis geometry)
GNA - linear elastic material
Geometrically non-linear -MNAthe shell geometry varies smoothly in shape
-change- linear insmall
the deflection
geometry theory
of the structure
• elastic
Materially nonlinear
analysis with analysis non-linear linear imperfect
-(without
the
GMNA discontinuities);
- imperfections
elastic of
buckling all
loadkinds arethe
ofand
the ignored
perfectplastic
• Geometrically and materially nonlinearGMNIA
imperfections (GNIA) - gives
analysis
- the
the
loads
plastic
have
limit
a smooth
load
distribution (without
-
structure
- the
gives
GNIA basis
the of the critical
geometrically buckling
nonlinear resistance
plastic limit
• Geometrically nonlinear elastic analysislocally
Geometrically and materially strain
-load
gives increment
withconcentrated
imperfections
the
evaluation
Δε
or included
elasto-plastic point loads).
buckling loads for the
non-linear analysis with non-linear - whereandcompression
the plastic strain
non-linear increment
or shear stresses
imperfect
"real" imperfect structure
imperfections (GMNIA) dominate in the shell
- elastic buckling loads of the "real" imperfect
structure
• Materials and geometry
• The rules in EN 1993-1-6 are not limited to steel shell structures
• The standard is valid for isotropic shells and shell segments made from any materials
that may be represented as ideal elastic-plastic
• For materials with no well defined yield point, 0.2% proof stress can be taken
• The material properties apply to temperatures not exceeding 150 ºC (otherwise see EN
13084-7, 2005)
• Where materials has a significant different stress strain curve, there are alternative
ways of representation of the material behaviour
• Bauschinger effect
• For austenitic steels (and aluminium alloys) at higher plastic strains, Rasmussen (2003)
curve is more appropriate than Ramberg-Osgood
• Materials and geometry
• The rules in EN 1993-1-6 are not limited to steel shell structures
• The standard is valid for isotropic shells and shell segments made from any materials
that may be represented as ideal elastic-plastic
• For materials with no well defined yield point, 0.2% proof stress can be taken
• The material properties apply to temperatures not exceeding 150 ºC (otherwise see EN
13084-7, 2005)
• Where materials has a significant different stress strain curve, there are alternative
ways of representation of the material behaviour
• Bauschinger effect
• For austenitic steels (and aluminium alloys) at higher plastic strains, Rasmussen (2003)
curve is more appropriate than Ramberg-Osgood
• Geometrical tolerances and imperfections
• Relevant tolerances due to the requirements of serviceability:
• out-of-roundness (deviation from circularity)
• eccentricities (deviations from a continuous middle surface in the direction
normal to the shell along junctions of plates)
• local dimples (local normal deviations from the nominal middle surface)
• Other forms of geometric imperfections:
• deviations from nominal thickness
• lack of evenness of supports
• Material imperfections:
• residual stresses caused by rolling, pressing, welding, straightening etc.
• inhomogeneities and anisotropies
• Wear and corrosion
• Non-uniformities of loading
• Residual stresses
• Geometrical tolerances and imperfections
• Relevant tolerances due to the requirements of serviceability:
• out-of-roundness (deviation from circularity)
normal to the shell along junctions of plates)
• local dimples (local normal deviations from the nominal middle surface)
• Other forms of geometric imperfections:
• deviations from nominal thickness
• lack of evenness of supports
• Material imperfections:
• residual stresses caused by rolling, pressing, welding, straightening etc.
• inhomogeneities and anisotropies
• Wear and corrosion
• Non-uniformities of loading
• Residual stresses
• Ultimate limit states in steel shells
• LS1: Plastic limit
• Identifies the strength of the structure when stability plays no significant
role.
• Covers two conditions:
• tensile rupture or compressive yield through the full thickness
• development of a plastic collapse mechanism involving bending
• The plastic limit load is also relevant to a buckling strength assessment
Rpl - the plastic limit load
Rcr - the elastic critical load
• The plastic limit load does not represent the real strength (even for
stocky structures): strain hardening of material, stabilizing or
destabilizing effects due to change in geometry
• MNA: often underestimates the strength very considerably
• Membrane theory calculations:

• If the stress state is entirely axisymmetric, it gives a close
approximation to the true condition at plastic collapse
• if the stresses are significantly unsymmetrical, this criterion

often provides a very conservative estimate of the plastic limit
load
• Linear elastic shell bending theory: This is commonly more
conservative than membrane theory calculation (is based on
the first yield on the surface)
• Geometrically nonlinear calculation (GMNA): problems arise
over whether the structure displays geometric hardening or
geometric softening
• The plastic limit load should be seen only as the ideal value of
the plastic reference resistance
• LS2: Cyclic plasticity:
• Repeated cycles of loading and unloading, eventually leading to local
cracking by exhaustion of the energy absorption capacity of the material
• Low cycle fatigue failure may be assumed to be prevented if the
procedures set out in the standard are adopted
• Methods of analysis:
• expressions in Annex C
• elastic analysis (LA or GNA)
• MNA or GMNA and find plastic strains
• LS4: Fatigue:
• Repeated cycles of increasing and decreasing stress lead to the
development of a fatigue crack
• Methods of analysis:
• expressions in Annex C (using stress concentration factors)
• elastic analysis (LA or GNA), using stress concentration factors
• LS3: Buckling:
• Caused by loss of stability under compressive membrane or shear
membrane stresses in the shell wall, leading to inability to sustain any
increase in the stress resultants, possibly causing catastrophic failure
• Three approaches used in the assessment of buckling resistance:
• GMNIA analysis
• MNA/LBA analysis
• Buckling stresses
• The strength under LS3 depends strongly on the quality of construction
• For this purpose, three fabrication quality classes are set out
• Design concepts for the limit states design of shells
• The limit state verification should be carried out using one of the following:
• Stress design:
• primary In general,
may beprimary
replacedstress statesresultants
by stress control LS1, whereas
where
• secondary secondary stress states affect LS2 and LS3 and local
appropriate
• Local stresses govern LS4.
• Direct design by application of standard expressions:

• the limit states may be represented by standard expressions that have been
derived from either membrane theory, plastic mechanism theory or linear elastic
analysis
• The membrane theory (Annex A) - primary stresses needed for assessing LS1 and
LS3.
• The plastic design (Annex B) - plastic limit loads for assessing LS1
• The linear elastic analysis (Annex C) - stresses of the primary plus secondary
stress type for assessing LS2 and LS4. An LS3 assessment may be based on the
membrane part of these expressions.
• Design by global numerical analysis

• Stress resultants and stresses in shells
• Stresses:
• There are eight stress resultants in the shell
• However, the shear stresses xn, θn due to the transverse shear forces qxn, qθn are
insignificant and they may usually be neglected in design
• For most design purposes, the evaluation of the limit states may be made using
only the six stress resultants in the shell wall nx, nθ, nxθ, mx, mθ, mxθ
• Where the structure is axisymmetric and subject only to axisymmetric loading and
support, only nx, nθ, mx and mθ need be used
• Modelling:
• Represention by its middle surface
• Nominal radius of curvature, imperfections neglected (excepting LS3)
• Eccentricities and steps if they induce significant effects
• Eccentricity at junctions between shell segments
• Stringers, corrugations, holes, depending on the conditions
• Boundary conditions
Boundary Simple Description Normal Vertical Meridional
condition term displacements displacements rotation
code
radially restrained meridionally
Introduction
BC1r to Design of Shell Structuresw = 0
Clamped restrained rotation restrained u=0 βφ = 0
Design ofradially
Steel Structures:
restrained meridionally Strength and Stability of Shells
• Stress resultants
BC1f restrained rotation free w=0 u=0 βφ ≠ 0
and stresses in shells
• Stresses:radially restrained meridionally free
BC2r w=0 u≠0 βφ = 0
• Thererotation restrained
are eight stress resultants in the shell
• However, the shear stresses xn, θn due to the transverse shear forces qxn, qθn are
radially restrained meridionally free
BC2f Pinned insignificant
rotation freeand they may usually be neglected w = 0 in designu ≠ 0 βφ ≠ 0
• For most design purposes, the evaluation of the limit states may be made using
BC3
Free edgeonlyradially
the six stress
free resultants
meridionally in the shell wall
free rotation n , n , n , m , m , mxθ βφ ≠ 0
w ≠ 0 x θ xθ ux≠ 0 θ
free
• Where the structure is axisymmetric and subject only to axisymmetric loading and
NOTE: The circumferentialsupport,
displacement
only v isnclosely linked to the displacement w normal to the surface so separate
x, nθ, mx and mθ need be used
boundary conditions are not identified in paragraph (3) for these two parameters.
• Modelling:
• Represention by its middle surface
• Nominal radius of curvature, imperfections neglected (excepting LS3)
• Eccentricities and steps if they induce significant effects
• Eccentricity at junctions between shell segments
• Stringers, corrugations, holes, depending on the conditions
• Boundary conditions
• Plastic limit state (LS1)
• The plastic reference resistance Rpl
• Where it is not possible to undertake a materially non-linear analysis (MNA),
the plastic reference resistance Rpl may be conservatively estimated from
linear shell analysis (LA) conducted using the design values of the applied
combination of actions using the following procedure.
• The three points where

obtain maximum
values for stresses
t f y ,k
Rpl 
nx2, Ed  nx, Ed n , Ed  n2, Ed  3nx2 , Ed
• Buckling limit state (LS3)
• To find out the design buckling resistance
• Defined as a load factor R applied to the design values of the combination of
actions for the relevant load case
• Different approaches have been proposed, difficult to generalise
• In EN 1993-1-6, a considerable effort to produce general procedures
applicable to all geometries, all loading conditions and all material conditions
• Buckling-relevant boundary conditions

• Buckling-relevant geometrical tolerances
• Stress design
• Design values of stresses σx,Ed, σθ,Ed and xθ,Ed: taken as the key values of
compressive and shear membrane stresses obtained from linear shell analysis
(LA).
• Design resistance (buckling strength):
λ0 - squash limit relative

slenderness
x  f y,k  x, Rcr
   f y,k   ,Rcr
  f y,k 
3  x ,Rcr
γM1 = min 1,1

• Stress limitation (buckling strength verification)
• The influence of bending stresses may be neglected provided they arise as a
result of boundary compatibility effects.
• In the case of bending stresses from local loads or from thermal gradients,
special consideration should be given.
• Following checks for the key values of single membrane stress components
should be carried out:
σx,Ed ≤ σx,Rd, σθ,Ed ≤ σθ,Rd, xθ,Ed ≤  xθ,Rd
• For more than one buckling-relevant membrane stress components,

interaction check for the combined membrane stress state should be carried
out:
kx k k
  x,Ed    ,Ed    x,Ed   ,Ed    x, ,Ed 
     ki     1
  x,Rd    ,Rd    x,Rd   ,Rd    x, ,Rd 
        
• Buckling design
• EN 1993-1-6 specifies three approaches that are approved for use in the
assessment of buckling resistance:
• Design by means of a global numerical MNA/LBA analysis
• Design by means of a global numerical GMNIA analysis
• Design by means of buckling stresses
• Buckling design by global numerical MNA/LBA analysis
• It is recommended for many applications
• It has the same basis as the traditional stress design buckling approach
• All relevant combinations of actions causing compressive membrane stresses
or shear membrane stresses in the shell wall shall be taken into account
• It involves the following steps, see left hand side figure

ov  f  ov ,  ov,0 ,  ov , ov ,  ov 
ov is the overall elastic imperfection factor,
ov is the plastic range factor,
ov is the interaction exponent and
ov,0 is the squash limit relative slenderness
• Buckling design by global numerical MNA/LBA analysis
• It is recommended for many applications
• It has the same basis as the traditional stress design buckling approach
• All relevant combinations of actions causing compressive membrane stresses
or shear membrane stresses in the shell wall shall be taken into account
• It involves the following steps, see left hand side figure
The lowest eigenvalue

(bifurcation load factor)
should be taken as the critical
buckling resistance Rcr

ov  f  ov ,  ov,0 ,  ov , ov ,  ov 
ov is the overall elastic imperfection factor,
ov is the plastic range factor,
ov is the interaction exponent and
ov,0 is the squash limit relative slenderness
• Design by global numerical GMNIA analysis
• Developed to exploit the full power of modern numerical
C1: The analysis
maximum load factor on the
load-deformation-curve (limit load);
• Application is more complex than for frame or platedC2:structures
The bifurcation load factor, where
• Several sequence of analysis: this occurs during the loading path
before reaching the limit point of the
• LA followed by a LBA to evaluate elastic critical load-deformation-curve
buckling resistance
C3: The largest tolerable deformation,
• GMNA to identify the elastic-plastic buckling resistance of the
where this occurs perfect
during the loading
structure path before reaching the bifurcation
load or the limit load
• GMNIA with different imperfection modes (the lowest value is selected)
C4: The load factor at which the
• Check the precision of the GMNIA by comparison with test
equivalent oratother
stress the most highly
relevant data stressed point on the shell surface
reaches the design value of the
• Methodology yield stress
• Action combinations causing compressive membrane stresses or shear
membrane stresses
• Rk should be found from the imperfect elastic-plastic critical buckling
resistance RGMNIA, adjusted by the calibration factor kGMNIA.
• The design buckling resistance Rd should then be found using the partial
factor γM1.
• Design by global numerical GMNIA analysis
• Developed to exploit the full power of modern numericalC1: The analysis
maximum load factor on the
load-deformation-curve (limit load);
• Application is more complex than for frame or platedC2:structures
The bifurcation load factor, where
• Several sequence of analysis: this occurs during the loading path
before reaching the limit point of the
• LA followed by a LBA to evaluate elastic critical load-deformation-curve
buckling resistance
C3: The largest tolerable deformation,
• GMNA to identify the elastic-plastic buckling resistance of the
where this occurs perfect
during the loading
structure path before reaching the bifurcation
load or the limit load
• GMNIA with different imperfection modes (the lowest value is selected)
C4: The load factor at which the
• Check the precision of the GMNIA by comparison with test
equivalent oratother
stress the most highly
relevant data stressed point on the shell surface
reaches the design value of the
• Methodology yield stress
• Action combinations causing compressive membrane stresses or shear
membrane stresses
• Rk should be found from the imperfect elastic-plastic critical buckling
resistance RGMNIA, adjusted by the calibration factor kGMNIA.
• The design buckling resistance Rd should
A conservative then be found
assessment using
of RGMNIA maythe partial
be obtained
factor γM1. using a GNIA analysis and C4 criterion to determine
the lowest load factor R
• GMNIA analysis
• Allowances for imperfections:
• geometric imperfections: pre-deformations, out of-roundness, irregularities at
and near welds, thickness deviation, etc.
• material imperfections: residual stresses, inhomogeneities, anisotropies
• EN 1993-1-6 requires that imperfections are explicitly modelled numerically,
not just treated as small perturbations in geometry
• They are introduced by means of equivalent geometric imperfections in the
form of initial shape deviations perpendicular to the middle surface of the
perfect shell
• The form of the imperfections with the most unfavorable effect should be
considered (the most unfavorable effect on the buckling resistance RGMNIA
of the shell); if practicable, they must reflect the constructional detailing and
the boundary conditions
• GMNIA analysis
• The analysis should be carried out for a sufficient number of different
imperfection patterns, and the worst case (lowest value of RGMNIA) should be
identified.
• The eigen-mode-affine pattern should be used (the critical buckling mode
associated with the elastic critical buckling resistance Rcr based on an LBA
analysis of the perfect shell)
• The amplitude of the imperfection form - dependent on the fabrication
tolerance quality class
• Imperfections
• The maximum deviation of the geometry of the equivalent imperfection from
the perfect shape ∆w0,eff = max (∆w0,eff,1; ∆w0,eff,2), where:
lg relevant gauge lengths

w0,eff ,1  lg U n1 lgx  4 rt t local shell wall thickness
w0,eff ,2  ni tU n 2   ni multiplier to achieve an appropriate tolerance level
0.25
lg ,  2,3 l rt
2
r Un1, Un2 dimple imperfection
• ni = 25 is a multiplier to achieve
an appropriate tolerance level
• t is the local shell wall thickness
• lg is all relevant gauge lengths
(see Dimple tolerances)
• GMNIA validation
• For each calculated value of the buckling resistance RGMNIA, the ratio of the
imperfect to perfect resistance (RGMNIA / RGMNA) should be determined and
compared with values of  found using the procedures of 8.5 and Annex D.
• The reliability of the numerically determined critical buckling resistance RGMNIA
should be checked by one of the following methods:
• by using the same program to calculate values RGMNIA, check for other shell buckling
cases for which characteristic buckling resistance values Rk,known,check are known;
• by comparison of calculated values (RGMNIA,check) against test results
(Rtest,known,check).
Rk ,know,check Rtest ,know,check

kGMNIA  kGMNIA 
RGMNIA,check RGMNIA,check
• GMNIA validation
• Where a known characteristic value based on existing established theory is
used to determine kGMNIA, and the calculated value of kGMNIA lies outside the
range 0,8 < kGMNIA < 1,2, this procedure should not be used.
• The characteristic buckling resistance should be obtained from:
Rk  kGMNIA RGMNIA
• RGMNIA is the calculated imperfect elastic-plastic critical buckling resistance;
• kGMNIA is the calibration factor.
• ANNEX A - Membrane theory stresses in shells
Uniform axial Axial load from Uniform shear

load bending from torsion
• ANNEX B - Additional expressions for plastic collapse resistances
Cylinder: Cylinder:
Radial line load Radial line load and axial load
A = +/- sx − 1,50
P
sx  x
fyt
PnR - plastic resistance (force per unit circumference)

• ANNEX C - Expressions for linear elastic membrane and bending stresses
Cylinder, clamped: Cylinder, pinned:
axial loading axial loading
• ANNEX D - Expressions for buckling stress design
• Unstiffened cylindrical shells of constant wall thickness
• Meridional (axial) compression
• Critical meridional buckling stresses
• Meridional buckling parameters
• Circumferential (hoop) compression
• Critical circumferential buckling stresses
• Circumferential buckling parameters
• Shear
Cylinder geometry, membrane • Critical shear buckling stresses
stress resistances and stress • Shear buckling parameters
resultant resistances
• Combinations of meridional
(axial) compression,
circumferential (hoop)
compression and shear
Finite Element Application
• Principles, simplified and advanced models, concentration of stresses,
stiffening
Hand calculation Simple computer Complete computer

calculations calculations
Formulas for: Computational evaluations: Computational evaluations:
Bifurcation load Bifurcation load Bifurcation load

Plastic limit Plastic limit Plastic limit
Imperfection sensitivity Interaction between different stress Perfect structure strength
Elastic plastic interaction components Imperfect structure strength
Interaction between different
stress components
Formulas for: LA GNIA
LBA GMNIA
Imperfection sensitivity GNA
Elastic plastic interaction MNA
GMNA
• Simplified FEM model
• Global analysis
• A simplified bar model using 1D beam finite elements, with or without
the account of global imperfection, using an elastic material law.
• Aim of analysis – find-out the internal forces needed for:
• The global check of the tower capacity;
• Refined local analysis of the relevant tower segments (between two
consecutive flange that assure the appropriate boundary condition, radial
restrained)
• Local analysis of a tower segment
• The 3D shell model should be build using 2D finite elements (plane), that
can be: either Homogeneous shell, either Membranes
• Simplified FEM model
• Local analysis of a tower segment
• The 3D shell model should be build using 2D finite elements (plane), that
can be: either Homogeneous shell, either Membranes
• The geometry can be perfect / ideal or taking into account of the
prescribed relevant geometrical imperfection:
• out-of-roundness (deviation from circularity),
normal to the shell along junctions of plates),
• local dimples (local normal deviations from the nominal middle surface).
• Advanced FEM Model
• Modelling the entire structure with 2D finite elements with a special
attention of details (connection flanges, welds connection, etc);
• Material behavior:
• Elastic
• Plastic
• Geometry:
• Ideal
• Imperfect
• Finite elements types:
• Homogenous shell
• Geometry and state of stress
• Homogeneous shell with bending stiffness, sections consist of a shell
thickness, material name, section Poisson's ratio.
• Membranes represent thin surfaces in space that offer strength in the
plane of the surface but have no bending stiffness. Membrane
sections consist of a material name, membrane thickness, section
Poisson's ratio.
• Material behavior law
• Elastic
• Elastic – Plastic
• Linear
• Multi-linear
• Continuous
• Powell,
• Ramberg-Osgood
• Load cases
• The load cases shall be determined from the combination of operational
modes or other design situations, such as specific assembly, erection or
maintenance conditions, with the external conditions.
• All relevant load cases with a reasonable probability of occurrence

shall be considered, together with the behavior of the control and
protection system.
• The design load cases used (IEC 61400-1:2005) to verify the structural integrity
of a wind turbine shall be calculated by combining:
• normal design situations and appropriate normal or extreme external
conditions;
• fault design situations and appropriate external conditions;
• transportation, installation and maintenance design situations and
appropriate external conditions.
• Load cases and design situations
• Load cases and assumptions for global analysis
• IEC 61400-1:2005
• Dead loads - Self-weight: tower head; tower body, installation etc.
• Wind action (EN1991-1-4)
• Seismic loads
• Temperature
• Ice
• Design situations
• Power production
• Power production plus occurrence of fault or loss of electrical network
connection
• Start up
• Normal shut down
• Emergency shut down
• Parked (standstill or idling)
• Parked plus fault conditions
• Transport, assembly, maintenance and repair
• Wind action using different load situation:
• Global analysis to find-out the internal forces
• Load of wind on tower (hub)
• Simplified distribution (a2)
• Surface distribution (a1)
• Load from turbine’s machinery
• Concentrated force and moment
• Equivalent load for buckling verification

• Axisymmetric pressure distribution
a1
a2
• Solver technique
• Pre-critical analysis and point results (Newton-Raphson)
• Post-critical analysis with deformation–to-failure (displacement
control, arc-length, modal analysis)
• The Static Riks step (based on arc-length solver) is able to find solution
during unstable loading response, when Static General step (based on N-R
solver) stops at limit load.
Plim
• Results: Stability of steel shell structure General Report – Herbert Schmidt
(JCSR 55(2000) 159-181)
Eigenvalue results pure bending
Eigenvalue results axial compression different discretization levels
different discretization levels
Eigenvalue results axial compression

different boundary conditions
• Stress concentrations
• Modelling stress concentration (door opening)
• Around door opening appear elevated values of the membrane
stresses. This stresses that develop within a band of width of 2 √rt
adjacent to a restrained edge need not be consider in buckling
calculations
Finite Element Application and Case Study
• Study Case
• General geometrical data
• Total height: 96.15 m
• Shape: cylinder (conic last segment)
stepwise variable wall thickness
• Base diameter: 4300 mm (t=39 mm)
• Top diameter: 2955 mm (t=12 mm)
• Divided in 5 segments of ~ 20m length
• Detailing
• Flange bolted connection (M48 - M27)
between segments and welded connection
between shells of different thickness
• Study Case
• Door opening details
• Ventilation opening
• Study Case
• Numerical models
• Simplified bar model
• Refined local segment model
• Relevant segments
• Door opening segment
• Entire model
• Verification procedures
• Analytical determination of moment capacity (LA) →MRd
• Characteristic buckling resistance (LBA + MNA)
• The plastic reference resistance → Rpl
• The elastic critical buckling resistance → Rcr
• The overall buckling reduction factor Rk  ov Rpl
• Characteristic buckling resistance (GMNIA)
• Calibration factor Rk  kGMNIA RGMNIA
• Study Case
• Expressions for buckling stress design
• Unstiffened cylindrical shells
• Critical meridional buckling stresses
l r l 20000
    119
r t rt 2150 13,14
t 12 N
 x , Rcr  0, 605  E  cx   0, 605  2,1105  0,943   669
r 2150 mm 2
r 2150 0, 2  t
  164  cx  cx , N  1   1  2 
t 13,14 cxb  r
0, 2  12 
cx , N  1  1  2 119   0,9434  0, 6
1  2150 
• Study Case
• Expressions for buckling stress design (hand calculation)
• Meridional buckling parameters
0, 62 0, 62
x    0, 25999
 w 
1,44 1,44
 10,51 
1  1,91 k  1  1,91 
 t   13,14 
1 r 1 2150
wk  t  13,14  10,51
Q t 16 13,14
 x 0  0, 20

   0, 60   p  0,806  x  0, 728  0,806
  1, 0

 0, 728  0, 2  N
  1  0, 60     0, 477   x , Rk  0, 477  355  169
 0,806  0, 2  mm 2
• Study Case
• Meridional stresses
Fx 1770 103 N
 x , Ed  
N
  10,92
2 rt 2  2150 12 mm 2
M 17702 106 N
 x , Ed   2 
M
 102
 r t   21502 12 mm2
• Study Case
• Critical circumferential buckling stresses
l r l 20000
    119
r t rt 2150 13,14
c t 1 12 N
  , Rcr  0,92  E    0,92  2,1105    9, 06
 r 119 2150 mm 2
r
1, 63   267,32
t
• Study Case
• Circumferential buckling parameters
  0,5 (Class C )
  0  0, 40

   0, 60   p  1,12    6, 26  1,12
  1, 0

 0,5 N
   0, 0128    , Rk  0, 0128  355  4,53
 2 6, 26
2
mm 2
r 2150 N
  , Ed  pn  1,5 1,102 103  0, 296
t 12 mm 2
• Study Case – Complete Model
• Imperfection amplitude
lgx  4 rt  4  4300 12  908,63 mm
w0,eff ,1  lg U n1
Fabrication tolerance Geometric tolerance

Value
Description normal to the
quality class of Un1
shell surface
Class A Excellent 0,010 9,086
Class B High 0,016 14,538
Class C Normal 0,025 22,7158
• LA / MNA at reference load results
• Local plastic zones around
connecting flanges
• LBA results Rcr = 5,877
Axial Shear Bending

Bottom
force force moment
section
[kN] [kN] [kNm]
LA 4323 1260 80088
MNA 4323 1260 80088
• MNA Rpl = 3,07
• GMNA Rpl = 1,77
• GMNIA Rpl = 1,49
3.5
2.5
Load factor R
1.5
1
GMNA
0.5 MNA
GMNIA
Reactions Axial Shear Bending
0 at force force moment
0 1 2 3 4 5 6
[kN] [kN] [kNm]
Displacem ent [m ]
R=1 4323 1260 80088

• Description of the complete shell model
• LBA - Linear elastic bifurcation analysis
• The elastic critical buckling load factor Rcr = 5,8772
• Imperfection shape
• MNA – Material nonlinear analysis

• The plastic reference load factor Rpl = 3,07
• The overall relative slenderness ov for the complete shell

3, 07
ov   0,522
5,877
• The overall buckling reduction factor

• Annex D EN1993-1-6 gives values for:
• ov is the overall elastic imperfection factor
• ov is the plastic range factor = 0,60
• ov is the interaction exponent = 1,0
• Study Case – Refined door opening segment
• Numerical model LBA results Rcr = 12.248
Axial force Shear force Bending moment

Section
[kN] [kN] [kNm]
Upper section 4413 1231 73988
Bottom section 4413 1231 88514
• LA Results
• GMNA Results
Bottom segment
2.5
2
Load factor
1.5
0.5
GMNA
0
0 50 100 150 200 250 300
Displacement [mm]
• Numerical model
Section Axial force Shear force Bending moment
[kN] [kN] [kNm]
Upper section 1586 409 13055
Bottom section 2015 666 23897
• LA Rpl = (126/355)=2,82
• Transformation of typical wind pressure load distribution
 c r   1 2150 
kw  0, 46 1  0,1   0, 46 1  0,1    0,514
  t   119 13,14 
l r l 20000
    119
r t rt 2150 13,14
N
qeq  kw qmax  0,5411430 1,5  1102
m2
• LBA Results • GMNIA results
• Imperfection afine first buckling mode
• Amplitude of imperfection 23mm and 17mm
corresponding to normal and high tolerance
Relevant segment
2
kGMNIA
1.8
1.6
1.4
1.2
Section Rcr
Load factor
Without EWL 6,8006 0.8
With EWL 6,8003 0.6
0.4
Quality class kGMNIA

0.2 GMNIA normal
GMNIA high
0
High (~17 mm) 1,79 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016
Rotation [rad]
Normal (~23 mm) 1,72
DESIGN AND ANALYSIS OF SHELL S1RUCTURES
SOLID MECHANICS AND ITS APPLICATIONS
Volume 16
Series Editor:
G.M.L. GLADWELL
Solid Mechanics Division, Faculty of Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3GJ
Aims and Scope of the Series

The fundamental questions arising in mechanics are: Why?, How?, and How much?
The aim of this series is to provide lucid accounts written by authoritative research-
ers giving vision and insight in answering these questions on the subject of
mechanics as it relates to solids.
The scope of the series covers the entire spectrum of solid mechanics. Thus it
includes the foundation of mechanic:;; variational formulations; computational
mechanics; statics, kinematics and dynamics of rigid and elastic bodies; vibrations
of solids and structures; dynamical systems and chaos; the theories of elasticity,
plasticity and viscoelasticity; composite materials; rods, beams, shells and
membranes; structural control and stability; soils, rocks and geomechanics;
fracture; tribology; experimental mechanics; biomechanics and machine design.
The median level of presentation is the first year graduate student. Some texts are
monographs defining the current state of the field; others are accessible to final
year undergraduates; but essentially the emphasis is on readability and clarity.
For a list of related mechanics titles, see final pages.

Design and Analysis
of Shell Structures
by
M.FARSHAD
EMPA,
Switzerland
~·
''
Springer-Science+Business Media, B.V.
Library of Congress Cataloging-in-Publication Data
Farshad. Mehd 1.
Design and analysis of shell structures l'fehdi Farshad.
p. cm. -- 1So11d mechanics and its applications ; 161
Inc 1udes index.
ISBN 978-90-481-4200-2 ISBN 978-94-017-1227-9 (eBook)
DOI 10.1007/978-94-017-1227-9
1. Shells IEngineer1ngl--Des1gn and construction. 2. Structural
analysis <Engineering> I. T1tle. II. Title: Shell structures.
III. Series.
TA660.S5F42 1992
624.1'7762--dc20 92-18175
ISBN 978-90-481-4200-2
Printed on acid-free paper
All Rights Reserved

© 1992 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1992
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
Contents
Preface xi
Chapter 1 - Introduction to Shells 1

1.1 - Introduction 1
1.2 - Uses of Shell Structures 2
1.3 - Geometry of Shells 3
1.4 - Classification of Shell Surfaces 4
1.5 - Summary of Classification of Shell Surfaces 8
1.6 - Outline of General Structural Features of Shells 8
References for Chapter One 9
Chapter 2 - Preliminaries of Shell Analysis and Design 11

2.2 - Thin Shells 12
2.3 - Internal Force System in a Shell 12
2.4 - Qualitative Description of Shell Behavior 15
2.5 - An Overview of Shell Theories 19
2.6 - Assumptions of Classical Shell Theories 19
2.7 - Force Method of Shell Analysis 20
2.8 - General Shell Design Considerations 21
2.9 - Stability Considerations in Shell Design 22
2.10 - Codes of Practice 23
References for Chapter Two 24
Chapter 3 - Membrane Behavior of Cylindrical Shells 25

3.2 - Geometrical Description 26
3.3 - Membrane Equations ofCylind1ical Shells 26
3.4 - Cylindrical Vaults 28
3.5 - Containment Vessels 30
3.5.1-Fluid Tanks 30
3.5.2-Pipes under Internal Pressure 30
3.6 - Membrane Defo1mation of Cylindrical Shells 31
3. 7 - Displacements of Cylindrical Vaults 33
vi
Numerical Example 3.1 33

3.8 - Qualitative Description of Cylindrical Shells Behavior 34
Problems 38
References for Chapter Three 42
Chapter 4 - Bending Analysis of Circular Cylindrical Shells 43

4.2 - General Governing Equations 44
4.2.1-Equilibrium Equations 44
4.2.2-Kinematic Relations 45
4.2.3-Constitutive Relations 47
4.3 - Displacement Equations of Circular Cylind1ical Shells 50
4.4 - Circular Cylinders with Axisymmetric Loading 51
4.5 - Some Axisymmetric Problems of Circular Cylinders 53
4.5.1-General Solution to Axisymmetric Problems 53
4.5.2-A Fundamental Problem 54
4.5.3-Thin Circular Cylinders under Internal Pressure 55
4.5.4-Analysis of Liquid Retaining Cylindrical Shells 56
Problems 60
References for Chapter Four 61
Chapter 5 - Design of Concrete Cylindrical Shell Roofs 63

5.2 - Geometric Design of Cylind1ical Shell Roofs 64
5.2.1-0verall Dimensioning of the Shell 64
5.2.2-Dimensions of Edge Beams 67
5.2.3-Profile, Central Angle, and Curvature of the Shell 67
5.2.4 -Rise of the Shell 67
5.2.5-Shell Thickness 67
5.3 - Reinforcement of Concrete Cylindrical Roofs 68
5.4 - "Beam-Arch" Method of Vaulted Roof Analysis 72
5.5 - Analysis of Cylindrical Vaults by ASCE Tables 87
5.6 - Design Examples of Reinforced Concrete Cylindrical Shell Roofs 89
5.6.1-Design of A Single Shell without Edge Beams 89
5.6.2-Design of an Inner Shell in a Shell Group 92
5.6.3-Design of a Single Simply Supported Shell with Edge Beams 95
5.6.4-Design of an Inner Shell in a Shell Group with Edge Beams 97
Problems 100
References for Chapter Five 101
Chapter 6 - Membrane Analysis of Shells of Revolution 103

6.3 - Governing Membrane Equations 106
6.4 - Rotational Shells with Axisymmettic Loading 108
vii
6.5 - Spherical Domes Ill

6.5.1-Membrane Forces Ill
6.5.2-Domes with Skylight 112
6.6 - Fluid Storage Tanks 115
6.6.1-Spherical Liquid Storage Tank 115
6.6.2-Cylindrical Tanks with Spherical Ends 116
6.6.3-Pressure Vessels 120
6.7 - Shells of Revolution with Nonsymmetric Loading 121
6.8 - Wind-Induced Stresses in Domes 122
6.9 - Displacements of Axisymmetric Shells 125
6.10 - Membrane Deformation of Spherical Domes 128
6.11 - Qualitative Description of Dome Behavior 130
6.12 - Conical Shells 133
Problems 137
References for Chapter Six 139
Chapter 7 - Bending Analysis of Axisymmetric Shells 141

7 .2 - Governing Equations of Axisymmetric Shells I42
7 .2.1-Equilibrium Equations 142
7 .2.2-Kinematical Relations 144
7 .2.3-Constitutive Relations 148
7.3 - Reduction of Shell Equations 149
7 .4 - Edge Effects in Axisymmetric Shells 153
7.5 - Analysis of Axisymmetric Shells for Edge Effects 155
7 .6 - Influence Coefficients for Axisymmetric Shells 158
7.7 - Force Method of Axisymmetric Shells Analysis 160
7.8 - Sample Analysis of a Dome 160
Problems 166
References for Chapter Seven 167
Chapter 8 - Design of Reinforced Concrete Domes 169

8.2 - Domes with Rings 170
8.3 - Force Method of "Dome-Ring" Analysis 171
8.3.1-General Methodology 171
8.3.2-Analysis of the Ring 174
8.3.3- Analysis of Dome under Edge and Distributed Forces 177
8.3.4-"Dome-Ring" Interaction 177
8.3.5-Summary of "Dome-Ring" Analysis Relations 181
8.3.6-Application of the Force Method 183
8.4 - Buckling Considerations in Concrete Domes Design 184
8.5 - Design Guides for Dome Geometry 184
8.6 - Design of a Reinforced Concrete "Dome-Ring" Roof 186
Problems 193
References for Chapter Eight 194
viii
Chapter 9 - Analysis of Shells with Arbitrary Geometry 195

9.2 - Membrane Theory of General Shells 197
9.2.1-Geometrical Description of Arbitrary Surfaces 197
9 .2.2-Methodology of Membrane Analysis of General Shells 199
9 .2.3-Equilibrium Equations of General Shells 202
9.2.4-Solution of Membrane Equations by Stress Function 205
9.3 - Bending Theory of Shallow Shells 205
Problems 212
References for Chapter Nine 214
Chapter 10 - Design of Hyperbolic Paraboloid Shells 215

10.3 - Membrane Analysis of HP Shells 222
10.4 - Description of Membrane Behavior of HP Shells 227
10.5 - Bending Filed in Hyperbolic Paraboloid Shells 232
10.6 - General Design Considerations of HP Shells 234
10.7 - Design of a Reinforced Concrete HP Shell Roof 239
Problems 245
References for Chapter Ten 247
Chapter 11 - Analysis and Design of Folded Plates 249

11.2 - General Features of Folded Plates 250
11.3 - General Design Considerations of Prismatic Folded Plates 254
11.4 - Methodology of Analysis of Folded Plates 256
11.4.1-Preliminary Analysis 257
11.4.2-Corrective Analysis 258
11.4.3-Compatibility Analysis and Superposition of Results 258
11.5 - Basic Steps in the Folded Plates Analysis 259
11.5.1-Schematics of Analysis 259
11.5.2-Shear Stresses at Plates Junctions 260
11.5.3-Distribution of Junctional Shear Forces 262
11.5.4-Summary of Analysis Procedure 265
11.6 -Analysis of a RC Folded Plate Roof 265
Problems 288
References for Chapter Eleven 289
Chapter 12 - Design of Liquid Retaining Shells 291

12.2 - Classifications of Liquid Containers 292
12.3 - General Design Considerations 292
12.3.1-Shape Design 292
ix
12.3.2-Serviceability Design 296

12.4 - Loading Conditions 297
12.5 - Axisymmetric Behavior of Circular Containers 299
12.6 - Force Method of Container Shells Analysis 304
12. 6.1-Influence Coefficients 304
12.6.2-Analysis of "Wall-Base" Interaction 305
12.6.3-Analysis of "Wall to Roof" Connection 311
12.7 - An Example of Cylindrical Container Analysis 319
12.8 - Design of a Reinforced Concrete Container 326
12.9 - Some Considerations on Reinforcements Detail 329
12.10 - Cylindrical Walls with Domical Roofs 331
12.11 - Design of a Cylindrical "Wall-Dome" Container 333
Problems 340
References for Chapter Twelve 341
Chapter 13 - Buckling of Shells 343

13.2 - Concepts of Stability and Instability 344
13.3 - Types of Loss of Stability 345
13.3.1-Bifurcation of Equilib1ium 345
13.3.2-Limitation of Equilibrium 349
13.4 - An Overview of Shell Buckling 349
13.5 - Methodology of Linear Stability Analysis of Shells 352
13.6 - Buckling of Circular Cylindrical Shells 353
13.7 - Buckling of Circular Cylinders under Axial Force 360
13.7.1-General Modes of Cylinders Buckling under Axial force 360
13.7 .2-Buckling of Axially Loaded Cylindrical Shells 362
13.7 .3-Buckling of Cylind1ical Shells under External Pressure 366
13.8 - Buckling of Concrete Cylind1ical Roofs 367
13.9 - Buckling Formulas for the Shells of Revolution 368
13.10 - Buckling of Domes 368
13.10.1-Buckling Modes of Domes 368
13.10.2-Buckling of Concrete Domes 369
13.11 - Buckling of Hyperbolic Paraboloid Shells 370
13.11.1-General Buckling Behavior of HP Shells 370
13.11.2-Buckling Formulas for Hyperbolic Paraboloid Shells 371
Problems 372
References for Chapter Thirteen 374
Appendixes 375
Appendix A - Formulas for the Membrane Field in Shells

Appendix B -Tables For Analysis of Circular Cy lind1ical Shell Roofs
Subject Index
Preface
Shell Structures present immense structural and architectural potential in various fields of
civil, mechanical, architectural, aeronautical, and marine engineering. Examples of shell
structures in civil and architectural engineering are: varieties of concrete shell roofs, liquid
retaining structures and water tanks, concrete silos, cooling towers, containment shells of
nuclear power plants, and concrete arch dams. In mechanical engineering, shell forms are
used in piping systems, curved panels, and in pressure vessel technology. Aircrafts,
spacecrafts, missiles, ships, and submarines are examples of shells used in aeronautical and
marine engineering. Shells are found in various biological forms such as the eye and the
skull, plants, and animal shapes. Thus, another application of shell engineering would be the
field of Biomechanics.
Shell structures developed since ancient times and now are being increasingly used in various
industries. Shells are used in the covering of large spans, liquid retaining installations, silos,
and containment structures. They are also used in the construction of light-weight vehicles,
pressure vessels, and space structures. Advent of such materials as ferro-cement, fiber-
reinforced concrete, composite materials, and reinforced polymers have all enhanced the
domain of shell technology. With the development of new prefab1ication schemes as well as
the need for recycling of materials, the potential of shell applications has further increased. In
addition to mechanical advantages, such as durability, high strength and stability, shell
structures enjoy the unique position of having extremely high aesthetic value in various
architectural designs.
In spite of all these features and potential applications, many engineers and architects are
unacquainted with shells as well as the aspects of shell behavior and design. The purpose of
this book is to familiarize the engineering and architectural students, as well as practicing
engineers and architects, with the behavior and design aspects of shell stmctures. The goal of
this book is to present three aspects: the physical behavior, the stmctural analysis, and the
xi
xii
design of shells in a simple, integrated, and yet concise fashion. Thus, the book contains three
major aspects of shell engineering. These are: (1) physical understanding of shell behavior,
(2) use of applied shell theories, (3) development of design methodologies together with shell
design examples.
To achieve these goals, simplified shell theories have been discussed in this book and have
been immediately applied to actual problems. In this sense, the book bridges the gap between
the elaborate theoretical treatments of shells, on the one hand, and, the practical aspects of the
analysis and design of shells, on the other hand. Being aware of a wide variety of existing
numerical routines for shells analysis, we have, nevertheless, made use of simple analytical
schemes of shell analysis so that the designer can understand the analysis procedure and to
perform parametric studies. The theoretical tools required for rational analysis of shells are
kept at a modest level so that engineering and architectural students, as well as practicing
engineers and architects, can grasp the fundamentals of shell behavior and, at the same time,
understand the related theory and be able to apply it to actual design problems. To achieve a
physical understanding of complex shell behavior, quantitative presentations are
supplemented by qualitative discussions so that the reader can grasp a "physical feeling" of
shell behavior. To make the book useful as a reference manual, a number of analysis and
detailed design examples are also worked out in various chapters.
The actual design of shells, involves the use of appropriate codes of practice. Thus, while
making use of some existing codes on shells, in order to provide a text that could be used in
various countries, we have attempted to present the designs apart from the existing codes. In
some cases, the common guidelines provided by several standards, including ACI, BS, DIN,
and IS, have been used.
This book can be used as a text book, and I or a reference book in undergraduate as well as
graduate university courses in the fields of civil, mechanical, architectural, aeronautical, and
materials engineering. It can also be used as a reference and design-analysis manual for the
practicing engineers and architects. To make the book useful to design engineers and
architects, the text is supplemented by a number of appendices containing tables of shell
analysis and design charts and tables. Metric system is used throughout this book.
The material of this book have been developed through many years of teaching at the
Universities of Shiraz and Tehran, University of Toronto, and the Swiss Federal Institute of
Technology (ETHZ) as well as through research and practical design experience by the
author. Thus, in the development of this text, various viewpoints and experiences have been
extremely constructive.
The author would like to thank the Swiss Federal Laboratories for Materials Testing and
Research (EMPA, Dtibendorf) and, in particular, Professor F. Eggimann and Professor U.
Meier and Hr. H. Fritz for supporting this project. Special thanks are due to Hr. P. Fltieler
who has given great encouragement and support in bringing this book to its present
publication. The author would also like to thank professor Gladwell for his useful comments
on the manuscript. This book is dedicated to my family (Gowhar, Anahita, and Mazda) who
have shown great patience during the long period of manuscript preparation.
M. Farshad
Switzerland
FORM AND CONSTRUCTION WITH HYPERBOLIC PARABOLOIDAL SHELLS
IN PLASTICS
A thesis submitted for the Degree of MASTER OF PHILOSOPHY
in the Department of Civil Engineering
UNIVERSITY OF SURREY
by
JOHN ZERNING (B.Sc.)(Industrial Design)
June 1970
ProQuest Number: 10803749
All rights reserved
INFORMATION TO ALL USERS

The quality of this reproduction is d e p e n d e n t upon the quality of the copy subm itted.
In the unlikely e v e n t that the a u thor did not send a c o m p le te m anuscript

and there are missing pages, these will be noted. Also, if m aterial had to be rem oved,
a n o te will ind ica te the deletion.
uest
ProQuest 10803749
Published by ProQuest LLC(2018). C opyright of the Dissertation is held by the Author.
All rights reserved.

This work is protected against unauthorized copying under Title 17, United States C o d e
M icroform Edition © ProQuest LLC.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, Ml 4 8 1 0 6 - 1346
S U M M A R Y
The research described has been undertaken in an attempt to give

an answer to questions often asked: Can glass fibre reinforced
plastics be used as a structural material? Can a structure in GRP
compete with conventional materials?
The relatively high cost of moulds needed for plastic structures

prevents many projects from ever leaving the drawing board. In
order to overcome this drawback a special "cocooning" technique was
developed which provides low cost moulds for highly warped anticlastic
surfaces. Moulds were produced by this process for the manufacture
of the GRP elements used in the experimental work. It is due to the
successful development of the process in the course of the research
that the investigation was able to deal with an extensive number of
configurations.
The research work is presented in three main parts:
(a) An outline of the material characteristics, with particular

reference to polyester resin reinforced with chopped strand mat.
.(b) An extensive study of the morphology of hyperbolic paraboloid

surfaces, illustrated with the aid of numerous models and drawings.
(c) An experimental method of analysis for a scalloped dome, made

as a single skin GRP laminate.
In order to study in more detail the behaviour of hyperbolic

paraboloid shells in GRP a large model was constructed. The model
had a clear span of 4 metres (13-123 ft.) and consisted of 26
hyperbolic paraboloid segments having an average shell thickness of
3 millimetres (0.1l8 in.). Gravity loads were hung from 1,352 points
to simulate various types of live loading. Strain measurements were
taken by means of electric resistance strain gauges.
The tests on the scalloped dome demonstrated its great rigidity

against buckling and showed that it is possible to build structures
of this type with an average thickness of shell to span ratio of
1 :1,500 .
2
A C K N O W L E D G E M E N T S
I am particularly indebted to Professor Z.S. Makowski for

his great enthusiasm towards this work which was carried out under
his direction and supervision.
Sincere thanks are extended to the technical staff of the

Civil Engineering Department, and in particular to Mr. M. Robertson
and Mr. C. Tidy for their help in building the various models.
Gratitude is also due to the technical staff of the Visual

Aids Department for their pains in producing photographs under
difficult conditions.
I am greatly indebted to a number of colleagues and

friends for their suggestions and help with this thesis, and in
particular to Professor E. Spira, Professor V.J. Meyers,
Mr. J. Butterworth and Mr. R.M. Davies.
Finally I would like to thank Shell International Chemical

Company Limited for financing this research project and for
awarding me a research scholarship which enabled this investigation
to be carried out.
•3
CONTENTS
List of Figures, 8
Foreward, 12
PART ONE Glass Fibre Reinforced Plastics as a Structural Material,
Introduction, 15
1.1 Materials and Moulding Methods, 17
Reinforcement, 17
Chopped Strand Mat, 18

Rovings, 18
Woven Cloth, 18
Resins, 20
Polyesters, 20
Fillers and Pigments, 20
Moulding Methods, 21
Contact Moulding, 21
Hand Lay Up, 21
Spray Up Moulding, 22
1.2 Engineering Properties of Laminates, 23

Directional Properties, 23
Shop Practice, 23
Environment, 24
Long Term Loading - Creep, 25
Factor of Safety, 25
Behaviour of Laminates, 26
Stress-Strain Relationships, 27
Final Remarks, 27
References on Glass Fibre Reinforced Plastics, 30

PART TWO Structural Forms: The Morphology of Hyperbolic
Paraboloidal Surfaces, 31
Introduction, 32
2.1 Brief Historical Notes, 34
2.2 Research into the Morphology of Hyperbolic Paraboloids, 42
Form in Hyperbolic Paraboloids, 44
Definition of Surface, 44
Transformation of Surface, 44
Organisation of Surface, 47
Magnitude of Surface, 47
Model Studies of Surfaces, 51
Construction with Hyperbolic Paraboloids, 102
Construction Principles, 102

Possible Application of
Hyperbolic Paraboloidal shells,106
Final Remarks, 112
References on Structural Forms, n 4

PART THREE Design and Construction of a Scalloped Dome in Glass
Fibre Reinforced Plastics, 116
Introduction, 117
3-1 Architectural Design, 118
General Description of Prototype, 118
3.1.1 Geometry of the Shell Segment, 121
Definition of Shape, 121

The Hyperbolic Paraboloidal
Triangular Segment, 121
The Hyperbolic Paraboloidal Canopy, 126
3.2 Engineering Analysis, 139
3.2.1 Approximate Design of Shell Segment, 139
Three-Pinned Arch Analysis, 140

Approximate Design of
Cross-Section Required, 147
3*2.2 Experimental Model Analysis, 150
Relationships Between Model

and Prototype Quantities, 151
Selection of Scale Factor, 152
Dimension of Model, 153
3.2.3 Construction of Model, 154
Mould Construction, 154
The Box, 154

The Cord Network, 156
The Skin of Vinyl-latex, 157
Moulding Technique, 159
Assembly of Model, l6 l
Concrete Foundation, l6 l
Steel Supports, l6 l
0. The Shell, 164
Steel Ring, 165
6
Loading Arrangement, l68
Test Programme, 173
Dead Load, 173
Snow Load, 174
Concentrated Loads, 175
Pressure and Suction due
to Wind, 175
Thermal Loads, Settlements
Loads and Dynamic Loads, 175
Deflection and Strain Measurements, 176
Measurements of Deflections, 176

Measurements of Strains, 176
Mechanical Properties of the Model Material, 178
Young’s Modulus, 178

Poisson's Ratio, 178
Ultimate Tensile Strength, l8 l
Evaluation of Test Results, 182
Dead Load, 182

Uniform Load Over Full Span, 182
Uniform Load Over Half Span, 183
Concentrated Loads, 184
Stress Trajectories-
Isostatic Curves, l84
Plastic Flow, 191
Failure Condition, 191

Comparison between Theoretical
and Experimental Results, 191
Final Remarks, 192
3.3 Cost, 195
References on Design and Construction, 197
7
LIST OF FIGURES
Part One
1.1 An approximate comparison of stiffness to weight ratio

and strength to weight ratio for various materials,
1.2 Mechanical properties of three types of glass fibre

reinforced polyester laminate,
1.3 Glass fibre reinforcing materials, (a) Magnification

of a chopped strand mat. Note that each strand is made
up of numerous filaments, about 6 /1 0 0 0 0 inch (0 .0 2 5 nun)
in diameter (see Figure b for comparison; notice also the
voids between the strands which will all be filled with
resin in the formation of the GRP laminate.
(b) Chopped strand mat; (c) Uni-directional roving;
(d) Woven roving; and (e) Woven cloth,
1.4 Stress-strain relationships; (a) Stress-strain

relationship of a glass fibre filament; (b) Stress-
strain relationship of polyester resins; and (c)
Stress-strain relationships of various glass fibre
reinforced polyester laminates,
8
Part Two
2.1 Framework of a windmill sail; an early application

of a "hyperbolic paraboloidal" construction,
2.2 A hyperbolic paraboloid (HP) stressed skin space frame,
2.3 Part of a double layer diagonal barrel vault formed

from four HP units,
I
2.4 A flat double layer grid fromed from twenty HP units,
2.5 An element of a spherical doublelayer grid formed

U
from fcjfr HP units,
2.6 An umbrella formed from eight HP units,
2.7 Geometry of a HP surface,
2.8-2.11 Transformations of HP surfaces,
2.12-2.15 Organisation of HP surfaces,
2.16-2.19 The macro, infra and micro modular units with reference
to a particular complex geometrical configuration,
2.20-2.43 Structural systems using HP surfaces,
2.44-2.68 HP segments with various perimeters,
2.69-2.80 Construction principles for HP surfaces,
2.81-2.91 Structural forms with HP segments,
9
Part Three
3*1 Scalloped dome, plan, elevation and section,
3.2-3-4 Geometry of a meridional HP shell segment,
3-5 Transformation of co-ordinates,
3-6 Contour lines for the triangular HP segment,
3»7 Free-body diagram of a meridional rib,
3*8 Three sections of a meridional HP shell,
3-9 Table for calculating the approximate stress in a

meridional rib,
3.10-3.16 Mould construction: The timber framework; the

cord network and the spraying of the vinyl-latex,
3.17-3*20 Moulding technique - hand lay up,
3.21/3.22 Section and detail through assembled model,
3.23-3.28 Assembly of model,
3.29/3.30 Steel ring at crown of dome,
3.31 General view of model,
3.32/3.33 Subdivision of the HP segment into 52 quadrilateral

elements,
3.34 Position of electrical resistance strain gauges and

dial gauges on the HP shell segment,
3.33/3.36 Interior and exterior view of model showing the

loading arrangement,
3.37 General view of model with the strain measuring equipment,
10
3-38 Apparent Young's modulus for the model material
expressed as a function of time,
<e
3-39 Computer output for the electrical resistance
strain gauge rosettes,
3-40 Dial gauge for measuring vertical deflections,
3.41 Wiring of an electrical resistance strain gauge

rosette,
3*42 Strain gauge extensometer clamped to the specimen,
3-43 Model on to which the "dummy” electrical resistance

strain gauges were fixed,
3-44/3.43 Vertical deflections through a meridional rib under

symmetrical and unsymmetrical loading,
3.46-3*50 Magnitude and direction of principal stresses in

a HP segment under different loadings,
3.51 Curves of deflection versus time at constant stress

under different loadings,
3.52 Curves of strain versus time at constant stress

3.53 Curves of stress versus time at constant stress

3.54 Stress trajectories in a HP segment, with

scalloped dome uniformly loaded,
F O R E W O R D
In this thesis an attempt is made to cover simultaneously

three problems: structural form, construction technique and
experimental model analysis.
There can be no doubt that the amalgamation of these three

disciplines (usually attributed to three quite separate professions,
architecture, building and engineering produces a most fruitful
starting point for exploring the rich potentialities of plastics.
The material science of plastics and the science of structural

analysis are generally treated as highly mathematical subjects.
The point of view taken here is that much useful work can be
performed by experiment, intuition and engineering common sense.
It is hoped that the many models and experiments produced in

course of this research and described in the following pages will
not only be a rich stimulus and help to architects, engineers and
builders in formulating new structural ideas but will also indicate
directions for further research.
The thesis deals with the structural application of glass

reinforced polyester resins which are otherwise known as 'GRP*.
In order to clarify the whole concept of their use it is

necessary to mention the kind of applications that there are
altogether.
Four basic alternatives exist in the adoption of a building

material, whether non-metallic or metallic:
(1) As a prime structural element. This case is one where the

building material is put to "work” 100 %; and the principal stresses
are carried by the material. In referring to a structural or load-
bearing application of GRP consideration will be given to the
situation where the principal load carrying element and the cladding
element become one.
(2) As a secondary structural element. Here the building material

has merely a cladding function, and forms a "skin” over a reticulated
construction. The loads are carried by the reticulated structure,
which is usually made of another material, and the cladding has only
to perform a secondary load-carrying function.
(3) As a contributing structural element. Here the building

material is used as an aid for forming another material, for example,
using timber or plastics as a formwork for a concrete shell.
(4) As a composite integral structural element, that is to say,

combining materials of different characteristics into one structural
member. For example, GRP is in itself a composite material made of
resin and glass fibres, hence a sandwich construction of GRP skins and
a plastic foam core can be looked upon as a compound composite
integral structural element.
At present the most common application of GRP is as a secondary

structural element. However, the unique characteristics of GRP, see
Figure 1.1 are not fully exploited in this application.
For reasons of structural economy GRP represents the most

practical plastics material for prime structural elements. This
conclusion is by no means new and for the last l4 years a number of
progressive engineers and architects all over the world have
experimented in this direction.
Research in the structural application of plastics is still in

its infancy and so understandably it is subject to a considerable
amount of controversy.
In this thesis a personal and intuitive interpretation is given to

the whole problem of structural’application of GRP. Time, with its
constant challenge for practical and useful solutions, will show
whether the approach taken here is capable of continued growth and
development and whether it has an objective character.
13
PART ONE GLASS FIBRE REINFORCED PLASTICS AS A STRUCTURAL MATERIAL
- their properties
Introduction
"It is precisely in the balance of

one factor against another that the
material engineer finds his challenge
and his satisfaction".
This quotation by Cyril Stanley Smith expresses the real

nature of the problem so often faced that there is no "ideal"
building material.
Glass fibre reinforced plastics, as its name implies, is a

composite material. The principle behind a composite material is
this: two different materials, one for example superior in tension and
the other superior in compression are joined to function together,
forming an integral new material which has the advantages of both.
Probably one of the oldest man-made composite materials in

building is a strawbrick, used by the Egyptians well over 4,000 years
ago. The most up-to-date composite product is a boron fibre laminate.
Other common composite materials that can be used structurally are,
papier-mache, plaster of paris reinforced with flax or asbestos fibres,
Ferro-cemento, and, of course, reinforced concrete. The family of
composite materials is large and increases every year. From the results
already achieved it is not difficult to predict that the real
"breakthrough" in the science of strong materials can only be made by
developing composite materials.
A very versatile and relatively inexpensive composite material

is glass fibre reinforced plastics. It consists of glass fibres
embeded in a matrix, which is usually a thermosetting resin.
Before going into detail about its use it is interesting to

compare the mechanical properties of GRP with other materials.
Figure 1.1 gives an approximate comparison of strength to weight ratio,

viz. tensile strength/specific gravity and stiffness to weight ratio,
viz. Young's modulus/specific gravity of various materials. This
shows that GRP is a good deal stronger than steel for its weight
but it is not nearly as stiff as wood. 15
Materials Specific Young's modulus (E) E Tensile Strength Tensile Strength
Gravity (S.G.) x 10^ lb/sq.in. S.G. x ^3 n,/sq.in. S.G.
Mild Steel 7-8 30.0 3.85 6 0 .0 7.7
Aluminium Alloy 2.7 10.5 3.89 5 0 .0 18.5
Wood along grain - 0.5 1.9 3 .8 0 9.4 1 8 .8

(Spruce)
Unreinforced plastics 1.4 0.5 0 .3 6 8.0 5.7

(PVC)
GRP random 1.5 1.0 0 .6 7 1 5 .0 10.0

reinforced 3 0 % glass
GRP unidirectional 1.9 5.0 2 .6 3 120.0 6 3 .2

reinforced 7 0 % glass
Carbon whiskers 2.3 8 0 .0 34.78 1,000.0 4 ,3 4 7 . 8
Figure 1.1 An approximate comparison of stiffness to weight ratio, and strength to weight ratio; for various materials.'
1.1
Tensile Compressive
Direction Shear
Moulding Method of stress strength Young's
% Glass Specific Tensile Poisson's Y oung%s Compressive Poisson's
and reinforcement:
by weight Gravity degrees perpendicular modulus Strength Ratio modulus Strength Ratio
contact moulding
p.s.i. p.s.i. p.s.i. p.s.i. p.s.i.
(kg/sq.cm.) (kg/sq.cm) (kg/sq.cm) (kg/sq.cm.)
2 02 .mat ^ 30.7 1.45 0,45,90 9 ,00 0 80 0,00 0 11,000 0 .3 2 900,000 15,000 0.42
(2 oz.per sq.ft.) (632 ) ( 56,240) (773) ( 63 ,270 ) ( 1,054)
25-27 O Z . 0,90 11,000 2,000,000 28,000 0.13 1,800,000 14,000

woven roving 55-7 1 .8 0 (773) (140,600) ( 1 ,968 ) (126,540) (984)
(25-27 O Z . per sq.yd.) 45 600,000 5,400 0.45 900,000 10,000
( 42 ,180 ) (379 ) ( 63 ,27 0 ) (703)
cloth 0,90 10,000 1,800,000 20,000 0.17 2,500,000 19,000

47.8 1.63 (703) (126,540) ( 1,406) (175,750) ( 1,335)
per sq.yd.) 45 50 0,00 0 8,400 0.44 900,000 10,000
( 35,150) (590) ( 63,270) (703)
Figure Mechanical properties are for short term loading tests. The values were determined experimentally at 0.1 hour, in room temperature, and
should serve for guidance only
1.2
16
C H A P T E R O N E MATERIALS AND MOULDING METHODS
R E I N F O R C E M E N T
In the course of manufacture glass fibres are made in a

continuous process by feeding glass marbles into an electrically
heated platinum crucible with a perforated base through which
the molten glass drips. Fibre filaments are drawn off on to a
revolving drum and then spun. Depending on the speed of draw the
diameter of an individual filament can vary between 0 .0 0 0 2 0 -
0.00100 in. (0.005-0.025 mm.) A freshly-drawn filament can
develop a tensile strength as high as 5 0 0 ,0 0 0 p.s.i.
(35,255 kg/sq.cm.) (This value begins to approach the theoretical
strength of glass). When dealing with a bundle of filaments,
however, this tensile strength cannot be obtained, one of the reasons
being that the mechanical connection between filaments in the
bundle does not permit the material to develop its full strength.
A bundle of about 204 such fine filaments are spun and

comprise one strand (Figure 1 .3 .a) and this elemental strand forms
the basis for various types of glass fibre reinforcing materials.
The glass commonly used to make the fibres is known as ME n glass;
it has a low alkali content, high chemical stability and good
moisture resistance. The glass fibre reinforcements used for large
size construction are: (l) Chopped strand mat; (2) Woven cloth;
and (3 ) undirectional roving.
The manufactured product of glass fibre in combination with resin

is commonly referred to as a laminate. Laminates usually contain
20 - 60 per cent by weight of glass. The strength of a laminate
depends primarily on the amount of glass content, since the modulus
of elasticity of nE n glass (10.6 x 10^ p.s.i.) is approximately 20
times higher than that of polyester resin (0 . 5 x 10 ^ p.s.i.).
The moulding process will also affect the strength of a

laminate.
Chopped Strand Mat
A mat consists of chopped strands of glass fibres, 1 to inch

long, randomly deposited to form a layer. The layers of chopped
strands are held loosely together by a high solubility resin
binder (Figure 1.3.a. and 1.3»b.). There are three standard
thicknesses of mats weighing 1 , l|- and 2 oz/sq.ft.
The advantages of mat reinforcements are: low cost, equal

physical properties in all directions, good interlaminar bond, and
they can be formed into complex surfaces.
The disadvantages of mat reinforcements are: laminate thickness

cannot be controlled accurately in contact moulding. A mat laminate
has a lower glass content than cloth or roving laminates which
results in a lower modulus of elasticity for equal thickness.
Laminates made with chopped strand mat in hand lay up contact

moulding are most suitable for the large shells suggested in
this thesis.
Rovings
Rovings consist of straight bundles of continuous strands.

Rovings can be used individually by laying them on the mould in the
desired direction as shown in Figure 1.3.c. Rovings are usually
woven into heavy coarse fabrics (Figure 1.3«d.). Woven roving
reinforcement is more expensive than mat reinforcement.
The advantages of woven roving reinforcement are: a high glass

content is possible, it has high directional strength and modulus
of elasticity, and it has an extraordinarily high resistance to
impact.
The disadvantages of woven roving reinforcement are: woven

rovings are difficult to impregnate with resin, can entrap air
bubbles and form voids, they can also cause resin rich areas
between individual rovings which become subject to cracking and
interlaminar failure. The heavy coarse weave becomes difficult to
mould over sharp and strongly curved surfaces.
Woven Cloth
Cloth is woven from strands of glass fibres (Figure 1.3-e.).

A range of weaves is available, including plain weave, satin weave
w $ w ®
d
'19
1.3
and undirectional weave. Cloth is one of the most expensive
type of reinforcement, but where consistency of high performance
and structural efficiency in terms of strength to weight is
required the high cost can be justified.
RESI NS
The resins commonly used in glass fibre laminates are

thermosetting types, that is to say, once cured they cannot be
remoulded. They include polyesters, epoxies, phenolics and
melamines. Polyesters because of their cost advantage and
versatility are most widely used.
Polyesters
The polyester resins suitable for use for laminates are

referred to as unsaturated rigid polyester resins. This means
that they are capable of being cured from a liquid to a solid
state when subjected to the proper conditions. In their natural'
state they resemble a syrup-like liquid. With the addition of
ak>oi*t Z-'/e b y tyGiyht of c a ~ t 11
about 2% by weight of accelerator/, the cross linking or poly
merisation process can then take place. This curing of the resin
can take place at room temperature without the application of
external heat. During the curing process heat is evolved and some
shrinkage develops.
Fillers and Pigments
Fillers and pigments can be added to the moulding resin in

order to reduce shrinkage, lower material cost, impart colour and
improve the surface finish. Of particular importance are the
additives that produce self-extinguishing resins and resins that have
class 1 and 0 surface flame spread.
(The classification 0 - 4 refers to the extent to which a flame will

travel along the surface of material subjected to radiant heat as
outlined in British Standard 4 7 6 . Class 0 denotes a superior
resistance to class 4).
M O U L D I N G M E T H O D S
The moulding methods can be grouped into two: contact moulding

and machine moulding. Most of the machine moulding techniques require
two matched moulds, male and female, whereas contact moulding requires
only one open mould, male or female.
There are several methods of producing glass fibre reinforced

laminates by machine processes: 1. Bag moulding; 2. Autoclave
moulding; 3- Flexible plunger moulding; 4. Matched-die moulding;
5. Vacuum injection moulding; 6 . Hot press moulding; 7- Cold press
moulding; 8 . Centrifugal moulding; 9« Filament winding; 10. Continuous
extrusion; and 11. Continuous laminating.
Each one of these methods has distinct advantages and

disadvantages. They will not be discussed here in detail, mainly
because none of them is suitable for the production of large prototype
shell constructions. Furthermore all these processes have been well
documented elsewhere (Ref.2).
The moulding process most widely used in large shell construction

is the contact method.
Contact Moulding
Contact moulding is usually used for relatively short runs, and

it is the one method that takes full advantage of the two unique
characteristics of polyester resin, viz. that it can be cured without
heat and without pressure. The contact moulding method can be
advantageously combined with the cocooning technique developed by the
author (Ref. 3 and 4).
There are two methods of making a contact mould:
1. Hand lay up; and 2. Spray up moulding
1. Hand Lay Up
Resin impregnated fibre glass reinforcement to the required

thickness are laid on an open mould (male or female) as shown in
Figures 3«l8 and 3-19 and allowed to cure at room temperature.
(See also page 1 5 9 )
2. Spray Up Moulding
The resin impregnated fibre glass reinforcement is sprayed

on to the open mould with a special spray gun, which has a twin
nozzle and a glass roving chopper. The resin is divided into
two parts, one of which is catalysed and the other accelerated.
The two streams of resin spray converge near the surface of the
mould together with the stream of glass fibres ejected by the
glass roving chopper.
Spraying reduces labour costs if a large enough continuous

production is possible. On the other hand great skill and care
is needed to control the thickness of the laminate.
C H A P T E R - . T W O ENGINEERING PROPERTIES OF LAMINATES
Basic engineering properties of reinforced laminates, such as,

specific gravity, strength, moduli, Poisson’s ratio, resistance
to impact, etc., are primarily dependent upon: (l) type of
reinforcement; (2 ) direction of reinforcement; (3 ) type of resin;
(4) method of moulding; (5) shop practice; (6) exposure to
environmental conditions; and (7) long term loading.
Due to these various factors it is not possible to give a

theoretical and general data for the exact properties of laminates.
For design purposes the physical properties have to be determined
experimentally.
Directional Properties
The important characteristics of different types of glass

reinforcement is the variation of their physical properties
with direction. A chopped strand mat reinforcement throughout is
considered as isotropic, that is to say, its physical properties
are essentially the same in all directions. Woven cloth and
roving reinforcements are orthotropic, i.e. their physical properties
are not the same in all directions. Maximum strength is developed in
the warp direction and minimum strength at 45° to it. Figure 1.2
gives the mechanical properties of three types of laminates (Ref.5).
Shop Practice
It is important to realise that GRP is essentially a craftsman

material, which does not produce the best results with hurried and
unskilled labour. The physical and mechanical properties of a
laminate can be considerably reduced by fabrication defects such
as voids, wrinkles, delamination, resin dryness or richness, crazing
and foreign inclusions. An excessively rapid curing process and
appreciable curing shrinkage can also greatly weaken the laminate.
However, with good shop practice that exhibits cleanliness and the
habit of conscientious care during the moulding process most of
these defects can be minimised.
Environment
Glass fibre reinforced plastics can be adversely affected by

unfavourable environmental conditions, like extended periods of
water immersion and exposure to extreme weathering conditions.
Laminates with lower fibre glass content have higher percentage
of strength retention when subjected to extended periods of water
immersion. Tests on mat-polyester laminates immersed in water
for one year have strength retention of approximately 80% for most
of the mechanical properties.
Coating the laminate with a pigmented gel coat (i.e. a resin

rich area on the laminate prevents glass fibres from coming
too near to the surface liable to be attacked) satisfactorily
resists the adverse affects of exposure to normal environmental
conditions.
Exposure to direct sunlight over a period of years will also

reduce the mechanical properties of a laminate. The ultra-violet
rays contained in the sunlight, which are the chief cause of
weather deterioration, can be screened out with a highly reflective
protective coating, incorporated in the gel coat. Up to now,
little reliable data is available on the adverse effect of exposure
to ultra-violet rays on a laminate over a period of many years.
Exposure to temperature change during the cycle of day and

night, will of course affect the dimensions of a laminate.
Plastics are particularly sensitive to thermal changes, and
thermoplastics even more than thermosetting plastics. The linear
coefficient of expansion of a polyester-mat laminate is about
—6
18 x 10 per degree F. (which is about 3 times higher than the
coefficient of expansion of steel). On the other hand, one should
remember that the relatively low modulus of elasticity of GRP -
which even decreases at moderately elevated temperatures - will
prevent the development of high stresses in the laminate due to
thermal change. Nevertheless, the effects of changes in dimensions
due to thermal expansion and contraction can be equivalent to large
loads, which may be particularly dangerous because they are invisible.
The design details of connectors and supports in a GRP structure
therefore demand particular attention.
Long Term Loading - Creep
Creep is a plastic flow or long term deformation associated

with the normally elastic range of the material. Creep in a
fibre glass reinforced laminate, under constant stress depends
on time and temperature, and is mainly due to the viscous flow
of the resin. Creep will therefore reduce the strength properties
of a laminate. Different reinforcements will exhibit different
creep characteristics. For example, cloth laminates have lower
creep characteristics than mat or woven roving laminates. The
creep characteristics impose a severe limitation on the safe long
term design load for a GRP structure.
Due to creep instantaneous test results cannot be expected

to define the behaviour of a GRP laminate. Values of Young's
modulus and ultimate stress used for design must be values at the
time equivalent to the expected duration of load on the structure.
The time dependent Young's modulus is generally referred to as

the apparent modulus, which is equal to: applied stress/observed
strain, as a function of time and temperature.
Figure 3-38 shows the apparent Young's modulus in tension for

rro
a polyester-mat laminate as a function of time at 66 F. and at a
stress level of 2,500 p.s.i. The curve was obtained experimentally
by using a strain gauge extensometer clamped to the speciment as
shown in Figure 3-^2.
In the shell roof structures suggested in this thesis, long

term continuous loading can only be caused by the dead weight of the
structure, which in turn produces very low stresses in the laminate.
Live loads such as snow or pressures and suctions due to wind, have
the nature of periodic loadings which are not ‘ cumulative and,
therefore, will not cause an appreciable increase in the stress level
over a long period of time.
Factor of Safety
Factor of safety (F.S.) is defined as the ratio of the ultimate

strength of the material to the allowable working stress.
< For structural materials like steel, concrete, aluminum,
masonry and timber, the allowable working stress is specified by
codes or by recognised sources of information. For GRP no code
is yet established, and the final selection of a factor of safety,
becomes entirely the responsibility of the designer. (it will
always be the designer's responsibility even when there is a code).
Experiments have shown that in general a safe working stress

level for a continuously loaded structure should be approximately
20 to 3 0 % of the short term ultimate stress.
For GRP structures the following factors of safety should be

considered as general guidance:
Static loads of short term duration
F.S. = 2 minimum
Static loads of prolonged duration
F.S. = 4 minimum
Cyclic loads
F.S. = 4 minimum
Repeated loads producing a reversal of stresses and fatigue
F.S. = 6 minimum
Severe impact loads
F.S. = 10 minimum
Behaviour of Laminates
The simplified structural theory, for a GRP laminate, as used

today by the practising engineer, is based on two assumptions.
1. That glass fibre and resin are firmly bound together and act
as an integral body which undergo equal strains under all loading
conditions.
2. The material is considered to be elastic and obey's Hooke's
Law, which states that the stress is directly proportional to the
strain and the relationship is represented by a straight line.
26
Furthermore it is assumed that the material will return to its ,
original shape when the load is removed.
Thus:
(7 = E (e) or E = - . (l.l)
e
where (7-;='.stress, E = modulus of elasticity and e = strain.
Stress - Strain Relationship
A glass fibre exhibits an almost perfect linear tensile

stress - strain behaviour up to failure at breaking point with
about 2% elongation at break (Figure 1.4.a).
Unreinforced polyester resin displays an almost linear

increase in stress - strain curve up to about 25 % of elongation '
at break. At 50% of elongation at break the curve shows a distinct
slope which becomes practically horizontal when failure at break
occurs (Figure 1.4.b). The degree of plastic deformation up to the
breaking point can be anywhere between 1,2 to 7% depending on
the type of resin.
The stress-strain curve for woven roving cloth and mat

polyester laminates exhibits a linear portion followed by a non
linear portion (Figure 1.4.c). It is similar to that of plywood
and it has no yield point like as for example in mild steel. There
is also no plastic reserve of ultimate strength which may be relied
on in the design of GRP structures.
It is important to note that the stress-strain relationship in

a polyester-mat laminate is not appreciably affected by the size of
the tested specimen. The various sets of specimens tested in
tension, which vary in width from 4 in. to \ in. (101.6 mm --12.7 mm)
and in thickness from 13/64 in.to 3/64 in,(5-2 mm - 1.2 mm) all
exhibited negligible variation in the short term Young's modulus at
a stress level of 3 0 % of the short term ultimate stress.
Final Remarks
It is necessary to point out that Equation (l.l) presupposes

that stress is independent of time and temperature. However, for a
visco elastic material like, for example, polyester resin this
relationship is not valid.
27
213,000 3,520
(15,000) (600]
7,100
(500)
142,000 5,680
(10 ,0 0 0 ) (400)
- Ib/sq.in(kg/sq.cm)
£ 4,260
H (300)
cr
in
cn
JSC
71,000 ass fib 0|i dig. 'T ' 2,840
polyester r<?sins
(5,000) .E (200)
cr
(/i
JQ
. 1.420
U) (100)
U)
st ress
0 2 3 4 0 0-5 1-0 1-5 2-0

st r a in % s t r a i n 0/#
8 5,2 00
(6,000)
G RP-roving
7 1 ,0 00
(5,000)
5 6 ,8 00
(4,000)
- l b/ sq. in( kg/ sq. cm)
4 2 ,6 0 0 GRP- cloth
(3,000)
u G R P-chopped strands
14,200
(1,000 )
stress
0 2 3 4
strain %
1.4
a
On the other hand the incorporation of time and temperature
in the stress-strain relationship will result in a very complex
analytical treatment. For practical design problems this
laborious mathematical treatment is seldom justifiable. In
practice the problem is simply solved by increasing the factor
of "ignorance" viz. factor of safety.
REFERENCES
On Glass Fibre Reinforced Plastics
1. Gordon, J.E., "The New Science of Strong Materials",

Penguin, 1 9 6 8 .
2. Benjamin, B.S., "Structural Design with Plastics",

Van Nostrand Reinhold Company, 1969.
3. Zeraing, John, "Das Kokon - Spritzverfahren und seine

c
Anwendung im Ausstellungsbau", DeutsJhe Bauzeitung, Mai 1964.
4. Zerning, John, "Das Glass - Polyesterharz - Spritzverfahren und

seine Anwendung im Leichtbau", Deutsche Bauzeitung, Oktober 1965-
5. Gibbs & Cox Inc., "Marine Design Manual for Fibreglass

Reinforced Plastics", McGraw - Hill Book Company, i9 6 0 .
6. Scott Bader & Co. Ltd., "Polyester Handbook", Wollaston,

Wellingborough, Northamptonshire, 1967*
7. Ogorkiewicz, R.M., (Editor), "Engineering Properties of

Thermoplastics", Wiley - Interscience, 1970.
PART TWO STRUCTURAL FORMS: THE MORPHOLOGY OF HYPERBOLIC
PARABOLOIDAL SURFACES
or how to learn from Nature
31
Introduction
"Each material has its own message to

the creative artist'1.
This quotation by Frank Lloyd Wright expresses concisely the

point that materials are not all suitable for use in the same way.
The problematic structural application of GRP in building can

perhaps be tackled best if attention is concentrated upon light
weight single skin roof structures of moderate size, since in this
application one can make good use of the high strength to weight ratio
of the material and also the ease to which it can be formed into
complex shapes. Furthermore the fire hazard for a roof structure
is minimised.
Which roof shapes are most suitable for GRP?
Essentially GRP is a sheet like material of high strength, but

‘ low stiffness. The material is costly and therefore has to be used
very efficiently.
An ideal application is that of doubly curved shell structures

in which the material is stressed in its own plane and bending stresses
are almost completely eliminated. Doubly curved shells are non-
developable and can be grouped in to two large families: (l) Synclastic
and (2) Anticlastic.
Synclastic surfaces are referred to as having a positive Gaussian

curvature, that is to say, planes passing at right angles to its
surface have downward curvatures. A dome is a typical example of a
synclastic shell.
Anticlastic surfaces are referred to as having a negative Gaussian

curvature. Two planes passing at right angles through the principal
curvature have one downward and one upward curvature. A saddle
surface is a typical example of an anticlastic surface.
An anticlastic shell even in a single skin construction can
effectively counteract buckling to which a material such as GRP wi
low modulus of elasticity is very susceptible. From the large
family of anticlastic surfaces, hyperbolic paraboloids have the
particular advantage that they can be generated by two systems of
straight lines, and hence are easy to build.
C H A P T E R O N E BRIEF HISTORICAL NOTES
Experience gained from hyperbolic paraboloid shell structures

built in reinforced concrete, steel, aluminium and timber can
fruitfully inspire the new field of structural plastics. A brief
general review on the history of hyperbolic paraboloids is, therefore,
of interest.
One of the earliest, large size, applications of "hyperbolic

paraboloidal" surfaces was in the sails of windmills. The large
windmill sails in Europe were wooden frames covered with sail cloth.
First they had flat inclined planes, later they were built with a
twist, like an aeroplane propellor (Figure 2.1). Their size from
tip to tip was over 71 It. (21.64 m ) . The sail was made of straight
wooden beams, and since it was a warped surface, it is according
to our definition, a hyperbolic paraboloid. This type of
construction could not have been improved in principle and remained
unchanged for many centuries; and it was only replaced by another
source of energy - electricity. In these forms there was perfect
unity between contemporary timber technology and function (i.e. a
body that will produce maximum thrust due to the wind force). Indeed
the warped surface was the "natural" solution to the problem.
Mathematicians found an interest in hyperbolic - paraboloids,

(abbreviated HP), only since the seventeenth century.
It is impossible to say who built the first HP to enclose

space, but the credit is usually given to Antonio Gaudi the Spanish
architect-engineer and builder in stone, who favoured the geometry
of HP surfaces, and used them first in 1909* For the still unfinished
La Sagrada Familia Church in Barcelona, he planned to use HP surfaces
for the vaults and walls of the nave. Gaudi saw in HP surfaces a
symbolic meaning as well as a suitable structural form.
In the search for new and suitable shell structures in

reinforced \concrete and in trying to enclose space with a minimum
of material, Bernard Laffaille, a French engineer, was perhaps the first
to understand the structural potentials of HP shells. In 1933 he built
3^
2.1
2.2
'35
in Dreux for testing purposes, a two sided cantilever structure,
made up of four HP segments.
A study of the structural analysis of HP shells followed in

1936, written by the French engineer F. Aimond.
After Laffaille the Italian engineer Giorgio Baroni between

1934 and 1938 built several industrial HP roofs for the Alfa
Romeo factory in Italy.
The great Spanish engineer Edwardo Torroja, was equally

intrigued by the structural and architectural possibilities of HP
shells, and used them in a masterly way for a sports stand at
La Zarzuela in 1935*
The outbreak of the Second World War brought a temporary stop

to these new developments.
(During the war only the Czech engineer Konrad Hruban built
"umbrellas" with HP shells for an industrial roof at Nove Mesto in 1943).
In 1951 the Spanish born architect engineer, Felix Candela, built

a small pavilion (known as the cosmic ray pavilion) for the University
of Mexico. This unusual building could only have been done by a
person who is capable of fulfilling the three functions of architect,
engineer and builder. It immediately received world wide publicity.
Many other exciting and daring roof structures with HP shells
followed. The restaurant in Xochimilco, Mexico, built in 1957 deserves
special mention. Very soon HP shells became synonymous with the name
Candela, and we can rightly call him 'the advocate' of HP shells.
Another contemporary pioneer and "virtuoso” in warped or

anticlastic surfaces, is the German architect engineer, Frei Otto.
It is interesting to point out that Candela is working with anticlastic
ruled surfaces, i.e. surfaces that can be formed by a system of
straight lines, whereas F. Otto uses for his tension structures
anticlastic-minimal surfaces, i.e. surfaces originating from the
surface tension of the membrane within its given boundary condition.
Minimum surfaces reach the limit in mechanical efficiency of strength
to weight ratio. On the other hand this advantage is greatly offset
by the fact that their geometry is difficult to define and hence

not easy to build.
An equally important contributor to the rapid development of HP
structures is the French engineer, Rene Sarger, who worked for many
years with Laffaille. Sarger worked mainly as a consulting engineer
and helped to materialize many ideas with HP surfaces.
HP shells did not remain only the domain of engineers; some of

the most prominent architects, Le Corbusier, Eero Saarinen,
Marcel Breuer, Kenzo Tange, Matthew Nowicki, and R. Rainer
incorporated them in some of their buildings. They have mastered the
rich and plastic tectonic potentials (i.e. the unity between form,
construction and material) and produced a powerful and significant
modern architecture.
Today, if an attempt were made to write an extensive bibliography

on HP surfaces, it would probably fill a few hundred pages.
The ever increasing interest and popularity of HP shells is

largely due to the following reasons: they have pleasing expressive
and dynamic forms, they are relatively easy to build, they are very
efficient in their strength to weight ratio, and the geometry of the
surface can be defined easily. On the other hand because of their
rich plasticity, they exert a powerful influence on the surroundings
and thus are difficult to integrate in the general design scheme.
For a long time reinforced concrete was the only accepted

building material for HP shells. Yet this monopoly position of
reinforced concrete is by no means justified. Other structural
materials like timber, aluminium and steel have also been successfully
used to build HP shells. Reinforced plastics, due to their lightness,
mouldability, freedom from corrosion, durability and variability in
colour and surface texture, form a specially suitable material for
the construction of HP surfaces.
One of the earliest applications of plastics was in the HP

umbrellas used for the American exhibition pavilion in Moscow in 1959*
The whole structure, roof and columns were made in single skin
GRP laminate. The architect was George Nelson and the consulting
engineer Albert Dietz.
A similar single skin HP umbrella system was used in 1961 to

cover the roofs for the markets in Ivres and Fresnes in France. An all
plastic HP vault roof was built in 1968 to cover swimming pools in
Lincoln, England and a similar one in Aberdeen Scotland. They were
both designed by John Vest Design Group Ltd.
Finally these brief historical notes are incomplete without

mentioning the recent developments of the HP stressed skin
space grids.
The principle behind the HP stressed skin space grid is this:

the two high points of the HP surface are interconnected with a
strut, similarly another strut is used to join the two low points
thus forming a tetrahedral unit with a pre-stressed skin inside
as shown in Figure 2.2. Units of this kind can be used to form a new
group of structures: a combination of space-frames and shells.
Generally the space frame is made of steel and the shell is in
aluminium or single-skin GRP. This principle was first applied in
1957 by Buckminster Fuller for an aluminium dome in Honolulu. HP
stressed skin space grids are very suitable for prefabrication and
mass production, can be quickly erected and are possible to
dismantle; they are very light in weight and have a high reserve
of strength. Stressed skin space grids are particularly suitable
for mobile structures. Some of the main disadvantages are in the
complexity of the connector details and in the difficulty to make the
many water-proof joints between the surfaces.
Full size experiments with HP stressed skin space grids in GRP

were performed in U.S.A. (Texas) in I9 6 I by Structural Plastics Ltd.
Their system consisted of HP shells in GRP 4 by 4 ft. (1.22 by
1 .2 2 m) squarein plan, interconnected diagonally at their high and
low points to form a tetrahedral unit. The units were arranged to
obtain a double layer diagonal barrel vault. Figure 2.3 shows four
such identical units, all the points A7and B touch a cylindrical surface.
A slightly different arrangement based on the same principle has

been developed by the French architect M. Chaperot in 1964. In this •
system six HP shells in GRP, and rhomboidal in plan, measuring
diagonally 7 (23 ft.) and 4 m (13 ft.) with a height of 2 m (6 .5 6 ft.),
are joined together along their edges, thus forming a flat double layer
threeway grid at one side and an hexagonal grid at the other side.
38
2.3
2.4
39
2.5
2.6
40
An assembly of 20 such HP units is shown in Figure 2.4 all the points
marked B lie on a flat plane. This system was applied by Chaperot
for a market cover in Lezoux and a garage in d'Arcueil France.
In Czechoslovakia the State Building Research Station in

Prague evolved in 1967 an even more sophisticated system. It
consists of HP shells in GRP quadrilateral in plan and measuring about
1*5 by 1.5 m (5 by 5 ft.). The points A-A and B-B are interconnected
by two systems of prestressed cables passing above and below the
surface (Figure 2.5)- The corners A and B have hinged connectors
to allow for the small necessary dimensional adjustments required
to obtain stressed skin space grids of double layer synclastic or
anticlastic form. In Figure 2.5 all the points marked B lie on a
spherical surface.
The Swiss engineer Heinz Hossdorf developed in 1964 quite a

different system. Eight HP shells in GRP are connected to form an
umbrella 18 by 18 m ( 59 by 59 ft.) square in plan (Figure 2.6).
By means of a central pyramidal truss composed of four hinged tubes with
a central prestressing device T ^ ? a constant prestressing force is
exerted on all the eight shells. This system was used at the Swiss
National Exhibition Lausanne to cover an area of 7,800 m 2
(83,958 ft.2).
4l
C H A P T E R T W O RESEARCH INTO THE MORPHOLOGY OF
HYPERBOLIC PARABOLOIDS
Research into structural forms, or for that matter into any

field of science, can be undertaken in two ways:
(1) Theoretical synthesis - a systematic accummulation of data

until the generalization is a matter of plain facts.
(2) Observation and experiments - a speculative exploration to be
backed later by reason.
Each method requires a different type of mind and so the

tendency for the one or the other approach depends on the individual
researcher. The author's preference l/fi s in the empirical approach.
The next step is to set out the research problem. Unfortunately,

very little research has been performed up to now in the investigation
of structural forms, hence there was no previous work to serve as
a guide.
A good starting point was considered to be the simple one of

breaking the problem down into a large number ofseparate questions
to be dealt with one at a time.
Given: (l) A material - glass fibre reinforced plastics; (2) A special

mould building technique - cocooning; and (3 ) A structural system -
consisting of HP shell element.
Required: Various space enclosing structural forms that will be

characteristic for these given factors.
Research into the various kinds of structural forms with HP

surfaces in plastics can be regarded as being divided into two
separate parts: (l) Form; and (2) Construction.
The first topic - Form - deals with the "idea", i.e. the geometrical
morphological aspects of HP surfaces.
The second topic - Construction - deals with the "physical reality"

of the Form; considering equilibrium, strength and stability, yet
without going into the structural analysis and building details.
42
It is possible to deal here only with general systems and
concepts of structural forms. In the final design process, Form and
Construction cannot be treated independently of each other. It
is the task of the creative designer to synthesize these two elements
to form one inseparable whole.
FORM IN HYPERBOLIC PARABOLOIDS
Definition of Surface
A HP surface can be defined as a translation surface. It is

generated by translating a principal parabola GOC (generator) with
an upward curvature upon a downward parabola AOE (directrix)
(Figure 2.7)- A HP surface can also be interpreted as a doubly
ruled surface. Two systems of straight lines intersect the non
parallel lines ABC-GFE and AHG-CDE called directrices. The directrices
are parallel to the plane XOZ and YOZ (director plane). Plane
sections, i.e. contour lines parallel to the Z axis, will be
parabolic curving positively or negatively depending on the direction
of the section. Sections parallel or oblique to the XY plane
give hyperbolas.
Transformation of Surface
The geometry of a HP surface can be 'transformed' in four

basic ways:
1. By changing the slope of the directrix one can vary the

curvature or twist of the surface and hence change the rise to span
ratio of the principal parabolas (Figure 2-8).
2. By varying the angle between the director plane one can get a HP
surface with two different principal parabolas (Figure 2.9)•
3. By tilting the Z axis with respect to a plane of reference one can

obtain HP surfaces with various quadrilateral and triangular
planes (Figure 2.10).
4. By cutting various segments from a standard HP surface one can

obtain HP segments with different perimeters (Figure 2.11).
2.7
45
2 .1 1
*46
Organisation of Surfaces
HP surfaces, or for that matter, any geometrical configuration,

can be organized according to two different concepts:
1. According to the synthetic concept, this is a process of adding

So
and putting together separate elements^/as to form a whole. Orderly
arrangements can be executed, for example, following the basic laws of
symmetry (Figure 2.12).
2. According to the analytic concept, this is the inverse of the

previous concept. Here a given whole is resolved into small elements
in order to be able to 'fit* them into HP elements. Four examples
are shown in Figure 2.13-
In the organization process it is sometimes difficult to

s
separate the two concepts of analysis and synthe/is and say which one
came first into one's mind.
The repetitive HP units can be identical or similar or quite

different from each other. The repetitive unit can consist of a
single element or be a compound element (Figure 2.14).
Arrays can be executed in one-dimension (linear), two-

dimension (planar) or three-dimension (spatial) (Figure 2.15)»
An arrangement can be of a limited or infinite nature. It can

be made of regular or irregular HP elements and can be compact
or loose.
Magnitude of Surface
Considerable difficulties arise if one tries to describe the

complex geometry of certain shapes suggested here, for example, the one
shown in Figure 2.16.
It is convenient therefore analogous to other scientific

disciplines, to distinguish between three concepts.
1. The macro modular unit-the prime structural form. It can be a

single isolated or compound form. For example, the macro modular
unit in Figure 2.16 is a synclastic paraboloid of revolution (Figure 2.
2. The infra modular unit-the repetitive element from which the
prime structural form is made. It is generally the prefabricated
element. For example, the infra modular unit in Figure 2.16 is a
compound unit composed of two HP segments (Figure 2.18).

organisation of hyperbolic para boloid units - s y n t h e t i c co n c e p t
b ila t er al sym m et ry translato ry sym m etry rotatio n a l symmetry

of r e f l e c t i o n
p o i nt of
the un i t
b i l a t e r a l + r o t a t i o n a l sy m m e t r y b i l a t e r a l * t r a n s l a t o r y s y m m et r y d ilatatio n
!
i a x i s of r o t a t i o n of t r a n s l a t i o n
2.12
organisation of h y p e r b o l i c paraboloid units - analytical concept
subt ra c tio n
superposition in te rp en e tra tio n
2 .1 3 '48
or gani sat i on of h y p e r b o l i c - p c t r a b o l o i d s ( h y p a r s )
t h e " r e p e t i t i v e e l e me nt a r y " unit ( wi t h a h i gher de gr ee of s y m m e t r y )
an elementary unit with 1h yp ar' su rfa ce two hypar' s u r fa c e s thre e ti y p a r ’ surfaces
f o u r ‘hy p a r ’ surface s fiv e 'hypar' s u r f a c e s six "hypar' s u r f a c e s
seven ' h y p a r ' s u r f a c e s eight ' hy p ar ' s u r f a c e s
2.14
linear arrangement
planar arrangem ent
s patial a rra n g e m e n t
49
m a c r o - m o d u la r u n it
a x is o f r e v o l u t io n
parall e]s_____
g e n e r a t r ix =
m e r id ia n
2.l6 2.17
m i c r o - m o d u la r u n it
in fra -m o d u ia r unit
axis of re v o lu tio n
meridian
2.18 2.19
'50
3» The micro modular unit-the 'structural pattern1 of which the
repetitive element is composed. For example, the micro modular unit
in Figure 2.16 is a honeycomb sandwich construction as shown in
Figure 2.19-
It is possible to develop a configuration in which a HP

surface can be used to form all three, i.e. macro, infra and micro
modular unit. Throughout this thesis the infra modular unit will
always be a HP surface.
Model Studies of Surfaces

s
The model studies, in this thesis, consist of three exercises:
1. This aims to explore the ways in which the perimeter of a HP

unit change if various 'segments' are cut out from a 'standard' HP
surface. To give this experiment a common denominator the standard
HP surface remained invariant. The only factors that were varied are
locations and perimeters of the segmental cut-outs upon the standard
surface. In other words, through cutting various segments with
various plans, from the middle or corner of the standard surface, it
was possible to obtain HP segments with different boundaries and
shapes. A number of identical HP segments were arranged - according
to the synthetic concept - to form a compound element. Particular,
consideration was given to configurations that can be applied to roof
structures. Some of the possibilities investigated are shown in
Figures 2.20 to 2.3^- They are grouped in two parts: 1. Having
straight edges; and 2. Having curved edges. The top left corner
in each figure shows the invariant standard surface (a HP over a square
plan) and the manner in which the segmental cut-out was chosen. A
plan, an elevation and a cross-section of the compound element are also
shown. A photo of a GRP model of the compound element accompanies
each figure. The illustrations are self-explanatory, so that no
further comments about them are necessary at this point.
2. This explores the organization of HP units-according to the

analytical concept. That is to say, a given whole-the common shell forms
like: vajult, cone, elliptical-paraboloid, paraboloid of revolution,
hyperbolic-paraboloid etc. - form the macro modular unit. The infra
modular unit is always an arch-like segment taken entirely from the
middle of an invariant HP surface. The possibilities investigated are
shown in Figures 2.35 to 2.43. The top left corner in each figure
shows the. invariant standard surface (a HP over a rhombic plan) and
the manner in which the segmental cut-out was chosen. A plan, an
elevation and a cross-section of the macro modular unit is also
shown. A photo of a GRP model of the same unit accompanies each figure.
3. In this experiment the procedure was reversed. The invariant

factor was the plan of the segmental cut-out (a square and an
isosceles triangle) and the variable was the so called 'standard'
surface. In this way it was possible to vary the contour lines of
the segmental cut-out edges. Figures 2.44 to 2.59 show the
segmental cut-outs having a square plan, and Figures 2.60 to 2.68
show the segments with an isosceles triangular plan. The left side
of each figure shows the plan and elevations of the variable
standard surface (a HP over a quadrilateral plan) and the manner in
which the segmental cut-out was chosen. A photo of a string model
of the segmental cut-out, accompanies each figure. The vertical
planes of the model, in their development, are shown on the lower
right of the figure. Again, the illustrations are self-explanatory,
so that no further comments about them are necessary at this point.
52
A-A B-B
2.20
‘53
A-A 8-B
2.21
2.22
'55
2.23
56
2.24
31
A
2.25
58
.
A-A B-B
2.26
'59
2.27
'6o
A-A B -B
2.28
6l
I
A- -A
q,
I
A-A 8-8
2.29
62
B
A-A B-B
2.30
63
2.31
6k
A-A B-B
2.32
■65
66
A-A B-B
2.34
67
B
2.35
'68
2.36
69
2.37
'70
2.38
71
2.39
72
A
A-A B-B
2.40
73
2.41
74
2.42
75
2.43
76
ii-./V:' it- . . *«
‘77
B A 0 c B
#J>.; k?p
-..' ' ' <^ y - ’X
,.Vv
-
■^'vv »
> 7 f
§
2.45
•78
iI l f i n n K r ^
wsh'ifW
A
b'
'80
6'
b'
C
a
-8i
b
C A
2.49
82
c
J k
2.50
83
d'
C1 A'
2.51
84
D
2.52
85
b‘
A'
2.53
66
;MS
> //» ■
2.54
87
c
b1
d’
C A C B A D C
88
c
d‘
2.56
89
V
m m ’v
2.57
90
b'
A’ A'
b'
A B A D C B
A"
91
'92
-
V *
§3
b'.d
A.C o’
2.61
9^
c
b.c
2.62
95
2.63
96
'98
2.66
:?-r
iXh.:
•99
3.cr
Ô q
c
2.68
101
CONSTRUCTION WITH HYPERBOLIC PARABOLOIDS
Construction Principles
Fundamentally all structures, as well as HP structures, can be

traced back to: solid reticulated; or surface construction. This
definition depends upon the relative dimensional relation between
length (l), breadth (b), and height (h). Therefore, in a solid
construction 1, b and h are of similar magnitude (Figures 2.69 and
2.70). A reticulated construction is made up of linear elements where
1 is always greater than b and h (Figures 2.71; 2.72; 2.73; 2.74 and
2.75)• In a surface construction 1 and b exceed h (Figures 2 .7 6 ;
2.77 and 2.78).
The integrated combination of reticulated and surface construction

form another group known as a stressed skin reticulated grid. As a
single layer system it has been used quite frequently; its application
as a double layer system is new (Figures 2.79 and 2.80), see also
Figures 2.2 to 2.6. (in a building one rarely finds any one of these
systems in their pure form, they are usually in combination).
These four fundamental principles can each be subdivided into

two systems:
1. Acting mainly in compression (Figures 2.69; 2.71; 2.72; 2.73;

2 .7 6 ; 2.77 and 2.79).
2. Acting mainly in tension (Figures 2.70; 2.74; 2.75; 2 . 7 8 and 2.80).

\
Depending on the size of the structure, one can build each system
in a single or multi-layer construction.
Let us look at a few examples:
solid HP construction (Figure 2 .6 9 ); pressurized HP construction with

membrane ribs (Figure 2.70);
warped single layer space truss (Figure 2.71);
warped double layer space truss (Figure 2.72); .f
interpenetration of arched trusses (Figure 2-73);
prestressed cable system (Figure 2.74);
prestressed network with discontinuous compression elements (Figure 2.75)5
102
anticlastic shell (Figure 2.76);
warped folded system (Figure 2.77);
prestressed tent system (Figure 2.78);
co-active pyramidal folded warped space truss system (Figure 2.79);
co-active stretched fabric prestressed cable network (Figure 2.80);
In Figures 2.69 to 2.80 the macro-modular surface has always the

configuration of a HP. The infra modular structure, viz, the
repetitive element, is shown on the right lower corner of each Figure.
Every system is unique in its own right. To express at present

a general and critical opinion as to which system to use for what
building is an ambitious and daring undertaking indeed. A structural
system can only be evaluated according to its total energy consumption.
One should always remember that the system of construction is only a
means to an end; if seen in isolation it is quite meaningless.
The overall design problem consists of a large number of

factors which includes: firstly, the human requirements, traditionally
handled by the architect, and secondly, the physical properties (the
domain of the engineer). The human and physical factors involve a
chain of problems that impose demands on the building. None of these
factors is independent of the other; there are definite relationships
between them, sometimes complementary and at times conflicting.
It is the task of the creative designer to separate and order these
factors so that they can be synthesized to form a whole.
103
2.69 2.70
2.71 2.72
2.73 2.74
104
2.75 2.76
2.77 2.78
2.79 2.80
'105
Possible Application Of Hyperbolic Paraboloid Shells
Structural forms in principle have not changed through the ages.

The macro-modular units like vaults, domes and 'umbrellas’, are
amongst the oldest structural forms. New materials with their
technological development have merely changed the infra modular
and micro modular unit, which in turn has enabled larger spans
to be built and reduction in weight.
Basically new are the large cantilever and saddle-shaped

structures which are made possible by building materials that can
resist high tension, compression and shear. Today, with the ever-
increasing trend towards prefabrication, the infra modular unit gains
particular significance and consequently gives the structural form a
specific formal expression.
HP surfaces if seen as a single isolated configuration, have a

strong 'individuality' and are generally difficult to integrate in a
general design scheme. By using a HP surface as an 'elementary
repetitive unit, viz. infra modular unit, its unique plasticity is
'neutralized' and made to fit practically, any imaginable enclosure.
‘A few examples shall illustrate this point.
Figure 2.81 Undulated diagonal vault roof system with an

undulated edge at its perimeter, and supported on exterior inclined
buttresses. Macro modular unit - a parabolic vault composed of
7+12 principal HP segments. Infra modular unit - a compound element
composed of three different segments. 1. One principal HP segment has
a rhombic plan and four upward curving parabolic edges, (see also
Figure 2.48). 2. The other principal HP segment is over an isosceles
triangular plan of which two adjacent edges curve parabolically
upwards. The third edge is slanted and hyperbolic (see also
Figure 2 .6 7 ). 3- The auxiliary HP segment that forms the canopy over
the entrance has two slanted arched edges, one is parabolic, the other
hyperbolic.
Figure 2.82 Undulated vault roof system with slanted arched ribs
having an undulated edge at its perimeter and supported on exterior
inclined buttresses. Macro modular unit - a hyperbolic vault composed
of 4+5 principal HP segments. Infra modular unit - a compound element
106
composed of three different segments. 1. One principal HP segment
has an "orange peel" like plan and two slanted hyperbolic edges,
(see also Figure 2.30). 2. The other principal HP segment has a
"bottle neck" like plan of which the two opposite long edges are
slanted and hyperbolic. The two short edges are also slanted and
hyperbolic. 3« The auxiliary HP segment is similar to the one
in Figure 2.81.
Figure 2.83 Single-folded scalloped roof system with half

d
cone-shaped ends having an urûlated edge at its perimeter, and
supported on exterior inclined buttresses. Macro modular unit - a
single folded system composed of 14 principal HP segments. The
transverse ends are intersected by two half-cone segments composed
each of three principal HP units. Infra modular unit - a compound
element composed of two different segments. 1. The principal HP
segment is over an isosceles triangular plan of which two adjacent
edges are straight and the third edge is slanted and hyperbolic,
(see also Figure 2.64). 2. The auxiliary HP segment is similar to the
one in Figure 2.81.
Figure 2.84 Scalloped cone-shaped roof system with an undulated
edge at its perimeter, and supported on exterior inclined buttresses.

Macro modular unit - a cone composed of 12 principal HP segments,
(see also Figure 2.37). Infra modular unit - a compound element
composed of two different segments. 1. The principal HP segment is
similar to the one in Figure 2.83. 2. The auxiliary HP segment is
similar to the one in Figure 2.81.
Figure 2.85 Scalloped dome roof system with an undulated edge at

its perimeter, and supported on inclined exterior buttresses. Macro
modular unit - a synclastic paraboloid of revolution composed of 12
principal HP segments, (see also Figure 2.32). Infra modular unit -
a compound element composed of twodifferent segments. 1. The
principal HP segment is similar to the second principal HP segment in
Figure 2.81. 2. The auxiliary HP segment is also similar to the one
in Figure 2.81.
Figure 2.86 Scalloped dome roof system with its crown pulled
down having an undulated edge at its perimeter, and supported on
inclined exterior butresses. Macro modular unit - partly synclastic
2.86
109
and partly anticlastic paraboloid of revolution composed of 12
principal segments, (see also Figure 2.43). Infra modular unit - a
compound unit composed of two different segments. 1. The principal
HP segment is over an isosceles triangular plan of which two
adjacent edges arch parabolically upwards. The third edge is slanted
and hyperbolic. 2. The auxiliary HP segment is similar to the one
in Figure 2.81.
Figure 2.87 Umbrellas over hexagonal plan, and supported on

their central low point. Macro modular unit - an anticlastic
paraboloid of revolution composed of 6 segments (see also Figures
2.34 and 2.4l). Infra modular unit - a HP over a deltoid plan of
which two adjacent edges are straight and two parabolically curved
downwards, (see also Figure 2.49).
Figure 2.88 Scalloped cone-shaped roof intersected by a

cantilever fan-type system, and supported on vertical Y-shaped
columns. Macro modular unit - two inter-penetrating cones composed
of 12+24 segments. Infra modular unit - a compound element composed
of three segments. 1. The HP segment that makes up the inner cone
has an isosceles triangular plan of which all four edges are straight,
(see also Figure 2.60). 2 & 3- The two HP segments that form the
outer cone have straight edges and are over a trapezoid plan, (see
also Figure 2.62). Their surface curvature is the same, one merely
has a left, and the other a right warp.
Figure 2.89 Twin umbrellas over rectangular plan, and supported

on their low points. Macro modular unit - a folded planar system
with high and low points composed of 4+6 segments. Infra modular unit -
a compound element of three segments. 1. The central HP segment is
over a square plan of which all four edges are straight, (see also
Figure 2.45). 2 & 3» The two HP segments that form the cantilever
part have one straight and three slightly curved edges and are over
a trapezoid plan, (see also Figure 2.63). Their surface curvature is
the same, one merely has a left, and the other a right warp.
Figure 2.90 Scalloped dome intersected by an undulated

’doughnut’ like shaped roof system with an undulated edge at its
perimeter and supported on inclined exterior buttresses. Macro modular
110
2.87
2.89
ill
tinit - a synclastic paraboloid of revolution composed of 12
principal HP segments (see also Figure 2.43) intersected by a
•torus' (with a parabolic cross-section) composed of 12 principal
HP segments. Infra modular unit - a compound element composed of
two principal and two auxiliary segments. 1. One principal HP
segment that makes up the dome is identical to the one in
Figure 2.85® 2. The other principal HP segment, that makes up the
'torus', is over a trapezoid plan of which two opposite edges are
parabolic and arch upwards, the other two opposite edges are slanted
and hyperbolic (see also Figure 2.44). 3 & 4. One auxiliary
segment between dome and 'torus' is identical to the one in
Figure 2.81 and the other being similar to it.
Figure 2.91 Scalloped tent system with an undulated edge at

its perimeter, and supported on an interior central column. Macro
P
modular unit - an anticalstic paraboloid of revolution composed of
24 segments, ( see also Figures 2.33; 2.4l and 2 .8 7 ). Infra modular
unit - a compound element composed of two different segments.
1. The principal HP segment is similar to the one in Figure 2 .8 7 .
2. The auxiliary HP segment that forms the canopy over the entrance
has two straight and one upward slanted arched edges, and is over an
isosceles triangular plan, (see also Figure 2.64).
Final Remarks
An attempt was made to illustrate with the many drawings and models
that HP surfaces present an ideal media, for the designer who welcomes
and searches for a geometric discipline.
It is commonly thought that the structural form can be derived

from the defined problem. In other words, first the problem is
analysed, then one searches for the proper form. This approach can
handicap the imagination of the inexperienced designer.
It is believed that research should be carried out also in the

reversal of this usual order. That is to say, that structural forms
should be developed of which applications are to be found later. In
other sciences most successful results have been obtained this way.
The derived 'formal vocabulary' in this manner serves merely as the
'raw material' which of course has to be modified according to the
given building problem, with the possibility that its origin might be 112
lost in the final work.
2 .9 0
2.91
113-
REFERENCES
On Structural Forms
1. Joedicke Jurgen, "Schalenbau", Karl Kramer Verlag Stuttgart, 1962.
2. Faber Colin, "Candela: The Shell Builder", Reinhold

Publishing Corporation, 1963.
3. Roland Con^rad, "Frei Otto - Spannweiten", Verlag Ullstein GMBH, 1965-
4. Exhibition Catalogue, Paris April 1 9 6 3 , "structures Nouvelles

en Architecture".
5. Catalano Eduardo F. "Estructuras de Superficies Alabeadas",

Editorial Univesitaria de Buenos Aires, 1 9 6 2 .
6. Burt Michael, "Spatial Arrangement and Polyhedra with Curved

Surfaces and their Architectural Applications",
Technion, Haifa - Israel, 1 9 6 6 .
7. Heger F.J. Chambers R.E. and Dietz A.G.H., "On the Use of Plastics
and Other Composite Materials for Shell Roof Structures", Proceedings
World Conference on Shell Structures Oct. 1962 San Francisco, California.
8. Makowski Z.S. and Zerning J., "Engineering Design Concepts as Applied

to Civil Engineering in Plastics Shell Roof Structures", Conference
on Engineering Design and Performance Testing, Plastics
Institute, Oct. 1968 Manchester.
9. Beveridge W.I.B., "The Art of Scientific Investigation",

Heinemann Educational Books Ltd. London, 1950.
10. Haberli Walter,"Beton Konstruktion und Form", verlag

Stocker-Schmid Dietikon - Zurich, 1966.
11. Catalano Eduardo, "Two Warped Surfaces", Student Publications of

the School of Design, North Carolina State College,
Volume 5 number 1.
1 14
12. Weyl Hermann, "Symmetry", Princeton University Press,
Princeton, New Jersey, 1952.
13. Wolf K.L. and Wolf R . , "Symmetrie", Bohlau-Verlag Miinster/Kdln, 1956.
14. Gross Roland, "Symmetrie", Zeitschrift DBZ 6/ 1 9 6 6 .
U-
13. Kuhn Walter, "Plastische und Raumliche Symmetrien", iysstellung
im Kunstgewerbe H.useum, Zurich, 1 9 6 8 .
16. Klee Paul, "The Thinking Eye", Lund Humphries, London, 1 9 6 1 .
17« Hart Franz, "Kunst and Technik der Wdlbung", Verlag Georg D.W.
Callwey, Miinchen, 1 9 6 3 .
18. Torroja E . , "Logik der Form", Verlag Georg D.W. Callwey, Munchen, 1961.
19. Siegel Curt, "Strukturformen", Verlag Georg D.W. Callwey,

Munchen, i9 6 0 .
20. Engel Heinrich, "Structure Systems", Deutsche Verlags -

Anstalt, Stuttgart, 1967*
21. Bennett J.D., "Hyperbolic Paraboloids", Reinforced Concrete

Association, London, 1961.
22. Angerer Fred, "Surface Structures in Building", Alec Tiranti,

London, 1961.
23. Esquillan N. and Saillard Y . , (Editors) "Hanging Roofs",

North-Holland Publishing Company - Amsterdam, 19 6 3 .
24. Ramaswamy G.S., "Design and Construction of Concrete Shell Roofs",

McGraw-Hill Company, 1 9 6 8 .
115
P A R T T H R E E DESIGN AND CONSTRUCTION OF A SCALLOPED DOME IN
GLASS FIBRE REINFORCED PLASTICS
- requirements for the determination of size
l l 6
Introduction
"In order to invent a structure and to give

it exact proportions, one must follow both
the intuitive and mathematical path11
(Pier Luigi Nervi)
The objective in construction with relatively expensive materials

must be to obtain "the most with the least". In other words, to
enclose a given volume with a structure that will use least possible
material. This aim is as old as mankind and is the ultimate goal of
modern technology.
The solution to this problem, at least so far, cannot be found

even with the aid of a computer.
The designer must first speculate intuitively and then test the
hypothesis by an experiment.
A e $ c r i'hc
An attempt is made here to introduce a method of research that
will assist in the achievement of optimum design, without the
disadvantage of facing the complex and lengthy processes of structural
analysis with the highly mathematical treatment usually associated
with such work.
It is hoped that the approach suggested will free the imagination

of the designer and encourage him to design and build new and
structurally efficient systems, regardless as to whether an analytical
method of design is currently available.
The design and construction of any structure, as well as the

prototype under investigation, involves three main problems:
(l) Architectural design; (2) Engineering analysis; and (3 ) construction
techniques. — .
In this research the engineering analysis constitutes the

principal problem and will have to be dealt here in three parts:
(l) Approximate design of shell segment; (2 ) Construction of a model;
and (3 ) Experimental analysis of the model.
117
SECTION ONE ARCHITECTURAL DESIGN
General Description of Prototype
While any one of the structural forms illustrated in part two

of the thesis be made the subject for a more detailed investigation,
the scalloped dome (Figures 2.32 2.39 and 2 .8 5 ) was chosen for the
prototype for three reasons. Firstly, a dome, being one of the
oldest structural forms known to man, has a very high degree of
structural efficiency. It is a composite "closed system", in which
rigidity is obtained by a combined action of all its parts, thus the
composite assembly is stronger than each of its individual parts.
Secondly, the configuration of a dome will make the large deflections
that are inevitable in using a material like GRP appear less noticeable.
Finally, it was considered that the elemental shape introduced provided
considerable structural advantages in comparison with conventional
dome systems. The disadvantage of a dome is, of course, its circular
floor plan, which is not always easy to integrate in a general
design scheme.
The scalloped small rise dome in Figure 3*1 which illustrates

the general shape of the assembly consists of a number of meridional
HP shell segments (n),in which longitudinal as well as transverse
cross-sections are parabolic (Figure 3*2). All segments in plan are
isosceles triangles. The cross-section depth decreases continuously
towards the top of the dome, and since the thickness (t) is small
compared with the depth of segment, the structure can be considered
"hinged" at the crown. Therefore the ribs of two diametrically opposite
segments (Figure 3-l) constitute a three-hinged arch, the number of
’three-hinged 1 arches being n/ 2 .
The shell segments supported at (n) ’points’ are connected to

a corresponding number of inclined exterior buttresses. The torsional
rigidity of the shell segment is very small, so that hoop forces cannot
be transmitted horizontally across the dome. The triangular HP shell
segments (due to their three-hinged arch analogy) have also the
advantage of being insensitive to differential settlements of their
abutments and also to changes in their contour lines due to thermal
expansion and contraction.
118
A— A ■ iiiii i ii iiiii-iJ l-
15 » t o
3.1
’
119
The rigidity of the thin free edges between the inclined supports
can be substantially increased by the inter-penetration of an
auxiliary shell segments which form canopies as shown in Figure 3-3*
This modification of the principal triangular segment is achieved by
cutting away a lower slice of the principal segment by means of an
inclined plane passing through the points ACF as illustrated in
Figure 3»^- This lower slice is then rotated through l80° and
connected again to the principal segment. A gapless fit is possible
after rotation because the surface has an axis of symmetry in its
axis of rotation.
The units are moulded with integral vertical flanges, which

can be bolted together to form the scalloped dome.
120
C H A P T E R O N E GEOMETRY OF THE SHELL SEGMENT
Definition of Shape
The first step in designing this scalloped dome is of course

to define its geometry mathematically. Since the configuration is
rather novel a more detailed treatment of its complex geometry
follows.
The surface of the shell segment is generated by translating a

principal parabola (2) with upward curvature, parallel to itself,
over another principal parabola (l) with downward curvature Figure
The parabola BO is called the directrix and the parabola AC is called
the generator. We thus obtain an anticlastic surface of translation,
and since both generator and directrix are parabolas, a hyperbolic
paraboloid (HP) is formed. The HP surface is over a rectangular plan,
and its perimeter is formed by the principal parabolas.
In the design of the scalloped small rise dome:
f
^ r- „
0.5 ? and —2 = 1.
See Figure 3«(t» ^ "" ... ........ ^ "
The Hyperbolic Paraboloid Triangular Segment
The scalloped dome consists of a series of meridional elements.

Each meridional element is a HP segment over an isosceles triangular
plan. The triangular segment is obtained by cutting the rectangular
HP surface with two intersecting vertical planes passing through the
points A A ’OO' and C C ’OO' as shown in Figure 3»^»
The horizontal projection of the space curve AO or CO on the XY

plane is the radius (r) of the dome. It is therefore convenient to
6
3.2 3.3
JC"
3.4
122
express all the dimensions in terms of r:
r cos
a
2 ’
r sm a
2 *
In the prefabricated dome n = 26.
The angle of the triangular segment, in its projection on the

XY plane is:
a= ^ = 13.768° or | = 6.884° ,
hence:
tan ^ = 0 .1 1 9 8 6 ,
r cos - = 0.99279 r 1
r sin - = O . H 986 r
If r = 10 metre (32.8 ft.) then each segment will have a width

of 2.4 m. (7*874 ft.); a size that can be easily transported on
standard vehicles.
1. The Skewed Straight-line Generators.
A HP surface is also a doubly ruled surface, that is to say, it can be

generated by two systems of straight lines (generators) that intersect
each other. Each set of straight lines lies parallel to a vertical
plane, called the directrix plane. The two directrix planes form an
123
Arbitrary angle between them. Since f / f ; / 90°--
1 eL
Thus an oblique HP surface is obtained.
To define the HP surface a Cartesian co-ordinate system X,Y,Z,

is used where the Z axis is perpendicular to the XY plane, the angle
o
between the X and Y axes is 90 , and where the origin of the
co-ordinate system is in the middle of the surface at point 0.
The general equation for a HP surface is:
z = h 4 - f2 4 • (3 .D
a b
2. The Angle U) Between the Directrix planes.
By inspecting Figure 3-^t it'can be seen that the directrix plane must
lie in the GOZ plane. The location of point G on the principal
parabola is not known. The straight line generator 0G is at right
angles to the Z axis, and thusat the point G, Z = 0.
The obsicissa of the point G,denoted by x can thus be found

G
by using equation (3*1) and setting Z =0.
2 2
XG y
° = fl ~2 “ f2 "~2 ’
a b
Since at point x , b = y , the equation becomes:

G
2
xr
0 = fl "a - f2 *
a
Solving for x
G
£2
fl
124
Substituting numerical values
*Q = 0.99279 r K § : “ 9 ?6 :9 9 2 7 9 " V) ’
0.488 r
Since ^ is the angle between OX and the straight line generator,
+
tan w
- _ b
- _ 0-11986
0 ^488 _ .
0.2456 , ,
G
| - 1 3 .8 °
3- Equation of the Surface.

Substituting numerical values equation (3-1) becomes:
2 2 2 2
0 .5 *_ . 3~ 3_
a b 0 .9 9 2 9 r 0 .1 1 9 8 6 r ,
j; (0.5036x2 - 8.3431 y 2 ) . (3.1.1)
4. Locus of the Space-Curves CO and AO.
The space-curve CO or AO form the perimeter of the triangular HP

segment. Denote its projection on the XY plane with £ thus:
£
x
------- 5 x - £ cos —
a ; y = x tan
a
—
* cos 0_ b 2 ’ J 2
2
Substituting these values in equation (3.1.1)
125
E
2
z — cos2 § [o.5 0 3 6 - 8.3434 (0 . 1 1986 2)] ,
2
I
= 0.377 ■ — .
Note that the space curve CO is a parabola with a lower rise than
the principal parabola (l) (Figure 3-4 ).
5. Check:
If g = r, then the above equation becomes simply
z = 0.377 r .
By inspection
Thus:
z = 0.4969 r - 0.1199 r = 0.377 r .
The Hyperbolic Paraboloid Canopy
1. Delimitation of Curve AFC

Let J3 =20^ be the angle between the intersecting plane AFC and
the XY plane (Figure 3-5 (2).) the projection of the curveAFC, on
the ZX plane, is a straight line, hence:
z = x tan p + (f^ - *2 ^ “ a tan 3 ’
which can also be written as:
z = (f^ - f ) - (a - x) tan p . (3 .1 .2 )
126
Substituting equation (3.1.2) in (3-1) gives the general formula
for the curve AFC in its horizontal projection on the XY plane,
■which reads:
(fa - f2 ) - (a - x) tan p “ ■ fi % “ (3.1.3)

2 b2
Substituting numerical values:
0 .3 7 6 5 r - (0 .9 9 2 8 r - x) tan p = - (0 .5 0 3 6 x 2 - 8.3431 y2 ) .
2. Location of the Point F.
At point F on the curve AFC, y = 0. Thus substituting the value of y

into equation (3 .1 .3 ) gives:
0 .3 7 6 5 r - 0 .9 9 2 8 r tan p + x tan p = ^ 0 .5 3 6 x 2 ,
0 .5 0 3 6 x 2 - 0.3640 x r = [0.3765 « 0 .9 9 2 8 (0.3640)] r2
Solving for x:
2 0.3640 0.3765 - 0 .9 9 2 8 (0.3640)

x - —— r x =
0 .5 0 3 6 0.5036
x2 - 0 .7 2 2 8 r x = 0.030024 r 2 ,
= 0.3614 r + \j 0 .1 3 0 6 + 0.030024 r ,
= (0.3614 ± 0.4007) r ,
= 0 .7 6 2 2 r.
3. Calculation of the z Ordinate of the Point F.
127
Substituting for x_ in the general equation (3.1.2):
F
z = 0 .3 7 6 5 r - [ 0 .9 9 2 8 (0.3640)] r + 0.364 x ,
= 0.0151 r + 0.364 x ,
Zp = 0.0151 r + [0.364 (0.7622 r )] ,
= 0.2925 r .
4. Locus of the Space Curve AFC.
To calculate the locus of the curve AFC in its horizontal projection

on the XY plane, equation (3.1.3) is used.
0.3765 r2 - [0 .9 9 2 8 (0.364 r2 ) ] + 0.364 m r2
= 0.5036 m 2 r2 - 8.341 y 2
(where m = x/r a dimensionless quantity)
Solving for y ;
2 1
y 0 .5 0 3 6 m 2 - 0.364 m - [ 0 .3 7 6 5 - 0 .9 9 2 8 (0.364)]
8 .3 4 3 1
2
y 0.364 m - 0 .0 1 5 1 2 )
With this equation the projection of the curve AFC on the XY plane can
be plotted as shown in Figure 3-6 .
5. The z Ordinates of the Space-Curve AFC.
Having found the XY co-ordinates for the space curve AFC its z
ordinates can be calculated by using equation (3-1.2) .
z = O .3 7 6 5 r - [0 .9 9 2 8 (0.3640)] r + m 0.3640 .
With this equation the curve AFC projected on the ZY plane can be
plotted as shown in Figure 3*6 . •
6. Equation of the Canopy Surface.
Three operations are necessary in order to obtain the surface equation

of the projected canopy (Figure 3*5). 1. Translation along the X axis.
2. Rotation about the Y axis through 20^. 3- Rotation about the
X axis through 180^ (reversal) .
7. Translation of Co-ordinate System X,Y,Z.
The point F is taken as the new origin of co-ordinates.

Translating the original co-ordinate system (X,Y,Z) to the new
co-ordinate system (X,Y,Z) with its origin at point F gives:
z = z ,
F ’
X X
(3-1.4)
y y
Thus:
z z + z.
F
X X + X.
F
Since x = O .7 6 2 2 r, and z = 0.2926 r .

r r
Substituting these quantities in equation (3.1.2) gives:
z + 0 .2 9 2 6 r = ~ [x + 0 .7 6 2 2 r2 ( 0.5036) - 8.341 y2 ] ,
129
- _ 0 .5 0 3 6 x2 + 0 .5 0 3 6 2 (0 .7 6 2 2 x r) + 0 .7 6 2 2 2 (0 .5 0 3 6 r2 )
z =
8.4315' - 0 .2 9 2 6 r ,
The equation of the hyperbolic paraboloid surface in the (X,Y,Z)

co-ordinates can thus be written as:
0 .5 0 3 6 x 2 « 8.341 y 2
Z = + 0.7677 X - --------------------- (3 .1 .5 )
8. Rotation of X,Y,Z Co-ordinates about the Y Axis.

o
Angle of rotation p = 20 . The new co-ordinate system after
rotation will be denoted by x Y Z • According to vector analysis
(Figure 3 . 5 (4).). P P P
x = x ■ cos p - z p sin p
z = x sin p + z p cos p (3 .1 .6 )
y = y
Hence:
X cos p + z sm
(3 .1.6 .1)
z = -x s in
m p z cos
P
Since: Sin p = Sin 20 = 0.342 ,
Cos p = Cos 20° = 0.940 ,
Substituting these quantities in equation (3.1*5) gives:
x 0.342 + z 0.940 = 5°-- (x 0.940 - z 0.342)2 +

P P P P
+ 0 .7 6 7 7 (0.940 X - z 0.342) -
p p
-2
8 .3 4 3 1 y i3
“ 1
130
-2
0.342 x + 0.940 z = 0.4430 ^-6 - ------- 8--- ^-- 3
p P r r
-2
z
^58l_JS + 0.7216 5
P "
!
8.3431 y2
- 0.2626 Z fi .
P
The equation of the hyperbolic paraboloid surface in
X Y Z co-ordinates is:
P P P
r -2 -2 -2
z _ x y
- 0.0589 ------ - 0.4450 P- + 3 .3 4 3 1 - P - +
r r r
ft z
0 .3 2 3 8 P + 1 .2 0 2 6 Z p - 0 .3 7 9 6 x = 0 . (3.1.7)
9. Rotation of the Surface about X axis through 180 (reversal)
In the new co-ordinate system after reversal
(3 .1 .8 )
p ’
x _ and y remain unchanged.

P P
Hence the equation of the canopy surface is:
=2 -2 -2
z x y
O.O 589 — + 0.445 — — - 8.341-------2.
X z
p p
- 0 .3 2 3 8 -- ----- - 1 .2 0 2 6 z + 0 .3 7 9 6 x = 0 (3.1.9)
P r
131
10. Location of Point B (reversal)
To find the point B (rev.) on the tilted parabolic free edge

as shown in Figures 3-3 and 3-5(2) in X,Y,Z co-ordinates requires
the following three steps.
(1) Translation:
B = ° ’
x = 0 .2 3 0 6 r .
B
Substituting these quantities in equation (3-1-5)
zB = [0 .5 0 3 6 (0 .2 3 0 6 2) + 0.7677 (0 .2 3 0 6 )] r,
= (0 .0 2 6 8 + 0 . 1 7 7 0 ) r ,
0 .2 0 3 8 r .
(2) Rotation 20°:
Using equation(3.1.6.l)
£ = 0 .2 3 0 6 (0.946) r + 0 .2 0 3 8 (0.342) r ,
pB
= (0 .2 1 6 8 + 0 .0 7 0 0 ) r ,
= 0 .2 8 6 8 r .
z = - 0 .2 3 0 6 (0.342) r + 0 .2 0 3 8 (0.940) r ,
PB
= (-O.O789 + 0 .1 9 1 6 ) r ,
= 0.1127 r .
132
(3) Rotation 180°:
For the point B (rev.)
x = 0.2865 r ,
z = - 0.1127 r .
P
Substituting these values in equation (3.1.6):
xD = [ 0 .2 8 6 5 (0.940) + 0 . 1 1 2 7 (0.342) 1 r ,
B ■*
= (0.2693 + 0.03854) r ,
0.30784 r .
z = [0 .2 8 6 5 (0.342)-(0.1127) 0.940 ] r ,
B 1 J
= ( 0.09798 - 0.1059 ) r ,
= - 0 .0 0 7 9 6 r .
11. The Free Edge Curve A B (rev.) C
Similarly, to determine the equation of the tiltedparabolic free

edge in X ^ Z ^co-ordinates requires the following three steps.
(1 ) Translation:
Using equation (3.1.1) and substituting x =a = 0.9928 r;
z = [ 0 .5 0 3 6 (0 .9 9 2 8 2 ) r2 - 8.3431 y2 ] ,
2
= - 8.341 ^ + 0.4964 r .
(2) Rotation 20°
Since z = z + 0.2926 r ?
y = y •
Hence the above equation in the X Y Z co-ordinates becomes:
2
z 8.341 - + 0 .2 0 3 8 r .
r
(3) Rotation 180^ :
The equation of the curve in i Y Z co-ordinates according to

equation (3-1.6) becomes:
P P P
2
x 0.342 + z 0.940 = - 8.341 £ + 0 .2 0 3 8 r ,
P P
and:
_2
z„ = - 0.3640 X - 8 .8 7 5 6 — P + 0.2168 r .
P P
The equation of the space curve in X Y Z co-ordinates according
to equation (3.1.8) becomes:
P P P
2,
z = + 0.364 x„ + 8 .8 7 5 6 — P- 0 .2 1 6 8 r. (3 . 1 .1 0 )
P P
12. Check:
Substituting in equation (3.1.10) the ordinate of point B (rev.) which
is known. The equation of the space curve AB (rev) C in X Y Z
co-ordinates using equation (3.1.10) and (3.1.6.1) gives:
- x 0.342 + z 0.940 = 0.364 (x 0.940 + z 0.342) +
-2
+ 8 .8 7 5 6 - - 0 .2 1 6 8 r ,
which can also be rewritten as
134
■\ z [ 0.940 - 0.364 (0.342)] = X [ 0 .3 4 2 + 0.364 (0.940)] .+
2
+ 8 .8 7 5 6 ~ - 0 .2 1 6 8 r ,
2
z (0.8155) = x (0.684) + 8 .8 7 5 6 - - 0 .2 1 6 8 r ,
r
2
z ■= x 0.83875+ 1 0 . 8 8 3 6 ^ - 0 .2 6 5 8 r . (3.1.11)
If x 0.30784 r ,
z = 0 .2 5 8 2 r - 0 .2 6 5 8 r = - 0 .0 0 7 6 r ,
which corresponds closely enough to the previously calculated
z =- 0 .0 0 7 9 6 r .
13. Equation of the Space Curve AB (rev) C Projected on the XY Plane.

The vertical projection of the space curve AB (rev) C gives a straight
line which passes through the points AC - B (rev) as shown in
Figure 3.5.(5) •
(z - z^) = tany (x - x^) . (3.1.12)
Since
y = - 50° tan Y (-50°) = - 1.1918 ,
\ = + 0.0839 r ,
x^ = 0.2306 r .
Hence
z = - 1 .1 9 1 8 x + 1 .1 9 1 8 (0.2306)r + 0.0839 r. (3 . 1 . 1 3 )
Using equation (3.1.11) and (3-1.13) one unknown can be eliminated
135
“2
O .8 3 8 7 5 x + IO .8 8 3 6 ^ - O .2 6 5 8 r =
- 1.1918 x + 1 .1 9 1 8 (0 .2 3 0 6 ) r + 0 .0 8 3 9 r ,
y2 = - 0.18657 x r + 0.0574 r2 . (3.1.14)
Translating equation (3.1.14) to the co-ordinates X' Y ’

Figure 3-5»(3) If x = x* + 0.2306 r
y'2 = - 0.18657 x» r + 0.014377 r2 . (3 . 1 .1 5 )
136
0 .9 9 2 8 r
-ox
0 .2 9 2 6 r
0.3 7 7 Or
sr—
0.119 9 r
0 .2 3 0 6 r
■0.0839r
0.11 2 7 r
0 .2 0 3 8 r
0.119 9r
0.11 2 7r
_ L1
0 .2 3 0 6 r
------------- 0 .3 0 7 8 r
(rev.)
X
0.1199 r
Y=50
A,C
->X
(rev.)
t* Xi = 0. 230G r — H
0.119 9 r
3.5
137
------- a= 0.99 28r
0*0167 r-
0*0000 -
-X
00000-
-OOMOr- - f - d=r/io —|—
0*3765 r
0*3660 r-
-f- b — f—
3.6
138
S E C T I O N T W O ENGINEERING ANALYSIS
C H A P T E R O N E APPROXIMATE DESIGN OF SHELL SEGMENT
In thin shells, as in our case, the buckling phenomenon has to

be considered, hence the buckling behaviour will determine the
stability of the shell.
An exact theoretical approach for determining the critical

buckling stress in the scalloped dome is at present not available.
In order to determine the mode of buckling it was necessary therefore
to construct a model. The important unknown factor in the model is
the required minimum thickness of the shell. The experimental model
technique can effectively be used as an iterative method, for
finding an optimum structural solution.
In our case this simply means, building first a model with a shell
thickness as thin as possible and loading it. From the observed and
measured deformations one can determine the critical buckling mode.
An unstable shell might have to be stiffended or by changing the
curvature of the shell or increasing its thickness.
Of course it would be advantageous and time-saving if one could

predict analytically, with a sufficient degree of accuracy, the
stresses and if possible also the deflections in the full-size
scalloped dome when subjected to the maximum live loads.
The three-pinned arch method will be used to predict approximately

the stresses in the shell segment. This assumption is valid because
the very thin scalloped skin of the dome can transmit only very small
hoop forces, and the dome is kept in equilibrium primarily by the
meridional forces along the stiff ribs.
The advantages of the three-pinned arch method are as follows:
(a) It brings shell analysis within the reach of those who are
unfamiliar with the techniques of advanced mathematics.
(b) It can be easily applied to shells with non-uniform thickness.
(c) It can handle shells strengthened by longitudinal and transverse ribs.
(d) It is a direct and comprehensible method.
‘1 3 9
Three - Pinned Arch Analysis
1. Vertical Loading
2 2
Only the live load (lb/ft or kg/m ) of plan area is considered, its
intensity is denoted by (q). The self weight of the segments, being
small, are included in q. With a roof load of q lb/ft , the load per
unit length of arch varies linearly from q^ at the outer ring of
radius r, to zero at the centre (Figure 3-7») •
2. Basic Beam Stresses

Considering first the three-pinned arch as a simply-supported beam under
vertical loading, q Jjj * spanning from A to B (Figure 3»7») •
2 TX r
qr = q ,
where n is the number of segments .
Similarly
_ 2 Tl g
96 = — 9
The reactions R equal:
2
RA - RB = -^r 9
The midspan beam - moment M at point 0 equals
M
2 2|
Tl r q3 r
o n n n
3
11 r q
n
3
II r q
3n
140
(-IE
N,
3-7
r=975em
■4o—
A= 12 0 c m 2
A = 132cm 2 A= 90cm 2
I 2 6 .5 1 = 5271cm 4 1= 374cm 4
48.5 I t = 1.8 cm
n.x.
I T
20 t= 0 .6 c m
_L 95cm
168cm B-B C-C
A-A
3-i
The bending moment at any point with abscissa ^ is:
_ j A ( g) _ jt A ( a r +v . Jt£a ( h ) ,
P n n 3 n 3
(r + g_2 r _g) _ J£l_a (i g )

n = 3 n 3 *
•j 'j
TCr q _ Ttl q
n3 n3 ’
( r3" S3 ) ,
_ 3 e3
3n ( i T A r3- > •
( + to the right of 0 ; - to the left of 0 )
The shearing force S at any point with abscissa £ is:
sS P - 2
- +
2 n
gq (I )
K 2 ' '
K2
r
+ TX 5 q
n
3. Arch Stresses
Horizontal thrust:
The thrust H is obtained by equating to zero the moments about 0 , i.e,
IM = 0 ,
o 1
which can also be written as
M H (f ) .
o o
142
Hence:
M
o
H
f ’
which is equal to:
-r r2 q
71
3 (0.3765) n
2
TCr q
0.8853
n
The bending moment at any point with abscissa ^ is
M M k - H (f - z) ,
f (b.b.r
(suffices (b.b.) denotes basic beam )

which equals:
M (1 + ) - 0.8853 - rn ^ 0.3765 r (1 - ) *
Tt r3 q
(1 ) - 0.8853 (0.3756) (1 - — - )
n
Tlr"1 q 1
n 3
TC r3 q
M (3 -2 .1)
n M
Where F,, is a dimensionless factor (Figure 3-9)

M
l zi 3
4. Axial Forces
N = + S sin <p + H cos ip , (Figure 3-7.) ;

§ s (b.b.)
N (+) = compression,
N (-) = tension,
where:
d z
tan ip d = 2 (0.3765 2 ) = 0.753 2 ,
cos ip
I1 + tan m w
1 + 0 .3 6 7 £
2 ’
tan tp 0.753 r
sin ip =
1 + tan *p
il + 0.567 ~
r
so that:
2
11 £ q 0.753 r
N = +
n
1 + 0.567
U r 2 c
+ 0.8853
n
1 + 0.567 ~
r
2
TC r q 0.8853 - 0.7530 r
n
1 + 0.567
144
2
ft r q
N
n
(3 -2 .2)
N
where F is a dimensionless factor (Figure 3-9)

N
5. Shearing Forces
S s= S Cos q) - H sin q) ,
» £ (b.b.)
so that
1 + 0.567
Tl r2 q 0.7530 r
- O .8 8 5 3
n
1 + 0.567
-
+ —
s
w 2
Tl r g
n
1 + O .5 6 7
Tl r2 q
(3-2.3)
n S ’
where F is a dimensionless factor (Figure 3-9) -

*S>
F F F
P t. E /r (g /r)2 (g /r)3 M N S
A 1 .0 0 - 1 .0 0 0 0 .0 0 0 0 0 1 .3 0 8 7 - 0 .2 6 6 2 5
o
o
i
l 1 -0 .9 0 0 .8 l - 0 .7 2 9 0 .0 2 6 9 9 1 .1 8 7 0 - 0 .1 7 3 9
2* - 0 .8 0 0 .6 4 - 0 .5 1 2 0 .0 4 2 6 6 1 .0 8 8 8 - 0 .0 9 1 4 1
3' - 0 .7 0 0 .4 9 - 0 .3 4 3 0 .0 4 8 9 9 1 .0 1 2 0 - 0 .0 2 0 7
4* - 0 .6 0 0 .3 6 - 0 .2 1 6 0 .0 4 7 9 9 0 .9 5 5 1 + 0 .0 3 6 4 2
5' -0 .5 0 0 .2 5 - 0 .1 2 5 0 .0 4 1 6 6 0 .9 1 6 9 + 0 .0 7 7 9 5
6' -0 .4 0 0 .1 6 -0 .0 6 4 0 .0 3 1 9 9 0 .8 9 4 1 5 + 0 .1 0 2 1 5
7' - 0 .3 0 0 .0 9 - 0 .0 2 7 0 .0 2 0 9 9 0 .8 6 1 6 + 0 .1 0 7 2 7
DO
«
0 .0 4 - 0 .0 0 8 0 .0 1 0 6 6 0 .8 7 2 0 + 0 .0 9 2 3 0
o
8'
O
•
9' ‘ 0 .0 1 -0 .0 0 1 + 0 .0 5 6 5 3
o
o
0 .0 0 2 9 9 0 .8 8 3 5
1
0 o .o o 0 .0 0 0 .0 0 0 0 .0 0 0 0 0 0 .8 8 5 3 0 .0 0 0 0 0
1 + 0 .1 0 0 .0 1 0 .0 0 1 0 .0 0 2 9 9 0 .8 8 3 5 - 0 .0 5 6 5 3
2 + 0 .2 0 0 .0 4 0 .0 0 8 0 .0 1 0 6 6 0 .8 7 2 0 * - 0 .0 9 2 3 0
3 +0 .3 0 0 .0 9 0 .0 2 7 0 .0 2 0 9 9 0 .8 6 1 6 - 0 .1 0 7 2 7
4 + 0 .4 0 0 .1 6 0 .0 6 4 0 .0 3 1 9 9 0 .8 9 4 1 5 - 0 .1 0 2 1 5
5 +0 .5 0 0 .2 5 0 .1 2 5 0 .0 4 1 6 6 0 .9 1 6 9 - 0 .0 7 7 9 5
6 +0 .6 0 0 .3 6 0 .2 1 6 0 .0 4 7 9 9 0 .9 5 5 1 - 0 .0 3 6 4 2
7 +0 .7 0 0 .4 9 0 .3 4 3 0 .0 4 8 9 9 1 .0 1 2 0 + 0 .0 2 0 7
8 +0 .8 0 ' 0 .6 4 0 .5 1 2 0 .0 4 2 6 6 1 .0 8 8 8 + 0 .0 9 1 4 1
9- + 0 .9 0 0 .8 1 0 .7 2 9 0 .0 2 6 9 9 1 .1 8 7 0 + 0 .1 7 3 9
B + 1 .0 0 1 .0 0 1 .0 0 0 1 .0 0 0 0 0 1 .3 0 8 7 + 0 .2 6 6 2 5
3-9
'1 4 6
Approximate Design of Cross-Section Required
The geometrical relationships described in Figure define

the cross-section of the arch segment. The vertical cross-sectional
depth f varies continuously from maximum, where f = O.II 986 r to
minimum where f = o. The arch segment was sliced into three
sections as shown in Figure 3*8. The minimum thickness of the shell
must now be calculated.
1. Position of Neutral Axis
The arch segment, even under its own dead weight, will produce
a state of stress defined as bending. That is to say, the bending
stresses vary linearly from a maximum compressive value at the top
fibre to a maximum tensile value at the bottom fibre. The axis
passing through the points where there is no bending stress is
called the neutral axis.
The determination of the neutral axis for the various cross

sections in Figure 3*8 was done by a graphical treatment using the
link-polygon method.
2. Moment of Inertia about Neutral Axis
The bending stiffness of a cross sectional shape is gauged by a

quantity called its moment of inertia, (second moment of area) usually
denoted as (i).
The approximate I for the various cross sections was found

geometrically using the full scale drawing of the respective cross
section in the model. The same drawing that served to find the
neutral axis was also used to compute I.
3. Check for Bending Moment
Vertical loading uniformly distributed over full span.
The bending moment (M)for the various cross sections can be

calculated using equation (3 .2 .1 ) hence:
The bending stress f at top of cross section (compression)
1 I
The bending stress f at bottom of cross section (tension)
- M h2
2 “ ~T~
For example: r = 1,000 cm.; n = 26;

2
intensity of load q = 15 lb/sq.ft (0 .0 0 7 3 2 kg/cm );
F^ (max.) at section A - A = 0.04899 (Figure 3*9 point 7);
and where h^ and h„are the distances from the neutralaxis

1 2
to the top and bottom fibre of the cross section;
4
and I at section A - A = 26,306 cm (Figure 3*8).
Substituting these quantities in the above equations:
M (A - A) = ^ ,'53° kg/ cm #
^ , 43,330 kg/cm2 (2 8 .5 cm) , 2 trr„ . \

f , * =--- 1------------r---------- = 47 kg/cm (6 6 7 p.s. 1 .).
1 ' 2 6 ;3 0 6 cm
f2 (A - A) = 43,330 kg/ cm (2Q cm) = 33 kg/cm2 (468 p.s.i.).

26,306 cm^
4. Check on Axial Thrust
The compressive stress(C)due to axial thrust is:
_ axial thrust_____________
(section) cross-section of area of segment .
The axial thrust for the section(A - A)can be calculated using

equation (3.2.2),
hence:
C (A - A) = 893 kg = 7.45 kg/cm2 (105.8 p.s.i.).

120 cm2
148
5. Minimum Thickness for Structural Stability
The maximum stress at section A - A due to the bending moment and

2 2 » 2
the axial thrust is therefore: 47 kg/cm + 7-45 kg/cm = 54.45 kg/cm
(773 p.s.i.).
Assuming an allowable working stress = 2,400 p.s.i.

2
(1 6 8 kg/cm ) (l/5 of short term ultimate stress), then the maximum
stress is well within the allowable limit.
Figure 3-8 shows the various values for t, in order to obtain

an arch segment of "constant bending strength", viz. segment with
approximately equal stress levels, at extreme fibres, at all the
cross sections.
149
C H A P T E R T W O EXPERIMENTAL MODEL ANALYSIS
Generally, there are two approaches available for studying

the behaviour of a structure: (l) Analytical analysis; and
(2) Experimental analysis.
The analytical approach is based on the idealization of a

physical problem expressed in pure mathematical terms, to which an
appropriate generalized theory was found; one really deals here
with a "mathematical model".
Alternatively, the experimental analysis makes use of a

"physical model".
Each approach has its merits and limitations. An experimental

method will be advantageous when dealing with novel shapes and complex
boundary conditions, which are not easily expressed analytically.
In most countries, nowadays, the results of a model analysis can
be offered in lieu of a theoretical calculation.
The model technique, which is very old (Michelangelo used it to

visualise the forces in the dome of St. Peter's in Rome) has recently
become a powerful tool for the engineer due mainly to the
development of modern electronic equipment and the refinement of the
electrical resistance strain-gauge.
The experimental analysis involves three technical problems:

(l) Construction of a model; (2) Loading arrangement for the model;
and (3 ) Measurement of deformations, evaluation of measurements and
their interpretation for the actual prototype structure.
150
Relationships Between Model and Prototype Quantities
In a model experiment, we are ultimately interested in

arriving at the stresses and displacements in the prototype based
on the observed results on the model. This means that we need
relations connecting corresponding quantities for the model and
prototype. In problems of statics the variables are (Ref.5):
1) 0* = stress in shell
2) a,b = linear dimensions of shell
3) t = thickness of the shell

4) E = Young's modulus
5) V = Poisson's ratio
6) y any displacement
7) p = a concentrated load
8) q = a uniformly distributed load
In the prototype we know. E, y , a, b, P, q and density of the

material.
In planning the model experiment, a convenient size (scale
factor) is chosen. The E and v values for the selected material
in the model are also known, similarly the appropriateloading
procedures. The remaining relations can be deduced.
Assuming: y^ = y^ the following relationships hold:
Deflection: y = y S, (3*4.1)
;p Jm ’
(Suffix m refers to the model and suffix p refers to the prototype)

where S is the scale factor.
E
Forces: F = S ■— F . (3*4.2)
p E m
m
1 E
Moments: M = S M (3*4.3)
p E m
E
Stresses: (j* = rr" . (3.4.4)
P m
™
a a
Dimensions: , (3.4.5)
D D
m p
am = ap , (3.4.6)
t t
m p
(These relations are just statements of geometrical

similarity, which means that all the geometric
dimensions of prototype and model are scaled down
to the chosen proportion. Hence:
a = Sa ; b = Sb and t «= St ).
p m p m p m
For a uniformly distributed load:
E
<1m “ q
p
—
E
(3 -4 .7 )
P
If we construct both model and prototype of the same material

and assuming that it is isotropic and that the E value does not vary
with thickness, then:
Stress in model = stress in prototype; /
Buckling stress in model = buckling stress in prototype;

1
Deflection m model = — deflection m prototype;
1
Bending moment in model = — bending moment in prototype.
S3
Selection of Scale Factor
For obvious reasons it was decided to build a model as large
as possible, and to make it in the same material and moulding technique
as the prototype. The available space permitted a model with a clear
span of 4 m. (13 ft).
152
Since the prototype scalloped dome had a clear span of
20 m. (6 5 .6 ft.) a scale factor of 5 was established, thus S =
Dimension of Model
The length 200 cm. (6 .5 6 ft.) and width 49 cm. (1.6 ft.) of
the HP triangular segment was determined by the scale factor.
The preliminary approximate calculations established a shell

with a continuous increase in thickness, as it approaches the
crown of the dome, between 6 mm. to 24 mm. (0.234 in. to 0 .9 3 6 in.).
In the model, therefore, t (min.) = = 1.2 mm; (0.0468 in.)

and t (max.) = — = 4.8 mm. (0.1 8 7 2 in.) (Figure 3-8).
5
153
CHAPTER THREE CONSTRUCTION OF MODEL
Mould Construction
The master (mould) for the production of large size GRP shells,
by contact moulding (hand lay up technique) is generally made with
plaster applied over a timber framework. This technique is far from
satisfactory, as the construction of the mould is complicated and
expensive. To replace this inadequate method the author has developed
a special technique known as cocooning which provides the profile
required for highly warped HP surfaces at low cost. In this technique
a sprayable quick-drying vinyl latex is applied to form initially a
"cobweb" which can bridge sizeable gaps up to 60 cm. (l.9 7 ft) wide
over any reticulated network as shown in Figure 3»l6. After completing
the spraying process a continuous seamless leather-like skin exists
over the sprayed area. During the drying process, the latex of
the "cobweb" shrinks, forming a prestressed membrane with "minimum”
surface (Figure 3'17)» GRP laiminates can be produced from the skin
covering the area without any difficulty to form a single element or,
alternatively,a more robust GRP mould can be produced first.
A brief description will show how a mould can be made. A mould

produced by this process consists basically of three elements:
(l) A large box made of plywood; (2) A course network of nylon cords;
and (3) A skin of vinyl-latex.
1. The Box
The three sides and base illustrated in Figure 3-10 are made of
4 " plywood sheets. (The "box" for the elements of the prototype
structure, will not be made in plywood as a lattice timber framework
construction is more practical and economical).
The calculations set out in Chapter One were used to determine the
exact contour lines for the panels. Figure 3»6 shows the dimensionless
values for the various points on the curves. To obtain a numerical
value the values merely had to be multipled by (r) the radius of the
scalloped dome model - 200 cm. (6 .5 6 ft).
15^
3.10 3.11
3.12
Figure 3.10 shows the panels before assembly. Note the
parallel line grid on the triangular base plate. These lines are
the horizontal projections of the HP straight line generators;
and will be referred to again later.
The assembly of the various panels is shown in Figure 3 - H -

Additional ply panels and a number of battens were built inside
the mould in order to stiffen the sides in their transverse
direction. The inner curved wall is made of l/l6 in. plywood. Its
purpose was to aid in locating the segmental shell cut-out that
formed the projected canopy. All the wooden components were glued
together with epoxy resin.
2. The Cord Network
Having completed the woodwork, the next step, was to locate

the positions of the straight line generators on the curved ply
edges. If the edges of the HP segment had been straight, any
arbitrary division of the edge into equal distances would have
determined the locations of the generators. With a curved edge the
procedure is more involved.
The locations of the straight line generators on a curved edge

can be found mathematically or graphically. Here the graphical
method was used. As already mentioned, the horizontal projection
of all the straight line generators were marked on the base plate
of the mould and it was then simply a matter of transferring them to the
curved edge. Only the points of intersection between the straight
line generators and the outer face of the boundary plane of the
mould were required for constructing the network. For this purpose
a large L square was used as shown in Figure 3-12.
It was desirable that the cord network should be stretched over

the sides of the mould panels flush with its surface. This was
achieved by using a flexible cord that could bend sharply across
the plywood edge.
Nylon cords were found most suitable. Figure 3-13 shows the
prestressing device - a lever - and the fixing scheme for the cord.
156
The fastening device consisted of a l/4 in. steel bolt & nut
and a fibre washer. The bolt had a fine hole just below its head.
While prestressing, the cord could pass freely through the bolt.
When the desired prestressing level was reached the nut was
tightened; friction between panel and washer held the cord in its
prestressed position.
3. The Skin of Vinyl-latex
With the whole of the network completed, (Figure 3»l4) the

spraying process that forms the skin can begin. Before spraying with
vinly latex, all the parts that will be in contact with the skin
were coated with a special adhesive (R.M.193)•*
A mixture of three parts of !tkoonkoten Plastic Type 3?* and one

part "Webbing Agent"* was sprayed on the network. This coat bridged
all the open meshes of the net forming a tissue-like surface as shown
in Figure 3«15» At this stage the tissue was rather weak and
quite brittle.
A number of separate coatings of Plastic Type 3 without the

Webbing Agent were then sprayed over the tissue forming a seamless
leather-like skin to a thickness of 30 mils (0 . 7 6 mm).
While the sprayed-on plastic skin dried on the mould it shrank

in all directions forming a prestressed membrane with "minimum"
surface (Figure 3»1?)« Note the diamond-like pattern on the surface
which was the result of the skin shrinkage between the skewed cords,
which thus formed numerous minimum surfaces for their particular
boundary conditions. The overall shrinking effect was so significant
that it deformed the cord network. Initially all the cords followed
straight lines, but after the skin was sprayed on they became catenaries.
Therefore, strictly speaking one is no longer dealing here with a true
HP surface.
* (All the materials for the spraying technique were supplied by

R.A. Brand & Co. Ltd., and in addition to adhesive R.M.193
the trade names of the other products were: "koonkote" Plastic
type 3; and a "Webbing Agent")
3.14 3.15
3.16
158
Two identical formworks were built in order to speed up production.
Standard compressed-air spraying equipment was used and it
consisted basically of a spray gun, a pressure feed floor container
and an air compressor.
Plastic Type 3 + Webbing Agent, was sprayed at a pot pressure

of 10-15 lb/sq.in. and at a gun pressure of 30 - 40 lb/sq.in. The
gun had a No. 203 (He Vilbiss) nozzle.
Plastic Type 3 without the Webbing Agent, was sprayed at a pot

pressure of 30 lb/sq.in. and a gun pressure of 70 - 90 lb/sq.in.
Thegun had a No. 7 65 (De Vilbiss) nozzle.
Moulding Technique
The vinyl-latex skin of the mould was ready for use after it had
dried thoroughly for about 48 hours. The first operation in producing
a GRP laminate was, of course, to wax polish and apply a release agent
on the mould surface. A floor wax and a release agent were used.
(As generally known, polyester resin will not adhere to a vinyl-

latex, therefore, if only a few laminates had to be made, the mould
surface did not need treatment with a release agent).
Several pre-cut triangular chopped strand mats, weighing 1 oz.

sq.ft. each, and impregnated with a fire retardent polyester resin
(No. SK19187/1 supplied by Bakelite Ltd.) were laid on the mould. The
required varied thickness of the shell was controlled by the number of
chopped strand mat layers and the glass to resin ratio. The glass resin
ratio was 1: 2.5 by weight = 1: 5 by volume; no gel coat was used. The
impregnation and consolidation of the laminate was done with a brush
and a in. steel washer roller (Figure 3«l8 and 3»19)» The GRP laminate
was moulded over the edge of the HP surface thus forming a vertical flange.
Two workmen needed 20 minutes to complete one laminate.
Two laminates were made each morning and left to cure at room
temperature. The segments were carefully removed from the mould on
the following day (Figure 3»20), and washed and cleaned with soap and
water. A portable diamond cutter, driven by compressed air, was used
to trim the edges of the flanges to a height of 3/4 in. Next small
holes were drilled, with a template, into all the flanges.
159
3.19 3.20
160
Finally, the auxiliary shell segment, that formed the projected
canopy was cut out from the principal HP segment, turned over l80°
and laminated to the same segment, A lr in. wide strip of chopped
strand mat impregnated with polyester resin formed a continuous joint
between the two shell segments.
Assembly of Model
The assembled dome is shown in Figure 3*31* It consisted basically

of four elements: (l) An unreinforced concrete foundation; (2) Inclined
steel supports; (3) A single skin GRP shell; and (4) A steel compression
ring at the crown of the dome.
1. Concrete Foundation
The polyhedral foundation was made of 26 truncated concrete solids

monolithically joined; no steel was used for reinforcement. The
forms into which concrete was poured into were made of -jj in. hard-
board, the abutting surfaces being jointed together with 2 in. wide
chopped strand mat strips (Figure 3-24). The radial internal partitions
consisted of \ in. plywood diaphragms. The hardboard walls were

stiffened horizontally with 1 by 2 in. wood battens. In addition two
steel cables were -wrapped around the outer walls to hold the entire
shuttering tightly together. To reduce the quantity of concrete
used, 32 hollow wooden boxes were placed inside the shuttering as
shown in Figure 3*23.
The concrete foundation was constructed in this manner merely for

convenience, the main reason being the necessity to raise the entire
dome a sufficient height from the ground to accommodate a loading
platform underneath.
2. Steel Supports
The design of the supports is of prime importance in a shell

structure. They have to fulfil a number of functions:
(l) Resist all the possible forces that might act, for example,
axial forces, downward, upward and sideways bending moments and
torsion; (2) transmit the loads to the ground in a most direct way;
l6l
------------------ i« i ------------------
-t- SIO t-
4
I
3-21
t / f > b e ll
3-22
162
3.23 3.24
3-25 3.26
3.27 3.28
163
(3) prevent high stress concentrations in the thin shell, at the
points of support; and (4) express the lightness of the shell structure
ideally it should appear to "soar11 upwards.
It is believed that the design chosen satisfied all these points.
The supports were designed in the form of a slender skeleton

tetrahedra, that is to say, broad at the base and tapering to a
"point11 at the top with a triangular brace at half length. The members
were of £ in. steel stubing with a % in. steel base plate (Figure 3-2l)
Figure 3-26 shows in detail the connection between the concrete

foundation and the steel support, before encasement in concrete.
It allowed for rotational as well as translational re-adjustment.
Figure 3*22 shows in detail the connection between the steel

support and the GRP shell. It consisted of two traingular steel plates
•g- in. thick, one of which had the shape of a slender isosceles and the
other an equilateral triangle; the latter was threaded at its centre.
The two plates were welded together along one side to form an L shape.
The L shaped unit was later to be laminated into the inside valley
between two segments, thus forming again a tetrahedra. A 3/l6 in - bolt
with two locking nuts formed the "pin connection" joint between shell
and support.
3. The Shell
Segments were bolted together along their upwards pointing

vertical flange, then turned over and laid on a wooden lattice
falsework (Figure 3-27)-
A strip of chopped strand mat tapering from 2 in. to 1 in. was

laminated over the full length of the joint. The L shaped steel plates
that formed the connection between shell and support, were also
laminated into the joint.
After the GRP strips had cured on the falsework, the segments were
again turned over, bolts taken out, and the projected vertical flanges'
were cut off with the diamond cutter (Figure 3-28). (For experimental
reasons it was decided to cut off the flanges, despite the obvious
fact that their omission would reduce the stiffness of the dome).
For easy handling segments were first jointed in pairs then in
fours and finally in two groups of eight plus one group of ten. The
three completed sets were put on their steel supports, and the
remaining joints were laminated in situ from inside.
The simple method used here for joining, ensured structural

continuity between the segments and was at the same time also easy
to make. Its disadvantage lies in the fact that it does not permit
easy dismantling.
4. Steel Ring
The bicycle wheel type compression ring at the crown of the

dome (Figure 3-29 and 3-30) was dictated by structural considerations:
(l) it had a high torsional rigidity which effectively stiffened
the somewhat weaker upper part of the dome against unsymmetrical
live load; (2) it was light in weight; (3) it served as a skylight
and fresh air circulator; and (4) it had the formal and visual
consistency with the supports and the rest of the dome.
The ring was made of 52 lattice tetrahedras, joined base to base

on to a polyhedral strip. The polyhedral ring was made from a steel
*1 1
plate "8 in. thick, and the lattice-work was all made of q in. steel rods.
Figure 3-30 shows in detail the connection between the compression

ring and the shell. The "pin-connecting” joint comprised a
"8 in. steel rod. The rods were laminated in situ into the inside
valleys between two segments. They had each four fine "branch" like
arms in order to ensure a better grip between shell and rod. A nut
at the open end of each rod allowed for small translational re-adjustments.
165
3-30
166
'167
Loading Arrangement
Various methods of load application to the dome structure were

considered; pneumatic devices, hydraulic jacks, screw jacks, dead
load with levers and dead load with lead shot "blankets”.
The dead loading method with a system of levers, as shown in

Figure 3«35 was found most suitable.
Since it was desirable to deal with hanging weights of equal

intensity throughout, it was first necessary to subdivide each shell
segment, in its horizontal projection, into a large number of small
quadrilateral elements of equal area. Figure 3-32 shows the
intermediate and final subdivision of the plane area of the HP segment.
It consists of 52 small quadrilateral elements; some of them are
quite different in outline, yet they all cover the same area in plan.
Next, the centroid of each quadrilateral element had to be

located. Having completed the lay-out of the grid, it had to be
projected upwards on to the sloping surface of the HP segment as shown
in Figure 3»33« This GRP segment served as the drilling template for
the scalloped dome, altogether 1 ,3 5 2 holes were drilled through the
2 2
shell which corresponds to one loading point per 93 cm (l4.4 in )
(Figure 3*36). Nylon ropes, 3/32 in. diameter, were threaded through
the holes. At their upper end they were fastened to a small circular
pad of plastic foam, which served as a load distributing device. To
enable any desired re-adjustment in the rope length, a strip of nPerspex
with three holes in it was fastended to the free end of the rope.
Friction between the tight rope passing through the three holes,
held the rope in the desired position.
To each set of four padded loading points, three levers were

attached, to which a single load hook was fastened (Figure 3*35)•
Various weights could be added to the hooks to load any part of the
structure as required.
In an unloaded condition all the hanging loads rested on a platform

under the model as shown in Figure 3*35- By lowering the loading
platform, with the aid of three hydraulic jacks, the weights were made
to hang freely and react simultaneously on the whole or part of the dome
3.32 3-33
si/
3.34
169
3.35
3.36
170
171
The loading platform was entirely made of timber. It consisted
basically of three triangular shaped braced trusses, made of 2 in.
by 4 in. joists, fastened together at their ends with \ in. plywood
gusset plates. A hydraulic jack was located at each of the
three corners.
Test Programme
Roofs must be designed to support in addition to their own

dead weight; (l) A live load, commonly due to snow; (2) Concentrated
loads due to special equipment and occasional maintenance work;
(3) Pressures and suctions due to wind; (4) Thermal loads due to the
expansion and contraction of the building material; (5) Settlement
loads due to uneven settlement of the supports; and (6 ) Dynamic
loads, for example due to impact load, resonant load and
earthquake motion.
1. Dead Load
The dead weight of one completed segment of the testing model is;
2
2 kg. and its projected plan area is 0.48 m . This corresponds to
4.2 kg/m^ (0 .8 6 lb/sq.ft).
The dead weight of prototype and model are related by the scale
3
factor, i.e: ¥ = ¥ (S ), (see page )-where the dead weiqht of
p m 7
the whole model ¥ with 26 segments is

m
¥m = 2 kg (2 6 ) = 32 kg.
The dead weight of the prototype in which the scale factor

S = 5 will be:
¥ = 52 kg (5 ^) = 6 ,5 0 0 kg. (l4,300 lb.).

P
The dead weight of the shell varies per unit area of sloping
shell but in order to simplify the calculation it was assumed capable
of representation as a uniform load per unit area of horizontal shell
projection. In a GRP shell where the dead weight per unit area is only
about l/ 6 of the live load, this simplication is not likely to cause
any appreciable error in the test experiments.
In order to simulate a uniformly distributed load between model

and prototype the relationship in equation (3*4.7) is used:
E
m
V “ % E" ,
P
Since model and prototype are made of the same material this simplifies to
173
In order to simulate the stress condition in the prototype, the
dead weight of the model must first be subtracted from the dead
weight of the prototype.
The dead weight of the prototype is:

p 2 2
4.2 kg/m (5 ) = 21 kg/ m (4.3 lb/ft. ) and subtracting from
it the dead weight of the model this gives:
2 2 2
21 kg/m - 4.2 kg/m = 1 6 .8 kg/m which multiplied by the
area of the model then gives:
[l6.8 kg/m^ (0.48 m ^ ) ] 2 6 = 211 kg.
Having 1,248 loading points on the dome (not considering the 104 points
on the small cantilevering portion of the canopies), the resulting
load on each point is:
211 kg _ _
17548 " °-169 k9‘
Since each loading hook is connected to four points, the load per
hook on the model which will simulate the dead weight of the prototype
will have to be:
0.169 kg (4) = O .6 7 8 kg.
2. Snow Load
The requirements of British Standard Code of Practice 3* Ch.5

for roofs of limited access, allow for 2 ft. of snow, 15 lb/sq.ft.
(73.24 kg/sq .m.) on plan. That is to say, auniform load per unit
area of horizontal shell projection. A minimum additional factor of
safety of 1.5 was agreed upon with the local District Surveyor for
the town where the prototype is to be erected and the design loading
thus works out to be 22^ lb/sq.ft (110 kg/sq.m.). (This factor of
safety is in addition to the normal factor of safety used in
the design).
To simulate the stress condition as in the prototype, due to the
maximum required snow load, the model required the application of a
load intensity of 110 kg/sq.m. Since the area of the model was .
1 2 .5 7 sq.m., the total load on the model was 1 ,3^3 kg., distributed
174
on 1,248 points, gives 1 .1 0 8 kg. per point, or 4.432 kg. per
loading hook.
Two loading cases were investigated: (l) A full span load i.e.
a uniform load over all the 26 segments; and (2) A half span load,
i.e. a uniform load over 13 segments.
3. Concentrated Loads
A concentrated load of minimum 200 lb (90.8 kg.) on any3 in.sq.

(32.25 sq.cm) had to be allowed for according to B.S.C.P.3 Ch. 5-
Its simulation in the model with a scale factor of 5 using equation
(3-4.2) is:
P = 200 l b / 5 ^ = 8 lb . (3-63 kg.) on any 0 . 2 in s q . ( l . 2 9 sq/cm

m
4. Pressures and Suctions due to Wind
A wind load is obviously a dynamic load and a complex phenomenon.

To simplify the design procedure it is usually simulated by an
"equivalent, static load".
/
5- Thermal Loads, Settlement Loads and Dynamic Loads
Although the thermal expansion and contraction of GRP is high,

it will not build up high stresses in the scalloped dome. Since one
of the characteristic geometric features is its ability to "breathe".
In other words, each HP segment is not restrained from expanding and
contracting in its longitudinal as well as transverse direction.
Possible uneven settlement of the supports will not induce high

stresses in the scalloped dome since two diametrically opposite HP
segments reproduce a "three-pinned" arch which is statically
determinate and hence insensitive to differential settlement of its
abutments.
The dynamic loads can be exceptionally dangerous if ignored.

Their mathematical treatment is involved and beyond the scope of
this research. However, with a dome of moderate dimensions, predictabl
dynamic loads will not endanger the safety of the structure.
Deflection and Strain Measurements
Having applied the required forces, the next procedure was

to measure the deformation on the model.
1. Measurements of Deflections
The vertical and horizontal deflections were measured with a

mechanical device, using dial gauges. Dial gauges are by far the
most convenient device for measuring small displacements up to 2 in.
on a model made of relatively non-flexible material subjected to *
"static" loading.
Figure 3»3^ shows the position of the 11 dial gauges on the

segment. Gauges No. 1 and 7 were positioned to measure horizontal
displacements and the rest of the gauges vertical displacements.
They were mounted on an independent framework (Figure 3»36). The
2 in. range plungers rested on perpendicular 1 in. by 1 in. plates
of "Perspex" glued to the shell surface (Figure 3,hO) •
2. Measurements of Strains
Stress, the quantity of prime interest, since it determines the

safety of the structure, cannot be measured directly. However, it
may be determined indirectly through the relationship between stress
and strain; the latter being measured.
Strain is the change of length per unit cf length of an element

in tension or compression, it is thus a dimensionless quantity, and
quite a number of mechanical, optical and electrical instruments exist
which permit measurement.
Electrical resistance strain gauges are most suitable for measuring

strain in GRP laminates. Standard linear gauges (Teddington Aircraft
Controls Ltd, Type SE/a / 5) were used. Three such gauges were arranged
to form delta and rectangular rosettes (Figure 3-^l) • The paper
176
backing of the gauges was bonded to the GRP surface with MDurofixn ;
the bond proved satisfactory throughout all the tests. Figure 3*34
shows the position of the 64 gauges on the shell segment. The
symmetry in the structure made the bonding of gauges elsewhere un
necessary. Gauges were only mounted on the outer surface of the
shell, hence only membrane strains could be measured. The shell
segment with the necessary ndummy gauges" is shown in Figure 3-43-
The gauges were connected to an electronic data logger, (Figure 3-37)»
which punched the strain readings on paper tape. The principal strains,
their directions, as well as the principal stresses were computed by
an ICL 1905F computer. A typical computer-printed output is shown in
Figure 3-39*
177
Mechanical Properties of the Model Material
To interpret the strain readings from the test model and to

convert them into stresses, it is necessary to know: (l) the value
of Young's modulus; and (2) Poisson's ratio for the material of the
model.
1. Young's Modulus
For design purposes, a chopped strand laminate is assumed to

have the same Young's modulus (E) in tension and compression.
For long term loading, the values for E will change with time
and temperature (see also page 25 )• It is practically impossible
to determine the time-temperature dependent E value (apparent modulus)
for a GRP laminate analytically. Figure 3-42 shows how the values
for the apparent modulus were evaluated. A strain gauge extensometer
was clamped to a \ in. specimen. The specimen was subjected to a
continuous constant stress of 2 ,5 0 0 lb/in (23 % of the short term
ultimate stress) for a period of 1 0 ,0 0 0 mins ( 6 days, 22 hours and
30 mins) and at a room temperature of 6 6 ° F . (19° C.). The various
values for the apparent modulus are shown in Figure 3-38. It was
surprising to find that the short term E value was 1.15 x 10^ ,
whereas the long term E value (after 10,000 mins at 6 6 ° F.) was only
0.88 x 106 ; a change of 23% • The time dependent E value was of
course considered in computing the principal stresses in the shell.
2. Poisson's Ratio
(1) The value for Poisson's ratio is determined by the ratio:

lateral strain / longitudinal strain.
(2) Generally, the Poisson's ratio is not so sensitive to time

effects.
(3) In GRP Poisson's ratio (v) in tension is not the same as in

compression. For a chopped strand laminate made in contact moulding,
178
0.6x10®
tO3 ID ay 6 Days
3.38
TApr NUt' B F P 2 5
N>J 2 E R0 TIME L OG
LOAD CASE NO. 7

NO TIME LOG
TWO-GAUGE ROSETTES
CHANNEL number PRINCIPAL STRAINS ANGLE PRINCIPAL STRESSES

0 1 17.0 -8.0 0.0 14.7 -0.9
2 J -44.0 79,0 0.0 -11.7 65.4
4 310.0 -17,0 0,0 327.3 122.3
A 7 7ft4. 0 -501.0 0.0 598,2 -194.6
A 460.0 -822.0 0.0 124.0 -679.5
in 11 112.0 -968,0 0.0 -318.3 -995,2
1 ?. 1i 385.0 -1978.0 0.0 -481.7
685.6
-1962.7
494,4
t4 13 537.0 232.0 0.0
43 - 90 ROSETTES
CHANNEL HUMBER PRINCIPAL STRAINS ANGLE PRINCIPAL STRESSES

;a 17 18 23.1 -6ft. 1 136,6 -5.1 -60,9
19 20 21 485.8 -785.8 83,6 168.3 -628.7
??. 23 24 406. 0 -719.0 84.1 112.4
-287.0
-592.7
-1302.7
,.’5 2ft 27 292.3 -1328.3 82,0
(Ift 29 30 6 46.8 -1069.8 82,2 213.4 -862.5
31 32 33 179.0 -501.0 142,0 -33.9 -460.2
EQUIANGULAR ROSE TTES
C HA N N E L NUMBER PRINCIPAL STRAINS ANGLE PRINCIPAL STRESSES

;;a 33 36 108.8 -162.2 158,7 44. 0 -125.8
37 3ft 39 298.9 -38.9 71 .3 305.3 93.6
AO 41 42 186.3 3.0 90.2 202.7 87.8
AS 4<- 45 179.7 -52.4 100.8 116.4 2.3
Aft 47 48 14.6 -21.3 127,5 6.1 -16.4
SO 51 32 226.9 -202.2 100,5 153.4 -115.5
S3 5a 55 307.5 -167.5 160,4 256.2 -41 ,4
Sft 57 3E 662.2 -476.2 5.9 499.5 -214.0
so 60 61 434.4 -165.7 144.4 394.2 18.1
02 63 64 -157.1 -367.5 136.5 -336.6 -468,5
3-39
179
3 .4o 3.41
3.42 3.43
l8o
the tensile Poisson's ratio = 0.32; and the compressive Poisson's
ratio = 0.42 (Figure 1.2 ) . The value of Poisson's ratio for the
specific model material, was not determined experimentally because
it is difficult to compress a thin laminate without introducing
bending.
Since the dome was mainly in compression the value v =0.42

was used to compute the principal stresses in the shell.
3. Ultimate Tensile Strength
The average, short term ultimate stress, for the model material
2
was 11,000 lb/in (773*3 kg/sq.cm) .
l 8 l
Evaluation of Test Results
A considerable amount of numerical data about deflections

and stresses has been determined and it is not possible to present
the whole of this information in a few significant graphs, but
particulars of Deflection and Stress Distribution are shown in
Figures 3*44 to 3*54. These curves show the vertical deflections
and stress distributions of the three loading cases which were
investigated.
1. Dead Load
The behaviour of a GRP shell under its dead weight, had to be

investigated in detail since it represents a case of long term
continuous loading.
Figure 3*46 shows the principal stresses at various points on

the shell, due to the simulated dead weight of the prototype. The
stress distribution throughout is fairly even, and it is predominantly
2
in compression. A maximum compressive stress of 87-3 lb/in (6.1 kg/sq cm)
2
and 165.3 lb/in (11.6 kg/sq cm) was measured at ERS gauge No.25 and
2
No.13; and a maximum tensile stress of 20.7. lb/in (1.4 kg/sq cm) and
2
49.3 lb/in (3-4 kg/sq cm) was measured at ERS gauge No.6 and No.l4.
All these values are very low indeed.
Point 4 (quarter span) on the shell was examined more carefully.

Curve 1 in Figure 3*51 shows vertical deflection curve of point 4 on
the HP segment, due to the simulated dead weight of the prototype,
over a period of 10,000 mins and at a room temperature of 68° F. It
is reassuring to see that the curve levels off after about 4 days.
2. Uniform Load Over Full Span
Figure 3*44 shows the vertical deflection curve of a meridional

rib across the dome under a uniform distributed load (U.D.L.) of
24 lb/sq.ft. at 67° F. Note, that the point 4 at £ span deflected
downward, whereas point 6 at 2 span deflected upward. The behaviour
182
is explained by the fact that although the load on the entire dome
was uniform, the load on a meridional rib was not uniform.
The un-uniform load on the rib continuously decreased towards

the crown and had its centre of gravity at a 5 the distance from
the outer support. This, as it were ’’unbalanced” load, produced
an uplift of the rib at the crown of the dome. (Only a concentrated
load at the top of the dome or suction due to wind could produce
a downward deflection of the crown).
It was most interesting to see that the low rise scalloped dome
behaved in a linear manner, in that doubling the load also doubled
the vertical deflections, and the stresses in the shell (Figures 3»51;
3 .5 2 and 3-53).
At a U.D.L. of 24 lb/sq.ft. at 67° F the maximum vertical deflection

of the rib at 4 span after 10 minutes was 0 .5 1 in. (13 mm).
Sustaining the load for 10,000 minutes produced a deflection of
O .6 9 in. (17.6 mm) it increased by 35% (Figure 3«5l)« Since deflection
in prototype = deflection in model x scale factor, the vertical
deflection in the prototype at £ span under the same U.D.L., will be
3.45 in. (8 . 7 6 cm).
Figure 3*47 shows the stress distribution, after 10 minutes and

10,000 minutes of continuous loading. Note that, for example, at
ERS gauge No. 25 the stress after 10,000 min at 67 °F only increased
by 4.8% (due to the change in the contour line of the rib) whereas the
deflection at the same point increased by 35%? this being a characteristic
behaviour of a material in which excessive creep is taking place.
It can be seen that the design criteria for a GRP structure is not
so much the danger of exceeding the maximum allowable working stress
of the material, but rather the excessive and continuously increasing
deflections that will take place over the life span of the structure,
as the value for E is reduced with time. Straight free edges in an
GRP structure should, therefore, be avoided.
3. Uniform Load Over Half Span
Figure 3*45 shows the vertical deflection curve of a meridional
rib across the dome, under a uniform load over half the dome of
24 lb/sq.ft. at 6 3 °F. A maximum vertical deflection of 0.97 in.
(2.47 mm) was measured at % span in the model .. : .
which equals 4.83 in. (12.35 cm) in prototype. Figure 3-48 shows
the stress distribution, after 10 min. and 10,000 min of continuous
loading.
A comparison between the fully and half loaded dome shows that
the vertical deflection of the rib at £ span after 10,000 min.
increased by 22%, in the latter case, whereas the stresses at the
same point only increased by 2.7%*
4. Concentrated Loads
A point load of 10 lb (4.3 kg) on the model (simulating

230 lb = 1 1 3 .4 kg on prototype) at l/ 3 span, produced a maximum
2
compressive stress of 399 lb/in . The same point load at \ span,
2
produced a maximum stress of 85 lb/in .
3* Stress Trajectories - Isostatic Curves
Having found the principal directions of the stresses, at the

various points on the shell, two families of curves parallel to the
principal direction at the various points can be drawn,giving an orthogonal
network (Figure 3«54). These curves are referred to as isostatics or
stress trajectories. A state of pure tension or pure compression
will exist at any point normal to the curves, hence no shear is
developed along the isostatics. They represent the stress-line pattern
which indicates the flow of stress within the structural element. They
are very useful in visualising the behaviour of a structure under load,
and also help to spot dangerous stress situations.
The flow of stresses in the HP triangular shell segment supported

on three points under U.D.L., is similar to the flow of stress in a
simply supported beam under U.D.L. Bending of the shell segment produces
a stress concentration at about l/ 6 the span, in which the rib at its
top fibre is in compression, and the valley of the segment is in tension.
184
A-A
B -B
v e r tic a l d e fle c tio n s - U.D L.2A lb/sq.ft.al67*F a f t e r 6 d a y s of c o n tin u o u s lo a d in g
3.
4-4
A-A
B -B
v e r t ic a l d e fle c tio n s - U.D.L.over h a lf s p a n , 2 4 lb /s q .ft.a t 6 3 *F a f t e r 6 d a y s o f c o n tin u o u s lo a d in g
3.45
185
Figure 3*^6 Stress distribution in the HP shell segment.
Load condition: 3*^7 lb/sq.ft. (17 kg/sq.m) * dead weight of prototype.
Duration of load: Top - 10 mins. Bottom - 10,000 mins. (approximately 7 days).
Temperature: 68 °F (20°C).
Information given: direction of principal stresses, stress in lb/sq.in.
(llb/sq.in. = 0 .0 7 0 3 kg/sq.cm.) (+) tensile (-) compressive.
186
.A
Cl
igure 3*47 Stress distribution in the HP shell segment.

Load condition: U.D.L. 24 lb/sq.ft. (117-1 kg/sq.m) over full span.
Temperature: 67 °F (l9.5°C).
(llb/sq.in. = 0 .0 7 0 3 kg/sq.cm.) (+) tensile (-) compressive.
187
•2017.7
Figure 3-48 Stress distribution in the HP shell segment.

Load condition: 24 lb/sq.ft. (117.1 kg/sq.m) over half span.
Temperature: 65 °F (l8°C).
Information given: direction of principal stresses, stress in lb/sq.in
(llb/sq.in. = 0.0703 kg/sq.cm.) (+) tensile (-) compressive.
188
Figure 3-49 Stress distribution in the HP shell segment.
Load condition: 29 lb/sq.ft. (l4l.5 kg/sq.m) over half span.
Temperature: 68 °F (20°C).
vllb/sq.in. = 0 .0 7 0 3 kg/sq.cm) (+) tensile (-) compressive.
189
Figure 3«50 Stress distribution in the HP shell segment.
Load condition: 29 lb/sq.ft. (l4l.5 kg/sq.m) over half span.
Temperature: 68°F (20°C).
(llb/sq.in. = 0 .0 7 0 3 kg/sq.cm) (+) tensile (-) compressive.
190
6. Plastic Flow
Figures 3*^6 to 3-50 show the considerable stress redistribution

that took place in the GRP scalloped dome as it was subjected to
continuous loading. This redistribution of stresses is due to the
"plastic flow" of the material. The less brittle a material is the
more it will be susceptible to plastic flow.
Plastic flow eliminates stress concentrations. It increases the

safety of the overall structure and contributes to its reserve of
strength. It can be seen that the lack of stiffness in GRP can also
be an advantage.
7. Failure Condition
The lack of sufficient weights for loading prevented the testing

of the entire model to failure. The various loading tests were all
carried out within the short term elastic range of the model, that is
to say, once the load was removed the model returned to its original
position.
The unit on which the guages were mounted and two adjacent segments
were loaded with a uniform load up to 48 lb/sq.ft. (234.24 kg/sq.m) when
signs of instability occurred. A maximum deflection of 1.57 in. (39*87 mm)
2
and a maximum stress of 3 1 0 ^ lb/in (2 1 8 kg/sq.cm) at t; span
was measured.
The "weak" area of the scalloped dome was found to be along the
rib at about f- the span from the support. Here the moment of inertia of
the cross-section was not sufficient, due to its shallow undulation,
to resist the bending moment and the axial thrust caused by the live
load. The downward vertical deflection at this point increased
continuously due to creep. It is considered that a live load of
48 lb/sq.ft. would eventually induce the rib to collapse inwards.
8. Comparison between Theoretical and Experimental Results
At a uniform load of 24 lb/sq.ft. (117*1 kg/sq.m) over full span,

the approximate calculations using the three-pinn arch analogy, gave for
example: at the top fibre of cross-section A-A (l/6 span), a compressive
stress of 1 1 2 7 *6 lb/in (79*2 kg/sq.cm), and a tensile stress of 191
o
834.6 lb/in (5 8 . 6 kg/sq.cm) at the bottom of fibre.
The stresses obtained experimentally, using electric resistance

2
strain gauges No. 25 and 5 6 , gave a stress of 1229*6 lb/in
2
(86.4 kg/sq.cm) compression, and 449.9 lb/in (3 1 .6 kg/sq.cm) tension.
At other cross sections, as for example near the crown, a close

correlation between three-pinned arch analogy and the experiment could
not be obtained.
Final Remarks
The various tests demonstrated the great stiffness and the high
resistance against buckling of the scalloped dome made of HP segments.
It was possible to establish a relationship between the free span
of the dome and the necessary minimum shell thickness. An average shell
thickness (t) to the span ration (l) of 1 :1 5 0 0 was thought to be
possible. (t/l of an average chicken egg is 1:100). In a dome with
2
a diameter of 20 m. the dead weight of the shell will be only 4.33 lb/ft
(21 kg/sq.m.). Despite the very thin shell the scalloped dome
is not susceptible to the phenomenon of Msnap-throughM buckling, and
due to its plastic flow it has a high reserve of strength. The
structure also demonstrated its efficiency from the material side,
by the fact that a stress level of 3 0 0 0 lb/in (210.9 kg/sq.cm)
compression in the skin was reached, without the occurrence of
buckling. If we assume that the ultimate safe working stress in a
chopped strand polyester laminate to be 2 ,5 0 0 lb/in (1 7 5 *7 5 kg/sq.cm)
then we have here a structure in which the material is put' 100%
to work.
192
d e fle c tio n c u rv e s for p o in t4 (q u a r te r s p a n ) u n d e r UD.L.
24.0
91.7
ve 4
66.8 18*C
13.7
I 0300
.5 (7.62]
m
a>
■C
u ,21.5‘ C
C 0200
8.6 71*F
i
in
c
o
17 kg/sq. ti 20’C
3.471b. ’sq.ft. ad we ght
1 Day 6 Days
duration of load - minutes

3.51
strain curves for electrical resistance strain gaugeNo.25( 1/6 span)underU.D.L.
(0.27.) 2,000
67*F
(0.17.) 1,000
63’ F
41.! kc /s 21.5'C
ght
ID a y 6 Days
duration of load minutes
3-52
193
s tre s s curves for ele c trica l resistance strain gauge No.25( 1/6 span)underUD.L.
1.500
(105.4:
1,400
117.1 kg 's l a 19.5* C
c u rve 3, VrO ttr scFt tr 3 67*P
1,100
(77.3;
91.7 Kg Is q- rr a 17*C
—
1.000 cc ve 4— « r6 Hr 'st |H t: a 63*F -
66.8 kg Is q n a 18*C
cL ve 3. 13./ 1b. 's it t
. a 65*F • _ _
..
Ib/sq.in. (kg/sq.cm.)
4- Ii.O v
cu ve 2. 8.6 lb' s<If t. a 71*F
-
17 k y/i 4-
c u ve 1. 3.47 lb. S( !•* t. = d ad we Ight a t 18 ■F
stress
1 2 3 4 5 6 7 8 910 io2 102 1Day 105
duration of load - minutes
3.5 3
3.54
\9k
S E C T I 0 N T H R E E COST
The dead weight per sq.ft. of the 20 m. (6 5 .6 ft.) prototype

scalloped dome is:
Weight of dome 14,300 lb . ^ ,

— ---------------t V — ---- -- -^p----- 7 , = ^ -2 3 lb/sq.ft. (2 0 .6 kg/sq
Plan area covered by dome 3i37o sq.ft.
If the glass to resin ratio is 1:2.5 than the laminate contains:
!t‘3 X . ill = 1 . 2 lb/sq.ft. glass,

3-5
and
4.23 lb/sq.ft (2.5) 0 ,, /

^ ---- -— =^' = 3 .0 3 lb/sq.ft. resin.
3-5
Based on a calculation that polyester resin (flame retardant)

mix = 2/8d per lb., and chopped strand mat nE M glass 2 oz./sq.ft.
= 4/- per lb., than the basic material cost per sq.ft. covered
area is: 4/l0d + 8/ld = 12/lld.
The cost of the finished manufactured laminate could not be

included, since it was found to vary up to 100 % with different
fabricators.
It should also be remembered that the material cost per sq. ft.
of laminate is directly related to the free span of the structure.
The ratio, cost per sq. ft. of covered area to the free span of the
dome, varies linearly for moderate spans and thus increasing or
decreasing the magnitude of the dome will also affect the cost of the
material per sq. ft. in the same ratio. Consequently with a relatively
expensive material as GRP, the question of dimensions gains particular
economic significance.
Economy is, of course, fundamentally important in present-day

KeW -todays . 0
structures. However, the cost of a structure^/is usually about 6 % T *7 )0 /0
of the entire cost of the building. Hence, even substantial reductions
in the cost of the structure only amount to small savings of the total
building costs.
It is believed therefore, that the argument that GRP is too

expensive for building is not such a real objection. 'j
Generally plastics have in their favour, over conventional
building materials a number of characteristic points.
The cost of a material can normally be identified in

three stages:
1. Production of the raw material;
2. Conversion, i.e. purification of the raw material; and
3- Manufacture.
In plastics it is possible to combine the second and third stage

in one, since plastics are usually delivered in a monomer state to
be polymerized in the same operation that yields the end product.
The general price trend for most conventional structural materials,

like steel or timber, for example, has been upwards, whereas the
price of the common plastics materials is decreasing. Plastics are
therefore becoming correspondingly more competitive. One of the
contributory factors is that the output of plastics has been
increasing at 10 - 15 % per annum resulting in more rational manufacturing
processes, and this in turn is helping to lower the price.
The predominant position held by GRP in the manufacture of small

boat hulls suggests that the ever increasing demand for light weight
or mobile structures, this material has an opportunity of securing
a similar monopoly position.
It might be said that in order to have more structures in GRP

the main hurdle to be overcome is not so much a physical one but
may well be an idealogical one the inertia of the potential
client who is slow to accept change. It is hoped that this research
will have contributed a little towards a change by illustrating another
type of construction and another method of manufacture.
196
REFERENCES
Design and Construction
1. Howard H.S. Jr., "structure: an Architect’s Approach”,

McGraw-Hill Book Company, 1 9 6 6 .
2. Salvadori, Mario and Helle, Robert, "Structure in Architecture

Prentice-Hall, Inc., 1963#
3. Salvadori, Mario and Levy, Matthys, "Structural Design in

Architecture",
Prentice-Hall, Inc. 1 9 6 7 *
4. Hossdorf, Heinz, "Eine Programmgesteuerte, Vollautomatishche

Modellmess - und Datenauswertungsanlage",
Schweizerische Bauzeitung, Sept. 1965*
5. Preece B.W. and Davies J.D., "Models for Structural Concrete",

CR Books Limited, London, 1964.
6. Cowan H.J., and Muncey, R.W., "Models in Architecture",

Elsevier Publishing Company, 1 9 6 8 .
7. McCurrich L.H., "Reinforced Plastic Applied to Building

Structures in particular to a Toroidal Shell Roof for
Prefabrication in Epoxy or Phenolic Laminates",
M.Sc. Thesis Dept, of Civil Engineering,
University of Southampton, 1965-
8. Hendry A .¥., "Elements of Experimental Stress Analysis",

Pergamon Press, 1964.
9. Davies R.M. (Editor), "Space Structures",
Blackwell Scientific Publications, 1 9 6 7 *

UNIVERSITY OF NIŠ
The scientific journal FACTA UNIVERSITATIS
Series: Architecture and Civil Engineering Vol.1, No 5, 1998 pp. 627 - 636
Editors of series: Dragan Veličković, Dušan Ilić, e-mail: facta@ni.ac.yu
Address: Univerzitetski trg 2, 18000 Niš, YU
Tel: +381 18 547-095, Fax: +381 18 547-950
ANALYSIS OF HYPERBOLIC PARABOLOIDS

AT SMALL DEFORMATIONS
UDC 692.1/.6:539.37:514.13:519.633(045)
Ljubica Velimirović1, Grozdana Radivojević2, Dragan Kostić2

1
Faculty of Science, Niš, 2Faculty of Civil Engineering and Architecture, Niš
Abstract. A hyperbolic paraboloid, treated from the constructional and mathematical

aspects, is analyzed in this paper. In the constructional sense, it is a thin shell of a
great bearing capacity and wide usability in spatial structures, either as a complete
form or in parts. In the mathematical sense, it is treated as a geometrical surface on
which it is possible to determine the rotation field and the field of infinitesimal
deformations, and it is rigid.
1. INTRODUCTION
The contemporary development of building mechanics is oriented towards
mathematics, thus offering a wide range of possibilities in the research of usually highly
complicated spatial-surface systems. The fact that the fundamental character of mechanics
cannot be ignored makes such an approach to the abstract analysis necessary even today.
In this, the properties of materials are not neglected at all, and they should be accorded
with the building mechanics that is mathematically oriented. The mathematical analysis of
these calculations is often complex and not adequate regarding the total time available for
the design of structures it is needed for. The approximation, that is the use of computers,
reduces long-lasting mathematical calculations and enables constructors to devote more
time to the design and construction. Analytical methods are supplemented by the tests on
differently sized models, which sometimes simplify significant assumptions of the
mathematical methods. The determination of forces and stresses acquired in the models
observing and measuring should not be underestimated, yet not overestimated, as well.
This paper represents a review of hyperbolic paraboloid (hereinafter referred to as
HP) shells as very frequently used roof structures and points out the possibility of the
mathematical analysis application in the case of small deformations.
Received March 23, 2000

628 LJ. VELIMIROVIĆ, G. RADIVOJEVIĆ, D. KOSTIĆ
2. GEOMETRIC CHARACTERISTICS OF HP
The ruled surface of HP is achieved by the movement of the line AB (ruling line)
along the two straight, mutually non-parallel lines AD and BC that do not intersect in the
space (Fig. 1).
Fig. 1. The HP geometry
A mathematically defined HP is a set of points in the space, whose coordinates,

compared to a rectangular coordinate system Oxyz, satisfy the equation
x2 y2
− = 2z .
a2 b2
HP originates in the slide of the movable parabola y2 = −2b2z, x = 0, over the
immobile parabola x2 = 2a2z, y = 0, so that the parabola axis and plane remain parallel
with the starting position.
If we examine the HP level sections, we shall obtain:
- the section of HP and x = h plane is a parabola parallel to y,z-plane
 h2 
y 2 = −2b 2  z − 2 ,
 x=h,
 2a 
- the section with y, z-plane is obtained for x = 0:
y2 = -2b2z ,
- the section of HP and y = k plane, the plane parallel with y, z-plane, is the parabola
 k2 
x 2 = 2a 2  z + 2 ,

 2b 
- especially, the section of HP and x, z-plane (y = 0) is the parabola
x 2 = 2a 2 z ,
Analysis of Hyperbolic Paraboloids at Small Deformations 629
- the section with the horizontal plane z = h, h > 0, is the hyperbola

x2 y2
− =1,
2a 2 h 2b 2 h
and with the plane z = h, h < 0, it is
x2 y2
− +
=1,
2a 2 h 2ab 2 h
conjugated with the previous hyperbola,
- the section of HP with the plane x, y (z = 0) are two lines
b
y=± x.
a
The equation of HP may be written in the following form
x y
+ = 2t
a b
x y z
− =
a b t
which shows that HP is a rectilinear surface. The lines in one family do not intersect,
while the lines of different families always intersect mutually.
HP may be given in a parametric form as:
x = aρ cosh v, y = bρ sinh v, z = u 2 .
HP can also be given by the equations:
x = aρ cos ϕ, y = bρ sin ϕ, z = ρ 2 cos 2ϕ

or in the form:
x = a(u + v), y = b(u − v), z = uv .
3. STRUCTURAL DEVELOPMENT OF HP
As spatial roof structures, HPs appeared in 1932 in France. At that time, they were
considered extremely daring for the contemporary theoretical approach, research level
and building techniques. Although HP is a spatial surface system with double curvature,
its realization is relatively simple, as all the roof-boarding elements may be placed in the
direction of the straight ruling lines.
Felix Candela, worshiper and author of attractive thin shell structures, particularly of
this kind (Fig. 2), greatly contributed to the affirmation of HP.
HP offers unlimited possibilities to architects and constructors in designing and
constructing. This warped surface may be applied over any foundation shape (rectangular,
triangular, circular, ellipsoid, and so on), and any building which is representative either
by its contents or by its position. A harmonious and daring structure of a unique form
enables creative expression (Fig. 3).
A HP surface may be continuous and homogenous over the whole foundation span, or it
may be assembled of multiplied parts that are copied by plane or axial symmetry. The
prefabricated construction with these shells is easily applicable, because the surface, even if
it is continuous over the whole foundation span, can be separated into integral prefabricated
elements arranged as strips or spatial rectangles that are compounded into a whole girder.
Fig. 2. Hyperbolic paraboloid shapes
Fig. 3. A combination of several united HPs the roof of a restaurant in Xolchimilco

(Mexico)
4. BEARING CAPACITY OF HP
Double-curved surfaces usually have satisfactory bearing capacities, while in HP it is
even greater, as the convex curvature stiffens in a way the concave curvature (Fig. 4).
Compressive stresses appear along the line a, and tensile strains are formed along the line
b. Having such stress conditions, HP is, righteously, regarded as a membrane, and the
calculation of the system bearing capacity is performed according to the membrane theory
with small deformations. In HPs, normal
forces along the ruling lines have constant
values, implying that there is no need for
the effect of the shell lateral forces on its
edges (edge beams), where constant,
longitudinally distributed shearing forces Compressive
are formed. stress
The shell principal stresses are created
alongside the vertical sections, which make
the angle of 450 with the ruling lines.
Placing edge elements or ribs that accept
the compressive or tensile stresses for the
Tensile
designed surface geometry most often strain
solves the reception of shearing forces on
the edges. Edge elements may be avoided
in highly curved shells, for spans that are
shorter than 30 m. In principle, care should Fig. 4. Membrane stresses in HP
be taken of the edge elements load, that is, their dead weight and its impacts. This is
particularly important in cases where they are asymmetrically loaded.
Many possibilities for the calculation of various forms of HP shells are given in the
literature. It is necessary to check the bending moments in the HP form that has a negative
curve in each point, or in the shells in which the relation of the span and structural height
cannot justify the rigid shell concept. Due to the multitude of forms, there are no rules that
could be introduced as generally applicable. Numerous theoretical papers that are
included in the references [1-8] offer clearly defined and tested recommendations for the
calculation of these shell forms. One of such recommendations is that the influence of the
bending stress may be neglected in the shells in which the ratio of the height and the
length of sides H/a is greater than 0,2 (9). In most cases, HP shells have great safety
against bulging due to their curvatures. In shallow shells, the size of elastic and plastic
deflection of console parts should be checked.
-------------- Tie
The forces in the

The supports are
edges are directed
posi-tioned in the
toward the supported
middle of the sides.
corners. The edges
The ties are not
are compressed.
needed.
The tie is necessary.
The forces in the

edges are directed
The supports are
toward the non-
posi-tioned in the
supported corner.
middle of the sides.
The edges are tensio-
The ties are
ned. An element for
necessary.
the reception of
pressure is needed.
The forces in the

edges are directed
toward the supported
corners. The ties are The tie is necessary.
necessary. They are
placed in the plane of
external walls.
The tie is not

Diagonal position.
necessary.
Fig. 5.
The areas around the HP shell edges - edge elements - are the parts with greatest axial
forces that are not in balance with the shearing forces. In order to prevent deformation, it
is necessary to place diagonal elements (ties or compression members, Fig. 5). The edge
elements are in many cases dimensioned not only for the axial forces but also for the
bending moments. Satisfactory results are achieved for the prestressing of edge elements.
5. A MATHEMATICAL DEFINITION OF INFINITESIMAL DEFORMATIONS OF SHELLS AND HP

The mathematical approach to the problem of infinitesimal deformations (ID further
on) can be presented as a part of the global differential geometry.
Many renowned mathematicians (Cauchy, Liebman, Hielbert, Weil, Blaschke) have
dealt with the problem of ID. One of the first works in this field belongs to Cauchy
(1813). He proved in it that closed convex polyhedrons are rigid.
In the case that
S : r = r(u, v) ,
is the vector equation of a regular surface S, and the surface S is included in the family of
surfaces St (S = S0), expressed by the equation
S t : r(u, v, t) = r(u, v) + tz(u, v)
where t ∈ R, t → 0, z -continuous differentiable vector function of the C (m ≥ 3) class,

m
defined in the points belonging to the surface S, which is the field of infinitesimal
deformations. The surfaces St, t ∈ R, t → 0 are the infinitesimal deformations of the
surface S if the difference in the linear element squares of these surfaces is an
infinitesimal value of a higher order compared to t, t → 0, i.e.
dst2 − ds2 = o(t).
This means that the curve arc length variation on the surface is o, δs = 0, in ID, that is,
the arc length of the curve on the surface is stationary in ID. The angles between the
curves on the surface are also not changed, as well as other elements that depend on the
coefficients of the first fundamental form.
The surface is rigid if it allows only for trivial ID fields. The deformation field is
trivial if it has the form of
z = a × r + b, a, b - constant vectors.
A necessary and sufficient precondition for the surfaces St to represent ID of S is that
the following is valid:
dr ⋅ d z = 0 (1)
where S : r = r(u, v) , z = z(u, v) is the ID field, and − denotes a scalar product, and x
denotes vector product.
This equation is equivalent to the three partial equations:
ru ⋅ z u = 0, ru ⋅ z v + rv ⋅ z u = 0, rv ⋅ z v = 0 .
There is a unique field y(u, v) for the ID field z(u, v) of the surface, so that:
z u = y × ru , z v = y × rv ,
i.e.
d z = y × dr . (2)
The rotation field for which the previous relation is valid is the vector field y(u, v) . As
the result of the ID surface, all its elements are subject to the rotation with the rotation
vector y(u, v) .
The field s (u, v) , determined by the equation
s = z − y×r
is the field of surface translations at ID with a defined field z(u, v) .

The derivatives of the vectors y u , y v of the rotation field y(u, v) are given by the
equations
y u = αru + βrv
y v = γru − αrv
where the functions α(u,v), β(u,v), γ(u,v) satisfy the system of partial differential
equations:
α v − γ u = Γ111 γ − 2Γ112 α − Γ122β
α u − β v = Γ11
2
γ − 2Γ12
2
⋅ α − Γ 222β (3)
b11 γ − 2 ⋅ b12 α − b 22 β = 0
where Γ ijk are Cristoffel's symbols of the surface r = r(u, v) , and bij are the coefficients of
the second fundamental form.
The solution of this system of partial equations determines the functions α, β, γ. The
fields y and z are determined in the following way:
Being that
dy = y u du + y v dv = (αru + βrv )du + ( γru − αrv )dv ,
is the total differential of the vector function y , by integrating we get the field y(u, v)
that is determined in a unilaterally connected surface S. With such determined field y ,
the ID z field should be further defined. Namely, we have that:
dz = y × dr = ( y × ru )du + ( y × rv )dv . (4)
As
( y × ru ) v = ( y × rv ) u ,
the right side of the equation (4) is the total differential, so the field z(u, v) is determined
by integration.
We shall examine ID of the surface z = xy, HP. The vector equation of this surface is:
r = r(x, y) = ( x, y, xy ), (u = x, v = y) (5)
or:
r = x e1 + ye 2 + xye3 ,
where e1 , e 2 , e3 are mutually perpendicular unit vectors.

As Cristoffel's symbols for this surface are
y x
Γ11
1
= Γ11
2
= 0, Γ112 = , 2
Γ12 = , Γ122 = Γ 222 = 0 ,
1+ x + y
2 2
1+ x + y2 2
the second fundamental form coefficients are:

1
b11 = b 22 = 0, b12 = , (a = 1 + x 2 + y 2 ) .
a
The equations (3) become:
2y 2x 2
αy − γx = − α, αx + βy = − α, − α = 0,
a a a
in which
α = 0, β y = 0, γ x = 0 , i.e. α = 0, β = ϕ( x ), γ = Ψ ( y) ,
so, on the basis of this
dy = y x dx + y y dy = ϕ(x) ry dx + Ψ(y) rx dy = (Ψ (y)dy, ϕ(x)dx, xϕ(x)dx + y(y)dy)
In the case y = (Y1 , Y2 , Y3 ) , the following will be achieved:
dY1 = Ψ(y)dy, dY2 = ϕ(x)dx, dY3 = xϕ(x)dx + yΨ (y)dy .
By integrating, we get
Y1 = ∫ Ψ(y)dy = λ(y), Y2 = ∫ ϕ(x )dx = μ(x ), Y3 = ∫ xϕ(x )dx + ∫ yΨ (y)dy .
The partial integration results in

Y3 = x ∫ ϕ( x )dx − ∫ ( ∫ ϕ( x )dx)dx + y ∫ Ψ(y)dy − ∫ ( ∫ Ψ(y)dy)dy,
Y3 = xµ(x) − ∫ μ(x)dx + yλ( y ) − ∫ λ(y)dy .
The rotation field of HP is
y = (λ(y), µ(x), xµ(x) + yλ( y) − ∫ µ( x )dx − ∫ λ( y)dy) ,

in which µ(x), λ(y) are arbitrary functions.
Applying (2), we will determine the ID field. As (5) for the HP is
dr = (dx, dy, ydx + xdy ) ,
based on (2), it appears that
e1 e2 e3
dz = y × dr = λ(y) μ(x) xµ( x) + yλ(y) − ∫ μ(x)dx − ∫ λ(y)dy .
dx dy ydx + xdy
The HP bending field is obtained by the following integration:

z = {y ∫ μ(x)dx + ∫[λydy − yλ(y)]dy , ∫[µ(x)x − ∫ µ(x)dx] − x ∫ λ(y)dy , − ∫ µ(x)dx − ∫ λ(y)dy}
where µ(x), λ(y) are arbitrary functions. It can be proved that the bending field is trivial,
i.e. that the HP is a rigid surface with regard to infinitesimal deformations.
6. CONCLUSION
The HP is a ruled surface by which a space based on any desired form can be covered.
It is suitable for the systems of roof structures either as a whole roof or as its part. The HP
is a thin shell of a great bearing capacity. In spite of its double curvature, its execution is
simple due to the possibility of laying the roof boarding in the direction of straight ruling
lines. Such structures can also be built by the monolithic assembly of precast units.
The mathematical analysis determined the field of infinitesimal deformations on a HP
surface and showed that this type of shell may reasonably be treated as a membrane, i.e.
that this surface is rigid.
REFERENCES
1. Tedesko A.: Shell at Denver-Hyperbolic Paraboloidal Structure of Wide Span, Journal of the American
Concrete Institute (1960)
2. Candela F.: Stress Analysis for any Hyperbolic Paraboloids Architectural Record (1958)
3. Candela F.: General Formulas for Membrane Stresses in Hyperbolic-Paraboloidical Shells, Journal of
the American Concrete Institute (1960)
4. Herrenstin: Hyperbolic Paraboloid Umbrella Shells under Vertical Loads, Journal of the American
Concrete Institute
5. Bongard: Zur Theorie und Berechung von Schalentragwerken in Form gleichseitger Hyperbolischer
Paraboloide, Bautechnik-Archiv H.
6. Bennet: Hyperbolic Paraboloides, Architect and Building News, London (1961)
7. Ramaswamy Rao: The Membrane Theory Applied to Hyperbolic Paraboloid Shells, Indian Concrete
Journal (1961)
8. Soare M.: Contributi la teoria incovoierii paraboloidului hiperbolic i sa Studii si cercetari de mecanica
aplicata (1962)
9. Parne A.: Elementary Analysis of Hyperbolic Paraboloid Shells, IASS-Bulletin Nr. 4
10. Rile H.: Prostorne Krovne Konstrukcije, Građevinska Knjiga Beograd (1977)
11. Sekulović M.: Teorija Konstrukcija, Savremeni Problemi Nelinearne Analize, Građevinska knjiga
Beograd (1992)
12. Velimirović Lj.: Beskonačno male deformacije toroidnih rotacionih površi, Magistarska teza,
Univerzitet u Nišu (1991)
13. I.Ivanova-Karatopraklieva, I.Kh. Sabitov, Surface deformation. I, J.Math. Sci., New York 70, 2(1994)
1685-1716
14. I. Ivanova-Karatopraklieva, I.Kh. Sabitov, Bending of surfaces II, J.Math. Sci., New York 74, 3(1995)
997-1043
ANALIZA HIPERBOLIČNIH PARABOLOIDA

PRI MALIM DEFORMACIJAMA
Ljubica Velimirović, Grozdana Radivojević, Dragan Kostić
U radu je analiziran hiperbolički paraboloid tretiran sa konstruktivnog i matematičkog
aspekta. U konstruktivnom smislu to je tanka ljuska, velike nosivosti, raznovrsne primene u
prostornim konstrukcijama bilo u celini svoje forme ili delovima. U matematičkom smislu tretira se
kao geometrijska površ na kojoj je moguće odrediti polje rotacije i polje beskonačno malih
deformacija.
THIDER HYPERBOLIC PARABOLOID SHELI,S SUPPORTED
------4-•·'--·---------------·
By
KAYODE AYI~illE SEGUN, B1Sc. (Eng), n.r.c.
A thesis submitted for the Degree of Doctor of

Philcaophy in the Faculty of Engineerine; of the
University of London.
Imperial College of Science and Technology

London. December, 1966.
T ·J MY lrTCTHER
2
Existing theoretica1 and experimental imresti.gations on th:ln

shallow hype:cbolic paraboloid. shells are reviewed Cilld dis01rnsed..
The finite element method of a:ialysis haG also bGRn :reviewed arnl :Ln
particular the use 01· the triangular elements for "'.:ihe analysis of
the hyperbolic paraboloid shell and the plate problems have beer,
investigs.ted. For dcfi:1:i.ng the stiffness matrix for the triangu.lar
element under i;ho actions of be:'lding forces 7 a new polynomial function
has been deriveds This function has been used in extensive coJ.1pu~
tatio:ns covering the plate and the hyperboJ.:i.c paraboloid shell on

fixed and simply suppo:cted boundary conditions.
In dealing with the hypei;bc.lic paraboloid shell on flexible

edge ::.;uppo:i:-ts; the shell type of actions i where the membrane stresses
are assumed present, has failed to predict the behaviour of the
shell under the actio:J.s of lateral load..l.:::Jg. Accordingly the shell
analysis has been modified to exclude the membrane effect 1 retain-
ing only the plate type of actions. •rM.s modifioat:Lon has been
called the Curved Plate Theory. This theory has been found to
p::-edict the behavj.our o:~ the i:ihell on fl0xil/i..e edge support.
Accounts are al20 g:>rc:::J of two experiments carried cut on a

tj_mber shell and a vyl1ak models. Certain design faults have been
pointed out in the ex:perimer:.t en. the ti'D.ber model which makes i t
difficult to give an adeQuate interpretation to the results of
the investigation. The expe1:ient on the vybak model, confirmed
that the shell transmits its load by the plate type1 of actions.
3
ACIQ\TOWLEDGENIENTS
·---.- . ....----
~----·
The au"'.;hor has had the previledp:e of conducting the research

under the general guidance of Pr8fessor S. R. Sparkes. Detailed
supervision was given by Dr. L. G. :Booth to whom the author is
indebted both for his interest and the initiation of the pL'oject,
'11he experimental work, which was conducted in the Structural

Engineering laboratory, would have been impossible without the
generous assistance of the members of the staff' in general and of
Hr~ Ja Neale in particular, and their help is gratefully
acknowledged~
The author gratefully acknowledges the financ.:ial support

given by the Universj_ty of Ibadan during the period of my stay
in the United Kingdom. I am also indebted to Dr. A. Oo Adekola
for the encou1;agerne11t he gave me to pursue a course on higher
education and his continued interests,
Finally, I must thank my wife on whom the onerous burden

of typing the manuscript fell.
4
CON'I1EJ:TTS
-----· •-"'""_.._
ABSTRACTo 2
3
CHAPTER l.
1.1 INTRODUCTIOr. 8
1 .. 2 Geometry and Des~rj.ptic.n of the Hyperbolic Paraboloid Shell 10
lo 3 Review of Existing 'l;heories. 11
1"3~1 Membrane Theory. 11
lo3o2 Bending Theory.
1. 3. 3 Approximate S·~lu-Cionse 19
Review of Literature on Finite Element. 21
1.4.1 lfatrix Formulation of the Finite Element Theory. 21
l.4o2 Pro1Jerties of Elements. 23
l.4n3 Application of Finite Element to the Analysis of Shell
Structure. 24
Review of Experimental VJor:k 25
The Elastic Properties and the Analysis of Tinber Shells 27
Scope of Research. 28
CHAPTER 2.
•l'HEORY 1.
2.1 General Introduction. 29

2.2 Definition of a Structure in the Finite Element Contexto 30
2.3 Some Fundamental Energy Principles. 31
2.3~1 Principle of Virtual Work. 35
2.302 Principle of Virtual Forces, 36
2.3.3 Principle of Virtual Displacement. 37
5
2.4 Watrix Hethods in the Finite Element Analysis. 37

2,4.1 The general Theory of the Force Idethod of Analysis. 38
2o4.2 The general Theory of the Displacement Method. 40
2o5 Finite Element Analysis in the Displacement 11Iethod. 41
2.5.1 Direct Assembly of the k 1fatrix. 46
2@6 The :Boundary Conditions 48
2,7 Calculation of Stresses. 50
CHAPTER 3,
THEORY II.
Element PropeTties. 51
Structural Idealisation. 51
Suitable Elements for Idealising Shell Surfaces. 53
Element Geometry. 54
General Theory of Derivation of Shell Element Stiffness 55
3.2.1 Natural and Cartesian Definiton of Displacement and
Forces; Elastic Properties. 55
3.2~2 Natural and Cartesian Stiffnesses. 58
3 .. 3 Displacement Functions. 59
3.3.1 Requirements of the Displacement Function. 61
3o3o2 Displacement Function Previously Derived. 62
1. Non-Conforming Functions. Bending Element 62
2. Confor~ing Displacement ~~notion 63
3~ 3., 3 Natural Modes of Defoi'I!lation. 64
3~4 Derivation of Stiffness Matrix for Triangular Element
in Bending. 67
3.. 4.l Element Coordinate Systems. 67
3.4.2 Derivation of the equation of the Surface. 70
3.4.3 Symmetric Functions. 71
3,,4~4 Antisymrretric Functions. 72
3. 5 Computational Molecle. 74
6
3o5.l Stiffness Matrix in tbe Car.tesian System. 76
CH.APTEB. 4.
COMPUTFJi PR'JGRIU!l.
Local Cartesian. a:;:iLs for Shell Element and the 86

Coordinate •rransformation Matrix.
4.2 Assembly of the Stiffness lfatr:i..x for Shell Element. 87
4,,3 Assembly of the;, Stiffness Matrix for the entire Structure 90
4o4 Svlution of the Systeill of Equation. 94
4~4-1 Failure of -Che Gauss Sidel Iteration Process. 94
4.4.2 Tfatrix Diagonalisation,
1
95
CHAP_rER 5.
TESTS ON THIBER .P.Jif.D ITYBAK I:IODEijS. 98
Object of the Experiments. 98

5o2 Factor Governing the Cb_1jice of Scale and Material
of the Hod8l, 99
Description of the S~ructure. 100
Corner Detail. 101
Loe.ding a~'ld Instrnmentation AiTangements. 102
Testing of Specimen. 103
Preparation of the Tension Specimen. 103
5.7 Presentation of Results for the Tinber Model. 103
5.s Evaluation of the Timber I-fodel set up. 104
5.s.1 Evaluation of the test results. 105
5o9 Details of Constructio:.:1 of the Vybak Hodel. 106
5.10 Corner Details 106
5,11 Loading and :::nstrnnentation Arrangement 107
7
5.12 Deflection and Strain Measurements. 108

5,13 Presentation of the Results of Tests cm Vybak Model. 108
CHAP'l1EH 6 ~
DISCUSSION AND COHGiiUSION.
6.1 Discussion of the Results of the Theoretical Computations 133

60 2 The Plate F:co'..:>lems. 133
6. 3 Hy-perholic P.3.!_-abolcid Shell proble:;ns.. 136
6.3~1 Solutions for Shell with fixed and Simply Supported 136
BoundarieR.
6 .. 38 2 Comparison of Solutions for. the liyperbol:i c Paraboloid
Shell with fixed and Simply Supported Boundaries 13'7
604 Hyperbolic Paraboloid Shell on Flexible edge surports. 139
6.,4.1 Application of' the Shell Jmalysis 139
6.4.2 Stresses in the Vybak Model. 140
6 .. 48 3 :Modification of the Shell Analysis to exclude the
t:lembrane Effect. 141
6,,4.4 Correlation of the test result by the Curved plate
Theo:::·y. 142
6.. 4~5 Comparison of the Curved Plate Solution and the flat
plate .Solution. 143
6.4~6 Curved Plate Analysis :for the Shell with Simply
S~pported Bou.ndar-J Conditions. 144
6~ 5 General Reinarks. 145
6,..5.1 Hesh Size and Hodel Size. 145
6.502 Lifting of the Shell Centre. 145
6., 6 C¢it~')lus.iou., 147

601 Suggestions for further Research3 148
8
1.1 INTRODUCTION
In the last decade or two, hyperbolic paraboloid shells have

gained a considerable amount of prominence among architects in roof
construction for several reasons. .Among these are, ease of con-
struction associated with the straight line generators, aesthetically
pleasing appearance, economy and the fact that from the design point
of view, the membrane stresses are very simple to calculate. The
hyperbolic paraboloid shell is also ideally suited to construction
using timber laminates as the constructional material because of
the natural advantages associated with ruled surfaces. Coupled
with this is the extreme rapidity of construction and the possibility
of factory produced componenents especially where small spans a.re
involved. Figure 1.1 shows the basic units of the hyperbolic
paraboloid shells with edges defined by the genera.ting lines. Figure
1.2 shows the various combinations and permutations possible using
the basic unit shown in figure l.lb. Thus large areas can be covered
by repetitive use of the basic unit.
Constructional problems associated with hyperbolic paraboloid

shells a.re now largely solved, but the design procedure is far from
satisfactory. It is possible to formulate the differential equations
of equilibrium and compatibility using the shallow shell theory.
The solution of the equations, ta.king into account the boundary
conditions, by the classical approach, seems to present some
difficulties. These eriae from the mathematical properties of the
equation which is essentially hyperbolic in nature. The solution
also accepts only a limited form of boundary conditions and often
the boundary conditions have to be made to suit the governing
equation, a procedure which results in unrealistic boundary conditions.
9
/
/
Q-"('
/
' '-/Jlf
/ 0...
/
"
/ . / IJ'4 c
'::J
""
/
/
~C" l)l...jrl r~ +,c; 1-tb.

10
1. 2 Geometry a!J._d_)1_~§.£.=r:iJ2.tion of t_f!.L!J;-'."}2_erbolic Paraboloid Shell
The hyperbolic paraboloid is a t:-mely curved surface and comes

..;_;'.,
mathematically into the general classification of ruled surfaces •
Further subdivision within the general class is possible depending
on the form of the ruling surfaces or generators. The generators
m~y be curved £i5. 1.1a in which case translational shell h~ve been
applied or straight fig. l.lb when ruled surfaces are appliedo These
distinctions are however superficial. Physically, both surfaces are
translational as well as ruled. If we consider a part of the surface
bounded by the straight lines AB and CD of one set of generators
fig~ lola and AC and ED of the other set, then the surface is
developed if a straight line traverses one set but is kept parallel
to the other set. On the other hand, the surface can be considered
translationalo For this purpose, we must realise that all planes
parallel to the plane of symmetry are parabolae of opposite curvatures.
The surface might therefore be conceived as that of the translation
of one curve on the other. Moreover the curvatures of the parabolae
being of opposite signs, the product of the curvatures known as
the Gaussian curvature is negative, 8<ld as such? the hyperbolic
paraboloid is part of the family of shells known as the shells of
negative Gaussian curvature. For the purpose of this thesis, the
above classification is adhe::i.·ed to and only hyperbolic paraboloids
defined by straight line gene:;~ators are treated.
If we denote by 0 the angle between the plane of symmetry and the

coordinate axis ~ say, then the general equation of the hyperbolic
paraboloid of the ruled surface type is given by
Z = lcxy Sin2e l.l
where
11
and r1, r 2 are radii of curvature of the parabolae.

In the special case when e = 45°, Ox and Oy are oxthogon~+, the
intersecting curves are rectangular and the surface is then called
a rectangular hyperboloic paraboloid. This is the conventional
shape usually adopted in roo~ constructions.
I£ a parallelogram is described on two opposit~ g~n~ratorg for

example .A13 and AC, fige l.lb, the fourth corner of the parallelogram
being E, the line DE rtllls parallel to the axis Oz. The length DE
is termed the linear distortion (v) of the hypar surface .A13CD. If
the projected area of the shell is Q on a plane perpendicular to the
axis Oz, then the ratio v/Q is term~d the specific distortion, and
is equal to the quantity (k) in equation 1.1. This is the practical
way of calculating (k). The rise of the shell is defined as the
arithmetic mean of the differenGe of the heights of the low and
high corner~.
1~3 Review of Existing Theories
In practice the design of hyperbolic paraboloid shell has been

based on the general assumption of a momentless state of stress in
the shello In this assumption only the inplane stresses are taken
into consideration~ The theory has therefore been appropriately
termed the membrane theoryc The governing equations are essentially
those derived by Purcher(l9 3 l)for shells of arbitrary geometry. In
his derivation, the static equilibrium of the projection of the
shell on a plane normal to the axial plane is considered, and the
resulting differential equation for hyperbolic paraboloid shell
results in the particularly simple expression
12
1.2
where ¢ is the Airy stress function with the usual meaning

the theory of elasticity ar.d
1 = -Z + Xz + Yx
k k
gives the load function.
Exact solutions of equation 102 have been derived by Tester(l9 4e)

and Rejssner has generalised one of Tester's formulae for a
constant load per unit area of the s~ell surface, rectangular in
plan. Practical design methods with tables have been prefaented by
Candela (l 960 )Parme (rl 9 5'e)and Portland Cement Association \19 60) •
Equation 1.2, though of a mathematically simple form however

has some limitations when applied in practice. Reissner~955)has
discussed the various limitations of the simple membrane theory
as applied to hyperbolic paraboloid, and Tottenham(l957)a1so on the
same problem concludes that the stress function ~ cannot in general
be prescribed along all the edges, and therefore recommended that
the effect of finite displacement should be taken into consideration
to overcome the defectso We also note that the basic non-linear
differential equations of equilibrium and compatibility for the
membrane state are presented in the same publication. Flugge ( 1960),
Flugge and Geyling (195~ have also shovm in addition that discon-
tinuities are developed in the internal stress system when the shell
is partially loaded~ The displacement consideration reveal further
limitations of the membrane theory.
To illustrate briefly the problems encountered in the

13
membrane theory of hyperbolic paraboloid shell, the stress and defor-

mation of the shell under a uniformly distributed load per unit area
is considered. Setting X = Y = O, we obtain
'~=Zk
.
-~
~x~y 2 2
where it has been assumed that Z = -q.
This equation has a general solution of the form.
4 = -l qkxy + f(y) + g(x) 1.5
2
The stresses in the shell are then given by
c~nstant throughout
the shell
·~2
Nx = rl.:.£{.tl , arbitrary 1.6
;) .2
'-'X
Ny ='~ 2 f(x), arbitrary

y2 v
In general, the forces Nx, Ny cannot be independently prescribed
to satisfy the boundary condition. A trivial solution can be obtained
if we put Nx = Ny O" Thus the shell is under a purely shearing
state of stress. A rough estimate of the load transmission can be
obtained if the shell is imagined to consist of a system of c1"l.ble3
and suspension systems. A half of the load under a uniformly distri-
buted load system can then be assumed to be carried by each system.
Because all diagonal cross sections are parabolic and are generally
part of the same parabola, the thrust in each cross-section is the
same as the tension in the suspension system. The stress distri-
bution in the shell can thus be described as one of tension along
lines parallel to the low diagonal. The stress distribution thus
results in shear stresses along the edge of the shell. Thus if
14
the edges are restrained by shear forces, then the shell should be
capable of bearing loads. It is thus essential to have edge beams
around the shell to provide the restraint.
The edge beam being under shear load, needs to be restrained

in order to balance these forces. The restraints of the edge beam
take several forms. The simpl8st type is the provision of a tie
rod between the lo~ corners and vertical supports at these corners.
The more complicated types are buttresses to take the thrust
resultants from the edge beams. 'rhese become heavy for large
shells or shells with high rise.
Thus far we have used only the conditions of equilibrium of

forces to determine the state of stress within the shell. To ensure
however the conditions of compatibil:i. ty of strain and displacemer. ts,
we must introduce the displacement components u, v, w. The stress-
strain relation and the strain-displacement relation will enable us
to determine the displacements. The stress-strain relation has
been derived by Reissner (1955) from a consideration of the minimum
potential energy principle~ In the publication, he showed that the
purely membrane stare does not allow the satisfaction of the
symmetry condition.
The deformation of the shell comprises a shear deformation Y

which if we assume the Poissons ratio v to be zero without any loss
of generality gives
y = nxy -.L_ 1. 7
Gt Etk
The shear strain is also given by the differential equation
y = ·;.!JL + J v - Zero 1.8

15
If we equate the two expressions 1 we find a possible solution in

the form
u = l/2yy
v = l/2yx 1.9
w = 0
The shell deforms into a form of rhomboid, with the high corners
moving out and the low corners moving in. To eneure a membrane
state of stress therefore, the supporting members at the edge have
to slide freely with the shell~ an arrangement which is difficult
to achieve in practice. Another mode of deformation is the
following
u :: V == O, LU = y /2k.
The possibility of this mode is present in principle 1 but in
general the required displacement of the edge will not occur.
1.3.2 Eending TheEEX.
As shown in section 1.31, the purely membrane state of stress

is not possible as a result of lack of compatibility between shell
edge and edge beam. Generally the shell would be accompanied by
bending moment which in general is expected to be maximum towards
the periphery of the shell.
The general theory of bending of shell has been in existence

since the nineteenth century. The work of Aaron and Love are
significant in this field~ However most work in this field for
doubly curved shells follow the th9ory developed by Marguere (1938)
and later by Vlassov (1944). The complete theory yields two non-
linear partial differential equations of the fourth order. No
direct solution of these equations are known, and there has been
a general simplification of the equations in order to achieve some
solution.
16
The general assumptions have in many respects been similar to

those used for deriving the differential equations for plate in
bonding. Plane sections are assumed plane before and after bending,
hence stresses in the plane of the shell can be integrated across
the thickness to give both the stress resultants and couples.
Strains are also expressed in terms of the displacement of the
middle surface, and thus the analysis of shell is reduced to the
study of the deformation of the middle surface. The foregoing
apply essentially to thin shells, in which category most shell roof
structures and allied surfaces belong.
There are two general forms which the differential equations

can take depending on the method of achieving the solution which
the analyst has in mind. The two alternative forms are given below:
1.. Margy.ere or Vlassov.~~E.
This comprises two partial differential equations expressed

in terms of the stress function (F) and the transverse displacement
(w). For shell whose middle surface is given by z = bey~ the resulting
expression is
-y-4p - 2Etk (;'. 2 w_ = 0
~ x,__'. y
1.11
\ 2
D\1'4w + 2k i) F = -q
~~y
where v4 and
and q is the uniform distributed load. The stresses and displace-

ments are then in the form
•
17
Nxx Nyy = 't.l 2 F , Nxy

i) x2
Mxx == -D ,f ~ 2 w + µ~,2'\'L, ·; titc

o(' - - ;,.
l ~x2 Jy2 l
1.12
'
Qx
-ox·~
= -D ( \t 2w) etc
u etc
A variation of the Marguere's equation is possible. Thus

using the Vlassov's stress displacement function ¢ defined such
that the first of equation (1~11) is identically satisfied thus
w = F = 1.13
the second reduces to a single eighth order equation
Dv
8
0 + 4Etk
2
w"· •11 = -q 1.14
an equation which reminds of a similar equation in the theory of
cylindrical shell.
2. Flugges Equation
By using the stress-strain and the strain-displacement

relations, in the equilibrium equations, three simultaneous diffe-
rential equations can be obtained in terms of the displacement
u, v, w. The resulting expressions are
2
yy + (1 -µ)u u ,xx ·· 27t(l - µ)w, x
2u = 0
(1 + µ)u,xy + 21t2 V,xx + (1 - µ)T 7 yy - 21t(l - µ)w,y = 0 1.15
(21tw - 11:2u,x - v, y ) + A(w, YYY'J + 21t2 w, - 1t2 w ) =Q

xxyy xxxx
where u, v, w are displacements expressed in non-dimensional forms

18
u = u/a v = u/b w =wk
TC = a/b 9 ;-... = t 2 /12(1 - µ)c 2 rc , Q = .:.tl_l - µ)a 2

Etc
and the subscripts after the comma indicate differentiation. The
stresses are also expressed as in equation 1.12 in terms of the
displacements. Equation 1.15 can be reduced to a single eighth order
partial differential equation in any of the unknown by suitable
transformations. The resulting expression is similar in form to
equation Ll4.
Equations 1.11 and 1.15 have been employed by various investi-

gators to analyse the h{,rperbolic paraboloidshell. Flugge and Conrad
(1959) hageused Vlassov's equation by expressing the load as a
double Fourier series and then following a method similar to
Ambartsumyan (l 947 ) separated the plate solution from the shell
solution, thus taking advantage of the rapid convergence of the
remainder. Flugge and Conrad 1:.ave also worked out examples of shells
of possitive and negative Gaussian Curvature under concentrated
load. They infer, in particuJar, that the stiffness of the hyper-
bolic paraboloid is roughly eq~al to that of the flat plate~ Levy
type solutions have been used by Bou~a (l959) and Apeland {l9 6l)
The later derived an i~fluence coefficient method for a shell sup-
ported on two opposite boundaries, thus enabling the actual boundary
conditions along the other two opposite edges to be applied. By
ploting the real roots of the characteristic equation against certain
shell parameters, it is shown that the edge disturbances in hyper-
bolic paraboloid shells are damped at a much slower rate than in
shells of positive Gaussian Curvature.
19
The set of partial differetial equations 1.11 and 1.15 are to

be solved in conjuction with the boundary conditions. It is then
possible using relations given in equation 1.12 to obtain the dis-
placement and stresses in the shell. A complete analytical solution
of the eighth order partial differential equation cannot be obtained
except in the case of the shell subject to pure bending - Maunder
and Reissuer (J957). Tuluch atten-tion has therefore been directed
towards the approximate solution of the equations.
The Levy type solution has already been referred to in section

1.3 0 2~ Some simplified solutions are based on the influence of the
roots of the characteristic equations resulting from the I1evy pro-
cedure~ The Levy solution ccns::'..sts of assuming that the stress
function J5 and the load q can be expanded in series of the type
OJ
0 ~, .. ePyx Sin yx
q = .4. qo
" ,.
~
r,: I ..,()
Sinyy 1.16
n; L
11:.1
where y = nn/b
which when ap~lied to the eighth order differential equation 1.14
yields the characteristic equation
( p2 - l) 4 - 1- p2 0 1. 1 7
p4
6
from which 4 rea.l a.11d 4 :hnagina:cy roots are obtained - Loof (l9 l)
pl-2 ± a1
p3m4 = ± cr.2 l.18
P5-8 = ± (a3 ± iµ3)

6
Loof (l9 l) has presented the roots in a graphical form and showed
20
that the values o.f o:2 arc sma~.l and can therefore be neglected.
He thus arrived at a sixth order equation 1 requiring three boundary
conditions on opposite edges and has shovm t:hat no generality is
lost. He also presented a series of approximate formulae on the
hyperbolic paraboloid shell supported on flexible edges. These
formulae have been investigated by Batchelor (l9 6 3) who showed that
results based on these formulae failed to be correlated by experimental
resultso By making assumptions similar to that of Loof, Bleich and
Salvadori(l959) have reduced the equation 1.15 to one similar to
that of a plate supported on an elastic foundation~ In a similar
manner., to BJ.eich ar.d Salvadori 1 Vreedenburgh(l95B/59) has reduced
equation 1.14 to a fourth order one similar to that of the elastically
supported bear:i..
Other approximate methods which have been used may be catego-

rised under the heading of numerical procedures. Among these are
finite differences, used by nas Gupta (1961) and the Variational
technique using the Kanta:c:ovitch and Galerkin procedures. Cheetty
(1961) has used the Kantarovitch procedure in conjuction with the
]~atrix Progression Technique developed by Tottenham (1962). In
this method? a solutio~ is sought i~ the form
ti
.0 :L fi(x)gi(y) 1.19
1
where gi(y) are the assumed\ functions which satisfy the boundary
conditions in the 3r-direction and f5_(x) are the functions to be
determined, The initial set of diffe~ential equation is thus reduced
to' the form
1!1i(0) = ? 1.20
where Fi is a set of ordir.ar;r differential operator and p are the
external reactions. The Kantarovi-~ch process is now used to minimise
the residual in the form
1.21
21
The advantage of tho method lies in ito flexibility to allow the

exact bounda,ry conditions to be satisfied and any desired accuracy
can be obtai:!.'led by successive use and improvement of the functions
assumed.
1.4 Review of Liter~ture on Finite Element
The basic idea behind the finite element formulation of

structural analysis has been known for a long time. The forerunner
of the finite clement analysis was the frame work e.nalogy pioneered
by Heronikoff (l94l). Ee conceived a structure as being represented
by a series of interconnected frame work, whose elastic properties
are suitably defined to correspond to that of the actual structure.
The structure can then be solved by conventional analysis like the
relaxation method and moment distribution. The accuracy of the
Herenikoff's analogy clearly depends on the success with which
the elastic properties can be defined. We shall call Herenikoff's
idealisation as a one-dir'!ensional element idealisation ..
The generalisation of ~he one dimensional element representation
into two - and three dimensional idealisation 'J"eo I"lttC•.=
possible by the advent of t~e electronic computer. Without this
the solution is not possible because the number of simultaneous linear
equations resulting is greater ths,n that the normal conventional
desk calculating machine can cope with.
L1;_.l Matrix for:nulation of the Finite Element TheoE:t
Generalisation and unifications of the various structural

theories have been tl1e sub,7ect of the work of many investigators
during the last t·vo decades. Attempts have been made to summarise
the past developments in structural analysis by me.ns of concise
methemo.tical s·catements, which are more convenient to write down
22
and also in such an easily processed form for machine computation.

One such formulation is the matrix algebra approach.
The most comprehensive contribution has been made by research

workers in the field of Aeronautical Engineering. This however is
not surprising, because, they have always been faced with more
complicated problems, especially as aircrafts become more complex
and refined and solutions, like the series solution which are all
embracing, are no longer adequate. Coupled with this, is the fact,
that the limited use of the early electronic digital computing
machines were confined mainly to the aircraft industry. The most
general and most systematic formulation of the structural theory
must necessarily be attributed to Argyris (1954/55) who expressed
the various phases of the analysis in a standard sequence of matrix
operations. Generalisation of the theory to make possible the two
dimensional element representation of structure is due to Clough
(1960) who defined a structure as being composed of an assemblage
of a finite number of finite size elements that are interconnected
at finite number of joints. It would appear that as a result of
imposing continuity requirements at only these few joints, the
flexibility of the idealised structure as compared with the actual
structure will be excessively increased. However, by restricting
the internal stresses to a few fundamental patterns that will
approximate the actual stress distribution, the structure will be
considerably stiffened$ Although there is no rational procedure
to determine precisely the relative and the resultant effects of
each of these idealisations, it is natural to expect that the error
resulting from the latter will diminish when the mesh size is refined.
How fast that error will be reduced can only be tested by solving
the given problem using different mesh approximations and observing
the convergence of the results.
23
1.4.2 Properties of Element

--------
The successful application of the finite element technique
depends on the accuracy witb which the approximating elements pro-
perties could be defined. Structures differ considerably in shape
and !':'O do elements., In fact there are infinite element types to
analyse a given structure. But for shell type of structure which
is considered in this thesis we are concerned mainly with elements
which are capable of transmitting both membrane and bending stresses.
Elements can roughly be categorised into displacement models and
equili1rium models. Earlier models were based on the parametric
representation of the displacement field (displacement model) or the
stress field (equilibrium model). The unknown parameters are iden-
tifiable with either the nodal displacements ( tramilation and rota-
tion) for the displacement model or the nodal forces (moments anc.
e.,,_u.1\;b,.-1um ( 1964)
shears) for the d~splaeement field~ The works of Pian and
Clough (1958) respectively represent some very significant contri-
bution in each fieldo We merely note here that the series representa-
tion is arbitrary. 1.111eoretically the series must be infinite, but
1
in practice it is finite and the way the infinite series is curtailed

is left to the judgement of the authoro The Argyris school have in
recent P'ublications - Argyris (Sept - Oct 19~5) come out with a more
rational approach towards defining element properties. The approach
is based on what has appropriately been called the Natural Modes~
These could be either of stresses or displacement. The significarlt

point about using the natural mode approach is that the guess, as
to which polynomial term to include or exclude from the infinite ·
series, is taken out of the formulation.
In general an element could be an n-sided polygon having (n)

nodes, but the general purpose element utilised in plate and shell
24
analysis is restricted to n == 3 - corresponding to the 'triangular ele··

ment and n = 4 - corresponding to the quadrilateral element. The
rectangular element which is a special case of the quadrilateral
element has received a considerable attention than the triangular
element although it has always been recognised that the triangular
element has an intrinsic advantage over other elements by virtue of
the fact that it has a great adaptability to meet the requirements
of most structure and in particular curved and irregular boundaries.
The formulation of the stiffness matrix for triangular element and
rectangular elements undeT in-plane forces, seems to have been given
exhansjve treatment in the following refreces: Turner et al (1956),
Clough ( 1958). The bending stiffnesb formulation however has been
confined so~ely to rectangular element (Zienkiewisz (1965) Melosh
(l9 6 l)j although recent ~ublication by Clough et al (1964),
Zienkiewisz (l9GG) show a trend towards the triangular element
stiffneos formulation.
L4 .. 3 Avolication of Finite Element to the Anal;rsis of Shell

Structure
It appears that the limited use of the finite element analysis

to shell problem has been confined solely to the analysis of cylin-
drical shells and domes. .b.dini(1961) has used the triangular element
to analyse a spherical shell while Zienkiewisz (1964) has used the
rectangular elements to analyse an arch dam. .Apart from these
examples, the finite element has not fou_rid a general acceptance in
the analysis of shell structure. There is no where in the literature
where the hyperbolic paraboloid shell has been analysed using the
finite element technique.
25
l:..5 Review of Experimental Work
A considerable nu.~ber of tests have been carried out mainly

in reinforced mortal of hyperbolic paraboloid shells as an aid to
design, Consequent on the nature of these tests no generalised
conclusions can be drawn as to the behaviour of the shell in general.
Much fewer tests have been carried out on Timber shell, and with
the exception of one or two, all the tests have been carried out as
aid to design. Hovrnver some of the tests which have been reported
have some significance for the problem discussed in this thesis and
it is p~oposed to rev~ew them here.
Leicester and ~fassey ( 1960) have described· teats on a 1/16

scalQn . polyester resin model of a system of hyperbolic paraboloid
shells consisting of six petals resting on six supports. They
indicated the presence of quite large bending moments even in
regions well removed from the boundaries of the shell. The same
conclusion was reached by Flint (1961) and Rowe (l9 6l). Lanletta
(l9 6l) reported tests ca=ried out on a brick hyperbolic paraboloid
shell supported on straight line generators. The tests were
intended to study the statics of the shell under uniformly distri-
buted loading. The author infered that the shell behaviour
differed considerably from that predicted by the membrane theory
and approached that of a plate. Another important conclusion from
the point of view of this thesis is that the test showed that the
state of stress was not influenced to a great extent by the change
in the rigidity of the edge beam. This is in support of the
observation made in the course of the present investigation. We
rowever ob$erve that this is not the view held by Batchelor (l9 63)
The most comprehensive·tests on shell with flexible edges

26
were carried out by Batchelor (l 963 ) o The shell was Bftx 8.Jt:x 3/41'1
with a rise of 141'1. The corners were all supported and tie rods
were provided. Arrangements were also made to vary the edge beam
stiffness. The model was of a reinforced concrete mortar. The
result of the investigation was very interesting. We observe that
the extensional or membrane theory failed to predict the behaviour
of the model, neither could the bending theory. A systematic
comparison was made between the shell behaviour and that of a plate
of the same dimension but with fixed and simply sup~orted boundary
conditions. It was observed that experimental results lie within
the solution obtained by treating the shell as a clanped or simply
supported flat plate. The result alsu showed a great susceptibility
towards the rigidity of the tie rod. With rigid tie rods, corrella-
tion with clanped plate solution was within 30°/09 whereas with
flexible tie the behaviour was nearer to the simply supported plate.
Batchelor concluded that the shell behaved more like a flat plate
since the deflections in the middle region of the shell was reasonably
predicted by regarding the structure as a flat plate.
A few test however have been carried out using fixed boundary
conditions by Ch&tty-( 1961 ), Brebbia( 1966 ). The deflections and the
stresses were observed to have been reasonably predicted by the
bending theory of shell. No tests however have been reported on
hyperpolic paraboloid shell on gable type or the simply supported
kind of boundary conditions although the possibility always exists
in theory.
27
106 The Elastic Properties and the Analysis of Timber Shell~
The problems of the analysis of Timber Shell by the cJ..assical

approach is necessarily complicated by its elastic properties.
Timber is a variable material and displays high degree of heteroge-
neity. It may be represented as a cylindrically orthotropic material
with the principal axes in the axial, radial and circumferential
directions. When boards are cut from a log, these boards maintain
the longitudinal axis, but the radial and the tangential directions
are differently orientated, which Corl;dition makes the application
of the theory of elasticity dubious. Radial and tangential elastic
properties however are sufficiently s~milar to enable us to consider
a piece of board as an orthotropic material, with the principal axes
along and across the plane of the board.
Tim~er shell roofs are made up of layers of boards placed in

different directioas and nailed and or glued together. The number
of layers and their arrangements in timber shell roofs vary very
little. Shells with more than four layers are rare, and the orien-
tation angles are normally 45° and 90° with the adjacent layers.
Fortunately the thickness of ~he layers are invariably equal.
Because the cross-section of the shell is not homogeneous, the

application of the theory of elasticity leads to a definition of
several elastic constants - one set for plane stress or st·rain and
the other set for bending. Assumptions in the shell theory also
lead to coupling terms between ex7.ensional forces and bending
moments• rrhese problems haYe been discussed at great length by
64 1961
Dong (l9 ) and Reissner and Stavsky < ). The general theory of
the anisotropic shell have also been Fresented in the same publica-
tions. Comprehensive tests reported by Knipers and Loof (l9 62 )
28
also show that tests on specimens of shell material do not follow

the theoretical prediction of the elasticity theory. The use of
average elastic properties based on experimental analysis is
envisaged in this thesis.
1•7 Sc9P,e of Research
1. Investigation into an improved displacement function for

the triangular finite element in bending.
2. Investigation of the stress and displacement distribution

of the hyperbolic ~araboloid shell whose edges are
unsupported by the use of the Finite Element Technique.
3. Feasibility study of the model of a timber hyperbolic

paraboloid shell using a table-sized model.
In cases 1 and 2 comparison are made between experimental results
and those obtained from knovm approximate solution.
29
THEORY I
-----
The object of a structural analysis, is to be able to evaluate

the responce of a given structure to changes in its environment.
'·
The changaJmay come about as a result of the application of external
forces, which may be 0f static or dynamic origin, or displacement,
as for instance in the sinking of supports or changee.::ln the thermal
conditions around the structure. As the changes: occur the structure
is displaced to a new equilibrium position with consequent changes
in the distrubution of stresses throughout the body of the structure.
The methods available for analysing any structure vary widely an~
depend on the complexity or simplicity of the structure. Basicall~,
hovrever, we are attempting in any method to sa.tis.fy the following
two requirements simultaneously namely (a) equilibrium and (b) cor:ipa-
tibility.
Ey satisfying the equilibrium condition, we are in effect

establishing a balance between the external and the internal forces
o~ the stru~ture. If the statical conditions are insufficient to
render the problem determinable, then we must cor.sider in addition
the question of kinematic or compatibility condition , which is the
relationship between the displacement of the structure and the inter-
nally developed strains. Two sets of equations expressed by (a)
forces i.e. the equilibrium relationships and (b) deformations 'i.e.
the compatibility relationships are produced. The link up between
(a) and (b) is provided by a third relationship which is quite
distinct from the mechanics of the analysis. This is the relation-
ship between the force and the displacement often identified with
30
the stress-strain relationship for the material of the structure

and is thus a specific p:r.operty of the particular structure. These
three requirements are recognised as normal ingredients in a shell
analysis program.
In this thesis one'method of analysis, the Matrix Method of

ct<r. "'i:'f>li~d
analysis>~ refQrreQ. to~ the Finite Element Method of analysis
is presented. This method of analysis has been extensively developed
and used in analysis df structures during the la.st decade in the
field of plane stress .and. related problems. E"..-::::tension of the .method
to inclu~e the effect of bending as in the problems of plates and
shells i:;i -0f recent origin. This thesis concerns itself with the
particular application of the method to the analysis of hyperbolic
paraboloid shells. It will be desirable here, however, to discus
the gener~l scope and features of the method.
2. 2 Definition of a Structure il.!.....the F:l,;&te Element Context
In the matrix algebra approach to structural analysis, a

structure is defined as an assemblage of finite number of finite
size elements, which may be of arbitrarJ uize and geometry, each
element is homogeneous and li.nearly elastic E:nd the connections to
each other are specifiable only in terris of some finite number of
Jllodes., :By finite size element is meant elements which may be imagi-
ned to be mechanically sliced from the given structure. When the
elements are assembled together in a given structure they must
satisfy the following conditions:
(1) Equilibrium: the internally developed forces, which
appear as equivalent nodal forces must balance the
31
external forces on the nodes.
(2) Compatibility: the elements must continue to fit together

in their deformed states under the action of applied loadso
In practice this is achieved only at the nodal points but
the theory assumes that the element deformation has been
described in such a way that connection along the sides of
the element is achieved.
Satisfaction of these conditions is generally approximate.
Customarily, one of these conditions is required explicitly, though
not nec0ssarily exact:y, whereas the other follcws implicity
according to the mechanics of the particular method being used. In
the direct stiffness method, continuity of deformation is an explicit
requirement, whereas the equilibrium requirement is implicit in the
aasembly procedure of the method.
2.; Some Fundamental Energy Principles
The consistent formulation of a structural problem in the finite

element concept is conveniently based on one of the various energy
theorems or variational principles. As far as the finite element
analysis is concerned the most widely used are the principles of
variations of displacements and the principle of variation of stresses,
also called the complimentary energy principle. More general princi-
ples exist which in theory allow simultaneous approximations on
displacements and stresses. (37) Since these theorems are so funda-
mental to the finite element analysis, especially in defining the
bounds to the solution, it is proposed to review here as briefly as
possible, the most fundamental of the theorems namely the principle
of the vitual work. In order to present the principle in its most
generalised form, it is presented in the context of a three-dimen-
32
sional continuum.
Consider a three-dimensi0nal body in space, whose position is

fixed with respect to some prescribed axes which in this particular
case has been chosen as the cartesian system of axes (fig. 2.l)o
The loading, the stresses and the strains are similarly defined in
that system~ For the equilibrium of the body in space we define the
following force vectors:
(i) Body forces/unit volume (fig. 2.?) w given by
w= { wx Wy wz } 2 .1
(ii) Surface forces/unit surface area (fig. 2.5) qi given by
{ qix qiy <pz} 2.2
(iii) Internal stress vector a given by
6 ::; { O'XX dyy O'ZZ <Jxy e5yz cJZX} 2o 3
where the subscripts in equations 2.1, 2.2 and 2.3 have their
usual meaning in the theory of elasticity and are in general
functions of x, y and z.
For an elemental volume subjected to a variation of stresses

(fig. 2.2), the equilibrium equations take the form
1_~ + _g_ 6!.l_ + 0 cJZX i· WX = 0
ox b y () z etc 2.4
In matrix notation equation 2.4 can be written as
a g_ 0
oz O'XX: Wx
a~ oy cryy
0 ~ 0 O"ZZ wy
== 0
I
'T
a~ 01' ~ <Jxy
<Jyz
c) d 0 wz
ai ~-~ ()x 6ZX
which can be represented more compactly as

DC5
+ w 0 2.4b
where D, is the rectangular matrix of differential operators.
. y
A~. !)./
I t'~
,L~'41. --
/ ~
/
Fi&. fl~.
34
On the surface. equilibrium relations of the following forms are

satisfied;
q>x = O'xx Cos(n~x) + O'xy Cos(n;y) + O'zx Cos(n,z)
~Y = etc
where Cos(n,c) is the direction cosine between the outward no:rmal
and the{c)axis. Equation 2.5 can be put in the form of equation 2.4b
if~we make the substitution
Cos(n,c) = ~ n
() c
and equation 2.5 then assumes the form
DnO' = <p
Thus far in deriving the equilibrium relationships we have

assumed no deformation of the element of the continuum and therefore
the equilibrium conditions apply strictly to the undistorted system.
This is the reason why there is a linear relationship between the
stress system and the load system. One consequence of this is that
we can have a superposition of equilibrium states~ This is particu-
larly significant in the derivation of element flexihility using the
force method, in that we can build up a complicated stress system by
combination of simpler equilibrium nodes.
To descri.be the deformation of the element we require the follow-

ing vector~:,,
(w) The strain vector in the body of the element given by

= {!xx <:.yy 'zz (xy l;yz ~zx} 2.6
(v) The cartesian displacement components
µ={u:vw}
The compatibility condition is then expressed by
~-"" ntµ 2.8
where the super-script indicates that the matrix D is transposed.
35
Again the strains are linear in the displacements and therefoTe we

can superirLLpose compatible deformation states which give rise to ·a
new state which is also compatible. This is an exteremely useful idea
when formulating the stiffness matrix of an element from pure kine- ,,
matic arguments.
_s.31 Principle of Vitual yt~
If we consider a stress system 6 which is arbitrary but is in

equilibrium with w and cp, also an arbitrary strain system £ which is
kinematically consistent with µ, these two systems being independent
of each other except in the context vf correspondence then the prin-
ciple of vitual work can be expressed by the scalar equation
Jot= f vwtµdv +.fr: cptu qs 2.9
where the integration is carried through the volume v and over the
surface y$ of the element. Equation 2.9 can be derived from equations
2.4b, 2.5b and 2.8 by the use of Greens transfo!'mation (25) Equation
2~9 can be interp=eted loosely by saying that the work of the internal
stresses and the corresponding strains is equal to the work of the
external forces and.the body forces and the corresponding external
displacements. It is significant to observe the following:
(1) ~ is not necessarily the true system of stress but includes
this special sta~e. It must be in equilibrium with the
external force system in other words it must be a stati-
cally equivalent stress system.
(2) ~ is not necessarily the same as the true strain system
but includes this special state. It must, however, be
kinematically consistent with the external displacement.
In its broad application to structural problems, two corrolaries

of the principles of vitual work are particularly important. They are
the principle ofvi.~tual forces and the principle of virtual displace-
ment.
2. 3,2 The Princi.E]..e of Virtual Forces
Equation 2.9 is now cast into the more usuful form suitable for
engineering application in the form
§tv = Rtr 2.10
where S is the stress resultant replacing ~ and ":!... is the strain
vector replacing €_, R is the external load vector replacing both w
and ~' while r represents the external displacement replacing ~1. To
express the principle of v::irtual forces, 2.10 now assumes the form
-t t
v = R r
S 2.11
where the bar below v has been dropped to indicate that it is the
actual compatible strain, R is now arbitrary. If in particular w9
form a statically equivalent system expressed by
S = bR 2.12
and Eubstitute 2ol2 in 2.11 we obtain
Rf"btv = Rtr 2.13
and since R is arbitrary
r = btv 2.14
which is the mathematic expression for the unit load theorem. Suppose
however that S = Se a self equilibriating stress system, 2.11 now
assumes the form
Setv = 0 2.15
which is a formal expression of the compatibility condition. Equa-
tions 2.14 and 2.15 form the basis of the force method or analysis.
37
2.~3 Pxinciple of Virtual Disnlacement
If S and R are tn the actual system and v and r in the arbitrary

system equation 2.11 assumes the form
vt S c rtR 2.16
which ig the formal statement of the principle. If in particular
from a kinematic analysis we obtain
v = ar 2.17
and substitute 2.17 in 2.16 we obtain
rtats = rtR
Since rt is arbitrar'J, we make it unity and thus obtain the unit
displacement theorem, the resulting expression being given by
R = ats 2.18
which is the equilibrium relationship established from pure kine-
matic argument.
2.4 Matrix Methods i.n the Fin~te Element Anal;rsis
The methods used in the matrix analysis of structure can be

classified into two categories namely, (1) The force method or the
flexibility method and (2) The displacement method or the stif.fness
method. These methods are fairly well established in the analysis
of frames and one dimensional structure in general. Extension of
the methods to two and three dimensional problems is made possible
by matrix notation, which enables groups of displacements or forces to
be represented by single quantities, and so enables the general formu-
lation of problems in all classes by the same matrix interpretative
scheme. The flexibility and stiffness methods are interrelated to
one another since one is the inverse of the other. The distinguish-
ing factor is the order in which we satisfy the equilibrium and the
compatibility conditions. The general formulation of structural
problems in the flexibility and stiffness method is now presented in
the light of the fundamental theorems presented in section 2.3.
38
Equilibrium conditions alone are not sufficient in the general

case to fix the stress matrix S for a given load R. But we can always
find stress system;;S which satisfy all the equilibrium conditions,
such stress system;we define as being statically equivalent to the
applied loading. Thus based on the undef ormerl geometry we can always
write the equilibrium relations in the form
S= bR 2.19
For a statically indeterminate structure, b is indeterminate. But
if however we aosign soae redundancies X and form a statically equi-
valent system boR we obtain
S = boR + b, X 2.20
where bo - is a rectangular matrix of constants having m columns,
whose ith column is a stress system in equilibrium with
Ri. =1
bt - is a rectangular matrix of constants having n columns)
the ith column of which is a stress system in equili-
brium with Xi = 1
X - is the (nxl) vector comprising the magnitudes of the self-
equilibriating stress system more familiarly lmown in the
literature as the redundancies.
Thus in the force method1 equilibrium is first satisfied by assigning
as many redundants as would make the structure determinate. We now
denote the deformation of the structural members by the vector v and
displacement of the structure by r. Using the virtual force or the
complimentary energy principles, the compatibility condition is
expressed by
bl tv =0 2.21
The deflection of the structure is now given by
r = botv 2.22
'
.J
'
f'ft, /1·11 ~l"~~T£b Sil6Ui' b/I) $1#11111tfJ.£N1 /'l~M. t.D~S

r~ liN1'1'f el-~'11Nf /llMl.ff'S
'
40
If the stress strain law for the structural member is now given
by a relation of the form
v = rs + H 2.23
where f is the flexibility of the member and
His a column of initial strains due,for instance,
to self straining induced by incorrect fit, manu-
facturing error etc.
Substituting for v given by 2.23 in the compatibility equation 2.21
we obtain
b 1 t[rs + H] =o 2.24
Now substitute for S from 2.20 we obtain
bl t[fboR + f'b 1 X + H] = 0 2~25
Expanding and rearranging equation 2o25 we obtain
x =-
(b• tfb 1 )-l[bl tfboR + b' tu] 2.26
Substituting for X in equation 2.20 we obtain
s
boR - b 1 (bl tfb 1 )-lbt tfboR -· b• (bt tfb1 )-lbt tH
=
= [bo -b 1 (b1 tfb 1 )-lb1 tfbo]R - b1 (b 1 tfb1 )- 1bt tH 2.27
s bR - bt (b1 tfb 1 )-lb1 tH
where b == [bo - bt (b,tfb4 )-1bl vfbo] 2.28
The displacement of the structure ca~ now be obtained from
r ==bot[fS + H]
~ [bot.fbo - botfb, (b 1 tfb~ )-lb 1 tfbo]R 2.29
[bot - botfb 1 (b 1 tfb 1 )-lb 1 t]n
+"
= FR + btH
where F = botfbo - botfb 1 (bf 'iifb 1 )-l,b1 tfbo 2.30
F is ·the flexibility matrix of the structure.
2,.,4.2 The General Theory of t~e Displacement Method
A similar argument to the one used in section 2.42 shows that

in general, for a set of displacements r, kinematic consideration£
41
alone are not sufficient to determine the strains in the members of

a structure. Eu.t we can always find a strain system v which is kine-
matically equivalent to r. If the r's ar·e small, then we can find
linear relations of the form
v = ar
Since the strain in the member is now uniquely defined, the stress
follows from the stress-strain relation given in the general form
S = krJV + JN 2. 31
where k:N is the stiffness matrix and JN is the initial stress conju-
gate to the initial strain H given before. If now we apply the
virtual displacement theorem we obtain
stv = Rtr
Substituting for v from 2.30 in 2.31 we obtain
star = Rtr
and since r is arbitrary
R = ats 2.34
which is the equilibrium relationship established from pure kinematic
argument. If now we substitute for S in 2.34 we obtain
R = at[kNv + JN]
= atkNar + atJN 2.35
= Kr + atJn
where K = atkNa
K is the stiffness matrix for the structure.
_&.j. Finite Element Analysis in th~ Displacement Method
In theory both the flexibility method and the stiffness method

can be used to analyse any given structure. But for automatic com-
putation, the stiffness matrix approach has proved so far to be more
convenient than the flexibility approach. The short comings of the
flexibility approach seems to be the difficulty of teaching the
42
machine to select a suitable redundant system which will result in a

well condition flexibility matrix. Considerable amount of research
is going on at present into the topological aspect with a view to
overcoming this difficulty. Before going into the detailed conside-
rati~ of the question of idealisation and the derivation of the
individual element stiffness matrix it is convenient here to present

in some details the sequence of matrix operations involved in the
finite element analysis using the displacement theory.
For any structural member we seek the relationship between the

forces (which could be static, dynamic or thermal) and the deforma-
tions of the member. In general a structural member may ha~e 'n'
assigned nodes or joints and both forces and displacements associa-
ted with these nodeso No loss of generality is involved if we
assume that to each joint or node there is associated only one force
and only one displacement. It is then always possible to establish
a relationship between the system of forces acting at these joints
and the corresponding displacements in the form
St= k11 v• + k12v.i_ •••••••••• k\jVj ••••••• k 1nvn
k2nvn
2.37
where Si is the force acting at joint i

vi is the displacement in the direction of the force Ri
k:i.j are the stiffness influence c::>efficients.
In matrix form this system of linear equations takes the form
.43
k1j • • • • kln
kzj • • • • k~
= 2.na
k .. k.
JJ Jn
knj • • • •
or more consicely as:
S = [k] r 2.37c
An observation of equation 2.37 shows that the geometrical configu-
ration, the stress condition and the elastic properties should all
be reflected in the stiffness matrix. Also if the material is
homogeneous, isotropic and linearly elastic, then by Maxwell reci-
procal theorem, the [k] matTix is symmetric about the leading
diagonal and positive definite. To conform with the finite element
definition, the numerical value of n must be kept finite. Figure
2.4 shows two types of triangular elements with their boundary
stress patterns, namely a triangular plate element subject to plane
stress, and a triangu.lar plate element subject to flexural stresses.
The system of joint forces in fig 2.4b will be assumed to represent
the resultant effect of distributed stress patterns shown in fig 2.4a.
By utilising suitable theories and techniques, and considerir.g

the various factors, such as the compatibility of the element with
each other~ it is possible to derive the stiffness relation in
equation 2o37. Several stiffness matrices have been derived for
plate elements in bending and we shall return to these later on.
44
A given structure will be idealised as composed of a finite

number of finite sixed elements. For the gth element of the struc-
ture we obtain a relationship of the forill
Sg = kgVg g = 1, 2.38
2 .•..•. N
where equation 2.38 is the same as 2.37c with the subscript g
attachedo 6, K, v are as def'ined before but for the gth element"
We now define the following vectors:
S = {~ §z • • • • Sj . • •
v~
V == fV, Vj
The relationship betwee:i. the vectors S and V is given by
s = [KJ v
where K is a super diagonal. mat.rix i,hose diagcnal elements are
composed of the stiffness matrices of the various elements arranged
in the same sequence as the Sg and vg occurs in the S, V vectors.
Its form is given by
I K, 0
•
2 .. 41
•
0
I!
I
'·
If the matrix multiplication of equation 2.40 is carried out we
45
obtain N equations of the form of 2.38 showing that the [K] matrix
is therunassembled' stiffness matrix for the structure. It is per-
tinent here to remark that the kg's need not be of the same size,
this is equivalent to formally expressing the fact that a complex
structure can be imagined to be composed of elements of different
geometry and elastic properties.
When the element is a part of a complete s·tructure its deforma-

tions are in various ways more constrained than when its free. The
disturtion of the element can be related to the actual displacement
of the joints8 This is accomplished by fixing all the nodes in the
structure and displacing one node at a time. Only members adjacent
to the displaced joints will be distorted and the components of this
distortion may thus be determined by melffi. inspection. Carrying out
this procedure at all the joints of the structure that will be
displaced under the action of the external loads and assembling the
results in one matrix equation will yield the following relationship
V ~ [a] r 2e42
where r is given by the relation
and r is the displacement cf joint i of the structure.
[a] is a transforn1a·aon matrix c.omposed of constants relating

the deformation of the elements \v} to the structure displacement
\r}. It is a linear transformation which implies that our displace-
ments must be small and the principle of superposition of compatible
deformation atatoo applies.
'The principle of virtual work can now be applied to obtain the

relationship between the load and deformation of the structure. This
follows from the fact that we know the strains in our structure
expressed by the nodal deformation matrix (V) and these strains are
compatible with the structure• s displacement expressed by the ( r)
vector. The resulting expression is
VtS === rtR 2.44
But 'f = ar and S = KV
Substituting for S and v in 2.44
rtR = rta"tl(a.r
Since this is an identity, we can equate the coefficients of rt on
each side of the equation to obtain
R = atK ar 2.47
== Kr
where K = atKa is the stiffness matrix of the sti:ucture
R is the loading of the structure and its form in consonance
with r defined by 2.43 is given by
R = { R 1 R2. R·J RN}
• ' • 0 2.48
'
where Ri is the loading on the node i.,
Equation 2 .. 47 represents a set of simultaneoue equations in N
unknowns and must be solved to obtain the displacements. In its
present form it may be singular, if we have included all the degrees
of freedom including the rigid body freedoms and mechanisms which
may have resulted from our idealisation. The rigid body freedoms
and mechanisms must be eliminated. If this is done, the K matrix
then becomes positive definite and possesses a unique inverse.
2a.5:' Direct Assem~!.Y of the K Matrix
It is extremely advantageous to include right from the begin-

ning all possible degrees of freedom in the derivation of the
stiffness matrix. The matrix [a] will then be seen to contain elements
which are zero, one or minus one. Because of the similarity of form
to the Boolean algebra, this matrix has been named the Boolean
47
Matrix [2 6]. The boolean mat:rix operations are simple and will be
illustrated under the section on computation.
Let the matrix [a] be partitioned in such a way that the trans-
formation atka is possible, thus (a] takes the form
a= \a1 a~.,. ag···· aN} 2.49
where N represents the total number of element. The stiffness matrix
is thus seen to assume the form
K = 'J-agtkgag 2.50
g==l
the summation being taken over all the elements.
However a more obvious procedure of assembling the stiffness

matrix can be obtained as follows. For an individual element, the
generalised nodal force incrementl\S required to maintain a set
of nodal displacement increment t~r is given by
Resultant nodal forces increments acting on the complete structure are

Ci R = '.t
r::._s g = Ke r 2. 52
g=l
where K the stiffness matrix of the complete structure is given by
the summation
K =>ke ·'~.
2.53
which provides an alternative basis for the matrix assembly, At
each joint, the components of the stiffness matrices partaining to
the elements surrounding the nodes are summed directly to obtain
the components of the overall stiffness matrix. Since any particu-
lar joint is only at·tached to a few elements surrounding it, this
means that the summation is performed over a few elements at a time.
As indicated earlier on, it is extremely advantageous for logical
nnd dynamic programming to include ab initio, all the degrees of
freedom including the rigid body freedom, that may be associated
with the element and structure. The final stiffness matrix is
then singular and possesses no inverse since the determinant is zero.
Both force boundary conditions and displacement boundary conditions
can be imposed. For the displacement type boundary conditions the
columns and rows pertaining to the specified boundary displacement
is colapsed with consequent shrinkage of the stiffness matrixo
A more systematic approach however introduces the idea of

partitioning. This is especially useful fo~ solving mixed boundary
value problems. Forces corresponding to a selected set of displace-
ments are assigned and the remaining displacements are required to
assume prescribed values. It is then required to determine the
numerical values of the displacements correspow!'ing to the prescribed
forces and vice versa. Elements of the displacement and force vectors
are rearranged as necessars and partioned thus
r {r, ro\
=
and R = J\R1 Ro J'
' 2.54
The subvectors R and r 0 are given and it is required to determine
Ro and r 1 • In conformity with equation 2.54 the stiffness matrix
K is rearranged and partitioned thus
K = rKII
LKot
KK.,0¢]
~ ~
2~55
Then the force-deflection equations may be rearranged in the form

of two matrix equations thus
R K r+K.r,.
· 1 1 1 ·1, · lo·.d-?
.~o - Ko1r1+ Xoo11r.-J
49
from which follows by direct elimination and substitution

1
r, = [K H r r I - [K lj
1
[K nJro r
and Ro = [Ko1][Ki1r1R\ + [Kuc-Ko,K-~, K10Jro 2.57
We also note the idea of the linear constraint matrix defined by
[1ii]r = 0
where the matrix M is essentially a rectangular matrix which the
programmer specifies. The number m' of constraints assumed inde-
pendent is equal to the number of rows of [M]. Its formulation is
extremely simple since the elements are either unity or zero. The
cons~raint conditions are to 'be used in eliminating a sat· of n''
displacements specified by the analyst. As before we rearrange
and partition the matrix M in conform:.ty with the first of equation
2~54, the elements of r 0 being selected for elinination. Thus
we have
'r i
[M1 : Mo] 1.- 1• ( = 0 2.59
r0l
or !,~ 1 r 1 + M0 r 0 = 0 2. 59a
Premultiplying the equation by :M0 -l we obtain
r0 == - M0 -~.1: 1 r, 2.60
and substituting for r 0 in the first of equation 2.56, the force-
displacement relations for the reduced system take the f orn
R = ( Ku - Kie) M0
-1M ~ ) r 2.61
1
Equations 2.55 to 2.61 asgume that the entire stiffness matrix is
already generated prior to partioning. A more systematic approach
can however be obtained if the partitioning is carried out directly
on the [a] transformation matrix. The elements of the transformation
matrix are rearranged and partioned in conforoity with equation 2.54
thus
v 2.62
"'· .
50
But K == atka
=i~~t~J k[a1 a0 ]
ao':l
=t· a ~ka; a,!ka 0 'j 2063
a. 0 ka 1 a 0 ka. 0 ,
==1" Ku Kio l
._Ko, Koo .I
The procedure is thus seen to lead to a gradual build up of the various
components of the K.matrix in their required form to enable the ana-
lysis to proceed as indicated by equations "2.57 and 2.61.
J. 70 Calculation of Stresses
The procedure outlir-ed above enables us to determine the displa-

cements by the solution of equation 2.47 thus
r = K- 1R 2.64
it being assumed that the boundary conditions etc have been imposed
to enable the K matrix to be inverted. Substituting for r in
1
equation 2.42 we obtain V9 the element strains and similarly sub-
stituting for Vin equation 2.40 we obtain the stress resultants or
the equivalent nodal force. The result of the successive back
substitution is
S = kar = k8.K- 1R 2.65
The theory thus far developed does not enable the individual ele-
ment stresses to be obtained, this is embodied in the derivation of
the stiffness matrix for the particular element. We merely note
here that if 6 denotes the element stress a relationship exists which
is of the fora
6 = QS 2.66
where Q is the stress matrix, whose explicit form will be derived
under the section on element stiffness matrix.
51
THEORY II
l.!_J;l Structu1:al Idealisation
The formulation of any structural problem often involves a

number of assumptions and idealisations. These assumptions may be
in terms of stresses such as the idealisation of the stress-strain
relations in theplastic stage of e..nalydis, or of beams capable of
carrying bending stresses only~-;tile shear stresses and shear
deformations are neglected. Often the idealisation involves only
the geometry. Thus a tappered beam could be approximated by a
series of uniform segmented or stepped beams, or a curved beam by
segmented straight beams. Experience often dictates what assump-
tions or idealisations are to be made in any particular problem and
there are no hard and fast rules to be observed. The objective on
the whole is to produce a mathematical model which is capable of
reproducing to a great extent the behaviour of the actual structure.
The quality of the analysis is to a large extent dependent on the
extent to which the mathematical model simulates the behaviour of
the actual structure.
A parallel idea is that of approximation. This manifests itself

in several ways and often cannot be distinguished from idealisation.
Thus in shell analysis, by neglecting the bending stresses, which
are regarded as secondary, one obtains an approximate shell theory
often called the membrane theory. Often approximations are made in
the final stages of analysis as for instance in the Shorer
52
type of approximation, to obtain a simplified solution. Solution

of the exact equations of the actual physical system by approximate
mathematical expressions such as the finite differences is still
another form of approximation.
In the finite element method of analysis, we are concerned

with idealisations which, involve physical approximation to the
geometry such as the idealisation of the elastic continua as
assemblages of finite continua or finite elements~ The central idea
is 'cutting' of the structure rather than substituting for the
structure an entirely different structure as for instance in the
beam or lattice analogy for plate problem. Whereas in the methods
of the substitute structure the elastic properties could to a
large extent be modified, no such approximations are called for in
the finite element method of analysis.
More severe approximations can however be made in the finite

element method of analysis. Thus we may replace a curved surface
by an assembly of flat elements. Such an approximation has a dual
advantage. Firstly it makes possible a general formulation of a
class of problems with similar geometry and stress pattern such as
plates and shells of various shapes and curvature. Secondly, it
ensures that one set of stiffness matrix shall be used, thus
obviating the necessity of deriving as many stiffness matrices as
there are elements. We merely add that the inverse idea is also
admissible, such as representing a flat surface by segments of
curved surfaces. Both ideas have been exhibited thus far in the
analysis of folded plates and cylindrical shells,
In accordance with the definition of a structure in the finite

element context it is first necessary to break up the structure into
53
suitable segments. The general shape of a two dimensional shell

type of structure is shown in fig. 3.la and fig. 3.lb shows its
possible idealisation. In general the element is a polygon having
N sides, but the numeric value of N must be kept finite both to
simplify the procedure by which the stiffness matrix or flexibility
matrix 0£ the element could be derived and to con£orm with the
finite element definition. Observation also shows that we are not
restricted to any particular geometry within the finite element
context. Thus we may use curved elements to follow the contours
of a curved structure and flat elements to follow the contours of
a flat structure.
3.1~2 Suitable Elements for Ideal~si~_j)hell Surfaces
For chronological purpose shells are divided in this thesis

into three categories namely:
1. Axially symmetric shells of revolution
2. Singly curved shells
3. Doubly curved nhells
While we are not concerned in this thesis with the analysis of
shell type 1, it is interesting to review a type of idealisation
which has been employed ~y Lu et al (l9 6 3) The idealisation
followed in that reference was that of the assemblies of conical
frusta and a spherical cap as shown in fig. 3.2. It is also inte-
resting to note that the same type of structure was analysed by
Adini (l9 6l) using the triangular elements. He took advantage of
the symmetry of the entire structure, and analysed an eighth
segment of the structure.
By far the most imp~rtant·elements for idealising shell types

0
54
2 and 3 are the rectangular elements and the triangular element

v7hen the shell has a single curvature as in cylindrical shells the
obvious choice is the rectangular element. With class 3 shells a
distinction has to be made based on the Gaussian curvature. Thus
for shells of positive Gaussian curvature like elliptic paraboloids,
rectangular elements could be used, for negatively curved shells
like hyperbolic paraboloid the choice is limited to triangular
elements because of the necessity of following as far as possible
the contours oi' the shell.
3.. UElemen_t Geometry
In de-riving the shell element stiffness we are faced with two

distinct possibilities. We may form an idealised model in which
the middle surface of the element coincides with the middle surface
of the structure and the element is thus curved. The resulting
element stiffness matrix is thus fully populated and both the
extensional and flexural actions are coupled. The stiffness matrix
in partitioned form may thus be represented by
ks =
( kr
I" kct
kc
k
d
l 3.1
where kd is the membrane or extentional stiffness
kr is the flexural or bending stiffness
kc is a coupling submatrix.
1I'he size of ks will depend generally on the number of nodal

points which has been used to define the geometry of the element.
Considering a rectangular curved element having four nodal points
the order of ks is 24 x 24. Moreover the matrix being defined
with respect to a particular geometry becomes unsuitable for any
other geometry and the versatility of the element is thus impaired.
55
The second possibility is the use of elements which do not

coincide with the geometi.'Y. of the shell except at the nodal point.
Thus the curved sides of the elernen·~ are replaced by straight lines.
The elarnent stiffness is thus independent of geometry within the
element and is definable solely by the nodal geometry.
The stiffness matri}: is of the form
0 ....
kr;
I
The absence of the coupling submatrix is apparent. Although the
order of the matrix is not reduced, the separate submatrices have
been reduced from three components to two and what is more from the
computational point of view can be ,senerated and stored independently
as
k8 == f kd kr j 3• 3
with considerable saving in computer storage. Also a general
purpose stiffness matrix results. This type of element has been
adopted for analysing the hyperbolic paraboloid shell in this thesis.
3. 2· General '11heory of...p_~J.~~2.n 0f Shell Element Stiffness
As pointed out earlier~ the stiffness matrix for a shell element

whose middle surface is flat can be bui:::.t up from two independent
stiffness matrices cor1:espon.ding to the bending stiffness for the
element and the in-plane stiffness for the element. Below the
general theory is presented.
3.2.1 Natural and Cartesian Definition of Displacement and Forces;

Elastic Pro_,Ee_J.'ti~~
Two sets of displacements are required denoted here as natura.l

and cartesian systems. 'N.atural' and 'cal!'·tesian' may be regarded
56
in what follows as being used in a rather loose sense in some

sections to indicate <1ocal' and 'global~ system of coordinates.
The more r.:!.gid use will oe explained later.
We start with the complete cartesian force and displacement

vectors which are arranged in the sequence
P ={Pa Pr} P = f Pd Jlr'f
(18xl) (18 xl)
The suff::.ces d and r stand for displacements (forces) and rotations
or slopes (Moments). The component vectors Pd and Pr are next
defined as
Pd ={P1 P.J. ~?:\ P,- = {52.., Q2. 52~ ~
(9 xl) (9 xl)
in which (see fig. 3.3)
P·l. ={Ui Sli = {Mxi
Vi Wi} Myi MziJ 3.6
Similarly for the displacement vector
Pd Pr
in which
Pi 1:1-i Vi w·J. lI
= { ilr. t
, J.. J 3.8
(3 xl)
Next natural forces and displacements are defined. The natural
force and displacement vectors are arranged in the sequence
PN = {PNM P1iJ:B i
PN = tP}m PNBi 3.9
where the subscript Nha.s been used to indicate the natural system.
The order of the matrix vectors PN and PN is influenced by our
definition of what we called the natural system. If we merely
ref erred to the natural system as mere transformation to a conve-
nient local coordinate system the P, PN,P,PN are of order 18 x 1,
assuming we include all the degrees of freedom in PN and PN• If
on the other hand, we neglect the rigid body freedom as explained
later and identify our Pi-..i and pN with straining forces and straining
57
modes of displacement, the orders of PN and PH are reduced to (9 x 1).

Thii~made up of
PNM ::: l
:& p ~ p y J i P1m = ~ TS TA ~ 3 • 10
in which Pp Py denote any three independent forces system which
P~
could be self equilibriating force system directed in some a, ~' y

coordinate system whioh in ref. (27) has been taken as the direction
of the sides. Ts and TA are given by
Ts = tTsa Ts~ Tsyi:1TA = {TAcx TA~ TAyt 3oll
where Tsi and r::-Ai are conveniently identified with the symmetric
and antisymmetric nodal moments acting at the vertices of the side
li, their vectorial direction being normal to the direction of li.
Ilifferentiation into s;yrilllletr~.c and a:r.tisymr.ietric components is not
essential to the theory.
Correspondingly, the natural displacement vectors P.mir and

P:r:m are a.escribed by
= {n~ P f3 Py} PUB =lPs PA~ 3.12
where = { <Psa <Ps~ <iisy} PA = {cPAo: cPxp <PAyf 3.13
Suffices M and N denote Memb1:ane and bending respectively.
It is pertinent here to comment on the order of PN and PN·

Ideally it should be (18 - 6) 12 x 1 matrix and not a (9 x 1)
matrix as given above. The reason for the reduced dimension is
because we ignore the inplane rotational stiffness at the nodal
point. As a result when investigating a flat portion of a shell
we have to ignore a rotational degree of freedom preferably that
associated with the axis nearest to the perpendicular of the plane
in question.
We also require in what follows, the relation between the

total natural strain and the natural component stress as defined by
58
the relation
-1
C.'N "" EN O'N 3.14
where EN is the 3 x 3 natural moduJus of elasticity and may include
arbitrary anisotropyo EN is of the
r K11 K12 K131
Eii =E I K21 K22 K23
l_K31 K32 K33_i
where E is the cartesian modulus.
3. 2, 2 Natural and Cartesian Stiffnesses
If the PN and ~a;..oe suitably defined then it is possible to

write doYm by mere inspection the relationship between eN and PN as
eN = aN PN 3.16
where a is a square matrix expressing the linear relationship
between eN and PN·· Further more we ca.~ write
eN = [em,1 eN13] 3.17
PN = [ PNM PNB ]
and since no coupling exists aN is of the form
aN ::: (aNM 0 ' 3.18
I
!.. 0 aNBJ
3.19
3.20
where
= f V fam;ft 0 -J EN f aml[ 0 1dV
l0 aNB L0 a:r._i:MI
= J Fm11t Ew amA: ; a:r:rn t EN aNBj dV 3.21
= rKM K13J
59
Relation 3~21 confirms that the stiffness matrix of the shell

segment can be built up of the membran8 and binding matrices
separately generated~
To obtain the cartesian stiffness we require a transformation

of the form
t
PN = CNP and P = aN PN 3.22
we now apply the congruent transformation
t
ks = CN kNs CN 3.23
to obtain the cartesian stiffness for the shell element. The
subscript S has been used to denote shell. If l~ has been described
in a local system which itself is ca.L.'tesian, en can be identified
with simple rotation of axes and CN is given bytl:J?. dic,go~l i.ie,trix
where
c cl
-
c
(ccx.x
=\cfjx
Cay
c {:ly
Ca·
c 13z
1
Cyy ~Cyx
Cyz
. /
I
and Cij is the direction cosine relating axis i to j.
3.3 Displace~ent Functions
The theory presented in the last section shows that in order

to establish the stiffness relationship 7 the strain in the element
must be known or determinable in one form or anothero It must also
be expressible by equation 3.16. For linear elements such as beams,
existing theories using the engineers theory of bending, slope
deflection equations etc enable us to do this. But with two
dimensional a.~d three dimensional elements these theories are generally
inadequate and we have to consider usi..~g the various energy theorems,
such as Ritz principles, virtual work etc. If for instance a
60
displacement field is prescribed in a continu'Ul!l by several unknown

parameters then the theorem of least total potential energy can be
used. The combination of the unknown parameters yielding a mini-
mum of the energy represents the best possible approximation.
Convergence of the result to the correct solution will be assured
if by increasing infinitely the number of parameters the tr~e
distribution of displacement could be achieved.
In the usual approach based on Ritz method expansion of the

displacement field in polynomial or Fourier series over the whole
field is envisaged. Thus for a three dimensional continuum we have
u = u 0 (x,y, z) + O'.ur ur(x, y, z)
v = v 0 (x, y, z) + O:vr vr(x, y, z) 3.26
w = w0 (x, y, z) + CXwr wr(x, y, z)
where
UQ, v 0 , w0 satisfy the kinematic conditions where prescribed
Un., vr, wr are linearly independent functions which
vanish at prescribed points
O:ur' O:vr, o:wr are unknown constants to be determined such that the
total potential energy of the system is a minimum.
In the finite element approach, the parameters (a) of the

displacement field are identified with the nodal displacement vector
which is in the literatµre referred to as the generalised displace-
ment. The nodal displacements prescribe the displacement field
within the element uniquely.
According to Vebuek .< 196 5) generalised displacements p are

defined according to the following rules:
(a) Connection of the elements with one another across a.
common boundary is to be expressed by equating two by two the
61
generalised displacement pertaining to this boundary.

(b) The whole displacement field along this boundary must be
determined uniquely by "!;he generalised displacements pertaining to
the boundary and conversely. In this manner, equating the genera-
lised displacement secures the continuity of displacement along ·
the boundary.
(c) The number of parameters (a) must be equal or superior to
the number of required generalised displacements. If superior~
convenient aduitional generalised displacements can be prescribed
so that the matrices a and p can be connected by a non singular
linear transformation T = (Tij) such that
-1
o:: = Tp and p = T a 3. 27
.2.!J~l Requirements of the D~splacement Function
In view of the prominence given to the conditions which the

displacement function must satisfy in recent pulications it is
proposed to review them here and comment on them.
(a) The displacement function must be continuous over the
element (34) and at no point of the structure should the strain be
infinite otherwise the field is no longer integrable ( 37 ). This
is the basic requirements of the potential energy formulation and
does not warrant further comment.
(b) It must be possible to represent the rigid body motions
of an element otherwise the equilibrium condition of the element
as a whole are falsified (3 4 ). As shown in the theory in section 3.?.2
the rigid body displacements are not essential to the formulation
of the natural stiffness matrix but is implicitly required in the
formulation of the QN matrix.
(c) It must be possible to represent states of constant stress
otherwise as the mesh of elements is finely subdivided, there is no
guarantee that the stresses will converge towards a continuous
function ( 41) This condition is no doubt reminiscent of the membrane
analysis of triangular E.lements. As pointed out in ref ( 2 7) it is
not possible to assume that the bending stresses or moments are
constant over the triangle in view of the necessity of transferring
shear forces, which require varying bending moments.
(d) Where neighbouring element abut between nodes there must
be no discontinuity of slope and displacement between adjacent
4
elements ( ~ While this condition is implicit in the theory it
does not appear to be a necessary condition and appears very
difficult to satisfy with elements in bending. Indeed the whole
basis of analysis of shells using e1~ments of planar middle surface
would be invalidated if this condition were adhered to. This re-
quirement appea:Vsto be the basis of partitioning of elements displace-
ment functions into the so called 'conforming' and 'non-conforming'
functions ~l ). The conforming functions are assumed to guarantee
the continuity of displacemento along the boundaries and continuity
of normal slope across the boundaries while non-conforming functions
are assumed to guarantee only the continuity of displacements, with
the continuity of normal s~ope sacrificed.
3~3.2 Displacement Fu.notion Previously Derived
J.., Non-Conforming l!.,_u.nctionso Bending Elements
Earlier displacement functions have essentially been in the

non-conforming category~ The various functions have been based
on the various interpolation formulae expressed by the scalar
equation
w(x, y) = [l x ·y x 2 x:y y 2 x?.y xy 2 x 3 ••• ]ex 3.28
or w = X t ex 3.28a
63
where the a 9 s are to be determined in terms of the generalised

displacements. For elements :i.n bending the generalised displacements
at the nodes are taken aJ the normal displacement and the two carte~
sian slopes e and p. Some differentiation of the function is involved

in order to obtain the slopes -@- and ¢
and generally these functions
are unsuitable for slope definition. For a rectangular element in
bending, Melosh (34) has improved this situation by using one of
the osculating formulae in which the amplitude (a) of the displace-
ment function w is expressed not only in terms of the nodal displace-
ments but also interms of the first derivative which is easily identi-
fied with e and ¢ a.~d so no further errors are introduced.
Furthermore in order to express the a uniquely in terms of

the p as given by equation 3o27, the polynomials have been truncated
quite arbitrarily and generally there has been loss of symmetry in
the polynomial expressed by the loss of the higher terms in the
general expansion of equation 3.28. Often by trial and error
certain terms in the polynomials have been arbitrarily coupled as
in the best formulation achieved thus far (43) for triangular
element given by
w = [l x y x 2 y 2
2. Conforming DisElacement...J'unction
The task of deriving suitable displacement functions which

1
satisfy conforming' definition has been difficult. These often
involve specification of additional nodes at the middle of the side
of the element boundary. At th8se points only is the continuity
of normal slope satisfied. In reference Gil ) a conforming displace-
ment function for a triangular element is claimed to h<l.ve been obtained.
But we note that in the same reference~ that the displacement function
while ensuring convergence 7 the approximation was not always as good
as that achivable with the simpler function discussed in the previous
eection, An example 0£ the application 0£ the various £unction~(4l)
yielded a maximum of 3 % for the non-conforming function and errors
of up to 30°/0 for the conforming function. This seems to show
that the requirements of conformity seems to impose ~:nnessary addi-
tional constraint on the element which the simper functions do not,
and in terms of comparison of the result with the exact solution the
simper functions are vastly superior. Moreover the conforming
solutions erred on the upper bound side of the solution while the
non-conforming functions erred on the lower bound side.
Both types of function have been used in the analysis in this

thesis and compared with a new function derived in this thesis.
hl,.3 Natural Modes of Deformation
After the initial trial and error efforts at deriving suitable

deflecting pattern for elements in general, research effort has
been directed towards finding a consistent and logical basis for
deriving suitable deformation patterns. Notable in this field is
the Argyris school. Their most important contribution is the
introduction of what they have appropriately termed the natural
modes of deformation.
An examination of equation 3.19 shows that as far as the

energy of the system is concerned we should concern ourself only
with those modes of deformation which contribute to the strain
energy of the element. Thus we could neglect thoBe d1splacement modes
which are associated with the rigid body displacementso But we
must have in accordance wjth requirements in section 3.31 a means
of prescr~bing the rigid body modes since this is implicit in the
assembly of the equilibrium equation of the entire structure as
given by equation 3~23. The importance of neglecting the rigid body
mode and deriving the stiffness matrix via the natural stiffness
matrix does not seem to have been generally recognised.
To fix ideas, let us consider a triangular element under the

action of inplane stress and inplane deformation. Clearly the
degreesof freedom of the entire elementare six, comprising the two
cartesian displacements of each of the nodea Thus the p vector
in equation 3.23is given by
p ={ux uy Uz 7.x vy Vz} 3.30
But included in the p vectors are three rigid body modes. Thus
the deformation or straining modes are three in number. Suppose
however that instead of considering the degrees of freedom we
have considered the force system P which correspond to these defor-
mations. By corresponding forces we mean forces which do work on
the displacement such that the relationship
W = Ptp
is satisfied. The six force system which correspond to the dis-
placement system may be represented thus
P ={Ux Uy Uz Vx Vy Vz~ 3.31
Now considering the equilibrium condition, included in the P vector
are three overall equilibrium conditions which consist of the
equilibrium of the forces in the directions of x, and y and rota-
tional equilibrium of the element about the normal. Therefore the
statical indeterminacy of the triangular element is three which is
also the kinematic indetermina0y. The entire nodal force system is
66
describable in terms of any three thus
p :;: If'pPN J = !,. 8Nl PN

a
l. 0 ... 0
where PN are the redundant forces and P0 has been used to describe
three overall equilibrium forces mentioned above. Since the Pn
and P0 system are to be only statically equivalent to the P system,
then we are free to choose the PN and P system as we please. And
0
in particular we may choose the PN system as self-equilibriating
system, and then the P0 system is wholly statically equivalent to
the external system. With this background in mind we may choose
the PN system in a manner most natural to the geometry of the
(2 61/
element. In ref the PN system has been chosen as the elonga-
tionsof the three sides of the triangle and the PN system as flange
loads in the tensile sense acting along the sides of the triangle
as shown in figure 3. 4 •
One important conclusion is that these modes enable us to deter-

mine the stiffness relationship in a manner most natural to the
element and any impositions about continuity etc if they exist,
manifest themselves in a very obvious manner. Exactly the same
procedure has been extended to problems of rectangular plate
element in bending and under planar forces and startingly accurate
results have been achieved by tbiis means.
Deriving a triangular element stiffness in bending based on

the above consideration does not appear to have received any
attention, and therefore attention has been focussed in that direc-
tion in this thesis.
67
3.4 Derivation of Stiffness Matrix for Jriangular Element in

~~~s:
2·4•1 Element Coordinate SY.stems
As pointed out earlier, the stiffness matrix £or a shell

element is derived in terms of what are referred to as a cartesian
system and a natural system. Also pointed out is the crude sense
in which the uatural system has been used to define a local system of
axis which could in fact be cartesian in itsel~ Fig. 3.5 shows the
various coordinate systems which can be used to describe the geometry
of the element. The local coordinate system can be seen in fig.
3.5(b) as being a translation and rotation of the cartesian axis. One
axis of the latter is made to coincide with one of the sides of the
element. The natural system shown in fig. 3.5(c) also referred to
as the area coordinate is here denoted by a triangular system of
axes. The coordinate of any point within the triangle is specified
by the magnitude of the three perpendiculars drawn from the point
to the sides of the triangle.
Consider an arbitrary iriangle i, j, k, arbitrarily orientated

in a cartesian coordil1ate system fig. 3.6. Then an arbitrary point
(o) whose coordinates are x, and y can be equally described by a
coordinate system which we may refer to as !:i 'f;j ·5k, where hn~n is
the perpendicular distance from point(o)on to the side(ln)of the
triangle facing the node(n)and(hn)is the perpendicular distance of
the nodeln)from the side(ln}•
By sur:mation of the areas of the component triangles oij,

ojk and oki, it is possible to present the relations between
68
3.33
where l1 has been used to represent the area.
Thus ~k = ~oi,j
6ijk
Similar expressions could be written for '5i, $j•

But 6 ijk = 6. oij + .6ojk + 6 oik = (~i + ;j + ~ k)Aijk 3. 35
therefore
This relationship can be used to represent any of the $'s in terms

of the other two.
To enable us to trans.form from one system to another we need

the relationship between the s's and x and y.
Now 6. ijk
I 1
X·l Yi
I~
1/2 Xj Yj
Xk Yk 3~37
1 x y
6. ojk = 1/2 1 x·J Yj
11 Xk Yk
where the vertical bars have been used to indicate the determinantal
evaluation of the terms within the brackets. Expanding equation
3. 37 in terms of the first row and denoting D.ijk by the scalar 0
then Si
=
6. ojk
6. ijk = <lxXkj Yjl
Ykl -x 1
11 Yjl +y
Yk
where ai and '3i have been used to denote the coefficient of x and
y, and Yi is the constant in expression 3.38 above.
Similarly 3.39a
and
3.39b
With respect to the geometry of the fig. 506(0) is a point nidway

between m and n, which we shall take as the moveable origin and it
is a function of Si· The coordinate of a point s moving between
m and n is derived as follows:
Since s moves between m and n, one of its coordinates is

specified by hiSi· If sl is the perpendicular distance to lj, then
sl = hjSj is the other coordinate. From the similarity uf triangles
snl and kjp we can write the relation
kp/ jk = sl/sn
therefore sn = sl 1 isj
But os = on - sn = z = li(l -~i)/2 - li~j

3.42
= 1i(1 -si - 2sj)/2
Thus z = l·(l
1
_-e.
~1
- 2f.)/2
...>J 3.. 43
This can be rendered non-dimentional by dividing through by li•

Similar relationship can be obtained by cyclic substitution of
i, j and k.
70
The sarnP. argument used before in section 3.~3 leads to the

fe.ct that for describing the natural straining modes for a triangular
element in bending with three nodal points we required only six
generalised displacements and three rigid body modes. Thus !JN is
of the order (6 x 1). Here we take as the generalised displacements 9
the generalised rotational vectors at the nodes whose directions are
normal to the sides of the triangle and in the plane of the triangle.
Thus the deflected surfaces are to be represented by polynomials
whose amplitudes are t~e rotation of the corners in a direction
parallel to the sides.
If two sides of the triangle are fixed with respect to the

vertical displacement only and the third is free to deflect, then
the deflected form of the surface can be derived as follows: The
displacement of the free edge in expanded into a power series
function as follows~
w (~·.:O] =a + bz + cz 2 + dz 3 + .,. 3.44
where z is the cordinate of a point moving along the edge. Since

however we require a third degree polynomial, as indicated by four
boundaTy conditions on the displacement curve, namely two displace-
ments and two rotations of the ends, we terminate the series as
shown above. The series can be separated into the symmetric
function represented by the even powers of z thus
-2
vr(s} = a + CZ 3.44a
and an anti-symmetric function represented by the odd power

function
'{l
w(a) bz + dz-3 3o44b
Thus approximately we can represent the deflection of the edge and

sections parallel to the edges by a symmetric parabola and an anti-
symmetric function. For the general section, the bar above z is now
dropped. The form of the deflection surfaces represented by the
symmetric and the antisymmetric functions, is shown in fig. 3.7.
Again the slope is expanded as power series of Si (say). Thus
we again terminate this series in the second power basingour argu-

ment on the fact that the slope is the first differential of the
displacement and must necessarily lag by one degree behind the
displacement.
¢ = a 0 will be rejected as an admissible solution from

purely physical argument since it implies that the slope at the
corner i is finite and has the same magnitude as the maximum
slope at the foot of the triangle. ¢ = a 0 can indeed be identified
with the rigid body displacement.
l.:..4•.2. __§X!PJI!etric Functions
The constants a and c in expression 3.34a can be found from

the boundary conditions of the curve thus:
at z = :I: 1/2, w = o, dw/dz = -<J

2
From these conditions c - ¢/1 and a= - cl /4 3.47
and w ¢1/4 [l - (2z/1) 2 ]

72
For the symmetric displacement we prescribe the linear variation
i~ equation 3.45. When Si = o, that is, at the edge ¢becomes ~~

and al = ~s. Then ¢ = ~( 1 - Si).
Substituting for ¢9 z/1 9 and 1
w psi1 - ·L.l21ir1 - c1 - si - 2~j} 2 I 3.49

4 i_ ( 1 - ~i ) 2
.J
which after expansion reduces to
w(s) 3.50
3.4.4 An~isymmetric Displacement
For the antisymmetric case we now take the remaining expression

in the polynomial
w ::; bz + dz 3
The constants b, and d are determined as in the symmetric case,

but now we use the parabolic slope distribution as given by
- -e 2
= ~a(l - Si)
The resulting expression simplifies to
3.52
With one edge displaced and the other two supported, we can now
write
73
for the edge facing node i a.isplaced.
Thus 3.53
By mere inspection we can write down the remaining expressions

thus
Wj = lj[~sj ~ k Si + ~aj ·gk -~i(Sk -5i)] 3a53a
wk = ik[~-sk ~ i ·5j . 1ak.Si sjC:~h -sj)J 3.53b
and
3.54
74
3.5 Computational Molecles
In contrast to the rectangular element, it is not possible or

appears rather difficult to present an explicit formulation of the
stiffness matrix for the triangular element in bending. However in
what followe, the main algebraic matrix formulation ia preeented in
a form which it would be assembled in a program for computer analysis.
The natural generalised displacement expressed by fs and ~a are

.i.:
represented by the column matrix PN thus:
1- ~ ~ . x I
l ~Si 2sj ~aj
r -
PN == 'i'.Sk ll!ai 'fak / 3.55
the rigid body displacements are represented by
Po = \ ~oi ioj ~·pk~
we also require the diagonal matrix 1 given by
l~ 3.57
1 == o)
LI,
where L == f li
lkj 1.
J
3. 58
and ln has been defined as the side length of the triangle opposite
to node n of the triangle. If now we denote by X the row vector
of the polynomial expansion in Si~j up to the third power thus
x =mi "Sj ~i~j si2 $j 2 si2 °Sj~i~j 2 ~i 3 Sj 3 ]3°59
then we can write the set of displacements derived in the previous
section using the equation 3. 36 to eliminate-~ k as
where c :iis a rectangillar matrix of coefficients which has as many

columns as there are generalised displacements pN and as many rows
as there are columns in X. It is obtained by expanding each of the
displacement components obtained in the last section in power series
involving only ~ i and Sj and grouping the coefficients of X as
column vectors in correspondence with the appropriate ~ for conformal
matrix multiplication. The c matrix is presented in table ~.l).
75
With respect to a rectangular coordinate system, the strain

component µx is difined by
2 "
12~ I
r
Idx2
II d --~!.
2
3.61
=
. a.y2 I
\2d2w 1·
i-
l dxdy)
Similarly with respect to the natural triangi:.lar coordinate system
defined in the previous section, we also define the functional µ
given by
r
2
,
'
I
I
d
clf
w
..... i·2
!
I
.. I d2 I
µ··
5 = w
d't.. J.2
u.2 w
I 3.62
d$.d:$.
l.. J.. J
By partial differentiat.tm of the displacement function Wll with
respect to x and y making use of the relation (3.49) between x, y
and~ i ~·j $k we obtain
µx = [a] µ0s 3.63
where the matrix a is given by
,- a·2 a·2
I J.. J
a =
I
. 2(o:i/3i) 2(o:j ~j)
'-··
f3i2
If we represent by ~ the rectangular matrix obtained by diff eren-

Pj2 3.64
tiating partially the row matrix X as indicated by µ , then we can
µx = aXc 1 PN 3.65
where the explicit form ofµ is presented in table (3. 2). The stress
matrix rix is def·ined by
76
O'x = ~ IIlx my mxy ~
then D'x = Ilµx = Da:Xcl PN

wh~re the D mat·:dx is the elastic relation defined before. To
obtain the natural stiffness matrix we now use the principle of
virtual vrork, giving
pN tpN = J µ.. t o dv
x t t-t t ..
= J 'f1T le X a: Da:XclpNdv
. . t
and sir..ce PN
where
for a material with an isotropic elasticity, k has been evaluated
and presented in table (~3). In the integration involved we have
made use of the relation
r. m nd,.. 26m! n!
JS$i sj : ; == (;+n:2St
where 1::::,. denotes the area of the triangle and the natural stiffness
matrix is given oy
_2,.5 1 1 .§tiffness Matrix in the C'ai~ian System.
To obtain the cartesi&n stiffne~:s matrix k, we now include the

rigid body displacement expressed by
gi
Poi + SjPoj + $kPok
WR = 3.74
and the total displacement w is now expressed by
w =WR+ WN 3.75
If p now expresses the cartesian vector of generalised displacement
given by
p
77
where w = ~w·
t l. Wj wk[
e == ~e·
\ l. ej Gk{
¢ }t <b.
' J. ..
¢; Cl>k}
and e = - dw/dy
cil = dw/dx I
making a coordinate substitution, we obtain

w [qiooJ Po
= 3~78
and differentiating w with respect to x, and y and again making a
c,oordinate substitution, we again obtain
T0 N TN}T T00 in the present analysis turn• out to be a unit matril{es

of rank equal to 3.
Solving equation 3. 78 we obtain
-1
Po = Too w 3.80
substituting for p in equation 3.79 and rearranging we obtain
-1
TNN PN = /~} - ToN Too w 3.81
'
= r - T oN T oo -l I 'J
l ( 6. x 3)
6 I
J
p 3.8la
=A p 3.. 8lb
The A matrix is presented ~n table (~4)
Now solve for PN we obtain
-1
PN = TNN A p = aNp 3.82
k matrix is now obtained by the congruent transformation
k = ~t 11i 8N 3.83
78
. -1 . . 1 .
-1 . . -1 • .
. 1 . . -3 .
l 1 -1 2 -2 .
[c] == l . . 3 . .
. . . -1 3 1
. . . -3 1 -1
. . . . 2 .
. . . -2 . .
TABLE 3~1 COEFFICIENTS OF POLYNOMIAL EXPANSION OF

THE DISPLACEMENT FUNCTION.
. . 2 . . 252 . 6!1 .
[x] == . . . . 2 . 2~1 . 6"%2
. . . 1 . 2gl 2'§2 . .
TABLE3 .. 2 X MATRIX IN EQUATION 3.65.

2D11
D13 033/2
26. 2012 023 \ 2022
A 8 c E
F G H J K
2011 013 2012 L M 3011
2012 023 2023 0 p 301212 3022

'
TABLE 3.3 COEFFICIENTS OF kN MATRIX
A =
2011/3 + 2 013
B = (D 13 +
D 3 3 ) /3
C = 2(012+023)/3
E = ( O11 + 03 3 + D13) I 3
F = 2(012+013)/3
G = (023 + 033)/3
H "' 2(022 +Dz 3 )/ 3
J = (033 + 012 + 2013 + 2023 )/fJ
K = (022 + D33 -t- 023 )/3
L = (D11 + 2013 )/2
M= (2012+ D13)/2
0 = (0 12 + 0 23 )/2
p = (2022+ 023)/2
80
4 4 _2 2 2 2
D11 = al + f31 + 2val 131 + 4yo:l !31
]) 2 2 2 2 [ 2 2 2 2]
12 = a1"0:2 + !31 f32 + v al 132 + a2 131 + 4Yo:1a2f.31P2
D13 = 2 c.:13 a2 + 2 1313 f32 + 2 v[a12 P1f32 + 0:1e<2f312 J
+ 4Ya1r31 Co:1l32 + a2f31)

4 4 2 2 2 2
D22 = Cl'.2 + f;2 + 2va2 f32 + 4ya f32
3 3 2
D2:; = 2a1a2 + 2 f31P2 + 2 [a{ P1f32 + a1i:x2f32 J
+ 4ya2f32[i:x1f32 + i:x2f31J
D33 = 4a1 i:x2

2 2
+ 4131
2
Pl + 8vi:x1a2f31/32
+ 4y(i:x1f32 + a2f31)(a1f32 + a2f31)
y = (1 - v)/2
v = Poissons Ratio
TABLE 3.3a COEFFICIENT Drnn IN TABLE 3o3
-f:l1 - 132 !31 + f32 1 . . . . .

-~ -!-32
n
pl + f32 • 1 . . . •
-f31 - f32 P1 + f32 . . 1 . . .
C(l Cl'.2 -a - C( 2
1
. . . 1 . .
C(l C(2 -al - C(2 . . . . 1 .
0:1 a2 -al - a2 . . . . . 1
TABLE 3,,.4 COEFFICIENT O'.:i' THE MATRIX A.

{Cl)
+:i ~. '8. I
82
z
X FORCE (disp)
FIG 3-3 FORCE AND DISPLACEMENT COORDINATE
7--------~\:---,o..
~-\\
{a) force mode (b) deformation mode
FIG3·4 NATURAL FORCES AND DEFORMATIONS

y y y
0 x 0 x 0 x
·(a) cart es·1 an coord. ( b) local coord. (c) area c oord.

Si. = l::,.ojk/Aijk
FIG 3.5 COORDINATE S'rSTEMS FOR THE TRIANGULAR ELEMENT.

y
n·I
~·h·
I I
r
} t .. k
lx·y·) I ( xkYk)
Jl
)(
F~G 3.6
ELEMENi
-
GEOMEi~Y.
-
00
0\
i<t.I
{a) symmetric deformation {b) antimetric deformation
·FIG 3.7 NATURAL GEOMETRY AND GENERALISED DISPLACEMENTS

FOR DERIVING A NEW FUNCTION FOR ELEMENT IN BENDING.
86
_ CHAP.f'ER_L
GOHPU~R PI:O~"F®!
As shovvn in the section( 3.,2 ) on. the derivation. of the stiffness
matrix: for a shell elemant, both the inplane stiffness matrix and
the bending stiffness matrix fo::;::o the element are required.. Quite
a number of acceptable fOfllls of the inplane stiffness matrix for
the plana:i.' trianglllar elel!l.ent are available (26,29,32) and the one
in refer~nce (29) has been chosen for inclusion. in the shell program.,
Of the bending stiffness matrices for the triangular element reveiwed
in chapter 31 two have been incorporated in-Co two separate shell
programs main:y for corr.pariso:.:l with t:r,e one derived in this thesis.
The details of these prograr.is are no-.v prese:..'lted.
A• l Lor::al cartesia..2_'.3-Xes for shell element and the noordinate

transformation matrix.
·----
A subroutine has been wtitten for the sequence of matrix
operations derived in referEmce (26) to evaluate both the
transformation mat~ix a...~d the local axes for the shell element.
In the prograr,1s, the side i .. j of the E;lc;"D.ent ~ fig 3. 5b, has been
chosen as the local X··e.xis 9 while the node i has been chosen as the
local origin of the element~ Based on the global geometry of the
nodes associated with the element, the program evaluates the direct-··
ion cosines of the axes x, y and z of the local system. This is
denoted here by c. If xn denotes the coordinate vector «xn Yn zn)
the local axis coordinates of the nodes are computed from the formular
Xq = C [ ~~q - Xi ]
where q takes the value of j and k respectively.
87
.1.• 2~~.§:..1'..IE.!z.....9f the Stiffness lTatrix for Shell Element
The inplane stiffnesR ma.trices and the bending stiffness

matrices provide the relation between vectors which are defineJ
as P·m P:m and P.b a..i1d P:b respecti·c1ely. The subscripts m and b · are
used to denote membrane and bending respectively. These vectors
ha«re been C.efined in section 3. 2. 'rl:.e relationships 2.re as follows:
I
Pm = ! Pmi Pmj Prak?
r
I
Pm = 1
1 Pmi Pmj Prok
a:ad
Pr,;. \Fti pbk~
Pb Pbk~
where = !Umn Pmn { ~m vmn ~
and = nri
i. xn
VJ l p
n! bnl n :·n
== ) 0
n \
n w \.
and n takes the value of i 1 j or k respectively.

Next we write two identification matriceo Mm and Mb which B::r:'e defined
by the relatioship:
Pm
where p
1 is of the same form as p in equation 3.30 but in the local
system.. The explicit form of r,\n and M:b are &s in table 4.1 and 4, 2
The uncoupled stiffness matrix k in the local system is then

obtained by the congrt:.~~t transformation
k =Mmtkm Mm+ :?~b kb Mb (18 x 18)
There io no ari thmet~_cal operation involved in equation
out
The operation involves the sorting~of the individual stiffness
coefficients into the appropriate location in the 18 x 18 matrix.
The operation could be performei as ~ow and column transfers using
the integer. tabulation and (~ table types of instructions, thus
making a considerable saving in computer storage and time.
U·
I
v.I
W·
I
-&·
I
~-I
~·I
u.
I
1 __J__
v. , -- - U·
.J
r
I V·
·-- J
U· 1 w·
J
V·
J
- , I
!
I
! I
-&·
j
J
c:
I
I ¢·J.. co
1 I
"kj
vk
I
, ~·
_l_
Uk
M vk
wk
-&k
~k
~k
s
TABLE 4·1 COEF OF Mm IN EQ. f..f.
LI·
I
V·
I
W·
I
-e. -e-.
I I
¢·I
~-l
>-----+-------· - __________ , _ __, _______________ ~ -------- ---+---·+--- +----!----+
~i
w.
I
-~---,- -~------ - · - ----1 U·
J
_.
-fr. >------>---+------.--·---+----+---
r··.
VJ
~ '. - .____,______,__,____--1~----~=-~·- --+----+----'--, -

·- -- ·- -·-··-
-----~-- -- ---r--- - ---+----+
W·
J
-e-.
J
c··
1..0
w.
J I ¢·
c---"-- - .-- ---- - ~-- ----~---+----- _ _ _ , _ _............._ __,___ _+ . . . - - - - . . i . - - - ' - 1 - - - t--------+-___j_---+-----i J
~j
Uk
vk
wk
-e-k
~k
~k
TABLE 4·2 COEF. OF M.b IN EQ. 4 ·4 - s
so
Next we set up the super dj.agonal matr:L.~ C (18 x 18) thus

. !
11
c c c c c ·
c ~ ~ ... ' 18 x 18
l he global s-tiffness matrix is now obtained 11y the congruent trans-
formation
k 4.7
As sho-0m in section ( 2. 51) t:'J.e corrrpletG stiffness matrix is

conveniently asserrt"bled by the use cf the [a] matrix the deriYation
of whion is now pr0sentod. 'l'he transpose of the [a] matrix wr1ich
e:x:presses the equilib:r:tUI!! e::iuation tetween the external load and
t:he r..0dal forces is used to ob·0a:i.n the [a] matrix.
For each element we red:i.fine the e:!.ement vector in the partioned

form
Pq = ( pi pj pk} q
where each subvector is of the orde~ of 6 x 1 and
Pn = \Un vn Wn Mxa Jvlyn i1Lzn1 4.9
wher9 n refers to the nod.al point (r.:; and takes the value i~ j, k
respeetively. We now form a super vector P which is composed of
the P vectors stacked one below another in their sequential form thus
-~
P P2 P P
==
3 4 l?1
••••••• PNJ 6 :z 1 4~10
fo:r a totc:.1 number of N eleme:11ts o Similarly we form a vecto,:- of

ex+,ernal nodal loads on the structure R also in the sequence of
noda'.:. arrangement thu.~
t
R = [ R1 E2 R3 •...•..• RQ] 4.,11
where each of the Rj's :is e.. 6 x 1 eq-:..iivalent nodal loads composed
of two :i_npla11e load.s 1 th:"'.'ce moments and one normal loads arranged
in the same sequence as lw thus
Rj = [ U j Ii ;j Wj
PI
R·I-1 •(1g..)
.... 4-13
PU
• IJ9o<•)
Pm
I 6 = 6x 6 unit rna.tri x ...
(. •)
Cl)= ·· nu(t ·· Prv

. (•""")
\..0
EQUATION 4-13 I--'
FIG 5-1
92
Then by considering the equilibrium of E:a.ch node in turn for a

total of Q nodes we f:_nd that the exter,,.,al load is equal to the
sum of the elencnt nodal forces meeting at the particular node.
This can be represented by 2. matrix A which is in fact the requ~i_red
(a t) matrix. We demonstrate t~is with respect to the simple structure

shown in fig. (5.1). The equilibrium equation at node i is to be
found.
The positioning of each of the unit matrix in equation 4.15 is

seen to corresponding to the columns corresponding to the element
nodal forces and in particula:;:- the ccrner identification symbols
i~ j ~ or k. Within the s1;.bdivision repi'.'esented by the .vertical lines
the unit matrix occurs only 011ce. By similar reasoning the other
rows of the at matrix are obtained. Since the rows of [a J are
sparcely populated. we partition the a· matrix as shown previously
and in equation 4.13 by the two horizontal lines and merely record
the positioning of the unit matrices by the i.ntc-ger ' e-c.d
table O
instructions. The rows of the e:-1tire matrix are novr assembled by
the algorithm
Ri ai \:i ai 4.14
The (ai) sub matrix can be set up automatically in the computer as
described. r.or a structure consisting of several elements which
could run into h,_mdreds, t-'J.e amount of data input for the integer
tabulatio~ becomes enormous. Therefore the whole ~recess has been
tu:.cned here to the computer itself to compile. This is computed
from the element nodal array. For each element there are three
numbers corresponding to its vertices i, j, k. These numbers are
presented as the input data in a sequential manner begining with
element 1 to element N. Also the number array is presented in the
sequence i, j, k according to the tabulation below
93
!?~~al j~~~n t :!:.£.~2.~i:\~'.!..:.._~· o ~2.1 em.en i,

!i_oG.~ __IT2.!.-__.i..._=:_.L._ ___j__~?- ..,.. __ _ls....::.._.2.
.,
.L
N
The entire set of 3 x N numberc is read in one go. For each node,
starting with number l~ the set of 3N figures in the above table
is scaned three at a time. If within any scan, a number ooeurs
which is equal to the noC.al n·..:1mber uncler the control, the element
~umber in which it occu~s i8 noted ac well as whether its position
ic 1, 2 or 3 corresponding to ~ 1 j or k. Also of im~ortance is a
counter which registers the total nw:z:.ber of elements meeting at
the l'!.odal point in question. This is 11seful in compiling equation
4~14 ; as well as finding the average stresses at the ncdal point
as will be explai~ed later on.
Assembly of the Equations:
The program uses the data generated in the previous section

according to the follow·ing c:.lgorithm. As a first step, the program
assumes that the stiffneas matrix for a particular element is
[~:]
I
partitioned by rows thus:
k ..
kii
'V:· .
·-Jl
kkj
kki kkk
lJ
kjj
kik]
kjk
[~~]
l
J
.th
.th
row
kth row
row
4.15
each submatrix kmn heing a 6 x 6 matrix. The row partitioning of

the stiffness matrix assumes that the force vector is similarly
partitioned but in the meanwhile the displacement vector is not.
ThEhwe set up a working space which has the dimension of 6 x 6M ,
94-
where 1!I is the total number of element in the structure. Using

the data gG:.:.1.era-ted in the previous section, we read the ele111ent
number, bring the element stiffness to a working sp8.ce where it is
partitioned ~-n accordance wiJ;h [4.15]. Depending on whether the nodal
"d en t":f"
'.i. . 1 , 2 ;:ir 3 we t rans f er th e i.th , J.th of kth
1 ica t•ion numb er is
row of the pari:;itionod. :;iubmatrix to the corresponding location in

the 6 x 6M working location. This is done for all elements directJy
attached to the node. The procedure described thus far correspond
to :forming the equation
Ri S!"'<eS
For the second stage 9 we now set up another 6 x 61,.., where 1 is
the maximum number of nodes. The mat:Lix 6 x 6IJI generated above
is now transposed and partioned into 6 x 6 units. The process of
addition of the stiffness matrix now takes place. Beginning with
the first element, we add the component submatrices to location
specifiei in the nodal connection data format.
Since there are six unknown compouents of displacement per

node to be determined the nt:<.li1ber of simultaneous equation to be
solved grows enourmously as the number of nodes increases. This poses
some problem as far as storage is concerned. However advantage can
be taken of the sparse population of the matrix. Two scheems were
tried.
4.41 Failure of the Gauss Sidel It~tion Pr~£§~
The Gauss Sidel iterative scheem is based on tr~ fact that we

can write the equilibrium equation in the :form
-l ,-R ·- k
ri = k ii I i - ,!.._ - ij rj - ~kij
L. 1~i i+l,N
95
. (n + 1) (n)
ri - ri 4.18
and ::; kii-l r;i - >kijrj(n+l) - ..:;>__ k l.J ~
· ·r J· ( nJ .
4 19
where n is the cycle of .aeration. In theory the algorithm ought
to work, but the result of the computation showed a divergence
after only a few iterations, and the method was therefore abandoned.
4~42 Matrix Diagonilisati9n
The structure is divided by rows of nodes as shown in fig 5·2

All members having nodes wholly in row i or in both row i and i + l
are said to be associated with row i. We now define the submatrix
r.J. such that each submatrix consists of node displacements for all
nodes in the row. The submatrix Ri is similarly defined as the
nodal loads in the row. We then see that for row-wise ari.angement
of nodes for a structure having q rows, K is of the form
,...
I
KJ.1 K12 '
I F£1 ~2 F£3
'
I
K 32 K 33 K 34
I
'
I
K 4.20
I
1.... Kqq.!
By symmetry considerations
t
kii = Kii
t
K. . l = K. l i·
J.' J.- J.- '
A set of equations for which the matrix of coefficient is of the

foregoing tridiagonal form may be solved relatively easily as
shown by Wilson (l'1~). We first write out equation (4. 20 in the
form of a set of simultaneous equations in the r~fthus:
96
== R 1
== R3 4~22
+ KqqWq
Kq q-lwq ... 1 ==Rq
The first equation is solved for rlo This result is then used to
eliminate rJ. from the second equation which may then be used to
solve for :r.·2. This may then be used to eliminate .r2 from the third
and so on. The sequen(!e of ope:cation is represented thust
Eqn. 1. r 1 == Kl l -l [ 111 ·~ Kl 2 r2 ]
1 _, 1
R1 - X11 -K12r2 == R1 - Q12r2 4.23
:::ubstitution of l.'J. in equatirm 2 give..;
K21 [R1 l - Q12r2] + K22r2 + K23l0 = R2 4.24a

multiply, and rearrange we obtain
which is of the same form as equation 1 and solving for r gives

I'2 = [K22
1 1
r
[R2 - K23~] 4. 24
= R21 .. 'i23~
and generally
~
rj == Rji - Qj)j+l rj+l
If j == q, rq+l = o and so we can make t~e backward substitution t~
obtain 'the va.l-4_C':S of the r's .. It will be observed. that at any stage
only Qij submatrices need be stored, thus leading to a g:re.a.ter
saving in space compared to the ruethod described by Paullin~ 6
( l9 4).
Machine size limits only the size of the submatrices Qij but not
the total number of nodes.
97
Evaluation oi StressP.G i:t the Structur~.
When the displacenent of the structure have been obtained,

.J
.. - the stresses in the elements are obtained by a bacbrn.rd s~bstj tut.ion
proce~s. The strains in the elements a:re obtai~ed by means of
equation 3.16; and usinp; the stress-str2.in. rel.~.tion f;"iven by e<J.ua.tion
)<>14 the n~.ri~Sscs are ·d~termined.. In general, stress jumps ac::::-oss
the bow1darles of the elements are to be expect~d. But by conslderin~
the· eq_uilibriurn of a n0de under the act ions of the forces contrib~1 ted-
by ec.:.ch element and the external force system, it -can be e3tablished

that the.s~ressea at a narticular node in the structure can be
oht::..ined by takin& the ae~n o.f .the. stresses~ at the corners of the
elements meet.in.g at the particular node.
97lrJ.) .
Fl(; 6"·1 StJr>ow1s10N ot= STIZVc7otf.€ 1111To ~1/'ls t:JF N~ES
/tNO '1~'1'€~S
er·
__,o
Experiments were conducted on hyperbolic paraboloid shells whose

boundaries coincided with the straignt line generators. Only the low
corners of the timber shell were supported and a tie rod also was
provided at these corners. All corners of the vybak model were fixed,
with respect to vertical displacements and corner '·rotations.
The objects of the experiments were as follows:
1. To investigate the behavior of the shell under a uniformly

distributed loading and flexible edge boundary conditions.
'I1he experimental results were to provide a basis for
comparison for the theoretical solution based on the
finite element a_~alysic daveloped in chapter 3.
2. To investigate the suitability of a laboratory table

sized riodel and the accuracy which may be achieved there
from.
3. To study the suitability of balsa as a structural model-

ling material for timber hyperbolic paraboloid shell in
particular and timber shells and platelik€ structuresin
general.
1
The objective described under 1' was primary, while those under
'-2' and '3' were secondary as far as the investigation... cnr3:-iod out
in this thesis is concerned.
_2~actor Governing the Ch~ice ~j.__:~ca1e and Material_~f.. the
----
Mc,del
~aron
(1962) reportei an exploratory test on a timber c0noid
shell roof, carried out at Southampton University. The object was
to compare the behavior of the model with that of another model which
was a half scale of the prototype& The material used for the labora-
tory model was balsa wood.; '11he choice of balsa was influenced by
the fact that it grows uniformly than any other species, thus ensuring
some degree of control on the density and hence the strength proper-
ties. The apparent success of the model in predicting the behavior
of the prototype in a rather qualitative manner prompted the choice
of the material for the model of hypar shell planned in the present
investigation.
A considerable number of experimental investigations have been

carried out on models of timber shells as an aid to designo Some
of these tests have been carried out on an eighth, quarter and half
scale models of the prototype. The rise in labour and material cost
has however diverted attention to the use of laboratory scaled models.
One of such tests by Baron has been refered to above. Kereszetesy
(l9 6 4~ 1 9 6 5) has also reported a series of tests carried out on a
model of timber l:y:perbolic paraboloid shell. This model was 5~t.
square in plan a...nd was planned to be a tenth scale model of a typical

shell ta...~en as 50ft. square in plan. It was felt that the use of
very small scale sized models should be further investigated to
determine their feasibility, and possible extension to more complex
timber engineering structures. Thus taking the typical 50ft. square
shell, a 1/25 model was planned. This scale was applied to the
membrane and thus led to the choice of l/32in. thick veneers. The
prototype would normally be nailed and or glued, but in the model,
we are restricted to gluing the veneers together.
100
A former was constructed using plywood and the forming surface

was shaped to the same curvature as the shell to be built- The
former was 24 in.by 24 in.in plan and the rise of the high corners
over the low corners was 4.8 in. Fig. 5.1 shows an isometric view
of the former. To prevent the glue from percolating to the top of
the former and so holding the finished shell down during the strip-·
ping off the fcrmer, strips of celophane papers were laid on the
former and held dovm by means of adhesive tape~
The vene.ers of balsa as delivered by Messrs Solarbo Ltd. 5 wexe

24 ino long, 6 in. wide and 1/32 in. thicko These had to be cut
into strips 1/4 in wide to conform to the model scale chosPn. The
cutting was done by means of a carving knife and a straight edge.
This proved a particularly difficult operation since the knife tended
to follow the grain of the wood. The grain was elightly inclined
to the straight edge which was aligned to coincide with the edge of
the sheet before starting to cut. The first layer of veneers was
placed on the former in the direction of one of the straight line
generators and held down by means of adhesive tape, while the next
layer was spread with glue and laid on the next layer. The veneers
were then pressed together and held down until the glue set. The
second layer was orientated in the compression diagonal direction.
The third layer was pointed in the direction of the other generator.
A general view of the shell and veneer orientation is shovm in figo
5G2$
Two types of glue were tried for the membrane. The first was
the traditional balsa cement normally used for gh1--ing pieces of
balsa together. The adhension was found to be poor and large
J_Ol
quantities of glue spread had to be used especially as the cement was

very poor at spreading., The second glue tried was the Tensol 7,
which wan found to be more suitable, and although rather slowly
drying had a good spreading quality.
The edge beams were made from the same veneers as the membrane
of the shell. Each beam was l/4 in. wide and 3/4 in~ deep. Eigbt
lengths of 24 in. each were prepared. After the shell dried, the
first set of four beams was placed arounC:. the top surface, and after
drying, the shell was trimmed and taken off the former. The bottom
set of edge beams was then fixed~
.2.~4--2.£rner Detail
The corner detail presented some problems. Since the material

was very fragile and would easily be crushed if a bolted connection
were used, reaction plates shown in fig. 5.3 were therefore glued on.
The reaction plates 1 in. in length were obtained from 1/4 in. mild
steel angle. The glu-ing of the reaction plate has two disadvantages.
First once glued on, it was very difficult to remove without damaging
the shell should there be any need to rectify a mistake. Secondly it "
was difficult to allign the plate prior to glue setting. The vertioal
reactions were designea to be transmitted to the frame by means of
1/8 in diameter steel rods scre1~at one end to the corner plate and
supported on the other by a ball bearing. The arrangement is shown
in fig. 5o3a. This arrangement it was hoped would rectify any imper-
fections in the fixing of the reaction plate. The low c0.rners were
restrained by means of a tie rod made from 1/16 in. diameter alumir.ium
rod.
102
An air bag was used for applying the load on to the shello
The choice of the air bag was influenced by the need to apply a uni-
formly distributed load and to obtain an arrangement which permited
a continuous incremental loading to failure, with minimum of inte:?:Ven-
tion with the test setup., The loading face of the bag, which was
supplied by P. B. Cow r.tdo, was shaped to the same surface as the shell
and it was enclosed within a box fitted to the frame within which the
shell was supported~ The enclosing box is of the same shape and size
as the former shown in fig~ 5ol except that it was completely hollow.
The air bag had the internal dimension of the enclosing box. The
al:i:.· supply was· obtained from an 80 psi supply using a reducing valve.
The p:r:·essure of the air in the bag was measured by means of a water
manometer connec·t;ed to the supply line going to the bag&
Attention was confined to the measurement of displacements

on the shell membrane and the edge beam. Dial gauges measuring
up to 1/10,000 in. were placed according to the distribution shown
in fig<> 5,,4a. lOmn Japanese strain gauges were placed on one edge
beam both at the top and bottom according to the distribution
shown in fig. 5.4b. No strain gauges were placed on the membrane
because of the uncertainty envisaged in interpreting the stresses.
Timber shells as constrQcted in practice are built up from

veneer pieces which are either nailed or glued and in the majority
of cases nailed and glued. It will thus be desirable that, in order
to obtain a perfect model of the prototype, the means of connection
should be reduced to the same scale as that of the model~ Clearly
103
there is a physical limit to which this can be done. In the

present investigation, the size of the model precluG.ed the use of
mechanical fastening; and hence the ven01::>:~s had to be glued~ The
absence of any mechanical glue spreader and presser made it inevi-
table to avoid the liberal ~se of glue. It was therefore necessary
to evaluate the influence of the glue stif'f'.aning on the shell
membrane,, A convenient model seemed to be the sinple tension test.
The balsa as delivered by the manufacturers was already in

prepared veneer form. Since there was no means of identifying which
veneer came from the same part of the log from which the veneers
were prepared 1 the matching of the specimens was not possible. The
selection of the plies for the tension tests was therefore made on
a random basiso The only degree of matching achieved was to ensure
that a test specimen was built up from only one sheet of material.
A jig and template was specially made. The template was used to cut
the profile of the specimen fig. 5.5, thus ensuring similarly
shaped specimens. The jig was used to allign the holes in the
bearing plates during the gluing process. Strains were measured
by two displacement transducers fixed longitudinally at midspan
one on each face. The gauge length used was 3 in.
'2!7 Presentation of Results for the Timber Model
The results of the experiment on the timber model are

presented in the following sequence~
a. The readings obtained from the deflection measurement for

the four loading cases are presented in fig. 5.14. The. readings
are recorded directly on the point where ·che deflections were
104
measured~ Those points have been indicated by triangles~ The

numbers at the base of the triangle ind'.Lcate the dial gauge numbering
corresponding to that srown in fig. 5.4.
b. The strain measu~ements are presented in fig, 5.22 and fig.

5.23 as the different~al of strains representing, a half of the
difference of the top and bottom stra:Lns, and the mean of these
strains repectivelyo From these ~aphs the bending stresses and the
azial stress may be found.
c~ The Young's moduli obtained from the tension test specimens are
p:r:esented in a graphical form in fig. 5.20. A typical stress-strain
relation is shown in fig. 5.18.
d. The plots of the displacements along the tension diagonal, the

edge bea~ are presented in fig. 5.24 and fig. 5.25 respectively.
The plot of the displacements at the centre of the shell correspon-
ding to the various loading stages is presented in fig. 5.19.
5.s E~uatioJ?- of the Timber Hodel s_et up
The··e were some design faults which affected the behavior of

the shell in a manner which was not easy to evaluate. These faults
a=e now set out below.
a. Bag Loading:
The original intention in the design of the test set up was to

load the model from the top. It was found however that the bag
could not be properly fixed into its box container. The bag tended
to slump on to the model causing the shell to deflect in an unsymmetry
manner- An unsatisfactory remedy was made by turning the test rig
upside down so that the model rested on the bag. This in turn
produced a secondary effect, The bag slumped into its container
and the reaction supports had to be given some initial pre-tension
to bring the model surface i:'lto contact with the bag surface. 'Phis
had the effect of giving the s:'J.ell an initial loading which is not
directly due to applied loading. This can be observed in f'ig. 5.19 1
where it would be observed that for zero deflecti01:., the plot of
the displacement does ni:it pass through the origin. It would also
appear that since the hyperb'Jlic paraboloid is not a developable
surface, the fabrication of the bag can at best be approximate to
the true shape~ Under full loading however one might expect the bag
to stretch to accommodate the surface of the shell4 The provision
of an edge beam also makes the use of a bag loading unsuitable espe-
cially in a shell of this dimension because of the constru0tional
difficulties involved.
be Su~port Condition
'!'he support of the shell also proved unsatisfactory. The

four coi-ners of the shell were originally supported. But during the
loading stage 1 the high corners were rendered ineffective by the
corners moving in the direction rather than against the reaction
supports. The lateral stability thus depended entirely on the
support provided by rubbing against the air bag., It was therefore
left to the low corners to carry the loads. The low corners deflected
excessively as can be seen in fig. 5.21. The scale of the model
did not permit the measurement of the diagonal extension of the tie
rod.
105
The Youngs moduli curves presented :i.n fig8 5.20, show that
superimposed on the variabi1ity of material, there is a definite
evidence that the glue has a considerable effect on the tensile
specimen.
Fig. 5.21 shows the plot of the displacement along the edge
beam. The actual displacements ha-ve been -plotted to show in parti-
cular the deflection of the low cornerg If however the deflection
values as shown in figo 5.14 are compared for the low corners and
the high corners 9 there is definite lack cf symmetry and the reduct~~on
of the displacement of the edge beam shown in fig. 5.21 cannot be

reduced realistically to exclude the effect of the reaction displace-
ment. However this reduction has been done in fig~ 5.25 based on
the dial gauge 1 only. The displacement trend can be observed from
this figure. The differential of the strains and the mean <~f the
strains on the edge beam are also shown in fig. 5.22 and fig. 5.23.
However to ensure that the shell still behaved linearly, the displace-
ment at the centre of the shell has been plotted for different
loading conditions and shown in fig. 5.19. From this graph it can
be observed that the shell still behaved linearly.
It was felt that there were too many parameters involved in

the test as carried out to enable a proper correlation to be made
and hence a decisioif;fuade to carrs out an experiment on a Vybak model
but drawing on the experiences gained on the timbermod.el.
l06
The material chosen was a type of plastic under the trade name
of VYBAK. It was chosen because it has a low softening temperature
of 90°c and could be easily formed to any shape. In contrast to
perspex, it dces not creep very much at room temperature. A~nroxi
mately 2 ft. square of the material, which was l/4 in. thick, was
placed in the steam bath and softened. The softened vybak was then
placed over the former fig. 5.1, weighted down and allowed to cool.
When cooled 1 the model took the shaue of the former. A certain
amount of local irregularity was obse:i.."'ved on one of the edges, but
this was soon cured by reheating and reforming as described above.
One quarter inch wide ~ieces of the same material for the edge beams
had meanwhile been cut and glued by means of Tensol cement which had
been recommended by the manufacturer~ The joint was found to be
structurally unsatisfactory, and other types of glue were tried
with little success. A bolted connection was therefore chosen both
for the ed~e beam and the corners. The edge beams consisted of
two 1/4 in. by 1/4 in. laminations symmetrically placed at the top
and bottom surface of the edges of the shell. The edge beams were
attached to the shell by 1/8 in. diameter studs at 2 in. interva~
along the edges. Fig. 5.6 shows the edge beams in position and
also the details of connection.
2.10 Corner Details
The corners were secured to the test frame by means of a rigid

bolted connection shown in figo 5.6. Steel plate pieces, pointed
in the diagonal directions, were rigidly fixed to the four corners
of the test frame. The high corner plates were on the same level,
but on different level corresponding to the shell rise, £rom the
107
lower corner plates. These plates were provided withcl/2 in. wide
horizontal slots to allow for free diagonal horizontal movement of
the shell if this is desiredo A Y-p:...ece plate was provided at each •
corner and formed in such a way that the V--prongs o.f the piece follow
the profile of the edge beams and was attached to the shell edge
beams by means of eight 1/4 in. diamet~r bolts. The tnil pieeP v:::·."· pro-
vided with two vertical slots, to ensure that before bolting to the
piece on the frame, the shell was not constrained in the vertical
direction. All the corners of the shell were fixed4 A general
view of the shell mounted on the test frame is shown in fig. 5.7.
The model was loaded by means of lead shot which were measu:L'ed
into cans suspended from the model. 1/16 in diameter holes were
drilled through the membrane on grid points at 2 in. inte~·al as
shovm in fig. 5.13. There were a total of 121 such holes. The c
cans were suspended by means of threads which were looped, passed

through the holes and tied into a knot at the top of the shell.
Thin pieces of wire rods fig. 5. 7 prevented the knots coming through
the holes when the shell was loaded. The loop in the thread engaged
the hook on the spring attached to the can. When the cans were not
loading the shell, they were supported on a wooden platform fixed
to a screw jack. The platform could be raised or lowered by means of
the screw jack. The loading arrangement is shown in fig. 5.8. ·
Strain gauges were fixed on the model both at the top and
bottom in accordance with the distribution shown in fig. 5.10.
Two types of strain gauges were used. They were the 5mm rosettes
and crosses. The crosses were used on the diagonals. The numbering
of the strain gauges for the purposes of interpreting readings
108
obtained from the Solartron are shovm in fig. 5~11 and fig. 5.12.
For the deflection measurement, dial gauges reading up to 1/10,000
in~ were used. The distribution of the dial gcuges are as shown in
fig. 5.13.
Deflection and Strain Measurements
The model size did not permit an extensive measurement of the

shell at a single loading of the shello This was due to the diffi-
culty of arranging the dial gauges side by side on the 2 in. grid
interval~ Hence the deflection measurements were carried out a little
a·;; a time. Vertical displacements were measured along the sections
marked A··A, B--B, C-C, J-J and K--K as shovm in fig. 5.14 for a load
of 23lb/ sq. ft. Foi' the same loading, the readings of all the
stI-ain gauges were recorded. The deflections of the centre point
of the shell were also measured for two other loadings of 14.4
and 32.4 lb/sq. ft.
5.1. 3 Presentation of the Results of tests on Vybak Mode:);_
The presentation of the displacement readings is similar to

that used for the Timber Hodel and is shown in fig~ 5.15. Here
however the numbering of the dial gauges is related to what has
been described as the possible position of the gauge. The positions
which are twelve in all are located as the mid point of the loading
points. The directions at which the displacements have been recorded
are indicated. The displacements for these indicated directions
are plotted as shown in fig. 5. 26 - 5.30 inclusive.
The presentation of the strain recordings is similar to that

of the displacements. Fig., 5.16 and 5.17 show the results of the
109
reduced strains fo:r. all the gauges at the top and bottom respectively,
Fig. 5.32, 5e34 inclusive show the plots of the strains at the
indfoated positions shown in fig, 5.16 anci fig. 5.17.
lowcornerr
high corner 6-48"
I-'
0
FIG 5-1 ISOMETRIC VIEW OF THE FORMER

111
.Sre. 7iru 1'RR()/,J'-ff 1'"rl'

Fue- &.IT1'11cf.
1(:,p lAyet. Mt1A1fl

GEw ,,~_...,."~
5,7~~ ~AyFtt ~~6

/,. f;N £It. ,vrt1/l
flll!Dt».E z.4y&i.. ,kDf.16.

t.""JPll.ES1101V btA"""'""-
Ftr.. 5.').
112
(a) reation plates. Mvertical re action ·

11 steel rom 1/gin dia.
direction of load
applicnt ion
( b) corner detail
FIG 5.3 CORNER· DETAIL TIMBER MODEL

113
~·
11
3" 3" 3 3" 3" 3" 3" 3" 3" 3" 3" 3" 3" 3' 3"
I I I I I I I I I I I I 1 I I I I I _ . ./
Hz ._15_ _ _~1._6_ _ _ _ _ _14-.Lz_ t H) 15 L2
(-r)1
J_ 16
H -1-
• 17
12
•
:t
: I
1
11 II 12 131114
11
•
10
• .a
9 ("f')I
•
._______...____-----tr--~-.~-,.---H1 ~1 -
5
-
7
- -
1234 5 6 7 2 4 6 8 10
(a) posit ion of dial gauges ( b) posion of strain gauges
FIG 5.4 LOCATION OF DISPLACEMENT AND STRAIN GAUGES.

TIMBER MODFL
0
reaction
plates
d isplace<rent
trans due er
-I I I 1a''
·1" thickness I
variable
0
1
FIG 5.5 TENSION TEST SPECIMEN
Tl M BER AND VY BAK MODELS.
Fl G 5.5 CORNER AND EDGE BEAM DETAILS VYBAK MODEL
FIG 5.7 LOADING POINTS AND STRAIN GAUGE DISTRIBUTION ON VYBAK MODEL
llG
FIG . 5 . 8 LOA.DING ARRA.~GEI-WT AND TEST FRALIF

VYBAK MODEL .
Fl G 5 .9 GENER AL VIEW OF TE ST SET UP VYB AK MODEL
118
Hi h Low
,(.
>- ~
>- ~
I
Iz,~ ~ ~ '"""
2"
I I
~
'""
~
>- -'
,(,. "'"' --" -' ~
"""' ~
'( y """""
y
.
y
'<cross }- rosette y
y
y
Low High
FIG 5-10 LOCATION OF STRAIN GAUGES

VYBAK MODEL
high
~0)- "' ~ti)- 'v'"'-

~~~16- -~-+~~-
41""" ,.,_.,"y.,o
79
?~,,., ~ 'r'b"y1f!
C
"'
-<
/()
t;.O
ff"'/)j
I I
7'~
1+
4-5.J..7,,,,_ I
I').
r,./
4
4/-"' 71.
~:!."""'
7'!
4-?> .....
'7+
C\ 1-b b 7 8 9
\( 37""";,e 3,5-'.,"' ~9"""'1c .fC..,.11 I
I '). ~ + 5
3'2.~
. li~ ~..l_.64 *'""-e; ~5'"" 61o 'Po.A..67
low high
FIG 5.11 CHANNEL NO. FOR COMPONENT GAU OE S

TOP SURFACE VYBAK MODEL
120
high low
low high
FIG 5 ·12 CHANNEL NO. FOR COMPONENT GAUGE"S

BOTTOM SURFACE VYBAK MODEL
121
high low
... + ... ... + +
+ ... ... ... + .... + +

0
+ ... ... ... ... + +
0
+ + + + .... ... ... + + +
0 0
+ + + + ... -t .... + ....
2''
0 + 0 + 0 + 0 + 0 .... e + + +
!<
2~
+ + ... + .... + ... + +
+
Ii. +
0
...
0
+ . ...
0
+ .... + +
0 + ... 0 + + ... 0 .. .... 0 + + 0 -t + 0

0 0
.,. .. + . . ... + + +
0
+ . + + ... +
0
...
low high
FIG 5-13 POSITION OF DIAL GAUGES(o) AND LOADING POINTS(•)
VYBAK MODEL
122
high low
~ 16_ ~
0
·O-'
• t:>f I
·0.2& I
0
- - - - - - !'JI
l'"'q
•/ID
-
·.11. .g:z
'"""
·DIS
•g"" I
-11 -·""'
·DIS
·DS'- '"''
·DGI.
.0•7 •/l"f
1l 10
- . b'10
'""7
•OIZ.
. 0~1
~
·DOZ
.. ,,, -·4'01. -·o•I
-·015
•.OH •ODO
• llOft
.010
-·PIS
• 11•1
-.11115
-.0111
."""
.001 ·'"'
.ooJ
•(10~
.."""
•oJ.t • 010
."'' .OID
""
2 3 4 5 6
low high
readings are for loads corri:>sponcl!ing to ~·~"}
5·/S f''f
J/J..·SO
,,.0 ·.50
FIG 5.14 READINGS OF DISPLACEMENTS
TIMBER MODEL
123
high low
l(q
"-s-. 0;· /
/~
~ . o-6;·
o'
"•
•Oil o~
(,
~·
•
·Db8
•
•071 •
·DOa
A-
.4178
•
.(J~~
• ,
.,57 ·D'J
•
·PJf
•
·D7~
• -··.A
• •
·071
,cf'/
•
"""'
.mo .c-15 .~61.
·"~ ·"::§8 .()16
8- • • • • • • -B
•
•tJ3'5 •
·c~
•
,Cf() ·~If
/
C-
low
.Ot15
-
.qlf
- -
.tJlf ·~3 .~""
"""6 -c
high
A
6
A
5
A
4
A
3
A
2
A
1
,,
·""- A
2
/\.
3
/\.
4
/\.
5
/\.
posit ion of di al gauges
FIG 5·15 READINGS OF DISPLACEMENTS

VYBAK MODEL
I"
124
-<.
S>
" l
•
~~
~ 3e\ -
- ---- -
~ ~
'.;>
J ~
·'~ .$'
I
.99
~
I~
-;;..
a ~
%-
,.- I ' ~
t
.9.9
/~ I~
~ ~ \
t ~ ~ I~ ~I
~ ~o,C
~ (S'
~ ~ ~ ~
/
c,
-iJ. ~ ~
1t U"
~o."\
low t
D
strains >< lCf 6 high'.-<.
FIG 5-16 REDUCE STRAINS FOR TOP GAUGES VY BAK MODEL

125
hig n low
~
,
'
I
~
/
~
.....07
I
,
....
~
I>
I
'
~ f
'"a
- - I
- - -
-1~ ~,\k -~
<.y ~>
'~ -~ I ~,I/'
"'\. ~
~1' -~ ~
IO
~ '>. ~
,
~ ~ ~ -~1 ~ <ir,, 4',g ~' ,~
~ :k ~ ~ ~'
o..:o...t
l( ~ I
!I.
)'
G ~.... ~
~ 11'
-~
....,. ~,,. ~ 3-
~
'
,1
low strains x 10-6 high'

-<.
FIG 5-17 REDUCED STRAINS FOR BOTTOM GAUGES VYBAK MODEL

126
tbOO i-------~-----..-------,------.,
Ill
Cl-
800
Vl
Vl
/
-
.... /
C1I
0
Vl /
/
/
_q,.I) iJ.O
3
strain 10-
FIG 5·18 TYPICAL STRESS-STRAIN RElATfON 3 ply.BALSA SPE.
0
20
......
Vl
a...
Ol
c 16
·-
"'O
d
0
~ 0
-
C1I
u
~ /0
....
::J
Vl
/ff ,.
/
0 of O·Z
displacement at centre (ins)
FIG5·19 LOAD-DEFLECTION RELATION FOR CENTER OF TIMB::R SHELL.
tens.ion specimen No.
127
"°',_____1_ ___:_i_ __::.o_ __,_4-_ __:_s_ _6_ _;,.1_ _a_~~--10_-.
0·6
1·5
/·0
two plie . <>

me n U 5icro
o·
x
Vl
:J
:J
"'O
0
l·O
E
Vl
CJ')
c:
:J
0 0·5
>-
thre pli s.
mea : 1·4- 0 x10'-
o.____;,___.....___,__~----L-~---'----1----1----1----'
FIG 5 .20 VARIATION OF YOUNGS MODULUS FOR 1,2, 3 PLIES. BALSA.

FIC.. 5.!J.f
5·0
~
~
I
~
~
I(
II)
0
.50
~ c---- l
-
'J- .iru·
't· ~·~
VJ
-100
F/6" !5·22 DIF"~E£.EN r!A~ ()F" SiU-1/U.5 Cf../ .Eb4E 8EA-.vf

T1t.41JE. le. Rcbet-
50 --·-
~~
'll
I
~
1·.20·5 ('$1
()
~ •JH ('I/·
)(
V)
~ -:z~
~
~
"'l -:o ---·
Fr (A 5".2:5 /..fE/rl../ or '/H$ 5/R,l.tA..!S /};...,/ FboE. SE,4-M
J;;.45£ R_. f..10blit..

128
~.'J.L__ _ _ _ _ _...i__ _ _ _ _- 1 - - - - - - - - - l - - - - - - - - '
'I
w
~ p
~
"-
~
f e·J
~
~
129
I 5
-- ....
;
~
~~
------- ~
·02 1--------~--ji---------+
.QB'-------'---------'-------+--:-~~~----'
__)--
~
---
_/""
~-
·CZ.
130
~ ..... -
P/)J/1"1t>N er GALI(,£~ PAI THE "'iy6Mt:. "1"1>6/,. .
t I 2. 3 4- 5 6
a
~
~ ·t/Z
'-.:.
"-
~ ·O+
~·
Iii
~ .()6
~
....
V\
~
·Ob
F1fA 5·.22 /)/,Sl'/...A~ElifFlll / /tl-D~/1. 'i#-6 &()lfPl!£S$1PIV P1.4t:oN,+z..
/)
.......
~ ·C"2
~
"-
i-
Iii
\J
·04
"I: .p(,
~
Ill
{S
•08
-......
! ':I ~·~
~
"'~
t tJ• I
~
...
\I\
~
~
~
tJ d ·o? t:NtJ
t:, t1"V I LA-1.. /:J/ SPJ.,+cE"'f6 N I {.IA./ S)
,C:-1 ti s-: 31
131
I 6
I
/P8 ------ I
:t=-=-----11:-------t----......=.~-1-·- .----- ---
+- - --- j.
I
-------r - I tI
-1'4 J · · 1;-:::,::::L
I
II
--T----
1
--r-
i
I
-~1-------- -f
() _ _ j____ I
'
-JDC
rit:. ~33
Jco
'()
/40
•
~ 0
x
~ -100
~
~ .J()()
"I ---
I 111 /l,fVu f
-3Co
J .80/10117
--~----·
6 _1lus
..f/6 5·:?f
132
240
2.10
/80
/50
---....z
-......, /20
\')
Q
~
\I\ 90
60
r;.szx10' tb/111 2
2
0·80 X fO' lb/M
OL--~~~~~~!__L:__~~~~~~~~~~~~..__~~~~~~~~~~
1400 1200 /300 /400

.STR..11-/N )( 10-b
133
DISCUSSION ANTI CONCLU~
6.1 Discussion of the Res~s of the Theoretical Computation~
A considerable amount of effort was directed towards establi-

shing that the theory as programmed for the computer could analyse
some problems for which solutions are kn0vm. This was necessary
to gain soffie confidence, in its application 1 as to the correctness
of the formulation and the approximations involved. More important
was the verification of the correctne~s and convergence of the
formulation of the new function for triangular element in bending
which has been derived in section 3.4.To this end, the problems of
the plate under lateral loading and with various boundary conditions
have been analysed. This is particularly relevant, because many
research workers who have developed similar functions for triangular
and rectangular elements in bending have chosen these particular
problems as the test case to verify their formulations and the
convergence~ The problems of the hype~bolic paraboloid shells have
also been analysed. The solutions obtained have been compared with
the various theoretical solutions available. V.fhere experimental
data are available, these results have been correlated by means of
the finite element solution. The details of these solution are
now discussed under the head of the plate problems and the hyperbolic
paraboloid shell problems.
6~2 The Plate Problems
Of the few functions available for triangular element in

bending, two have been chosen for comparison along with the function
134
which ha.s been deri>..-oJ. in i:secti.on Jo 4 These functione a.re

identified as Clough and Zienkwi::wic·z fu:n.ctions resp.actively and the
function derived in section 3. ~ is refered to as Segun. The
test problem has been chosan as a squa:!'.'e isotropic plate with all
the edges fixed. A further check is obtained by comparison with an
ex~ct solution. By exact is meant a solution based on an infinite
series representation taking as mar1y terms as possible with little
changes in the solution after some terms. This solution is taken
from the work cf ~imoshenko (l94B). Fig~ 6.2 shows the plot
obtained from the va.rious solutions.. 'I'he solutions based on the
finite element functio.ns and carried out using a mesh division of
16 x 16are shown in figa 6.2. The Cl~ugh's function and Segun's
f1L.~ction can be observed to give solutions which converge to the
Timoshenko's solution from below. The Zienkiewicz function can be
observed to give solutions which is abo-.re the Timoshenko solution.
Zienkienwicz (l9 6 5) has however shown that the solution obtained
using an 8 x 8 mesh absolutely converged to the Timoshenko's solution.
If this 8 x 8 mesh solution is considered along with the 16 x 16
mesh solution obtained here, it would thus seem that the convergence
is away from the exact solution.
The convergence of Segun~s function is further exhibited in

table 6.1. Here the results of the computation, using 8 x 8, 12
x 12, 16 x 16 and 20 x 20 mesh division, for the central deflection
of the clamped plate using this function are presented. The accuracy
of the function com~ared with the other functions is also clearly
exhibited in this table. For the 16 x 16 mesh division~ the Clough's
function gave an error of +13 °/o , the Zienkiewicz function -10 °/0 ,
while Segun' s function gave an error of + 2 %;

135
.
AUTHOR MESH CENTRAL DEFL. o/o of Exact
SEGUN 8 x 8 1,300 103

12 x 12 1.288 102n5
16 x 16 1.283 102
20 x 20 1.279 101. 5
CI, OU GB 8 x 8 1.520 120

16 x 16 10420 113
ZIENE:IEWICZ 16 x 16 1.130 90
TIMOSHENKO EXACT' 1.27 100
l/IULTIPLIER q 1 4/D x 10-3
TABLE 6.1 Comparison of solutions for a clamped

square plate.
Fig. 6.1 shows the comparison of solutions for a simply

supported square plate. The sol:i.d curve shows the result of a
16 x 16 mesh finite difference solution. Because the curves practi-
cally coincide, the finits element solution using Segun's function
and a 12 x 12 mesh division has been shown as dots. The value of
the exact solution is 4,06. The finite element solution gave 4~1
an error of 1 fo .
Fig. 683 shows the result of the computation for the plate on
only four corner support~ This problem has also been discussed by
Timoshenko. The exact solution has been given as 25.7 and the
finite element solution for a 20 x 20 mesh is 26 again showing an
error of +l cro.
136
The diverse nature of the problems investigated on plates and

the accuracy which has been achieved demonstrates the capability of
the f'ini te element using the triangular element stiffness formula-·
tion. In particular, it can be observed that the triangular
element stiffness in bending based on the function derived in section
)s~ yielded consistently more accurate and rapidly converging
results compared with the other formulations discussed.
§.::. 31 So:J:l!_ti~- f~'.E....§he.11 with li'ixed and Si!!!.:ply SuPF,orted Boundaries
Of the various solutions available, the most aboundant are for

the shell with the boundaries fixed. The reason is simple. For
analytical solutions of the Rayleigh - Rit:6 type, deflection modes
have to be assumed and this can be achieved far more easily for the
fixed boundary condition, tran for any other type of boundary condi-
tion. Even the simple su~;port type of boundary condition seems to
present some problefl in defining suitable functions. Problems
originating from extrapolations at the boundary and corner points
also make it easier for the fixed boundary conditions to be tackled
more easily by the finite difference method. On the experimental
front, it appears that more reports have been presented on the fixed
shell than on the simply supported shell. This appears to be due
to the fact that the fixed shell is more easily set up than the
simply supported shell. The practical shell however is the flexible
edge type. This has been investigated exhausively by means of
models, but no theoretit::al sc.-::.ution ifl .-1vailable.,
Since no exact solution is available for the shell with the

fixed boundaries, the finite element solution for this problem has
been compared with two approximate solutions. The first solution
137
is based on the work of Chetty (l 962 ) who used a versatile method

based on the Kantarovitch approximation. Lasioally this is a series
solution., He also compared his solution with 2nother series solution
(
presented by toof 1961) and the results of the experiment performed
by him on an araldite model. The second solution is based on the
work of Brebbia (l 966 ) who used the finite difference method to solve
equation 1.15 He also compared his solution with the experimental
results on a model, he.tested. For the solution of the simply sup-
ported shell, c0mparison has been made with the solution obtained by
( 1 966)
Michael - • He used a beam vibration function in the Gerlerkin
approximation process.
These examples were chosen with two purposes in view. First

to ensure that the finite element method could be applied to the
analysis of hyperbolic paraboloid. Secondly, it was necessary to
have contrasting examples to demonstrate the convergence trend of
the solution, so as to give an indication of the element size
necessary to assure convergence for varying shell dimension.
6~32 _Qomparison of_§_~~tions for the It'f.2~rb9.lic Paraboloid Shell

\fith the fixed and simply SU,EJ?orted Boundaries
For convenience of making cross refrences 9 the comparisons made

below.has been designated as problem 1, 2 and 3. Problem 1 deals with
the comparison with the work of Chetty, problem 2 that of Brebbia,
and problem 3 that of Michael (l9 66 ).
The data for problem 1 are as follows: length 12 in., rise

lc304 in. and thickness 1/4 in. Fig. 6.5 shOl'!S the comparison
of the deflections along the centre line of the shell. We shall
not concern ourself with the experimental result which had been
138
dealt. with at great length in the reference from which it has

been detainedo Presented along the finite elel.!lent solution are
results of two other solutions referred to as Loof Solution and
K. G ~ a shortened form for Kantarovitch-Garlerkin method.. The
finite ele1'.l.ent is based on a shell progran incorporating Segun's
function. The m0sh size used in thiB particular proble~ was B x 8.
The excellent comparison between the solutions can be observed.
Fig. 6~4 shows the plot of the displacements for shell type
of problem 2. The data for this problem are as follows: length
100 cm, rise 10 c:rJ. and thickness 0.8 cm~ We note that the 8 x 8
mesh which was quite adequate for probleB 1 is no longer accurate
enough for problem 2. Thus we had to employ successively 12 x l~
and 16 x 16 nesh subdivisions. In the figure comparison has been

made with a series solution using eight harmonics, and we have also
presented a solution using Clough's function. It can be observed
that the solutions are of the same order of magnitude, but towards
the centre of the shell and in general where boundary conditions have
been applied, the finite element solutions show some divergence
from the series solution. But within the shell agreenent is reason-
ably good. Again we observe that Segun's function appears to give
for the same mesh subdivision a better agreement at the centre than
Clough's function.
The data for problem 3 was taken as that in problem 2. Fig.

6.6 shows the comparison between the finite element solutions and
the series solution using the beam vibration function. 8 x 8 mesh
and 16 x 16 mesh divisions have been used in this problem so as to
exhibit the convergence trend. The large discrepancy at the centre
using the 8 x 8 mesh can be observed, and the resulting i!"lprove:rient
using a 16 x 16 mesh subdivision can also be observed. Within t}'ie
139
shell however, the discrepancies are not so pronouncedo It can

be observed that as the mesh division proceeds, the finite element
solution can be seen to be approaching the series solution.
In continuation of the procedure which has been followed

thus far, it would be ideal to compare the result of the finite
element solution with that of a series solution or solution by any
other approximate method~ for the shell on flexible edge supports.
No solution of this problem has been found in the literature thus
far. Hence the computational results are compared with the experi-
mental results. This experiment has been reported in section 5.9.
In order to eliminate the effect of the edge beam stiffness, the
comparison has been made, as a preliminary step, for a shell with
zero edge beam stiffness. This case being expected to be the
liI'liting case in comparison with the identical plate problem discussed
in 6.2. The corners of the shell were assumed restrained at the
four corners with respect to the vertical displacements and at the
lower corners with respect to horizontal movements. These conditions
represent the least constraints to ensure stability. Fig. 6.7 shows
the plot of the displacement along the central line of the shell.
8 x 8 and 16 x 16 mesh divisions have been used in the computation 0
We immediately observe the lifting of the centre of the shell. This

lifting of the central portion of the shell has been associated with
lack of convergence and has been previously noted in the discussion
on problem 2 in section 602. The subdivision into 16 x 16 mesh,
which had in the previous calculation on problem 2 been sufficient
to give improved result which can be compared with other solutions
140
and the experimental result, does not seem to be adequate. But more
disturbing i8 the fact that when the theoretical values are compared
with the experimental values, both results do not agree and are not
of the same or1er of magnitude. The experimental result is shown
on this g:raph but to a scale ten times that of the theoretical curves·
scale~ Table 6a2 below giYeB an idea 0£ the order of' I!lagnitudes
:Lnvol ved. The displacement has been compared at A - the centre of the
shell, B - a point midway between the centre of the shell and the
edge and C - the edge po:!.nt on the central lineo
0
M:E:Sll DIS PL. "!0 of EXl'T DISPL.
I
I.u of EXPT. DIS PL. °j0 of EXPT,

A B c
8 x 8 ~0032 4.26 .0058 9,0 .0038 22.4
16 x 16 00046 6.14 e0065 l0.3 .0050 29.6
Ex pt .0760 100 .0640 100 .0170 100
TABLE 6.2. Comparison of experimental and theoretical results

using the shell theory.
Although the convergence problem seer.ls to have obscured the

results, our experiences on the results of problems 1-3 would suggest
however that convergence alone cannot account for the discrepancy in
both results. Table 6~2 also shows some convergence trend although
not as fast as would be desirable to approach the experimental result.
It was therefore necessary to examine the state of stress in the shell.
Hence the design of the Vybak model.
6.42 Stresses in the Vyb~k Mode1_
The strains measured in the Vybak m,)del have been shown in

141
fig. 5~32 - 34 inclusive. The actual strains recorded by the top

and bottom gauges have been plotted. Also plotted are the mean of
the strains and the differentials of the strains from which the
direct streAses and the bernling stresses ca..."'l be obtainede Within
the limits of the er.ror in reading of the digital strain recorder
which is of the order of ± 5 microstrains, hardly any direct strali1s
could be observed in the model except along the tension and compre·-
hension diagonals. Even here, we observe that the direct strains
are negligible and could be considered as secondary. One would
theTefore justifiably be led to conclude that the shell transmits
most of its load by pure bending and that its behaviour approximated
more towards the plate .rather than tl1e shell actions~
hl2,. Modification of the @]lell_!nalysis to eJS_clude_ _i!!e Membrane

!!~i
WHh the experimental results indicating that the shell

behaved more like a plate rather than a shell, it therefore was
necessary to modify the mathematical illodel to take into account
this fact. From the finite element point of view, we have ascribed
more degrees of freedom in the mathematical model than the physical
model possessed. In the theoretical formulation of the shell theory,
we have assumed that the strain energy of the shell is of the form
...
U == fv e N" <fr-! dV
where ~· N and e :N have been defined by equation 3.17. t N and <S N

were defined as being composed of flexural and extensional compo-
nents of strains and stresseso If the extensional components are
nul~ the theory is not invalidat~d. If therefore a null vector is
substituted for the extensional strains in equation 3.21
and the corresponding displacements u and v are suppressed, the
142
resulting transformation giYes the stiffness matrix for an element

in bending. The geometric transformation matrix [a] then ensures
that the structure is analysed as a curved plate,. r:l.'he resulting
theory has bE:en captioned the Curved Plate Theory. In essence it
is the analysis of a plate but with the coordinate transformation
matrix [c J taking other values than. unity.
Fig., 6.8 shows the comparison between the experimental result

and the theoretical result based on the curved plate theory. Segun's
function has been used in the curved plate anaJ.vsis and solutions .
have been obtained for 16 x 16 mesh €nd 20 x 20 mesh divisions. The
edge beam effect has also been included in the computation~ The
results show a better agreement than those obtained when shell type
of actions are assumed.
One peculiarity of the solution is however worth noting.

When the computation was carried out with the edge beam effect
included the computer result for the deflection at the centre of
the shell was 500.253. When the computation was repeated without the
edge beam effect, the result was 500.376. When the results were
reduced to obtain the actual displacement in inches, the differences
cancell out in the number of sigr::.ificant digits retained. In fact
the solution presented in fig. 6.8 also represents the result of the
shell without an edge beamo It was therefore felt that the computa·-
tion should be carried out for the case in which the edges remain
undeflected representing the case of the shell with an edge beam
of infinite stiffness. The result is plotted on the same graph and
captioned curved plate S.S (indicating simply supported). If the
values of the displacements at the centre of the shell are compared,
143
the reduction in the dis~lacement based on the zero edge stiffness

case is of' the order of' 306 %, Generally within the boundary of'
the central portion of' the shell, the shell is insensitive to the
presence or absence of the edge beamq Towards the edges however
the displacement is affected by the presence of' the edge beam.
This effect seems to have been borne out by tests conducted by
Ba·bchelor ( l9G3) on a concrete model. An extract f'rom his thesis
is presented in f'ig. 6.10. It will be observed that the ef'f'ect of'
replacing an eage beam made of' channel section having a sectional
moment of' inertia of' l.78in 4 to a box section having a sectional
moment of inertia of' 3~56in4 and a greater torsional rigidity has
not produced signif'ican~ changes in tne distribution of' displace-
ments in the shell. It would thus appear that the rigidity of' the
shell is more important in determining the stif'f'ness of' the shell.
Test will howev-er be requir.·ed. to show that this still holds f'or a
large span thickness ratio •
.§...d-.2 Comparison of' the curved pla;te solution and the f'lat plate
solution
In f'ig. 6.8 the computation using the shell data f'or a simply
supported plate has been plotted. Comparison between the simply
supported plate solution a.rid the curved plate analysis shows that
the shell stif'f'ness is roughly of' the same order as that of' a
f'lat platee This f'act has also been pointed out by Batchelor (l9G3)
The results of' his investiGation on a concrete model with boundary
conditions similar to the one conducted in this thesis also £'ailed
to be correlated by the shell theory and he therefore compared the
behaviour of' the shell with that of' flat plate under fixed and
simply supported conditions. When he used a flexible tie rod
his results compared with a clamped plate solution, while with
144
a more rigid tie rod, the results compared with that of a plate
which is sin;ply supported. His broad conclusion that the shell
behaviour could be approximated by the clamped plate behaviour must
be open to eq_uestions. Clearly the results must be governed by the
shell rise. If the warping constant is reduced to zero, the shell
tends to the plate on four corner support to which a solution has
been given in section 6 .. 1. Thus with very low shell rise one might
expect the result to tend towards the solution of the plate on four
corner support~ The little information available cannot make a
generalised conclusion to be drawn between the solutions., However
one can say that the shell does not behave as a plate under identical
boundary conditions.
6.4.6 Curved Plate A..~alzsis and the Shell ~na~xsis f~e

Shell with SimnJ:.Y.: Sup~~1ed Boundary Condi tio~
The result for the simply supported condition obtained from

the curved plate analysis raises an interesting problem~ We observe
from considerations of problems 1 and 2 that, when the edges of the
shell were fixed, the assumption of the presence of membrane stresses
and the consequent use of the shell theory, enabled the test results
to be correlated. Also based on the shell theory, we have assumed
the presence of the membrane stresses in problem 3, leading to the
displacement pattein shown in fig. 6.6 and again this solution
compared very well with the series solution. However for problem
3, if we ignore the presence of the membrane stresses, we obtain
the result of the curved plate S.S. shown in fig. 608. These results
are shown again in fig. 6.9. It will thus appear that there are
two possible configurations for the shell which is simply supported
along the boundary depending on which assum:ptions on stresses is
used. 'Which of the conf:Lgu.ration the shell will assume can only
be decided by experimental analysis.
145
6• 5 GENER.AI, RBMAR...TCS
&,o.5 .. 1 Mesh Size--·-----·----

and :tfodel Size
As already pointed out in the discussion on the results of

problem 1 7 the 8 x 8 mesh division of the structure was sufficient
to assure the convergence to the result obtained by other methods.
The model size was 12 in x 12 in thus giving the leading dimension
of the triangular elements as 3 in. When the model size was
50 cm x 50 cm for problem 2, we find that the 8 x 8 mesh division
giving the element leading dimension of 6 cm was not sufficient
to assure convergence. The 16 x 16 mesh division giving the leading
element dimension of 3.125 cm was required. Even for this we
observe from the result that convergence has not fully materialised.
It would thus ap-pear that as the numerical dimension of the shell
becomes very large we will require more divisions of the structure
to obtain convergence. However more subdivision into finer mesh
entail greater number of nodes and hence very large systems of
equations to be solved. lfost of the cost of computing is however
accounted for by the computing time. In the present computation, it
was found that with the program organisation adonted, the computa-
tion time varies rougly as the cube of the mesh division as shown
in fig. 6.11. It thus ap~ears therefore that the finite element
idealisation of shell in its present form is not an economic propo-
sition for the everyday use compared with the other form of analysis
of the problems considered.
§.5.2 Lifting of the Shell Centre
The cause of the lifting of the centre of the shell, observed

in problems 2 and 3.and also on the problem of the shell on
flexible support is not very easy to trace. The computations carried
146
behuviour and
.
out on ·the shell with the flexible edge
the
b,:r assuming the shell type
curved plate action seem to suggest that it is
associated with the membrane stresses. The portion can be taken
as the slowly converging 'Jortion of the analysis. As shown in the
comput~tions on problems 2 and 3 its effect is minimised by division
into finer meeh. I'he wor;:iening of this e!'f'ect in the computation
on the shell with the flexible edge support however suegests that
the boundar.1 conditions could also have some part to play. 'l'his
effect is totally absent on the plate problem~ and the curveO
plate analysis.
6.5.3. Bending moments in plate and shell.
The results of the com/.lutation of the bending moments are

preseritcd in fig. 6.12-6.15 inclusive. In general, for the plate
problems, the bending moments are reasonably predicted bJ the finite
element method using the trLmgular elements. Errors in th(
bending moments are maximum towards the edges and to a lesser
extent at lines of symmetry. The results for the clamped plate
and the simply supported plates are shown in fig 6.14 and 6.15
respectively. The same remarks apply to t'1e problem of shell
fixed along the boundary. The results, compared with other solutions
and experimental results, have been presented for Problems 1 and 2
and shown in fig 6.15 and 6.14 respectively.
147
6.6 CONCLUSIONS
The various problems discussed in section 6.1 - 684 have

demonstrated the ability of the finite element analysis to tackle
the problems of bending of vlates and shells under the fixed and
simple sup~ort conditions. In particular the use 0£ the triangular
elements have been demonstrated as capable of producing the same
degree of accurarcy as solutions based on the series solution
especiall3r for plate probleI'ls ~ For shell type of problems, the
finite element solution has been shown to suffer some convergence
trouble which can be improved by division into fine mesh. A new
algebraic function for aefining the stiffness properties of the
triangular element in bending has been developed. '.!.'his function
captioned Segun's function has been shown to give a consistently
more accurate solution than Clough's and Zienkiewicz's functions
for the e.nalysis of plate problems. Its convergence has also been
shown to be better than the other functions.
The shell theory has been shovm to fail to predict the beha-
vioUI· of the hyperbolic paraboloid shell on flexible edge supports.
This failure has been explained by the fact that the shell does not
develop membrane stresses as assumed in the shell theory for this
particular boundary condition, An experiment conducted on a Vybak
model established that the shell transmits its load by means of
bending stresses which are associated with plate type of actions.
A modified shell theory which neglects membrane actions has been
advanced to explain the behaviour of the hyperbolic paraboloid shell
on flexible edge support. This theory captioned the Curved plate
theory has been shown to predict the shell ·behaviour under the
flexible boundary conditions.
148
The large number equations which result from analysis based on

shell type action, coupled with the convergence problem makes the
finite element analysis less attractive than the other methods for
the fixed a~d simple support condition.
6 .. 7 Suggestio;i:i_~_!_<?r further Research
The test reported in Chapter 5 on the Vybak model has shown

that the hyperbolic paraboloid shell tended to hehave like a curved
plate~ Based on this test, the curved plate theory has been proposed.
It will therefore be useful to investigate the extent to which this
behaviour still holds when several parameters of the shell are
variedj particularly the rise/span and the span/thickness ratios.
The curved plate analysis has also demonstrated the relative unimpor-
tance of the edge beam stiffness for the test data obtained in the
present investigation. Since the greater part of the cost for this
type of structure is taken up by the provision for the edge beam
it will be useful to determine the maximum ratio of shell/beam
stiffness beyond which the beam stiffness does not contribute
significantly to the stiffness of the shell. The tie rod influence
could also be simul t.aneously studied in the same program.
A test of the shell on a simple type of support will also be

useful to verify the findings of the curved plate and shell theory
discussed in section 6.3.6. This could be used to simplify the
analysis and design of the hyperbolic paraboloid shell by separation
of the edge influences.
More investigation into the basic properties of glued veneers

of balsa and relation to full scale specimens would appear to be
necessary if the results of model studies on timber shell were to
be used to interprete the behaviour of the prototype.
14-9
y
t TI
I '-----,/'---->'--~'----,l''-----,1'~----JL---IL--.I -
I
I X·
f-71'--7!'-----71"--,.J"--,.J"--,JL____-,l"-----J,L__J.L---.j'I
I l'--,!----7r----,,~'---JL___J"---JL----J."---~~ ~;
~I { f--7!'-----?f----7:"-?~"---j"--)L___J.'--+~ ~.
~
t---- - £ - - -~
~~ ~____,,____,, i
f· /../~ -+ i~ fl no. or .f'4'vit1- 'x 1
(_t"i) Ccott>/AlloTG sr~rEM

.SH£1..1.. /t1VI:> --PLArli 'Pfi!cJ,L£1~
x:
150
·-=7~-----·---+--------+
- - /b"/b 1J.JIT£ />1f:"Fl(l>.Jer

-+- e, . .~ Nt1'£ fii.&!"ffJ<J1
{ .SF6t/IJ)
---------- --~--- ---·
g,5 - - - -
__.
--------+----
I 7t!"ftJSf{£1(K.O - -- -
1
._ a. aour.H - -o- _ 1
.
J 'r!#lf(.!EWIC;t - -
J.~,_____-_-e_-_-_--_~-------~---4~_$_E(~•U _______~_
F'/4 G·:2 .Stf/U!-1!..E Pt.A 7£ Nt7# ti."1f'£b E~&U. IJ~!!. (),/J. /.. (Q)
c~fvffiJ,£/S~IJ "F R.6Sfk.T.S IJ$ir.J~ Vl'cR.ID/IS F'VM:TIC#..16 r~ /Ct-..6u.:~
,cu•~N7 ~ th"'tb #/£..SH 1>1YrS11!1Aif 1>r 7~ PJ.ATF
-'7 ')(/t..r (J (,J. :Z~·1 (1iJ.f6~116.AJ1'0)

I
-i
--'
--·
-
-
flt. G·3 5~lJMU PlAT5 (:N Fcv~ e~EA!.. JuPP6R/S
.f1A11r£ £1...Eg.;"'1 S"1.v11"AJ 111.1w.f. 1't""'1~v~A1(' Iltt1'Jtl?(s { Je4t.1111)
151
l
Or---------,--------,--------....--~~~-::=---.-
()./ ~-------
--
(}/}. -
03'--------->-------~
·-- 5t/.1E$ 5oJ."'· ((!, f/,t.,r2J.ft>AJI(,>)
Ct.r;11~µ}
SE6UN
16 Kl6 1-fE.># .f'iv11f. !wrl/.Jr
+ £?,1.oge,/A : EJtp'f. (£.!Il.!Lr5
F/6 6.4- Hyf'£R..8"t.1c Plrf'.i4f>oLOt7> 5RcL.I... Mn1 Fl'x.£D ,fo&£s

ccPl'fP.41<.tStJAI O.t:° SoLl)/ IONS ()N UNT~E LIN& Rzo8LEAf 2
()
--.,
"'~
-.::,
......
~
~ O·OI
"~
"
~
c:;
f"f 6. 0·5
~
/, .
/
·001 ----------'+-------------+---------+------,~"-T-------+-
v .
,//~~
--
---------~·-·---·
----
\..!.J
/" .~ L
OOit===;;i-::e::-~-====::.;···t=-==-=-=-~--~...;.--:e=::-~~~=------,-?i"i..JEif
-rr- :t.
s
-r1A11rr
_ _ _...___ _ _ _ _ _ _ _
"
1&1i1..,.;.
61.6"'11NT
,, ~
~-· MESK(jf&tN
i•·r~ .. I
-------<
.r:i6 6· b
t
o <
)\
~ "I
~ I-.
·~ 50
t-
!--'
~
~
~
~ ~ \J1
"'...
~
'U
$)
~
~
~
['$
ft."""
'II'! :\
~
~
.Q
">
~
'l 1·5 ~ '
<l
'.I.I
\:. ~
•41 .._
ti.I. ~
~
. ______,____ L__1.~- - - - ____________ -~ /. S°¥/'I· fes11f1. Vy&A* ~r:>£L ~
'><
l- !2. .f. E. 'htrt 1'.1~i ( lb•l~rtt4 '4
g. " p (f-$ • )
N ---------- 10
FIG. o.]
--+-------~~--
f-'-==-'---+---'-l--=f_<t.:..c..,~-"'~"':'!'!"_lFS0,1'$ • W&.-:- MM~l.

i f2 cui.v~~ 1i111~ Aiqr..ystf fL6!UMS ' ' " ~,,,.
~ 5 " • S1mfol'f ;,,ff-1•1 Elf's
'<ii
~ ~ldO + ,, fL&I~ EtU ~,,,,,.
~ '""''- "4 Ft.~., 11 • s,,,,p, Jiff"'"' a,.,
4~VE/!; PLAT.If 1P6N.tSATli1A/.

SHELL CEA! r..<.E LIIV.£ (!A'S'
154
~ 5 10 11
I I
]: ·00!251--------+------l---+--- ----~-__,,,_
1 1
~ L I
~
~
~
"'>:.
~
~ ·P"5"1--- --+-------+-----~- __ _..,,o~
~ I
~ I I
l~
--.1
·POJ5"1-----+------+---_J____
I
-i---+-
I
~
V\
1
1 ~
I
SHELL l~kt-1.S~TfOA/
,
1
{i.2) CtAv•t:> ~r{ 1
1
_ _ _J_~_J______j _______
I I I
J_ __
, PIO'-----~--
/.I yfEll..dl>/1.1 ~ f1tt:A804.Dl.l>

'
I 3 +
I
II
• I
';'
~ P·I 1-----~-;----- _ _ _ J_
I
'-.J I
k
~
~
"
. 'I;:
CJ·~
~
~
.()
O·'/J
Au 6. /{)
155
If
I
I
/:J. I
I
I
/0
8
H6 '"' THE S/2.N!it:ET /NIM<.M'S
No of f;fltflJLTKN~,,s £teNI ll'lr'OLV£a
0 8 /6 .2+
f!G. . 6.J/
15'6
'4
3
~
~
~
.. 1
""-l
~
t--
"'-....;
~
0
"'::.
\II 0·(
~ -I
:::...
• ia,,.lll. i:'t~l'Tti E.L;~EAJr l'UIJN)
-2 - IC>wl~ ~rU•i& S.tFFirQ.6MC:'
-o
0 ~.J.lj ~.5
·/
~
....... ·J
..~le
,,-...,
""- ·i
~
~
·II-
, /~w../2 ~J..11TI kWflA/1
~ f~1'/0 .f/AJJT£ b1~nle#t'~
~ ·S
:l
1 -b
8EAJb/AI' ~#fEAIT
~16. 6·13 $1.AfPI..'(_ .(IJ~l'P• T#b -'1Ulf£fi l'l.,,r$ ()l./fJI~ P.Altf:"DllJ.f UM 'f_
(j) 11>1/1 "/Esll ~JI.JI TE ELE"f6AJT
(g} £;tpT. /!..£U/,.7!, (&Ea&IA)
~·---
I
I
(LJ tK.PT. R.E.SIJI... T {.t:ll&"TTY)

® K.. G.
@ Loo~
•I
© FINITE 6'Lf.'1EAIT (!:lx/2 /rf£~)
® -----
;
BIBIJIOGRAF.dY
1 ..
1.. FLUGGE 9 W and GEYLIUG 9 ~ !!'. T.

A General Theory of Defo:rmation of Membrane Shells
(Memoirs of the Internatior.al Association of Bridge and
Structural Engineering, Vol. 17, 1957. pp. 33-46).
2e Parme , .A. L •
Hyperbolic Paraboloids and other Shells of Double Currature
(Journal of the Sb.·uctural Division. P:cocs. of American
Society of Civil E:J.gineers. Sept. 1956. pp. 1057-1-32).
3. VLASOV, V. Z.
The Basic Differential Equations in the General Theory of
Elastic Shells
(National Advisory Committee for Aeronautics Tech. Memo.
No. 1241).
4. Munro, J.
The linear Analysis of Thin Shallow Shells
(Journal of the Institution of Civil Engineers, July, 1961
VoL l pp. 291··306).
5. Amba:r.tsumyan, S. A.
On the Calculation of Shallow Shell
(National Advisory Committee for Aeronautics Tech. Memo.
no. 1425).
6. Flugge, W. and Conrad, D. A.

A note on the Calculation of Shallow Shells
(Journal of Applied Mechanics, Dec. 1959 . pp.683-86)
/59
7. Bouma~ A. z.
So:ne Applications of the Bend:i.ng Theoix regarding nouJ)ly
Cured Shells
(P:t:oCo on the Theory of Thin Elastic Shells. Delft Aug .
1959. VP• 202-239).
8. Apeland, K. and Popow9 E. P.

Analysis of Bending Stresses in Translational Shells
( Pr:ocs, of the Colloquim oc>. Simplified Calculation Meth•icls.
Brussels, Sept. 1961. pp. 9-43).
9o ApelMd K.
Stress Analysis of Translational Shells
(Procs. of the American Society of Civil En~~ineers? ·
Engineex·ing l11echanics Division, ]'sbo 1968. p:p. 111-139)
10. Reissner, E,
On some a.spe".}t of the Theory of Thin Elastic Shells
(Jou.r.!1al of the Boston Society of Civil Engineers, VoL
42 19?5).
11. Vreeden~uygh, G. J.
'I'he Hyperbolic Paraboloid Shell and its Mechanical Properties
(Phillips Teclm:lca~ Review No •. 1 9 V«Jil,., 20, 1958/59)
12. Loo:f,H.W.
Eenvoudige FoiY,'!Jules voo:r de Buigings-Storingen in H"yperschalen
die volgena beschrijvenden zing begrensd.
(Rapport 8-61-3-hr-l Stevin Laboratory, Delft, March 1961)
13~ :Proceedings of the Symposiu.111. on Shell Research
(:CeHt ~ Augast 1961).
(a 'i
• I
RJwe ., R. Eo
,
Te8ts on Four Typ!~s .::.f .5ype:r:bo1ic Shells
Statics of Hyperbolic ParabGloidal Shells s·bu.died by !'leans

cf models
( c) 1~1mro ~ J. and Ahuja 9 :B. M.

An Inves·tigatio:i of the Strain Distribution in Reinforced
Concrete Shallow Thi.n Sb.ells of Negl':t.tive Garosian Cn:cvdt'Ut'P.
14~ Jones, Lo L.
'.fosts on a One-tenth Scale Model of a Hyperbo::.ic
Pa~aboloid Shell ~oof
( Magazine of Concrete Research, 1'-e.rch 1961).
15. Ahuja, E, M.
Stress Distribution :L:n Sh.ells cf Negatitre Garssian Curvature
(Ph .. D Thesis, London Ur.iversity Oct. 1961).
16. i'j.moshe:::.ko 7 S. P.
1
·Theory of Plates and. Shells

(McG:r:avr Ilill PubJ.ishing Co.) &
17. Tottenham.1 :n.

The .Analysis of Hy?erbolic Paraboloid Shells
(Research Report E/,:"tR/5, Tinber Development Assn. July 1958).
18. 1iargaere, K.
Zu~ Theo~ie der Gekrummten Platte Grosse Fornar.derung
Proceedings of the 1":!.fon Inte:r.l:..s.t:!.onal Congres8 i.'or applied
Mech (1938) pp. 93-101.
/bl
19. Purcher~ A.
13<?.itrag zur ·rheory '.I'ra.gender Jn.::,,0hen
(:Dissertation Tedmische Hoo1.8chule g:rzz, 193l).
20., Tester Kc
Bej_tral? zur Be:cec:hungden Hy}Jerbolishen Pa:caboloidshale
Ingeniew ·· Archid 16 1 (1947/48).
21. Candela, F~
Gene::::al Formulae for Membrane Stresses in Hyperbolic

Paraboloid Shells
A., C, L Journal No~ 4 1960"
22.. Tottonham, H.
T~1e Anal;;rsis of H;n:ierbolic :Paraboloid Shells
I MeI!lbrane (Hone~1·i;le.=~s) f.:'l'llysis
'i'irnber Dev-elopmer~t Association Lond.on 1957.
23, Flugge, W~
Stresses in Shells
Springer - Verlag 1960 pJi~ 181-184.
240 Green A. E. and Zi;;rna, W,.

Theoretical Elasticity
Oxford Glai·end.on Pres::; 1954-
25,, Argyris ~ J. IL
EneI·gy Theo-r·ems and Structu!.·al A:::Je.lysis
Butterworths 1960~
~6. A?.:gr:~is J. H.
Hecent advances in natri:z: me,:;hod of Structural analysis
Perg8Jiloll Press 1964.
Elastfo ... plastic :Matrix Displa.c13;:r,2nt Analyc:5.0 of three
Dimensiona.1 Contirn1a Sept., 1965.
Tria:r . gular Elenents vdth linearly varying stra;i..n for the
Matrix Displacement Metl:.od Oct .. 1965.
R~info11 c~d Fields of Triangular Elements witt1 Linearly
Varying Strain; Effect of Initial Strains Nov. 1965
Matrix Displacement Analysis of Anisotropic Shells by
Triangular Elem.e11ts lfov~ 1965
Jourr.al of the huyal Aeronautical Society, 1965.
28. Argyris, J. JL
1foie:T1 E'usela(?,'e lmalysis and tbe EJ.astic Aircraft
:Butte;:worths, London 1963.
29,, Tl.A.!."!ler, 11. J., Clough, R, J., Martin, H. Ce ancl Topp, 1, J.

S~iffness and deflection analysis of Complex Structures
J. Aero. Sci., No, 23 S2pt" 1956.
30. Turner M. J.
The direct Stiffness method of Structural analysis
l~ga.rd 72 Aachen Sept .. 1)590
31. Meloah R. J.
A stiffness IfiE..trix fo:c the .P..nal:n=ds of Thin Plates in :Rending
Journal AIAA 1960.
32~ Clough, R. W.
The finite element method in pla.vie stress analysis
Proc, Second Cor..f. on EJ.ectronic Computation, A.SaC.E. 1960.
33. Adini, .i.
Analysis of Shell Structures by the finite eleTient method
Ph.D. Dissertation., Univ. of Ua1ifo:;:nin., Berkeley 9 1961.
34. llelosh, R~ J.
Basis for deriYation of matrices fm: the <lirect stiffness
net hod
Journal .A.IA.'\ No .• 1. 1963.
35. Zienkiewicz, O. C. and Cheung, Y. K.

'.Phe fi..'1.i te element Method for A:r.i.alys is of Elastic Isotropic
and Orthotropic Slabs
Proc. Inst. Civi} Ei'.g~.neers. Au5u.1;:;-I; J.964, Feb. 1966.
36., Herenikcff 1 A*
$oJ.ution of problernc of e 1.astici ty by the framework method
J. Appl. Mech.~ J.941.
37. Fraeijs De Veubeke, B.

Displaceraent a.'1.d Equilibriu,~ node ls in the finite element
method
Stress Analysis, ea. Zienkiewicz 3.nd Hulwter J. Vliley 1965.
Strecs Di::::tributi0n in some shells of negative Gaussian

Curvature with particular reference to Fdge Effects
Ph.D. Thesis, Univ. of London 1963 ..
Zienkiewicz, O. C •• ar.d Cheung; Y. K.

390 Plate and Shell problems finite ele~:ent Ilis:pl~cot!ent
method
(Univeroity of Newcastle - Unun ···· Tyne Rep. No. C/R/42/66).
164. ,I
4c. Irons, B and Draper, K.

On the Inadeqnacy of nodal connections in a Stiffness
Solution for plate bending.
(l~.· I. A. A.. Jour-nal, 1965, 3(lfAY), 1961)
41. plouch, R. N. and Tocher, J. L.
~nalysis of thin arch dams by the finite element method
l
(Int. 3ynp. on ·the Theory of Areh Dams Sou thar:1pton 1964)
42. L~, Z. A., Penzieri, J and Popov, E. P.
Finite "Slenent Solutlon for· thin Shells of Revolution.
(University of Ca2~ifornia, Herkeley, S~pt. 1963)
43. P.s11llin€", J. :t.
The analysis of Cor~2plex Ship Structun~s by the Finite
\
Element Technique.
44. Wilson, L. B.
SoLrLio!"~ of Certain Lar,:-e sets of Eq 1r.+.ions on Pe{;asus
u3ing.~atrjx tlctho~s.
(rr'J1e Computer Journr:..l, Vol. 2, 1959)

45. Baron, H. n.
An ir.vesti;:,ation into U1e uses of Structural TiMber Models
(B.Sc. :Pht"!sis, Udversit.1 of Southarrnton 19G2)
46. Chet t~,·, S.
An inves-tig-;:.lion into Lin~ar AnA.lysis of Hypet'bolic
faraboloid Sh~lls •
(Ph."J. Thesis, Univet'sity of Southarrnton, 1961)
47. Brebbia, S. A.
,. '
Pnr~bolaid Shells with par~l~ulnr refer~nce to edge effect~
(Uni Yer.d t,y of Southan::;ton Dept. Report cr;.;/2/66)

~8. Kercszte3y, L. L. )
An experinen tal investigation' in to the St:rcs:i 2~nd Dj snlace-

ment in Hype:-boloid Ti.:iber Shell Roofs
I
ANALYSIS O? A HYPERBOLIC
PARABOLOIDAL SHELL
by
MELVIN L, ~URDORF
B. s , , Kanoao State University, 1958
A MASTER 1 S REPORT
submitted in partial fulfillment of the
requirements for the degree
MASTER OP SCIENCE
Department of Civil Engineering
KANSAS STATE UNIVERSITY

Manhattan. Kansas
1963
Approved by1
11
\:./15
D~o"J,:,_:1- TABLE OF CONTENTS
~ <.n-ts
SYNOPSIS ••••••• •• ••••• •. •• • ••••••••• ••. • •• • • ••••• • •••
INTRODUCTION •• • • •• •••• • • • • ••••• • •••••••••• • ••• • • • • • • • 2
METHOD OJI' STRESS ANALYSIS ••••••• • •••••••••• • ••••• • •••
General Definition
Surface Definition
Conditions of Equilibrium 5
APPLICATION OF STRESS ANALYSIS 16
Surface Definition • ••••••••••••••••••••. , • • • • • • • 16
St re a a Conditions ••••••••• , • • • • • • • • • • • • • • • • • • • • • 18
Secondary Stresses .•.••.••.• , • , •••••••••••••••• , 22
DESIGN EXAMPLE
Shell Design • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 27
Edge Beam Design •••••••••••• , • • • • • • • • • • • • • • • • • • • 29
Detailed Drawing • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 33
CONCLUSIONS • • ••••••••••••• • •••••• •. • • • ••• • ••••• • • •• ,, 35
ACKNOWLEDGl'IENT , • , •,., .... , •• ,,, ••••• , •. •,., • •• , ••• ,.. 37
BIBLIOGRAPHY , • , •• , , • , , , .. , • ., .. , , •• , ... , ....... , .... , 38

1
AllALYSIS OF A HYPERBOLIC
PARABOLOIDAL SHELL
By Melvin L. Burdorr, 1 A. M. ASCI!
SYNOPSIS
Many articles have been published on hyperbolic parabo-

loidal shells. However to the knowledge of the author, no
complete simplified analysis covering various loading
conditions has been developed to date .
The shell analysis explained is a re•ult of reviewing
several references using the membrane theory solution with
uniform vertical loading. By the membrane theory. no bend•
ing moment is allowed within the shell, therefore the shell

acts 11&inl7 by axial forces . Applying this theory, the
basic equations are developed for one quadrant of the hyper-
bolic paraboloid . A consideration of secondary stre&ses,
which cannot be included in the membrane theory solution,
is given . The basic equations and the conaideration of
secondary atreases are then applied to a practical design
of an inverted• single support reinforced concrete hyper-
bolic paraboloid.
1 Graduate Student, Department of Civil Engineering, Kansa•

State Univeraity, Manhattan, Kanaaa .
2
A brier detailed drawing is presented showing the

dimensions and reinforcing result ing from the design of
the hyperbolic paraboloid.
INTRODUCTION
Increasing interest and acceptance by the general

public or shell roof construction haa led to the wide
usage of reinforced concrete hyperbolic paraboloidal shells .
Through their use the designers can depart from the system
or linear members confined to three perpendicular planes
to a curvilinear system for more imaginative and graceful
structures. Some advantages other than beauty resulting
from the simple hyperbolic paraboloid are economics or
design, economics or conetruction, and low maintenance coat .
By membrane theory analyais, the hyperbolic paraboloid
carries a uniform load 11111inly by direct axial compression or
tension, uniformly stressing all material or the cross-
aection, thus giving an economical design . Economy or
construction is gained by the straight-line simplicity or
forming the surface and by the repeated use or constructed
rorwus . Maintenance on reinforced concrete hyperbolic
paraboloid roofs is usually low . By nature or the material,
it is virtually fireproof. Al10 being constructed as a
3
oingle unit reduces maintenance . Other advantagea 2 relating

to settlement. wind and earthquake rorcea influence a reduc-
tion in maintenance costa.
Realizing their growing usage , the author reviewed
available literature concerning the stress anal7sis and
design or hyperbolic paraboloids in order to gain a practical
working knowledge or the simpler casea . Therefore this
paper contains a summary or aeYeral references which explain
the stress and design analysis . When a oection i• developed
mainly fro• one or two references, proper acknowledgement or
each reference will be given at the beginning or the section .
METHOD OF STRESS ANALYSIS
General Definition
A structure whose thicknesa has a much smaller magn1-

tude than its other two dimen•ions (length and width) is
called a laminar structure . The 1hell la a laminar structure .
Flugge3 cites two example• or shells which otructurallJ
behave very differently. Other references clasoify the
difference in structural action or shells as either a proper
or an improper shell. q
Structural Applications of Hyperbolic Paraboloidical

Shells, by Felix Candela, Proceedings, ACI, Vol . 26,
January 1955, p . 397 .
Streoses in Shells, by w. Flugge, p. 8.
Understanding the Hyperbolic Paraboloid, by Felix Candela,
Architectural Record, Vol. 124 No . 1, July 1958, p. 192 .
The main structural act i on or the improper shell ia
perrormed by bending and twisting momenta combined with
noraal and shear forces. An example or such is a clyindrical
shell formed by rolling a sbeet or paper into a cylinder and
aecuring the ends together. It ia easily observed that a
small lateral force applied to this shell causes considerable
deformation which is reaisted mainly by bending and twisting
momenta .
The atructural action or the proper shell ii performed
predominately by direct stress (normal and ahearing forces) ,
An electric light bulb ia an example of a proper shell. Even
though ita material ia very thin and rather fragile, a large
lateral load can be applied without any visible deformation
or thia ahell . Since the deformation and thickne11 ia very
small, one could assume that the bending and twiating momenta
are almoat negligible . Flugge mentions that a detailed study
shows this to be true .
Surface Definition
Geometrically, the shell is bounded by a doubly curved

surface . The two curved aurracea are defined as the faces
or the ahell . Tbe middle aurrace of the shell ia defined aa
the aurface paaaing midway between tbe two races. By know-
ing the ahape of the middle aurface and the thiekneas, a
ahell can be geometrically deacribed ,
5
Referring to the two shell examples g1ven previously,

the middle surface of an improper shell la developable .
Thia means the middle surface can be formed by an initially
plane flexible sheet without stretching the sheet at any
point . The proper ah<tll h... a non-developable •iddle
surface, which cannot be formed from an initially plane
flexible sheet without stretching at some point .
Conditions of Equilibrium
Using the approach of Flugge 5 to define the stresses

in a shell, tirat an x, y, z coordinate system aa shown in
Fig . 1, is aet up to describe the position of a point on
the middle surface. The x and y coordinate axis 1a tant1ent
to the middle surface with the z axis being normal to the
middle surface. Then an element is cut from the shell by
two pairs of planes whicb are normal to the middle surface
of the shell.
The forces formed by removing the element are reaol ved
into three components positively si~ned as shown in Fig . 1.
Referring to Fig . 2, each of the three forces shown in
Fir . 1 is the resultant of either normal atresaea (crx, ~y),
shear stresses parallel to the middle surface C1'xy-'I"yxl,
S Ibid . , pp . 3-9.
ffil. STRESS RESU...TANTS AND LOADS
z
t
I
~i
)
\FK3.2. STRESSES ACTN3 ON A DFFERENTIAL aEM:NT
8
and shear stresses normal to 1t ('!'xz' TY•). Therefore the

three forces can be termed stress resultants.
Us1ng the terms illustrated 1n F1g. l and F1g. 2, the
stress resultants are defined as followsi 6
it/2 r -z it/2 rx-•

TX • (f {..L-)dz, Ty • <fy<--> dz,
t/2 x r 1 t/2 rx
Txy •
j_t/2 r 7-z
'I' {--)dz, Tyx
{t/2
•
rx-•
'I' {--)dz,
t/2 xy ry -t/2 Yx rx
Qx •
it/2 r 1-z
'I'xz{--)dz, Qy •
it/2 rx-•
'I' {--)dz,
t/2 ry t/2 yz rx
M
x
it/2< (-
• r 1-z
-)zdz, My
f_~/2~ <--lzdz,
•
rx-•
t/2 x r1 t/2 1 rx
Mxy
{t/2
•-
ry-z
'l'xyC--)zdz, Myx •i~/2'f (--)zdz,
rx-•
t/2 r1 t/2 yx rx
The factors { 1 -•) and {x-•) are assumed to be equal to
6 Theory of Platea and Shells by s. T1moahenko, and

W. Wo1noYsky - Kr1eger, p. ,29.
unity aa the thickness of the shel l is generally small aa
compared with the radii of curvature .
Since the load on the proper shell is carried mainly
by normal and shearing forces, the moments may be assumed
negligible in the streas analysis . Thia assumption leads
to the membrane theory or shells . Fig. 1, therefore shows
all the necessary forces acting on the sides of the shell
element . By moment equations concerning the element. it
can be proven7 that Txy•Tyx and Qx•Q7•D . With the•e
assumptions, the ten unknown stress resultants are reduced
to only three unknown stress resultants (Tx• Ty• Txy> •
From the three equations of static equilibrium, the three
unknown stress resultants can be solved .
Using the same coordinate axis and transforming the
actual forces or the element into forces acting on a
projected area as shown in Fig. 3, considerable simplifica-
tion or the analysis can be gained . The following portion
or analysis follows closely that set forth by A. L. Parme . 8
The stress resultants Ty• Tx• S (S•Txy•Tyxl represent
the skew forces being tangential to the element'• aurface,
The stress resultants Typ• Txp• Sp represent the projected
forces in the xy plane . The skew forces T 1 , Txt S are
1 Stresses in Shells, by w. Flugge, p . 9.

Hyperbolic Paraboloids and Other Shella of Double Curva-
ture, by A. L. Parme, Proceedings, ASCE September 1956,
82 ST5 .
10
l
Tj;p
FIG. 3. Sl£LL ELEMENT AND ITS PRCU:CTD.I

11
measured in pound• per unit lenf!'.th or line ele11ent. 87

geometry it ia 1een that
dp coif• dy (la)
dq co•~· dx (lb)
Multipl7ing the normal forces by the 0011ne or the angle

between force and the xy plane yield• the x and y component•
ot such. Th• x component• are
TxpdPTxcoa<t> dp,
Spdx•S ooa<t> dq,
which by aubat1tut1on becomes
T dy•T g.2.!..t. dy,

xp Xcoa qi
Spdx•~ dx,
co~~
and the y component• are
Typdx-Tycoa 'I' dq,

Spdy•S coa 'I' dp,
wblcb bJ aubat1tut1on becomes
T dx•T COi ~ dx
J'P 7~ '
spdY·~m ~ 41.
12
The new atre11 resultants in terms or the projected

element are
Txp • T £2il_ (2a)

Xcos If'
Typ •T~
YCOS lj>
(2b)
SP • S. (2c)
Deriving the element'• area dA in term• or the projected

element area dxdy gives
dA -~ ~sin Ill• dxdy(l-sin2'f' sin2d>)l/2 •

cos <1> co• 'I' cos cp cos 'fl
Aaouming only a vertical distributed load per unit area Wz
acting on the shell element dA, the transferred distributed
load per unit area Wzp acting on the projected element is
~. (l-sin2 If' sin2 d> )112. (3)

..,.-- cos 'I' coo $
The load w,p can be assumed to be uniform for most hyperbolic
paraboloid shells of moderate rise . The sin2 'f' sin2tP term
can be neglected since it will become small for the moderate
rise case . Using this assumption Equation (3) becomes
wzp • wz 1 (4)
cos 'f cos 4>
With the forces acting on the element varying from the
one race to another, equilibrium of forces in the x direction
13
expressed in terms or Txp' TYP and SP givest
a_!u~ o (5)
ax+ay.
Likewise equilibrium in the 1 direction yields
a...:u + ~. o.
() y <IX
(6)
Equilibriu• in the • direction involves all or the shell

element•a etre1s resultants as they all contain a vertical
component. The Tertical component or the normal force Tx is
and by substituting ror Tx and dp from Equations (2a) and

(la) yields
cos 4' sin~

Txp .A.11.d cos 'l' • Txp tan~ dy • Txp~
u
dy,
The vertical component acting per unit or length along the

y axis 1s
A similar expression tor the vertical component of T 1 per

unit or length along the x axis i• round to be
The vertical component or the shear force along the y axis is
S dp sin 'f'
and by substitution from Equation• (2c) and (la) gives
S !!.L_ sin o/ • Sp tan

p cos 'l'
ljl dy • Sp #" , dy
which per unit of length along the y axis is
Similarly, the vertical component or the shear force along

the x axis is
Taking into account the differential increments of all

the forces, the equilibrium equation in the z direction yields
{7a)
By differentiating the products, Equation {7a) becomes
;Jl'xo + &)e! _ {~ + ~)£.! _ Wzp, {7b)

-fax <) y ax ay ax ay
Substitution of Equations (5) and (6) into {7b) gives
(7c)
15
The three Equations (5), (6), and (7c) are the basic
membrane theory equations for this case, Using the defined
shell's middle surface, a direct solution or these equations
may be tried. Since we have three "dependent• variables
Txp' Typ, Sp, each depending on two "independent" vari-
ables x, y, the solution becomes very complicated. By
assuming the stress resultants described by a single stress
function F of x, y, instead or Txp• Typ• Sp, the equations
are reduced to one second order equation. The stress
resultants are derived from the "Airy stress funct1on• 9 F by
differentiation, as follows:
Txp ·- a2p
ay2
(Ba)
• a2 '1t'
Typ (8b)
ax2
SP --~
axay
(8c)
By substitution, Equations (5) and (6) are satisfied and

Equation (7c) reduces to
(9)
For nost cases, the algebraic solution or this differential

equation is quite difficult, however for a h7perbolia para-
boloidal 1hell loaded under uniform loading, the solution
1a fairly simple,
9 Advanced Strength or Materials, by J , P, Den Hartog, p. 174,

16
APPLICATION OP STRESS ANALYSIS
Surface Def1n1t1on
As previously stated, the hyperbolic paraboloidal shell
is bounded by a doubly curved surface. This surface may be

defined in two ways. either as a surface or translation or
as a warped parallelogram. The surface of translation can
be visualized by translating or moving a vertical parabola
or upward curvature over another parabola with downward
curvature. The translating parabola's plane 11 perpendicular
to the other parabola and ls at all times parallel to !ts
original poa1t1on, thus giving a saddle-shaped surface,
The second method or description ls described as a
warped parallelogram and ls shown in Pig. 4. The remaining
portion or this report, including application and design
will be based on this method or description. This surface
is constructed bJ moving a straight line along the x axis,
remaining parallel to the yz plane at all times. but rotating
about the x axis as the outer end or the line slides along
the straight line ABC. A similar generating line can be
established along the y axis. The resulting surface formed
by the grid of straight lines may be described at a point by
the intersection of two such lines contained in the aurrace,
17
E' c
FIG. 4. Slff'ACE DESCRPTION

18
The "warped parallelogram" connotation for the surface can

be visualized as the horizontal plane A1 C1 E1 G1 1B warped by
vertically depressing corners A' and E' to new positions A
and E, and allowing corners G' and c• to rise vertically
to new position• G and c.
Streaa Condition•
A basic quadrant ABOH of the •warped parallelogram•

surface taken from Fig. 4 is shown in Fig. 5. Continuing
as A. L. Parme10 hae done in his paper, any point on the
aurrace can be described as a function or x, y and z.
Referring to Fig. 5
Likewise,
*. f or c • RZ • (lOa)
(lOb)
Substituting into Equation (lOb), the value of c found in

Equation (lOa) gives
z • c~ • chf>i • xy C~al •
Letting k • ~b ,
z • kxy. (11)
lOHyperbolic Paraboloids and Other Shells of Double Curvature,

by A. L. Parme, Proceedings, ASCE September 1956, 82 ST5.
19
)(
FIG. 5. BASIC QUADRANT

20
Upon substituting Equation (11) into Equation (9), the

second differential of Equation (11) equals zero and the
remainder yields
a2F
- 2 am k. - wzp
which by Equation (8c) reduces to
sp -~.
2k
(12)
Using the above value for SP' it is easily seen that
~-
ax
and
3Sn
n· o.
Substituting these values into Equations (5) and (6) yields
(13)
By Equations (12) and (13), it can be observed that the

edges of a hyperbolic paraboloid shell resist only a uniform
tangential shear, since the normal forces equal zero.
Using Mohr's circle, this pure shear resolve• into
principal stresses of equal and opposite magnitude acting
on sections at 45 degrees to the shear plane (see Fig, 6).
Identical parabolic arches are formed by the sections taken
•5 degrees to the coordinate axis aa ahown in Pig. 6. The
parabolic arche• or parabolas parallel to OB curve downward
while those at right angles to OB curve upward.
21
z
0
(B)
FIG. 6. ~ABOUC ARCHES
~--~-~~--.......- --~-- __., -- ··--- -----

22
Assuming the uniform load is equally divided between

the two sets or perpendicular arches as shown in Fig. 6,
the arches parallel to OB exert an outward thrust R on
the edge members by the nature of their curvature, while
those perpendicular to OB exert an inward thrust R.
Since the arches have curvature. they exert both
vertical and horizontal forces at their ends on the edge
members. The net effect as shown in Fig. 6-b 1s that
the normal force components to the edge member cancel out
and the vertical force components cancel out leaving only
the pure shear acting along the edge member. Fig. 7 •hows
the net result or the pure shear acting on the edge members,
or in this case. the edge beams, tor two quadrants forming
a portion or a complete structure.
Secondary Stresses
In using the membrane theory or analysis, only the

equilibrium or forces has been considered, neglecting the
compatibility between stresses and strains. The following
discussion on secondary atreeaea is baaed mainly on P.C.A.'a
publication of "Elementary Analysis of Hyperbolic Paraboloid
Shells", 11
ll Elementary Analysis of Hyperbolic Paraboloid Shells,

Portland Cement Association, 1960,
23
FIG. 7. 9-£AR FORCES ON EDGE BEAM

As the shell becomes flatter (less rise to span ratio)
axial strain within the perpendicular parabolic arches
becornea important. This axial strain will set up secondary
bending moments within the arches . For the usual rise, h/a •
1/5 or h/b • 1/5, the effect of axial strains is relatively
small and can be safely neglected . When the shell flattens
out, 1.e. h/a decreases, the effect or axial strain must be
investigated. Using a two-hinged parabolic arch subjected
to uniform load as a comparison to the parabolic arches in
the shell, the horizontal component or the reaction ror a
given span decreases as the ratio of rise to span decreases.
This decrease in the horizontal component is coupled with a
subsequent increase or axial strain, thus resulting in
secondary bending moment . The limiting value or secondary
bending moment would be the case or zero rise in which it
would equal the simple-beam bending moment .
Fig. 8-a represents the results or a study12 concerning
secondary bending moment . This graph is also based on the
tact that the curved surface can be represented. by sets of
perpendicular parabolic arches, whose shearing forces and
normal forces on the two opposite races can be neglected.
Also the assumption that the ends or the arches are not
free to move is made .
12 Ibid ., p . 16 .
25
~I)
fl
l300.-----.,_._,.-=r---,--,-~-;-~.---,
·~
f .50
~001---+--f-+=:--lr---+~+----t~-;--.
~
~50 1---l--li~::--+~..-t-~-+-~;-~r---i
~
<:;;, 100 1---~~~-f~orl-~;:-t--;--..:t::---i
~
1so~~~~~+:::::::'"i-=:::=-f"'-=:::t:=::::I
~ o0~~LJ:J~~~~~~~~iij.
10 20 30 40
~
(B)
ma SECONDARY MOr\ENTS AND STRESSES

26
Using the appropriate ratio ot ~ coupled with the

dimensionless quantity ~. the secondary bending moment can
be round for a strip or L length. In studying Fig. 8-a, it
ia seen that aa f decreases to zero, the bending moment
approaches the simple beam bending moment, which is not in
line with the membrane theory analogy. Aa f increases, 1,e,
the strip of L length is farther from the corner, the
secondary bending moment decreases, the rate of which is
a function or ~· As the ratio or !l* becomes larger, the
rate or decreaee or the secondary bending moment becomes
larger, Using a shell thickness of 3"• the normal ratio
or rise to span or 1/5,
ht • 1 p~~
iO x ). 0 • 010 •
From Pig. 8-a, for ~ • 0.010, and a distance of 5 feet or
more from the corner, it is observed that the secondary
bending moment ii becoming less important.
Pig. 8-b, representing the secondary bending moment
as secondary flexural stress, brings out the importance of
curvature on the magnitude or secondary streaaea, 1 3 For
example, a square shell whose thickness is 3 inches and
~ • 1/5 has a secondary stress in terms or w equal to -o,
at a point 5 feet from the corner. Another square shell
l3 Ibid., p. 16.
27
with the same thickness and at the same point from the
corner, except having ~ • 1/7 has secondary stress in
terms of w equal to 115. From comparing the two values
of stress, the importance of curvature is quite evident.
DESIGN EXAMPLE
Shell Design
This design is for a square roof unit having a shape

commonly called the inverted umbrella. Fig. 7 shows a
partial unit with the exterior edges being horizontal. The
plan dimensions are 30 x 30 feet with a 3 foot vertical
rise. Since adequate reinforcement coverage controls the
shell thickne•• rather than stress limitations, a shell
thickness or 3 inches will be used. This gives a uniform
dead load of 37,5 psf for the shell. Live load is equal
to 30 pef plus 15% of the live load to allow for the weight
of the edge beams. The total design uniform load la
therefore 72.0 psf .
From Equation (12), the pure shear per foot acting
within the parabolic arches of the shell is
sp • :t ~
2k
• :t !'.!.P. ~ • :!:
2 h
72 (15) (15)
2 (3)
+ 2700 l /f
• - b. t.
28
The compressive concrete stress is
f • ll.22. • ± 75 psi.
c 3(12)
No steel is needed ror concrete in the compressive parabolic

arches, however a nominal amount should be used tor tempera-
ture and shrinkage stresses. The tensile concrete stress is
also 75 pal. Here again, no steel ls theoretically needed
for loading stresses, however steel will be included to
take care or all tensile forces. The amount being
A8 • ~ • 0.135 oq. in. per ft.

20,000
Recalling that the parabolic arches are rotated 45 degrees
from the shell'• edge, the reinforcing should be placed
likewise. If for ease of placement, the reinforcing ls
oriented parallel to the shell's edge, the amount of
reinforcing must be subsequently increased. For this
example, the amount of reinforcement placed parallel to
the edge• equals
As • 0.135 x 1.414 • 0.191 sq. in. per ft.
Therefore No . 3 bars placed at 7 inches apart are suffici-

ent. This same pattern will also be used for the
compressive area.
29
Considering secondary •tresses in the flattened

corners. Fig. 8-b showa the maximum secondary stresses
occurring at x/t equalling approximately 22 for
~ -~l . 0.0033 .
For the 3 inch thickness, this gives the L length strip
placed out an x distance 5' - 6" from the corner . For the
D. L. plus L. L. or 72 . 0 psr. the maximum secondary etress
equal a
f
c
• 175 x 72. 0 •
144
a1 • 5 pal
•
While the magnitude or flexural stress is not critical,

five No . 3 bars placed 45 degrees to the edge beams will
be added for tension in the immediate range of maximum
flexural stress.
Edge Beam Design
According to the membrane theory analysis the horizon-

tal edge beams are loaded by pure shear from the ahell.
causing tension ranging from zero at the corners to a
maximum value at the center . Maximum tension in the
horizontal edge beams is
H • 2,700 x 15 • 40,500 lb .
30
for which the area of steel required is
.
A • .
~Mgg • 2. 025 sq. in .
Two No . 6 bare and two No . 7 bars are suff icient .

Excessive corner deflection or the horizontal edge
beam has been experienced ror this type or structure. 14
Recalling the discussion of secondary stresses, as the
shell flatten• out near the corner• the validity or
membrane theory analysis decreaeea . The parabolic arch
stripe approach simple beams in which the horizontal
edge beams take the vertical reaction. This in turn
causes a downward deflection. To offset this deflection,
the centroid or the edge beam is placed above the applica-
tion or the shearing action from the shell . Thia results
in an eccentric loading which tend• to lift the end or
the edge beam. In addition, a small amount or steel can
be placed in the top or the edge beam to take tension
stresses .
The sloped edge beams are designed for axial compression
resulting from shearing action on both sides ranging from
zero at the outer edge to a maximum at the valley of the
shell . The max11tum amount ia
2H 15 . 30 • ~ x 40,500 x 1y530 • 82,620 lb.

15
14 Building for Economy with Hyperbolic Paraboloids, by

G. Madsen and D. Biggs, Journal ACI, Vol . 32, No. 4,
pp. 373-383 .
31
Since the •loped beam is subjected to axial compreosion with

amall eccentricity, it may be designed by column formulas,
Another approach1 5 to the design would be to consider the
sloped beam aa a flange with the shell acting as the web.
This approach permits using allowable compressive stress
in flexure. Since the membrane analysis does not include
the effect of strains occurring in the edge beam, the
conservative column formulas will be used to reduce
possible strains.
The standard formula for lateral tied columns ia
aubstituting PgAg • A8 , the above formula becomes
from which
Ag• ~225 f'c + Pgf 0 )
Using a percentage or steel Pg • 0.01, r 8 •. 20,000 psi and

r'c • 3,000 psi, the required gross area at the valley for
the sloped beam is
82 620
Ag • 540 + 20,000 x 0,01 • 111 ' 6 aq, in,
l5 Elementary Analyo1a or Hyperbolic Paraboloid Shello,

Portland Cement Asaooiation, 1960,
32
For an 18 inch wide beam, the depth required ie
d • 18
111 . 6 • 6 , 2 inches,
Due to possible unsy!Rlletrical loading causing increased

bending atreaeee, an increaae or depth to 9 inchee will
be uaed , The required As for the eloped beame equals
A8 • 0 , 01 x 111 . 6 • 1 , 116 sq, in,
Four No . 6 bars will be used ,

33
5· "'3@ 6"cfrs. incr. IJvm

1:0" fo 10~6" plbced htlow
"3@7" m11f.
~·3•7"aachwd!J
mw "in3"8/ab
c-*6•20~0" -..,.-;1>1
~*6•30~0"
4'' Roof drain

A
f.._..._lHfi~====~====:E;=rtl..=:J
S!lction A-A
FIG. 9-A. DETAILED DRAWING

3Ifi
'*"cir.
S12ction C·C
&ction 13-B
FIG. 9-B. DETAILED DRAMNG

35
CONCLUSION
By membrane theory, the doubly curved surface or the

hyperbolic paraboloidal shell will support a uniform load
mainly by direct axial compression or tension forces. The
edge beam stiffen• the shell at the boundaries, and 1o
loaded by pure shear from the shell, according to membrane
theory. However, since stresses developed by this theory
are dependent upon shell curvature, sections or the flatter
portion or the shell near the corners tend to act as 11mple
beams, thu1 developing bending atreaaea. Accompanying the
bending stresses ot' the shell is a vertical reaction on the
tension edge beam. Thia in turn produces bending momenta
in the edge beam.
Analyoia by the membrane theory t'or stresses in the
ahell under uniform load doea not involve complicated
mathematics. However, under unsymmetrical loading, the
mathematics becomes very complicated. With unsymmetrical
loading and even uniform loading to a leaser degree, there
is some question as to how much or the shell is conforming
to the assumptions made for the use ot' membrane theory. llo
allowance by the membrane theory is given t'or compatibility
ot' stress-strain relationships. Consequently, any deflection
which the structure obtains will introduce a need ror atreas-
atrain consideration. The uniformly loaded structure
35
treated in th1s paper will have alight derlection, especially

near the shell boundaries . However the stress analysis
resulting rroa this deflection ia beyond the scope or this
paper.
In studying the available literature, the author
railed to uncover a complete general solution of hyperbolio
paraboloidal shells. Several approaches have been uaed. 15
Model studies with simplified mathematical analysis appears
to be one or the simpler, while a more complicated approach
would be to apply a combination of membrane, bending,
ultimate load, and buckling theories to the particular
shell being studied.
15 Thin-shelled Structures, by D. P. Billington, Civil

Engineer, December 1961, p . 57.
37
ACKNOllLEDGf'ENT
The author wiahes to thank Dr. J. a. McEntyre,

professor or the Department of Civil Engineering, tor
bis help and guidance. Also the author expresses sincere
thanks to his wire, Charlene, for typing the report.
38
BIBLIOGRAPHY
Bleich, H, H, and M. a. Salvador!

Bending Moments on Shell Boundaries, Proceedings
A.s.c.E. 85 ST8, paper 2223, Oct. 1959, pp, 91-101,
Billington, D. P,
Thin-Shelled Structure•, Civil Engineering, Dec, 1961,
p. 57,
Candela, Felix
Structural Application• of Hyperbolic Paraboloidal
Shells, Journal A~er. Cone. Ins. V 26, Jan. 1955,
pp. 397-415.
Candela, Felix
Underatanding the Hyperbolic Paraboloid, Architectural
Record V 124, No. l, July 1958, pp. 191-195.
Design of Cylindrical Concrete Shell Roof•, A,S,C.E,
Manuals of Engineering Practice, No, 31, 1952,
Elementary Analysia of Hyperbolic Paraboloid Shell•,
Structural and Railway Bureau, Portland Cement
A•sociation, 1960,
Flugge, w.
Stre•sea in Shells, Berlin, Springer-Verlag, 1960,
Gerard, P. A.
The Analysis or Hyperbolic Paraboloid Shell Roofs,
Trans. Eng, Ins, Canada V 3, No, l, April 1959,
pp. 32-42.
Hartog, J. P. Den
Advanced Strength of Materials, New York, McGraw-
Hill, 1952.
Madaen, Gordon and Dutton Biggs
Building for Economy with Hyperbolic Paraboloids,
Journal Amer. Cone, In•. V 32, No. 4, Oct, 1960,
pp, 373-383.
Parme, A. L.
Hyperbolic Paraboloids and Other Shella of Double
~:~~~t~~;~.Proceedings A,S,C,E, 82 ST5, paper 1057,
T!moshenko, Stephen
Theory of Elasticity, New York, McGraw-Hill, 1934.
Timoahenko, s. and w. Woinows ky-Krieger
Theory of Platea and Shella, New York, McGraw-Hill, 1959.
ANALYSIS OF A HYPERBOLIC
PARABOLOIDAL SHELL
by
MELVIN L, BURDORF
B. s., Kansas State University, 1958
AN ABSTRACT OF A ~ASTER'S REPORT
submitted in partial fulfillment or the
requirements ror the degree
MASTER OF SCIENCE
Department or Civil Engineering
KANSAS STATE UNIVERSITY

Manhattan, Kanaaa
1963
Approved byr
J.aJor Professor
Many article• have been published on hyperbolic
paraboloidal ahells, However to the knowledge or the
author, no complete a1mplif1ed analysis covering various
loading conditions haa been developed to date.
The shell analysis explained is a result of reviewing
several references using the membrane theory solution with
uniform vertical loading. By the membrane theory 1 no
bending moment is allowed within the shell, therefore the
shell acts mainly by axial forces. Applying this theory,
the basic equations are developed for one quadrant or the
hyperbolic paraboloid. A consideration of secondary
stresses, which cannot be included in the membrane theory
solution, la given. The basic equations and the considera-
tion of secondary stresses are then applied to a practical
design of an inverted, single support reinforced concrete
hyperbolic paraboloid .
A brief detailed drawing is presented showing the
dimensions and reinforcing resulting from the design of
the hyperbolic paraboloid.
'lliE STRESSES DI HYPEROOLIC PAP.AOOLOID
SHEUS USIJ."'(j 'IliE MEMBAANE 'IliEORY
by
Yousi! Kellow Dawood

B.S., University o! Baghded, 19$9
subllitted in partial fulfillment of the
MASTER OF SCIE!CE
KANSAS STATE UlilVERSITY

Ma.Mat tan, Kansas
1965
Approved bys
LO
?..&b<t:,
4 ~ .:
I q"
DJ 7 I TAIU OF CCllTEnS
c. 2.
lWCPSlS • • ••••••• .. ... . .. . . . . . . . . .

' • • • • • • 1
• • • • • • • • • • ••• • • • • • • • • • • • • • • • • •• 1
•••••••••••• •• ••• • • •• 7
ES FOR A OOFOltlLY DISTRI11JtEO LOnL> • • • • • • • • •• • • 11
. ..
ltUS • • • • •• • • • ··- 19
DESillJ OF 'Di~ OF-OIJ:O VAULT • • • • • • • • • • • • • • • • • • • • • • • 20
lSCUSSlOI OF .. UL1S • • • • • • • • • • • • • • • • • • • • • • • • • • S.3
C<X.LUSIC. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • Sb
•• • • • ••••• • •••• • •••• • •• • •• •• • SS
by
YCA11 lt it. ood
Syncp•lt
"lhlt tr l:l::l • dis $l¢ft o! th• ll•ltatlOUI ot th• theory.

Defhution •. gtOMtry O'! th• h:,ipet'boltc perabOlota sur!'lce J.t• d.ltcuH o
The r1 ten o! th• part 1•1 <11!f t.1•1 a u-,e otr 1 ln d0\bly-
c !4'V stl'U'Jturn 1a prn ed. Ula• ot a tre ~t!on in caleul:at.lng •
bi'tne str• lt given. Eq:atlo:w o! 1tt•u ot •~ ec tel cH• are
d r!..:tld. A atutr o! the e: •?¥1 Ull of th e typ• ot 1tr\l:tUffl 11 • •·
'!he ded;n o! • h)tpefl>l:>llo nbololdall grolntd vault la t.akc."l • en 1: 1•.
DTlQllCTIOI
Th• IC tane ~eery 1n m1ny cNat glveit fi cl »1' lnslght into the 1trc11
dittrlbutlcn ot oubly curved ~Ills. It 1s • good by-pGH to avoid the
•l•borat. c1lculat1on1 •J¥:0Uftttrcd ln Ulln th• lng theory.
FlU;s•l ntlonld that 1htl11 mich are e;g-eh 1d or eltctrlc-bulb-

haptd. arc much strca er truan cyllndr!eQl on •· lb• tvo toraer c•n 11lthtt1rd
r ... rYJlbl• fore:., without bending •ni vithout undtr~olng vlalbl• de!orottlons.
1hcre!ore 1 the mcch1ntsa ct the loed-corry!nG cepecity of e\l:h shell• ccnsl1t1,
11nt111lly, ot nor ·1 t:d '11c1 r !nz tort c only.
1tres1
--
T1JQC»henko2 ct al. stated that 1n
can be neglected,
ru prabl- ot shells th•
only ttr scs du• t.o stnln !n th• alt!dl• au.r•

ndl~
fie• ot th• lh•ll• nHd to bl cmalder d. Th• doubly curved 1h1lll hW• been
can1ldcrcd 11 ex~les of th••• proble• \here bcniing atre1H• hlV• ntgllolhl•

el!ect.
C•nd•1•3 Hid that equaUcwa o! •*8n• 1treH 1 arc 1uff lc lently eccu•
rate to dcterain• th• 1treHa ln oubly cwrved 1urr1e... That 11 bee• • th•
d1lor•tlon1 In 1w:h 1tructura are very s:all. FltX\11'• 1tru111 are po11lbl•
only 1t 11r;1 dctct•tlon1 oscur. However, bencSlng atHHN •y bl expected
ln 10M ca111 ln th• vie lnlty ot th• eds;• beHI •nd at ••11 zon• nar th•
·~ort.a.
lwtlhela Fl • gc, Strq ln Shtll!, Springer-Verlag, Blrlht/Oottlnge/

Hdddberg, 1962, PP• 6-§.
2s. Tf.aoebtnko ands. VolnOW1~-Krle;te, Dl•o1} f.Jl§ttu am Sh•l11,
Ml!Graw..fflll look CCll98f\Y1 lac.,
2nd !d., 19.$9, PP• 32 •
lrlllx Candela, •strlllture Appllc1Uon of HyperboUe P•rtboloidlcal
Shills," ACl Joyrnal, Jen. 19.SS, Vol. 26, lo. S, PP• .392•
2
0
·. ~
Fig.1- DEPENDENCE OF THE EQUILIBRIUM OF THE MEMBRANE

FORCES ON THE ~OU NC>ARY CONDITIONS.
Fig . 2- EQUILIBR I UM OF MEMBRANE fORCES UNDER

.' A CONCENTRATED LOAD.
•'
3
cording to t"• eh, oubly curved ccrv:rete •11• vi th
c<m.lnu loid pdr ip lly by direct st.r HI. He also • ntloned tht pOI•
11bllity of th• ccci.:rre~• o! locellzed ~ing -'*lits n er th• ed~ts d..c to
th• ettect ot th dlspl c ot th• .. rs• but for th• est p•rt the
r.hdlt are tree o! £1~1.tHl tore . •
Uou~cr, the j.lSti!ic11tion o! th• ue;:bran th~ory ;;Di its 1u;~ • 111r~
cl tly C\.tlnCCtcd. with 1a e l tor.s \t\ic:h " se:v• c1:c!ul 1t.uGy. 1h•• tac.tori
the bo· ~conditions, t:;pc of loodlng, n1 rlsc-t.o-
pen ot lh•ll ratio.
n.1 cttcct of. the boundary c0!¥1!tlons on the • '"'In th ory 1JQ
vls.slic•d DJ considtr!ng a lqlle ~:J?l• given by P!l\jgc:S. In tlg. lb

ccly v rt1ca1 re ct.lon:s trc possihl , but the ·:nlt r.o 1 !'ere• 111" In th;
ell ls Un in the d!rectlon ot the et1d1en. Thua, th• ""e de: r.t cen•
not be in equllfbr1 beceu., the react ice "Jr-I prod.ti:e~ a c on r.t that b
r.ot. l to th• rdddlt surt1e• •nd 1• not i-11necd by 1 c er !orce !n th• ehell.
The l\:Y;>ort 1hown f Flg. la would 1ntro:1u:e re ctiot\6t Into the shtll !n nccor-
r¥:• with th• n~ th~ory.
lb• CHt of tr.• go • dlsCUIHd by Can:1ela

6• c 11ld that th•
exllt•n:• ot th••• tree u h., • ;rat etftct on th• tnt1rn11 1tre11u

obtained by th t nn• theory. 1beretcre. readJt.11taent should be •d ct
J;A. L. tar••• "HyperboUe P rebolold aid other Shella ot Double Curvature,•

Pree e f1 ot the SSC£-ST$, Vol. 28, Sept. 1956, PP• 10S7/1·1l·
SA Ji'!lt!ger, l••l'Y Static• of She It
E.nglllh tflnelltlca by lrvln
Gal•n&•Y• Dodg• Coq>orat0n:1. V., 2nc1 td., I 1, PP• l-1e, 91-93·
F. Candela, "General ForllUlH tor n• Str111n In HwerboUc
Parabololdlcal Shells,• .ACI Journal, Vol • .32, lo. h, O:t. 1960, Pi>• JS.3•371.
th• internal 1tr1ss11. Su:h rndJustaent .-y dllturb th• cordltlon of equl•
Ubrl in th• bell. Accordingly, certain types of 1q>port1 thould be

provided at th• restrained cdgtl \blch hive th• ablllty to absorb the ncri-
bllam&d tore.a.
Ac:cordl to th• •bav• dl1cusslco, a ;en ral atatu nt can be •411 th•
ane theory l• ap;>liclbl• only 1t th• 1>0t.11n<1ary condltlcns are c 1tibl1
vlth the ccndltl o! quUl.btl • Hovw1:, r tult1 of testl
nbolold shell• ow a loc•lh: d ccn:entraU ot flexural 1tre11t1 n•r the
•• r rib 1tltfcncn1,8. 1b1rctore, ec 111 attention 1hould be talttn
ln the dctlg ot this part ot th• 1111. Curves that show the He ry
- . nt n ar th• • are pt11 cc1 by th• Portland Cacn Assoc latlcn8.
"nl• etfect ot the type ot load! d11eu111d by Ptl .:?• A1 own

th• load 11 p•rp lc:ular to th• alddl• 1ur.race, that 11,
In th• dlrtctlon of the z-u11. Th• ell el nt en "1lch th• !Old 11 lied
cannot posllbly bl !n equtllbrlm. It th• el · ant were ct lnflnlt••l•l • nl•
t c, lt vould be cm:elvabl• to 11U1ty the conditions o! quUlbtl . by
111 lng thlt Ix •nd w would produ:• c on nt1 of 1uUie lent nitude to
1
lane• th• 1 •1,• becl\.9t the curvature of th• lhdl 11 In th• dlrtctlon of
the cor¥: ntrat loed. Obvl ly th n, cone ntrated 1
1urt11ee are not. co1p1tlbl1 vlth the a 1 • theory. However, by 1 •tthod

•nalog to that Of th• two-bl •d para.bollc arch, lt can be proved that a
ttol'llly d11trlbutl'd load will 11tl1ty thl• condltlcns.
1n. E. Rowe, "Ttst ot Fout T.YP••

of Hyperbolic Shells,• rccce t ct th•
S)'llp l Shell 1nrch, Dll!t, Aug • .30-Sept. 2, 1961, PP• 1 - •
ortla C111tnt A11oclltlcn, 11 &1 cntary Antlytla of Hyperbolic Paraboloid
Shells,• 33 West Grand Avenue, Chicago 10, Ill. ~·
Th• rise-to-cp n o! hdl ntlo 11 me ot .t h• t. ~ortant !actors \hlch
 n cf ..ii.ii ntlo, th• •f'feet of axlal ·1ttliin1 11 unilport1nt •1¥1 can be
. .; '.
11tt1y ignored. Howw r,, llhcn th!t ratio dee~c111c,, · th• ef!cot c! axial •tnlnl
• 4 I t' • 7
beglnl to acrt e ' dord I .lnt1uenc• 'on tht .b9hav!Ol' ot th• lh•ll· 1h• d ar-
tute !n bthtvlor fr• that lmtuted tw. th• llCllbranc th•ory 1~ •nel~oua to
, • it I •
th t oeeun-1 ln the -~Vo-hinged p1rabOUC1 arch llhen it 19 •ub.Jected to .unl!

load vlth • decreue ln th~· ntto of rise to _,.n. In th• _p1rabollc: 1rch, the
. '
horlaoftttl COllpOftent dtUeuts II_ the ntlo ot· t!H to •n deer..1t1. With no
. .
rise th'e .horh: 11 : ~~on• d~rtHes to t:ero,, thw th• HCondtf1 bending due .
" .
to U1al :atfl Ina 1pproechtl tht l111plt• baft' bending aOMnt • 1hc Pert.land
+ . • •
c...nt
AHoclaUcn e wt• that "1.tn' th• rlsc-t<>-.pan ot. lhell rat.lo. it on••flfth or
f I .C,
.
aore, the effect ' of ~' extal. 1tralne
..
11 · unlaport&nt. and can ·M 1itdy ignored. In
other vord1, tor 1w:h tlee-t.o-ep1n o! ihdl r1Uo th• Mnding aoaitnt haa •
negligible ettect.
th• h)'r>•l'bollc ·p •rabOlolds au ih•ll•

' f
or anUcl81tlc dowbly c..rv1d atru:•
I
tut.. , that 11, lhdls vith n•gtttive curvature. 'lb•• tw•• ot lhella develop•
atate of 1tre11 i.vorahle to the
'
uH ,ot cC1¥!ret•. 'Ibey .anov re•rklblc res 11• '
• r
tance to cxploeton; ~r~, U~cluake. little HnUUvltY to !o tlon
1ctt1C1Mntl• . 1h~ ar1 •Clft•ic~l d.ua .to ·~dr ruled sur~a \blch •them
"' ~. "' • ·' t • - .,. , '.' .. • .
11ore ... uy. tor•d by straight plink 1'.or.t. Also. th•H lhcll1 give the archt•
~ ' ' .
tect.I I chance to .depart frca th•: conventional prectlee ot beea ind col
fl ~ • ' I
"' • t
building• to aore L._ghwUve •nd gr11<:etul m;>u of, ttru:turu. There!ore,
they ire vlddy !ltvored today.
1hc scope o£ thi1 peptr lnol\d th• ..crlptlcn· ot the surface ot th•
6
hyperbolic p1nbolold, the del'tnUon o! the gcncr•l •ne 1tr111 equatlorw,

dwel~llMl'\t of atreu ec;umttont tor sou special eaaa, dl1ey11lon ot th•
boundary conditions end • ••mrl• d•tlgn. A dalgn ot th• Groined Y1ult t11lng
th• ....wane theory •nd in aceordllrv:• vlth th• C.rdele6 Ket.hod vlll bl prff•ntld
•• •n catpl•. 1h• econodc Hptct1 ot the bypubollc paraboloid lhlla vlll

also he d11eus1ed.
7
Sur.f' c• DtllniUor1 and Ge etry
Th• h)?Ctboll~ arabolo!d surtau can be dctintd as • 1ur!acc ot tr•na-

l•tlon or s • "'9tped 1rallelogt1:a. nte surt1ee ot tnnd1Uon b to ed by
llCWl • vutlc•l p nbola, opine uplillrd, over tnoth t vertlctl pare.bol
cpenld dew rd • In Flo • .)b.
1be ~'lrped p talldograa ti tor
parellcl to th• XOZ plane (Jl'ig. )I) alcng the two slHJ£ t, norparnlltl,
nonlntcrtectlng Unes .. tht Y-axl• and ltnc AB:. Th• retultlng rftce 11
r ruent•d in Ftg. 31 by th• grid Un • hn •rd ln' llhicb l"JH provision 1)¥
ne d the generators. tvery point on th• surt;ce y b• c(¥'lsidercd to be th•
intersection ot two • h 11 • coatlln•' ln the surtsce.
Th• equation ot the surface ln blra:ttnguJ.Ji'r coerdinates cen ba deriv
by conatdGrln the bflsic quadrent, HO , shown in f'ig. 1:. In tr!engle HA•A
5.. !.
h I
ot
c. r.
l
Also, ln trl .ngl• Ed'd
cz.Tb'.
z•-
lb XY
• KXY. • • • • • • • • • • •• • • • • • 1
nts th• 1 l ltsi pc:+!lt lhl• aqumt ton o1 th• 11c ~ d grH
,.
/
,. I
~
..,, Cl.. Cl
\ ,.--
,.,.-
Ul
(.., I (
I I \
,~· \
(/\
I
c
.,,
;llJ
)>
l)
m
\
\ .
\
0
m ,. \
..,, \
,. \
z
....
\
0
a m--- --1m,.
I
',, z I
(/l
I
N ,. I
I
~ I
I
I
I
I
I
I
, /
<-\
I
\
\
I
\
I
I /
~~-
\
- --
\
\
9
Yllch •crlhce th• 1urt1e• of the ~trbollc paraboloid.

10
z
Fi g . 4 -GEOMETRY
I vy-+~·dy
?Jy
s~Tic
!
.I
I
z I
Fi g.5 - MEMBRANE F 0 RC ES
il
iCJll.branc Strr,su1 !or a Unl.t'o1•llly Disti-ibut Suttee• Load
Fig. S shows • tcpracntattve amall element of • doubly curved •h•ll
tol'lled by tour lntct1tctlng generators. The direct stresses (1ttesscs U es

the shell tblckntH), Tx •ni T measured in p0W¥11 pu unit length, are
1
posltlva lilhen they create tension. The shearing atte11, s, 1110 measured in
pounds pct it length, 11 polltlvc lillm it creates tans! o in th• diagonal
direction Of lncras i values o! X and Y. Th• load per unit ot projected
area, wz, ls ccnsldercd positive en downward.
To sbpllty the upresslcns tor cquillbrl 9, th• actual stresses arc
transtormed into tlctitlous 1tr1s111 acting on the projected. area ot th• upper
el ant in Fig. s.
Ft Oil geometry
ds1 cos "fJ • dx • •• • • • • • • • • 2

and
ds2 c ~ • dy~ • • • • • • • • • • 3
The horizontal cc.ponant of the nor 1 1tresse1, Tx, 1s
In order for the projected de ent to havs th• same total strc11 11 th•
actual en•
Cl'
12
Vx • T c •
X COll(i • • • • • • • • • • • • • • b
• •laUat pr • ut• one ctn snov that
Vv•T ~.
" y cos p •••••••••••••• s
Fr Eq. l
~•lY
1> x
Ind
h.xx •
1jY
1htflfcre,
V • T
x x
-f 1+K2XJ-
l+K.2.'f.. • • • • • • • • • • • 6
In. •1•1111'. t
~-~1~~
" . . . .. . . . . . . 7
Since,
thtrc.tou,
S • T. • • • • • • • • • • • • • • • • • • 8
111bti of atrcas per unit 1 ngth • flcUU •1•-
t 11 th• Y-exl1, con1ld1tl th• varl•Uon of th• forcu troa th• n•r
CM to the f'lf cm, I only vertical le11c1 thl lhtllt I ;tltl~ I
'0() Jx +']..I_
;;;Y • o• • • • • • • • • • • • • • •9
Sl•Uarly, qulllbl'lua alo the X-t!xls glva
~v ]1
~. °1)-X. o. • ••••••• • • • • • .10
'?h• vert lc•l c ont a el ong th• Z-exl• arc 11 11 1
'the v1rtlc11 c~oncnt ot th• nor•l 1tra1 Tx 11 equal to Tx llntf dl2•

tltu lng the Vllu 1 tor Tx •Di d12 •• given by • 3 Ind 4 gives
Therefore, the vertical c onent of T.x per unit lc~h ale the Y-exll
~
Yx •· ~·
S'l•Uarly, th• vertical c~o.nint ct Ty prr nlt ct length along th• X-exls la
()z
v~ ~ 1)V.
Th• vertical comonent o! th• shear 1tre11 •lo th• Y-exl• ta

le er unit ot lt th 1
Slailatly, the v rtlc~l C:ot!i'Cr~nt ct ch tr Ill ' th• x..xla is
Ccr.sl~crlrsiJ the vr.rlttlon in the aagr.ltua; ot streasu ftoa to snot.her,

t."it SU U~ t stttHU in t.~s Z-dlrectlcn
1..<Yx1>Z) + /() (Vy~Z.) + 'l) (t'?lz) + '() (TO ) • ·=1ln w.
() x 0-X ?J-r -o-Y -o-X 7J 7J""Y ~.
DU!crentlatl ard collactl~ ttr. ylel
1.! {W, +°-.V +~@Vy + W 2
+ Vx 'Q z +Vy o 2z + 2T? 2z •
{)x ox OY 7JY7JY 7Jx 'O'lf. ol2 oil>t
S Ututlng 1. 9 •rd 10 ln th• &bcn• equation yleldl
1-~z '02z 2
Vx~ + Vy ~ + 2T LL • ·z • ln ..., • • , , • c • • 11
{IX ?J Y'- '15~-Y
Tn• u•• ot • 1tr· s tunctlon3 de!lnlng ~11 three strc11 c nt s vlll f'ac:ll•
lt•t• the solutlcn of the dlf!erenttal au on. This 1tr 1 !'unction, F(x,y),
reducll th• th.tee 1lll\lltln'l Eqs. 9,, 10 end 11 to one equation, the coa-
patlblllty 1tl , with only one unkno~n.
It y be 11 that •· 9 1.nd 10 1r1 ldentleall)' 1atl1tl by teklng

?; 2, 7J2p :02F.
Vx ~-yr, Vy •?>yr' Ind T .r~.
q. 11, therefore,
lS
1bc fora cf • 12 11 ot gu1t ~ortarw::. in f lndl th• lnl ltHIHI
of• dc:iubly Cutv 0111 llh•r• th• algsbrale 1olutton of th• ditter t11l
· ~·ri;al pro::c.dut •Y be used. However, In
th• caH ot th• hyperbollc paraboloid shelll, subjected to 1 ltora aurface
1 d ltor.ly diatribe: load • horizontal projection, dlrect lnte-
ration i1 relatively 1~10.
Fr the Uon of th• 1ur.f8Ct ot th• hyperbolic pat'lbolold, Eq. 1,

lt an be own that
---oz -o . ~ o2
z • KXY'OX • KY,o • KX,o-XW. K'o"X7. o, ...... '-'t""'[;" 7
Su?:Jltltutlng th st valu s ln Eq. 12 gives
7},
-2K~ • 'z aln w,
r:~
0 XQY.
Thtrafot•,
T• ~ sin w. • • • • • • • • • • • • • • lJ
For hyperbolic paraboloid ell• havln~ a gr.. t rlao, the dead loe c•n•
not contidet unl!oraty distributed on th• h bontal projected ara.
'nlete!ora, •n re111Q\ tot th• dtl loa t any point ln ter ot Wz should
be duiv •
Vo !1 th• load per unit r a at P~ point, th n

• • • • • l!i
th• low r •rt ct Fig. S1 lt c1n seen th•t
(d11)2 + (<192)2 ..2 •1ds2 c ti( -~. • • • lS
16
Z1 • • l{J 1, a 22 • lln<( 2• z1-~ • •ln1( 2-etnf 1•

Fr th trl ngll, , obtains
L2 • (lln1f dt2-aln tf dl1)2 + (dx) 2 + (cty)2 -2 dx dy c v. • •• 16
E tin Eq. 11' to 15 a •illpll.fYlng, It la t that
cot IX • 1lnl{ 11nf.~ • ~

. "91 cl.S2
~ ,ff)~0 ~
c2x • see 'f' •0 x - 1 ittt r. • • • • • • • • 11
av 2_
A
• 1ec-i.r
~ -~y •ji+x.2?.
C/ e e e e e • e • 18
1h t!o:e,
•In tX. ./t1n2vtK2itft2il-2 2XY c v • • • • • • 20

1 (1~2x2)(1+il¥)
Ut • 20, 17 and 1 into .Eq. l4 glw1
17
Finally, ·bstltutl Eq. 21 Into 14• 13 ruults ln
T • ~((i • • • •• • •• • ••• • • • • 22
Dlf! rentltltlllSJ tq. 22 with rHpect to X 11nd Y, then s;ubstltutlng the••

dlltcrentlals In E • 9 Ind 10 1 Incl lnt•gratlng the resultl Ui>Hlllcnt vith
r S?tet to X •1¥1 Y, th• cqu1tlon1 tor Vx and Vy are obtained.

'OT • lWc m2x-2x~Y COi w
"O'f n- 2 Y"4? • ••••••••• 23
• •••••• •
Then by 1 tltutl £q. 2h ln Eq. 9 th• tollovltJQ can be •in da
+waj 'fR x-21i"ffx c91 y

v.x. rz-
2 2
dX
..wj2
• _.s. 2K Y sin2\l+2K2Y cos2v-2K2x c01J t1 d.X
2K 2 {ff
2 £. 2 2
+We l,...,JcOI w(21\2rc v-2K Yl dX:fK ~ln v • ~
·a- f11
Howwer, f(Y) 11 a constant of lnt.egratlon t.hlch can have 1ey val
1~1'1.iln zeto.
1'11ualng fer convenlenc• that
'lh rcforc,
~?~ cos wv(/•

,fM w
r r:!SI c w .1?(
Yx •
._
Y stn2w log ya
..,/
aln v yl.J\2 1
+ t1(Y). • • 2.S
18
logcusly,
r w • 1~
1
.Ji w y: c
Vv • ~! C:OI v{cf • ~ X s1Jw log -/
2 2 + 1'2{X). • 26
., " it; 11nv l+KX
'lbc i·bltrnry !Unctions o! lntcr-;£Uon, .t1(Y) •td t2(X), allow us to

aati•ty certain edge coniltlons.
'nle tor cf Eqs. 2S ~ 26 Cetn be 11.tJpUfied tor I • lplCld CIHS o!
hyperbolic panbolold lhdl1. 1b• • t c on sb;>ll!leatl area

(•) the case of • nc:tangul1t hyper»oUc p•t•bolold shell, !.1., w-90 d •
w
T• :tk.f(i5 • • • • • • • • • • • • • • • • • 27
Yx • ~ ucs{~:};} t 1(T) • • • • • • • 28
ti
We
Vv • -:;- X log
..
~KY+
_J
.Jl+K2~
fi] + t2(X). • •• • •• 29
(b) the ase of th• uniter.lily istrlbutad lcsd on a horizontal projection
We 1ln v
T• 2K • const11nt. • • • • • • • • • )>
Yx • t1(Y). • • • ••••• • ••••••• Jl
Vy • 12(X). • • • • • • • •••••• • •• .32

19
i;ccinca~ •rd ..., o! .YP•tboUc Paraboloid 1111
di and Slr;g110 stated that "1h• b»>trbollc paraboloid

roct c t l1e1, use. lus than halt 1 •ny columns, f er ntly h.. gr t
1tr 1 will rc1ult in lower •lnt ~· c t and 11 tires £e.• Thq agreed
that Fdlx C.nctll1.3 rt t 'Mb n h• 11ld 1 Hllli• 11 th• aost ec teal thl
1h• econ le aspectl at tbia tw• cf 1tt\lltut• erivcs tr th• f1Ct th•t
it dcvdcp1 • 1t1t• ot 1trc11 favc •bl• to oncrtte, 10 that thin saetl art
anticipated. In 11ao1t all th• 1hdl1 c~tructcd, th• • Uons ted vary tr
tvo la;, ct to f l~t1. Their ru.lcd s~1'aa11 • th• aorc tily to cd
~ atrlight pl• tcr • In 1ddlt1on, If\! hyperbolic p raboloid yp )root
c~otcd ot llU!ti• it1 ot 1laller hyper llow1 the rt•uslng of the 1 •tor•
ln c trti:tin~ ucb unit ot th• structur , ~1~"'1 low rs thii av r 11 cost.

lbuerora, thctc an n l 1: any doubt thtt thi• typt o! c tru:Uon la
th• oet ea oataal ot •11 tra fag 111te in th• rang• of 1p1
'lb• h;yptrbollc p1tabolo!li ~hdl• c nb t.t3ed •• root•, s1.1:h a th•
alla I thl wtolnci VJUlt roofs, or I £<K1tlf\G! for Col
u Uy c lttru:te ln th• fora ! •n !nvut~ elh.
~ol\kln e " ton tr , " Udf tar Econ vith ttyperboUc

P tabolold1,• .I.cl Jour r, 1960, I • 4, V. 32, PP• 37.3-383·
A groin d v ult 1quarc in plan, tor d by th lntlracctlon ot two hypers
vlth vertical axe~, 11 dt,lGn d 11 an 11 (Flg. 6).
D1 enston t
Side ct th• qunrs • 70.00 1'eet.

ld~t o! th• crown • 20.00 tect.
i • tan •l ¥• t1 •l j. •
1b.cfe!or•,
J• 26 d g. J4 ainutc1.
D lg Dlt11
t•c • ))00 pll.
ts • 20000 p1l., d1!or blrs.
v• 60.oo pat.
Codtt
1h• ICI code (.ACI 316-6.l) •rd th• racaa.tr¥1SUCC'll of th• .ACI c lttH
)J41l IH Ulld.
Prcacdure ot Dc1lgn1
Sl=• tne plan o! th• strax:tut• 11 • 1 re, 1ymctl")' pu11lt1 to ln•

v tls•t• only one-et th ot th• atrt.t:tur., 11y th• triangle 1-8-22·1 ~ ln
Fig. 6c.
l11te1 Co.atttce 334 1 "Cor¥:r•t• Shell Str\J:t es-i>ractlc• 1 C ntary, 11

/.Cl our lo. 9. • 61, Sept. 1964, PP• 1091•1108.
21
~
._t -_ _7o_~oo
1
\ ...,
PLAN
fv (a)
I
_ £LE V~TION
t
(bJ
Fig.6-THE GROINED VAULT

22
'nlc sign prcx:e ure e n h• s r !ztd th• following steps•
Sttp la
Th• coo dlnatca o! th• bo ry points 1 to h2 of th• 9r ld should bt t •

'lhls ay be ace lllhcd e1111)' by exprc11l th• XY-coordlnat11 ln tcr• ot
rect ul1r coordlnatea. Jtttcrrlng to Fig. 7•
y .. -
.. • • • • • • • • • • • • • • • • 33
•ln f
a• v-n •V-L tin f• ........... .34
t 2•t+ii •••••••••••••• • 3$
Frca Eqs. 33 1 34 and JS, wa obtain
)!
v+dtan 2
y. - -• • • • • • • • • • • • • • • 36
2 1in f
A110,
v •n •L
1ln 2"• X• COi ¥• • • • • •• • • • • ••~
Ft • JJ, .34, 3S, J6 and 37 lt can be lhown that
utan r .,,,
x. 2 l JI. • • • • • • • • • • • • • • • .)8
•n2
Col 1 •nd 2 or Tsblt 1 lhow th• valu ct X •Di Y of th• ?>ounda•ty
point• 1 to h2 Which ara obtained by lng Eqs. 36 a )a. 1n der to flnd

th• z..va1 , th• c t• , K, of Eq. 1 ould be c uted.
It '" X2~2'" •l§,i%b.f '" -0,0174,

ColUllR 3 ot Tlblc 1 th• value• o! z.
23
P(><, Y)
y
v ··:
fig.7_ COORDINATES SYSTEM
fig.8- CONDITIONS OF EQUllBRIUM

tep 2i
th• lOld th• bl11s ot horlzcat•l proJtletl ould be calcula ed.
Ass th• thickness of the htll ts thr • l~ht,t, tha

o. L. •fr x 1 x 1 x 150 • 37.5 lb. p sq. ft.
lloot!ng • 12.s 1b. per sq. tt.12
Tot1l D. L. • $0.0 lb. peit aq. tt.

1h• d d loa per lt 1 ·th ot tdg• 11
2x So • 100 lb. per tt.
• .1 a. per tt.
Fraa 1. 22, 2$, •nd 26, th• .bsilc nluc1 of T, Yx, and Vy n• c uted.
To sult the tal>\•litU repr scnt1t1cn, th • cquatl •7 be p 1n re
•brldgecl tcrat
T • • .. K Wc-j{f
Yx • •T Cot v + A 1 + t1(Y)
ltlere1
v ·-fll
• 11n v ll+K2Y2
c• iwc x 11n2w
2)
'Jh• l'l lt lch uc n"nsaty tor this 9t ar1

4 to .. Table 1.
ibc ary polnt• in thi• •t 11re tr tad lill 1nt1rlor in Wilch
t1(Y) · t 2(x) are 111 to be eq 1 to zero. Detu:aln•Ucn th• cnct
Vil I of tb H functl •lil r djuabttnt ct Yx end v1 ii den• in later at s.

St 31
Thu 1tr111 S, Tx 1 and Ty ahovn ln Fia. S ire p1r11lel to tht gt r1tcra.
H u, th• str 111 •lo th• bo nd;ry n d t.o be lrN~st lw•t.-'• Tx and t1
are oblique ttrtases, 10 e 1c ru::U ot Hcbr•1 Circle c ot be ed
hU•· Flg. 8 °"s th• c ltlon ot equUlbrlU!ll or an al nt 1 th• tirat

· tvo lntusect Ing a crat • and s arbitrary • ti Vhieh
MkU •n 1ngl1 fi with th• p011Uve dlrectlcn of th• x-g trator. Frcaa trig-
try It an be th t
!..!-- • !f.
•1n • n tt•til) . .JL..
iln-A<
Fr \blch Is 11
1ln (8~)
"2 • •ln D<. •
ot th• lldts 11'¥1

resolv thtH t s ln a dlrtat.lon • 1 lnd. parallel to the 1ect.ton1 the
values o! th• no l s t•ngcnUal 1tr.•• , Tb r¥i Sb~ are obtain •
1tn
11> • 1x sin rx.
2
e
+ 2T s!nA!ln
sin 0c.
v.aj + Ty tln2otn"'
CB-~)
• • • • • • • 39
26
Tablt l. E~s!c vglues o! stre11e1.
K2y2
t-olnt
l
2
r
0.0000
2.8000
x y
2
0.0000
2.8o0o
l
0.0000
E
0.0000
7.8400
y2
r:
0.0000
7.8400
K2x2
6
0.0000
0.0024
1
0.0000
o.o02li
- 0.1363
) s.6000 s.6ooo - 0.5451 )1 • .)600 31.3600 0 .009$ 0.0095
4 8.LOOo 8.1&000 - 1.2266 70.$600 70.$600 0.0213 0.0213
s 11.2000 u.2000 - 2.1806 125.4400 12s.hhoo 0.0319 0.0379
6 14.0000 14.0000 - 3.4071 196.0000 196.0000 0.0592 0 .0)92
7 16.6000 16.8000 - 4.9063 282.2LOO 282.2400 o.OSSJ 0.08.SJ
a 19.6o00 19.6000 - 6.6700 .334.1600 )8h.1600 0.1161 0.1161
9 16.8000 22.4000 - 6.S417 282 .21.00 $01.7600 o.08Sl 0.1516
10 14.oooo 2s.2000 - 6.1329 196.00CO 6)S.0400 0.0592 0.1919
11 11.2000 28.0000 ... s.hSJ.4 12$.4400 784.0000 0.0379 0.2,369
12 8.4000 )J.8000 - 4.4974 70.;600 948.61!00 0.0213 0.2867
1.3 S.60oo 33.6000 - J.2709 31.;soo 1128.9600 o.OQ9S 0.3411
lh 2.8000 J6.4ooo - 117717 7.4800 1324.9600 0.0024 o.4004
lS 0.0000 39.2000 0.0000 0.0000 1$36.6400 0.0000 o.h~.3
16 - 2.aooo 42.0000 2.0!Ji3 7.4aoo 1764.0000 0.0024 o.S.330
17 - s.60oo 44.8000 4•.3611 31 •.3600 2007.0400 0.009$ o.606S
18 - a.4000 47.600\) 6.9So6 10.S6oo 226.$.7600 0.0213 o.6847
19 -11.2000 so.Loco 9.6126 12.$.4400 2S40.16oo 0.0379 0.7676
20 -14.oooo 53.2000 12.9472 196.0000 28)0.2400 0.0.)92 o.esso
21 -16.Booo S6.oooo 16.3$43 282.2400 31.36.0000 0.0653 0.9476
22 -19.6ooo SB.Booo 20.0000 .384.1600 JJ67.1:400 0.1161 1.1608
23 -18.1000 56.oooo 16.20h4 349.6900 31)6.0000 0.1057 0.9476
24 -11.1800 S3.2000 16.lt4.30 316.8400 2830.0000 0.09$7 o.8550
2s ·16.Booo so.0400 1417184 282.2hOO 2540.1600 o.08S3 0.7676
26 -lS.6800 47.6000 13.1381 252.8000 2265.7600 0.0764 o.6847
27 .14.esoo h4.8ooo 11.S6S1 220.$200 2007.0400 o.o666 o.6o6S
28 •14.0200 42.0000 10.2300 196.$800 176h.oooo 0.0594 o.S.330
29 .13.0800 39.2000 a.9127 171.0900 15)6.6400 0.0517 o.h64.3
X> -12.1800 36.bOOO 1.1069 lh8.,3SOO 132h.'600 o.0448 o.h004
31 -u.2100 33.6000 6.S828 127.0700 1128.9600 0.0.384 0.3411
32 -10 •.J)OO )).8000 S.Slh6 106.0900 948.6400 0.0321 0.286?
.33 - 9.3600 2a.oooo 4.5sss 87.6100 78h.OOOO 0.026; 0.2369
34 - 6.J600 2s.2000 3.7019 11.4000 63$.0400 0.0216 0.1919
3S - 1.s100 22.4000 2.9244 S6.booo so1.160o 0.0170 o.1S16
J6 - 6.saoo 19.6000 2.2419 43.3000 384.1600 0.0131 0.1161
37
38
- s.6lm
- 4.7200
16.8000
14.0000
1.6SS9
1.1488
31.8100
22.2800
282.2400
196.0000
0.0096
0.0067
o.OSS3
o.os92
39 - 3.7700 11.2000 0.7362 14.2100 12.).hhOO 0.0043 0.0379
40 - 2.8400 e.uooo o.4116 8.0700 10.S6oo 0.0024 0.0213
41 - 1.9100 s.6ooo o.16S8 3.6500 31.3600 0.0011 o.oo9S
h2 - 0·2;!22 2.eooo O.O!i2~ o.8600 z.el.ioo O.OOOJ o.002h
27
Table 1. (Cont.)
RX coe v l<Y cos w 2K2XY co v

feint u 1h
l v.OCMJ .~ O.O:.N'O ;, {·tr.,-.~
.... ..,vvv O.O'.J'.JO 0.0000 o.oc.;00
2 7.8400 -0.0487 -O.Oli87 0.0024 -0.0292 -0.0292 0.0029
3 31 •.3600 -0.0973 -o.onJ 0.0095 -0.0564 -0.0584 0.01!4
4 70.-Sco\> ..o.11;6o -o.l.460 O.O~lJ -0.0&76 oo0.0876 0.0,56
~ l~S .t.t-oo -0.1947 -o.1s1L7 0.03~9 -o.1166 -0.1168 o.ohS5
6 196.0000 -o.24.34 -o.2431~ o.os·S12 -o.lb60 -o.l..460 0.0110
7 282.2400 o-0.2920 -o.~920 0.08$3 -0.1752 -0.1752 0.1023
6 J84.Uco -o.Ji:07 -o.JL07 0.1161 -0.20~4 -0.2044 O.lJ9J
9 .376.3200 -0.2920 -0.3394 0.1137 -0.11.s~ -0.2336 0.1364
10 .1)2.8000 -0.2434 -o.4.381 o.1ce6 -o.~60 -o.2628 0.1279
11 313.0000 -0.1947 -0,4867 0.0948 -0.1168 -o.2S2 3 0.1017
12 2.~a. 1200 -o.lll60 -o.$354 0.0782 -o.01376 -o.~12 0.09,)8
13 188.1600 ...0.0913 -o.S84l 0.0568 -J.05e4 -0.3,04 0.0681
14 101.9200 -O.Oh87 -o.6J'l7 0.0308 -0.0292 -0.3196 0 .0370
l) 0.0000 0.0000 -o.6814 0.0000 0.0000 -o.4083 o.ooco
16 •l17.6o0o 0.0487 -0.7):)1 -o.o.;55 o.o29'l -o.4JBQ -o.0426
17 -2.$0.8800 0.0973 -0.7788 -o.015a 0.0584 -o.4672 -0.0909
18 •399.8hOO o.1h60 -o.8274 ..0.1206 o.0876 -0.4~ -o.U'49
19 -S64.b600 0.1947 -0.8761 -o.17o6 0.1168 -o.5256 -0.201'7
20 -1U.6000 o.21J.34 -0.9248 -0.22!)1 o.1460 -o.5548 -0.2701
21 -940.6000 0,2920 -0.9735 -0.2843 o.11s2 -o.5841 -o.3411
22 •1152.4800 0.3407 -1.0221 -o •.3463 0.2044 -0.6132 -o.4179
23 -1047.2000 0.3251 -0.9135 -0.3164 0.1950 -o.58111 -0.3764
24 •948.$600 o.JJ91 -o.924a -0.2666 0.1654 -o.SS48 -0.3439
:?S -840.6700 0.2920 -0.8761 -0.2$40 o.11s2 -o.$256 -0.3048
26 -756.6800 0.2160 -o.e274 -0.2286 0.1656 -o.4964 -0.2743
27 -.565.2800 0.2581 -0.7788 -0.2010 o.154a -o.4672 -o.2hl2
26 -588.8400 0.2.431 -0.7301 -0.1719 O.lh62 -o.h.380 -0.2135
29 -su.1aoo 0.2274 -o.6814 -o.1ss2 0.1.364 -o.h088 -0.1862
JJ -4l1j.3S00 0.2117 -0.6327 -0.1.3!.io 0.1270 -0.3796 -o.16o8
31 -380.7000 0.1959 -o •.$841 -0.11;0 o.111s -0.3504 -0.1380
32 -319.))00 0.1790 -0.5354 -o.096S 0.1074 -0.3212 -0.1158
3.3 -263.700o 0.1627 -o.4867 -0.0797 0.0976 -0.2920 -0.0956
34 -214.h60o o.1h69 -o.4381 -o.0646 0.0881 -0.2628 -0.0777
JS -169.S?oo O.l))S -0.3894 -o.os12 0~0783 -0.2336 -o.0614
)5 -129.9500 O.llhJ -0.3407 -0.0393 o.0688 -o.204h -0.0471
37 - 9$.0.300 0.0980 -0.2920 -0.0287 o.osaa -0.1752 -o.o.31.:4
3 - 66.9))0 0.0820 -0.2434 -0.0202 0.0492 -0.1460 -0.021,.2
39 - 42.6800 o.065S -0.1947 -0.0129 0.0393 -o.116s -o.01ss
1'0 - 2h.2500 O.Oh94 -o.l.h6o -0.0073 0.0296 -0.0676 -0.0087
bl - 10.900o 0.0332 -0.0913 -0.0033 o·.0199 -o.osah -0.0040
L2 ... 2.600o 0.0162 -o.oL87 -0.0007 0.0097 -0.0292 -0.0008
28
Table 1. (cont.)
2 I 2
l+K2 ~ l+K2y'l •in v l+K2.x2 sin v l+K Y2 ccs w 2' sin w
16 18 21
Point
1
2
o.&s
o.&>20
1.0000
1.0020
i.Moo
l.0020
o.tiooo
o.eo16
12
0.6000
0.8016
20
c.4boo
o.4812
0.0000
0.8960
3 o.eoso 1.0040 J.• 0040 o.6o.32 o.eo32 o.4829 1. 7920
4 0.6110 1.0100 1.0100 o.eoeo o.eoeo o.4866 2.6880
s6 o.8180
o.8,J>O
1.01so 1.0150
1.0290
0.8120
o.e232
0.6120
o.82.32
0.4908
o.49eo
.3.SeliO
4.4500
1.0290
7 o.e420 1.0410 1.0410 o.8.328 o.8326 0.$158 s.J16o
8 o.ssoo 1.0550 1.0$50 o.e.s24 o.6.$2h o.Sl.36 t.2720
9 o,8600 1.0410 1.0110 0.8238 o.8408 o.Sl60 s.316o
10 0.6140 1.029'J 1.09$0 0.8232 o.s44o o.s244 4.4800
11 0.9020 1.01.so 1.1100 0.8120 o.aaso o.s411 3.,S8Lo
12 0.9290 1.0100 1.1320 o.soao 0.9356 o.SS73 2.6880
1.3 0.9600 1.001J0 1.1sao o.6032 o.926h o.s1s9 i.7920
14 1.0000 1.0020 1.1610 0.6016 0.9Lh8 0.6000 0.6960
1S 1.0500 1.0000 1.2090 o.eooo 0.9672 0.6299 0.0000
16 1.1000 1.0020 1.2350 0.0016 0.9680 0.6599 -o.8960
17 1.1610 1.0040 1.26.30 0.8032 1.0104 o.696S -1.7920
18 1.2210 1.0100 1.2990 0.8080 1.0392 o.132s -2.6880
19 1.2850 1.01.so 1.3300 0.8120 1.0640 0.1109 -J.S840
20 1.3.)00 1.0290 1.3600 o. 232 1.0680 0.8099 -4.4800
21 l.4l$O 1.0410 1.3950 o.8328 1.1160 o.6489 -s.3760
22 1.446() 1.0780 1.4720 o.BS2h 1.19.36 o.867S -6.2720
2.3 1.43ao 1.049() i.3950 0.8392 l.116o 0.8627 -!).93l:C
24
25
l.JSfO
1.3)6o
l.OhSO
1.0410
1•.)600
1.J))O
0.8.360
o.8.328
1.0880
0.0640
o.a13s -s .6896
0.801.$ .5.3760
26 i.2940 1.03)0 1.2990 0.8280 1.0392 o.7763 -5 . 0016
21 1.2440 1.0:330 1.26.x> o.8264 1.0104 0.71£6.3 -4.7520
28 1.2010 1.0280 1.2.JSO 0.6224 0.9880 0.1205 -4.1'864
29 1.1.$90 l.:'\2)) 1.2090 0.8184 0.9672 o.69)3 -4.18$6
)J 1.1160 1.0200 1.1810 o.816o Q.9448 o.669) -J.8976
31 l.076o 1.0180 1.isao o.8144 0.9264 o.6h55 •.3.606li
32 1.0370 1.oiso 1.1320 0.8120 0.9356 0.6221 -J.29£0
33 0.9760 1.0120 1.1100 0.6096 o.eeeo o.S85S -2.9952
34 0.9660 1.0100 1.09So o.808o o.6Wio o.S79S -2.7040
J.5 0.9340 1.<X:>70 1.0110 o.BoS6 o.8408 o.S603 -2.1.io32
J6 0.9040 1.0040 1.osso 0.8032 o.e524 o.S423 -2.1056
31 0.8770 1.00~ 1.ouo 0.8024 0.8320 o.s262 -1.8048
.38 o.6S60 1.0020 1.0290 0.8016 0.6232 o.Sl.36 -1.5104
39 o.8.)So 1.0010 1.0150 0.8008 o.a120 o.so16 -1.2064
40 0.0200 1.0000 1.0100 0.8000 o.80&> 0.4920 -0.9088
41 0.6090 1.0000 1.0040 o.eooo o.eo32 o.46S4 -0.6112
42 0.80~ 1.0000 1.0020 0.8000 0.8016 0.4818 -o.22z6
29
Table 1. (Cont.)
!2 sln2w c
A IL D ~ .12:iW2.
? o!nt
l
2;:
0.0000
2j
0.0000
"!1
1.0000
B:
0.0000
26
l.0000
21
0.0000
2(j
0.0000
2 o.09(;o o.oa96 0.9761 o.. os90 0.9i61 -0.0243 ...0.0243
.3 l. 7920 0.1792 0.9538 0.1792 0.9533 -0.0471 .0.0471
4 2.6680 o.~688 0 .. 9314 0.2688 0.9314 -0.0710 ..0.0710
s6 3.$840
4.4eoo
o.J.$84
o.4480
o.911Ji
o.e911
o.35a4
o.4480
o.. 91lh
0.8911
-0.0932
-0.115h
-0.0932
.0.1154
1 5.3100 0.5376 o.8708 0.5376 o.a1oa -0.1.383 -0.1383
a 6.2720 0.6212 o.84hJ 0.6272 o.8443 -0.1707 -0.1707
9 1.1680 0.1168 o.9s34 0.5376 0.7839 -o.0481 -0.2h33
10 8.0640 0.6064 1.oses o.4h8o 0.7069 o.osa2 -0.3467
11 6.96oO o.8960 i.12SJ 0.3554 o.6S.S3 0.1088 -o.4231
12 9.86)0 o.9865 l.1802 0.2688 o.59SS 0.16;$ -0.5184
13 10.7520 1.0752 1.3095 0.1792 0.5h07 0.2700 -o.6149
14 11.6!:80 l.16h8 1.4086 0.0896 0.4946 o.34.36 -0.7032
15 12.sW.Lo 1.2544 1.5083 0.0000 0.3999 0.4121 -0.9163
16 u.W+oo 1•.3440 1.606o -0.01.!96 0.4250 0.4731 -o.8551
17 11.i.3.360 2.4336 2.828S -0.1792 o.4031 o.6o03 -0.9088
18 25.2320 1 •.52.32 1.7065 -0.2688 0.3787 o.s344 -0.9702
19 16.1280 1.6128 1.6847 -o.35a4 0.3597 0.6337 -1.0217
20 11.02ko 1.7024 2.9744 -o.ueo 0.))92 o.6eos -1.0817
21 17.9WO 1.7920 2.0.s.30 -0.5376 0.3198 0.7193 -1.1394
22 18.6160 1.8516 2.0106 .0.6272 o.2s1s 0.6936 ..J..3567
23 11.9200 1.7920 2.1032 ..o.$984 0.3211 o.6981 .1.13JS
24 17.0240 1.7024 2.2256 41().$689 0.3309 0.7990 -1.2347
2s 16.1280 1.6128 2 .031.aJ ..o.SJ76 0.3493 0.1102 .1.0S26
26 1s.2320 1.s232 1.9931 -o.$082 0..5961 0.6881 -1.0J.k7
27 l!i.))50 l.h3J6 1.9488 -0.4752 0.$714 o.6678 -0.9795
28 l~-4400 1.,3440 2.9009 -o.h486 0.6418 o.6416 ..0.9288
29 12.s440 1.2544 1.8538 -o.4186 o.61S3 o.6153 -o.8734
30 11.6480 1.1648 1.6070 -o •.3898 0..$91) o.s913 -o.828.S
.31 10.7.520 1.01s2 1.16.54 -0.3606 0.6231 0.$678 ..o.1as.;
32 9.86SO 0.986.S 1.6958 .0.3296 o.S,306 0.$.306 ..0.7318
.33 e.9600 o.8960 1.6104 -0.2995 o.h762 o.4162 -0.7339
34 ~.8064 o.8806 l.S798 -0.2704 o.b574 o.1674 -o.61S2
3S 7.1680 0.7168 1.)173 -o.2403 o.4187 o.4187 -o.SS34
)6 6.2720 0.6272 1.h1S7 -o.21o6 0.3471 0.3471 -o.4894
37 s.3160 0.5376 1.3830 -o.1eos 0 •.3241 0.3241 -0.4232
)a 4.4aoo o.Wiso 2.3178 -0.1510 0.27.)2 o.21s2 -o.3S81
39 J.Sh8o O.JS48 2.2s20 -o.12o6 0.22!6 0.2245 -o.2890
40 2.6880 0.2688 1.1861 ..0.0909 0.1697 0.1697 -0.2200
41 2.7920 0.1792 l.12h2 -0.0611 0.1129 0.1129 -o •.lh91
h2 o.6960 0.0896 1.04$9 -0.0298 0.0438 o.0438 -0.0693
.30
T""..l:>le l . {Coiv.:l.)
-
~.].cs C lgs D
.)2 ..
T .~ft. T COlf 1'
-~
vx ';</ft.:,.
lit.~~~·
?oi t :i2 ___}} ,?:!: ...... ~
l,.. o.oJOO 0.0000 • 2.29tu l •.;l93 i.3793 1.379.3
-0.0022 -0.0022 2.)016 1•.3011 i.3;69 1.37e9
'3 -o.ooas -o.ooes 2.3100 1.,3561 i.:,776 1.3776
u -0.0191 .JJ.0191 2.3268 le.35'73 l.J78~ 1.3782
s ..v.o.358 -0.03$0 2.3449
2 • .)8:;5
l.4o69 1.3711 1.3111
6 ..o.0;16 -o.OS16 1.4.;ol 1.;7s5 i.,;78S
7 -0.0744 ..,.01U 2.4166 l.b512 1.;;768 l • .)768
e -0.1011 -0.1011 2.4600 1.4160 1.)689 1.)689
~ -o.o.344 -0.1~2 2.4658 1.479S 1.Lw.s1 1.3493
10 0.0388 -o.1SS4 2 •.$160 1.5100 1.$~96 1.3$54
11 o.om -o.1s1~ 2.$6~8 1.55;;9 1.6514 1.40~7
12 0.163.) -0.1393 2.6610 1.))66 1.7601 1.4573
l3 o.:scs -0.1102 2.1ss1 1.6552 1.9457 1.5450
14 0.4000 .o628 2.b1$ 1.72.41 2.1241 1.6613
1s 0.5160 0.0000 3.0189 1.81lJ 2.3273 1.8113
16 o.6JS1 0.0767 3.1600 1.8$60 2.5311 1.9727
17 0.6611 0.1629 3•.)4£;1 1.%41 2.76S2 2.0680
18 o.e100 0.2616 J.$011 2.1007 2.9107 Z.3625
19 1.0196 0.3660 3.68)9 2.2139 3.2335 2.5799
20 1.1.$8.S o.4850 3.6788 2.3273 3.4ess 2.6123
~1 1.2810 0•6221 4.06.)h 2.4392 3.7202 J.0613
22 l.))4S o.as1e 4.1soo 2.4700 J.7945 3.3hl8
~) 1.2.Sll o.6686 4.1.354 2.4812 3.7323 J.1498
.:4 i.;oos o.701e 3.68)7 2.J3Ja .).694.3 3.0356
2S l.l.484 o.S6SS 3.S.)C9 2.~21 J•4.SCS 2.8676
26 1.0499 0•.5160 3.;210 2.2326 J.2825 2.?486
27 0.9)99 o.4659 3.s1a9 2.lh?.3 J.1072 2.6132
28 o.8643 o.4166 J.16co 2.0700 2.9~3 2.4866
29 0.7718 o.36.58 3•.)2.).} 1.99,52 2.7670 2.3610
.:0 o.6~ 0.3229 J.j)JO 1.99ao 2.6800 2.3209
.31 0.6182 o.2a2e 3.oass 1.CSl3 2.h69S 2.1.341
32 o.s2.;s o.241e 2.s191 1.7578 2.3113 2.0.:96
33 o.4267 0.219$ 2.eooo 1.6800 2.1067 1.8995
34 o.403S 0.1661 2.7689 1.661.3 2.o648 1.8374
JS o•.J>OO 0.1338 2.679'.1 1.6o77 1.9077 1.741)
J6 0.2810 0.1029 2.5942 1.)$65 1.837) 1.6594
37 0.1731 0.0754 2.s1a1 1.s109 1.6840 1 •.$863
38 0.1234 o.osJ.i2 2·!:579 1.4747 1.!)981 l.$289
39 0.0796 0.0349 2.4000 1.4400 1.;196 1.4749
4C> o.o4S6 0.0200 2.1)00 1.hloo l.hS.56 1.4300
41 0.0203 0.0091 2.3199 1.3919 1.1'122 l.1&010
g2 O.OOJ2 0.002J 2.3)00 2.~00 l·J§J2 1.~2J
31
Val • ot a 1• /> •Y be obtain ~a to that UI ln

•lnf.n; 1nglt rx. ot £q. 19.
'Ihcrcf'Ql'e,
KY 'lb • cos W}>
COi $'!I .,~ 2) • • • • • • • • • hl
y(l+r... 1-) (1~
• ere
\fb 11 t? • ptoJ Uon of. ~11)?, m th• XY49hnfl.
~b 1• the Ulgon tr!cal tang t or i.hc ,..mle !orMd "1 th• •trdG}lt lin•
repr s nt th• s tlo vlt.h the XY~lAne, ~. •·•
ere a it tht ptoJt<:t~ f.nte~ l.
A 1lgn c 9'\U tor the prc4uct ltY <::o should bt cptc her. In rdtr to
avoid alsta ••• The velue KY · 11 podtiv. en both sldca ot •~1• v are
'lb• tree the groin vault c nnot vitbtlt11 •1>Y nor 1

t v.nUMl &trna. ~, si~• th'r• l• no 1tructunl cl nt. wh!eh l• 1 1• to
l'altt thea. 1h•ntcrc, • Ung 1. 39 nd LO to Z$l'O • 1olvi for Tr
1 Tr• th• tollcvl ta obt•lnt4•
""-
"X • .. ,
,.. 1 !p (,!'.-es,l. I
sin J1 t • • • • • • • • • • • 42
't): • 'I"
•T ' n fl
(§::otj • • • • • • • • • • •• h.l
U, wlth llb 1 to 116 ~ · iU .34 !nut. , tt
valu~ •r~ shown in ol 12 of Ttible 2.
arc lhovn 1 col 1 a 19 of Table 2.

j3bl• 2, StH3 e It tht ffQ~ Gd t•
~in::~~~
KY.g'b+
Lb. ~ KY•li-
6
cos "aa.
Point 2
0 • lol 0.0000 O.OOYJ .oooo i.o.s 0 0 .. 0000 -o •. 7
9 1 .. 11.i; o.osos 0.0026 1 .. 0011 1.0722 -0 .. 0197
...().0716
..t>.4669
-o.s1es
10 i.12s1 0.16.)S 0 ..0261 1.0125 l•lOSJ
11 i.1266 0.2726 0.0743 i.oJ6S l.l.SOS ..0.1327 ..0.5799
12 l•lbJl 0 • .)816 o.. US6 i.0703 1 .. 2116 -o.20hl -o.6$1)
l.) i.1626 o.h906 0.2407 l .. llhl l.29'Jl ...o.2a6$ -0.7338
14 1.16.34 o.sm o.»96 1.16S6 1.3766 -0.3794
-o.h629
-o.8266
lS 1•2090 0.7087 o.so22 1.22s1 1.4819 .0.9)Jl
16 1.237s o.8177 o.6656 1112916 l.S9Sl -o.sno -1.0442
17 1.2681 0.9267 o.esss 1.)634 _i.7220 ·<>.7217 -1.1669
18 1.3120 1.01)8 1.0729 1.4397 1.8702 -o.es10 -1.))42
19 1113499 hllc48 1.3106 1.s201 2.0217 -1.00~ -1.4s02
20 1.mh 1.2SJ8 1.s120 1.6037 2.1810 -1.1595 -1.6061
21 1•4S22 1.)628 1.es12 1.690) 21113$82 -1.3267 -1.7739
22 1.$868 1.12 , 2.1266 h768l 21o6029 -1.L90? -1.9317
'hble 2. (Cent•)
DwH• !!n• Oeq ••Un. E.!fL• !il.!l.•

COIA$ ~ ~ 8-~ 11
10 1 2 1
o. o. 1 -o •. 0 1-0 -0 • 1
9 0.7136 o.6400 -o.Jim so-12 116-06 6S-S4 o.e91
10 0.106$ 0.6269 -o.b699 Sl-o2 118-o2 67-00 o.87h9
11 o.6947 0.6166 -o.s04o Sl·.56 120-20 66-34 o.86ll
12 0.6781 o.S9Jl -o.S377 Sl-37 122-.32 68-SS o.Sb31
13
lh
o.6$67
o.6.x>7
o.S64B
0.5329
-o.S668
-o.S916
ss-:n
S7-h6
12u-4o
126-42
69-0J
68-Sh
o.e22s
o.eo1e
lS o.sm o.4962 -o.6276 6o-J5 12a-s2 68-37 0.7786
16 o.S64h 0 .1661 -o.6546 62-.33 lJJ-Sl 68-20 0.7)60
17 o.s221 o.b117 ..o.6788 6S-b1 130-l6 674 0.7343
18 0.4791 o.3652 -0.6973 68·35 lJL-13 6S-2 0.7167
19 0.4293 0.3180 -0.7173 71-h9 13S-SO 64-01 o.6967
20 0.3748 0.2678 -o.1YJ1 74-28 137-27 62-59 0.6162
21 0.3157 0.2173 -0.7.$22 77-27 138-li7 61-20 o.6589
o.2z1z o.&2S6 -o.zhYl §0-22 1~:28 ~8-o6 0.6562,
JJ
Tabl& ~. (C .}
!"' ~-~
-!.~ ~i~
0.9074 1.005 0.997 • f)S -2.. ·Sb 1.00JO
9 0.9128 1.0222 0.9780 -2.s~22 -2.4158 0.9754
10 0.920$ i.os2s 0.9501 -2.6152 -2.35$5 0.9321
11 0.9.)J 1.0784 0.9273 -2.7$'28 -2.4015 D.9lh4
12 0.9331 1.1067 0.903S •2.~9 -2.4042 0.8922
13 0.9339 1.1.3S4 o.6807 -3.1~2 -2.la96 o.8670
14 0.9329 1.16,?> o.859S -3.3433 -2.h69 o.e4e4
15 0.9312 1.1960 o.a,;61 -).6!06 -2.S2hl 0.6271
16 0.929~ 1.2294 o.Bl.34 -J.68li9 -2.5703 o.s113
17 0.9210 1.2S!il 0.7973 ..Q.169S -2.6631 0.7949
18 0.9097 1.2693 0.7878 ~.Lh39 -2.7582 o.111S
19 o.C989 1.2902 0.77)1 ..Q.7607 -2.8600 0.16.32
20 o.e909 1.31-,s o.1S90 -.s.1103 -2.91&40 0.7S66
21 o.a774 l.J.316 0.7$10 -$.hl.3$ •J.OS31 0.7462
22 o.84 2 1.222~ o.zz~ :2·~26 ·J· 2 o.z~2J
Table 2. (Coi¥:1.)
l+l
~· Av)/tt. llYyd{ft.
,. +k
2I
-J.~77
n I
22 't
a
9
i.oooo
i.0299
-~.w;~
-2. 661
-2.~M
-2.L811
-J.ih
.3.9112 .3.e.301.i
10 i.062S ..2.~7S -2.s;ea .4.0111 -).8935
11 1.09)5 ~.ssn -2.626) .1~.2os1 -4.0290
12 1.120 -2.6274 -2.69~6 -1..i.)87S -L.1$19
-2.f.C~J .-!i.6613 ~ •.;uu
~
1.1)34 -2.71$6
1.17 -2.6J6S -2.9109 -li.9606 -la.s122
l5 1.2090 -2.9863 •J.OS16 ,.3136 -h.8629
16 1.232s •3.1Sl8 •J.1679 -S.6f-29 ,.lho6
17 1.2seo -J.3.JX> •).,1,;02 -6.09S2 ,.h182
18 1.2861 •J.hSS1 •J.S47l -6.36$8 ..;.909
19 2.3103 •.3.633b •J.747S -6.6669 -6.3274
20 1.3217 -J.866$ -3.e911 -7.3$23 -6.70~
21 1.3400 -4.0J9S -4.0911 -7.7$97 -1.1s2
22 ·~2 -J·SZQ - ·.ltl22 -z.z212 ·Z·Z.lQ
Step hs
Ue l E.qs. 6 and 11 th• valu.1 of Vx- • Vy are t 22.
1heH ar• own ln colmns 22 1 23 cf Table 2.
lh• lues of Yx v1 shown ln col 33 • 34, Ta lt 1 tor point•
-22, ire th• UIUllpU that f1(Y) •ir.1 f2(X) are l to ctto.
"lbt 1tra1u Yx 1 1, ah.own ln col..- 22 • 23 of Ta le 2, are •ln9d

en Tb• 51, 1 th• n 1 and t•~tntlll atr ,. , are Ht 1 to cero •
1htr1tore, to l•av• • 8-22 ca1plct•l1 ttH o! •tr•HU' • llt Of •tr....
1• cppotlt• to th• dltfercnecs bctw• th• 1tc val Vx
• called VJ' • Y'l need. to be a11l d t th• conat•nt• t
lntcgrattan, l. c.,
Table 2.
Sttp 61
1h• sttHHI AVx • ! Vy lntr ed Jn 1tep S disturb th• equUU:d of

th• atructur•• ?htte:f'ore, r dJuat Hqu1H that thl ta
streu• hould be lntroduc.c.t at th• oln. 1b11 can bt ecc 111 Id ~
ot h g1ntrat01" to th• cppotit•
cncs, 11 lt ICh generator were • tie or 1trut. Such tren1tora1t1on v111 pro-
d c, ot c:ours1, •ltcratl ln th• •t•t• ot 1tres11s at the lntcdOl' points ot
the 1utfk•• Hovevtr, thtH •ltcratl are lnalgnlticant 1in:1 tb• l>Oul¥lary
point• i;overn tht dttl;n.
Accord! 1y, t.'°\1 atr•ssca at th• tr•
.3S
tra formed to the groin, ;>olnta 22•42•1. Th• values obtained frc this step
ere shoa-n in colwans 1 and 2 of T;ble 3.
Step 7a
Th• aodl!lad values of V~ and Vy at the groin, shown ln colmns 3 and
41 Table J, are obtained. by adding the basic values o! Yx and v1 in columns

33 ind 34 or Table 1 to those presented In col 1 and 2 of Table J.
Step 81
Using Eqs. 6 •nd 7 the values or Tx •nd Ty- arc found tor points 22-42-1.
These value• arc shown in colUllnS 7 mnd 8 of Tabl• .).
Sttp 91
As in step J, V1lu11 of 1tresH1 along th• other boundary line, 22-42•1,
need to be lnvcstigattd. This can be ecc Ulhed by using Eq. 41, with Wb
equal to 71 deg. and 34 minutes in ordH to !11'¥1 th• angle fi tor th• point•
22-42•1. then, by ualr.g Eqs. 39 and hO, V1lu11 of TiJ and Sb can b• obtained.
The necusar,y ca.putations tor this step are shown in colu:uns 9 to 41 of Table
J. Values of Tb a~ Si, are shown ln col I.a and b3 ot Table J.

Step 101
Stresses Tb ind Sb ate stresses per unit length o! dge. To obtain the
forces concentrated at the points 22-42•1, i.e., FTb sand f'Sb, these atrusu
t be 11Ultiplied by the cottespondlng re 1 intervals. These intervals y
b8 computed by the following for•ula a

1 • /cz1-z1+ 1)2 + c2
lbere
I • interval
C • th horizontal projection of th• interval, t.e., 2.357.
the real interval • Ir • Ii +It + 1,

2
36
• ). St tSHt at th• groin.
~Vyk/tt.
Po 2
2 .7.72 -1. JJ ·J· i 0 • o. J2J • •
23 ·1·1S91 -1.sJ&> -b.0274 ~.)862 o.7S20 1.329 -S.35S6
24 .7.JS23 ..7.Jh$2 -J.6SBO -Li. )'.)96 0~7684 1.))14 -lJ.760S
2s "'6.8669 -1.1S2h -3.3164 ..u.2~e tJ.7880 1.2690 -1J.2oes
26 -6.365 -7.0028 •J.08)) -ls.2Sla 0.7968 1.2ss1 -J.869
27 -6.09S2 -6.8S31 -2.96eo -1i.2399 o.8166 1.221£6 .3.6$91
28 ,.6829 -6.70.Jh -2.7466 -1.i.2168 o.8))4 1.2QL3 -J.)101
29 -s.3136 -6.5781 -2.Sh66 -h.2171 o.Bh41 1.1838 .3.0147
l> -b.9606 -6.1621 -2.2726 -h.1)16 o.66.37 1.1578 -2.6312
31 ~.661) -6.327b .1918 -u.1933 o.a s2 1.1297 -2.4761
.32 -L.:;a1s -6.1882 .0762 •h.1Se6 o.e91a3 1.11~ -2.3216
33 -h.20s1 -6.0090 -2.0964 -h.1L9$ 0.9036 1.0988 -2.30S7
1l -4.0171 -s.9098 -1.9S2l -ll.0724 0.9266 1.0792 -2.1069
» .3.9112 -s.1L6o -4.0035 -4.()()4S 0.91:02 1.0636 -2.1.J>9
J6 -J.8341.& -s.sa21 •1.9969 •J.9227 0.9524 1.(1.)02 ~.0971
J7 .3.9112 -?.b182 -2.2272 •J.832.S o.964h 1.0369 -2.))94
38 -h.0171 .5.3257 -2~190 -:h1968 0.9725 1.0279 -2.486)
y; .2os1 -s.2331 .68.SS .3.15e2 0.9833 1.0170 -2.7312
40 -4.JS?S -s.14o6 -2.9319 .3.7106 o.m1 1.0100 -2.9612
41 .6613 -S.01460 •J.2h91 .3.6470 0.9970 1.00.l)..3.2se
IJ2 -h.9606 -b.sisss •J.S767 -J.h806 0.9980 1.0020 •J.S8~
1 -~·Jl~ •.8622 -~·~J ·J·k ~ !·0000 .oooo -J·2lil
J7
Table 3. (Cont.)
Tyk/ft • 1-(l~~~l gb2 1+~2 fH+~?;t

Point
22
8
-2.6151
(1+' )
l.~868
~
o.,gu -!1
KYffb
-0.1156
-
12
o.5115 l.~~12
U:• l 4
1~
l.8457
2.) -3.2999 1.4634 0.7499 -o.7289 o.S602 1.2499 1.7411
24 -3.311.; l.4212 o.734S -o.677S 0.)399 1.2400 1.6892
2s -3.3764 l •.37$2 0.612.s -o.5690 o.4S23 l.2021 l.604.S
26 -3.3897 l.34L5 0.6700 -O.S.$49 o.Uh69 1.2008 1.$588
27 -J.462) l.)J67 o.S6So -o.4398 0.3202 l.1496 l.4526
28 -J.S016 i.2727 o.5632 -o.hll7 o.316o l.1424 1.4148
29 -3.$622 l.2.)89 o•.;124 -o.,3465 0.2628 1.1269 l.,3647
:30 -J.5686 l.2046 o.1.i1as -0 •.3024 0.2297 1.1066 1..3099
.31 .3.7119 l.1707 o.45h6 -0.261:) o.2o63 1.0979 1.2626
32 -3.7190 l.1s20 o.Lo12 -0.2207 0.1678 1.0807 1.2266
33 -3.7495 1.12.;3 0.,3621 -0.1761 0.1316 1.0641 1.1833
34 -J.773$ 1.1001 0.3311 -0.lhSl 0.1096 1.os34 l.1462
35 -J.7650 l.01BS 0.2900 -0.1135 0.0841 l .OlJ.38 2.1179
J6 -J. 1359 1.0612 0.2497 -0.0848 0.0623 l.0327 1.0916
31 -J.6961 1.0431 0.2160 -0.0646 o.4660 1.0212 1.0620
38 -J.693' l.0,320 o.11s1 -o.0426 o.o.306 1.01)8 1.0463
.39 -J.4000 1.0190 0.1362 -0.0267 0.018,$ 1.0092 1.0274
~ -J.6739 1.0100 o.09ao -o.ol.45 0.0096 l.0048 1.0148
41 -J.6361 1.00~ 0.0591 -o.oose 0.0034 1.0019 1.0049
L2 .3.4736 1.0020 0.0192 -0.0009 0.0037 1.0018 l.0038
1 -J.~806 1.0000 0.0000 -0.0000 0.0000 1.0000 1.0000
Tabb .3• (C .)
...
zfxr-tcos v K~q b 'fCOI Vb ~ • !!.!l• D .• t.ln. £!.a•l'!!n.
~ ;r- 4!-~
ol
22 o.~17 -o.ei84 o.HS6
COi
:o.!fss d {>(
2
ro
10414L 1
2J•J2
23 o.28JS -o.h127 0.1937 -0.2.)68 78-50 103-42 2h-$2
24 0.3133 -o.3613 0.2204 -0.2139 77-16 ~02-21 2S-OS
2S 0.3459 -0.2726 o.2s1s -0.1700 7S-26 99-48 24-22
26 0.371.) -0.2367 0.2762 -o.1SJ2 13-S1 90-li? 2h-S2
27 0.3989 -0.12.36 o.JOSJ -o.oeso 72•13 94-5.3 22-40
28 0.4220 -o.o9SS 0.3316 -o.067$ 10-38 93-Sl 2.3•13
29 o.44li1 -0.0323 o.35e9 -0.0236 68-58 91-21 22-2J
):> o.!:6S9 0.01.38 O.)e68 o.01os 67..J;.h 8'~ :2-10
31 o.h849 o.os11 0.4lh2 0.0409 6S·32 81•39 22-07
32 o.so.34 o.o9SS O.h370 0.0100 64-0S S•.32 21-27
JJ o.s202 0.11+01 o.4623 0.1181 62-.26 eJ-n 2o-45
l4 o.SJSl 0.1711 0.4660 0.1492 6o-SS 81-2S 20-)J
JS o.5487 0.2027 0.5088 0.1820 S9•2S ?9•30 20-0S
)S o.S6o6 0.2314 o.s2a3 o.2us ;s-01 77-l.i7 19~0
37 o.s112 o.2S16 0.5476 0.2,368 56-48 76-18 19-.30
JS o.S797 o.21lS o.5617 0.2617 SS-So 7h-SO 19-00
3~ o.sa10 o.269S 0.5761 0.2820 si.-u9 73•37 18-48
kO o.S926 0.))17 o.S867 0.296S Sk-ok 72-47 !8-lil
hl o.S966 0.3104 o.s9ue 0.3091 53·.JO 72-00 l8•JO
h2 o.S992 0.3153 o.s9 0.31)8 Sl-16 71-Ul 18-27
1 o~~ O.Jl62 0 ·222.2 o.~162 2~ z1-~ lS-26
T&bl l• (C ->
l l /){_
Point 1
22 20 ~ o. J o. o., o. 0.1 2
23 207-2.b 128-34 0.9511 0.9716 0.9442 o.h20S 0~1768
2li 204~ 127-26 o.97S4 0.9769 o.9S4S o.~239 o•.1796
2s 199-36 l2S•J9 0.9679 o•.96S4 0.9712 o•.4126 0.1701
26 197• 12h-h1 0.9610 0.9882 0.9762 o•. 20$ 0.176
27 169-46 117•33 o.ss22 0.9964 0.9910 o.38S4 o•. t46S
23 187-.42 116-0Z. 0.9434 o.wn o.99so 0 •.3942 o.1SS3
29 162-l,2 113-44 0.9334 0.9997 0.9990 o•.3808 o.1hS1
.» 178 111·.34 0.9221 0.9999 0.9999 0.3773 o.1426
.31 17S•18 109-L6 0.9102 0.9992 0.99 0.376.S 0.1416
32 171-oh 106-59 o.en 0.9910 0.9940 o.36S7 0.1337
33 166-26 103-SS 0.8861 0.9930 o.9e6o o.JS4J 0.12.se
3b 162-?0 101-SS o.e139 0.9888 0.9781 0.)$02 0 •.1226
.3S 159-00 99•.lS o.&)09 0.9832 0.9668 o.3434 0.1179
)5 lSS· 96-27 0.8491 0.9773 o.9S41 0.3)66 0.1132
37 152·36 SS-h o.8366 o.971s 0.941 O.JJ.)8 0.1111
38 149-40 9.3-50 0.8274 o.96S2 0.9316 0.3256 o.1os9
3? 147-14 92-2S 0.8173 o.9$94 0.9200 0.3223 0.1037
ha 116-.34 91•)) 0.6077 o.9ss2 0.9124 0.3209 0.10.30
bl "44-00 90-ll o.so39 o.9!)11 0.9016 0.3173 0.1008
li2 143-26 90•10 o.eo14 0.9h9S 0.9015 0•.316$ 0.0999
1 II ~ 9Q=OQ 2· 0.2!i§l o.am O.J162 0.0228
Tlhl• 3• (Cont.)
1ln/3 lln 2T 1fn

1tn,2B•oc.l ,e.cc.2 008 .O{
!JclM
olnt 22 ~J l 2 2
22 0.1~61 O.JtlltO -0.2328 o.9 o. 0 • 4! .1 .)
23 o.1a19 O.hlOS -0.2304 0.9073 0 •.3819 ..,.o;6; J.hooo
24 o.79i,.1 o.L~2 -o.2oe6 0.9057 o.)SJS .Q.ShSo 3.14h9
25 o.e126 o.1&001 -0.16?.3 0.9110 0.3159 -L&.089() J.06e$
26
27
o.e223
o.e866
o.uss
o.;;e19
-o.1517 0.9013 0 • .381 -.). 7800 3.0943
-0.0861 0.9228 0.3'76 •).6206 2.7300
28 o.e9a3 0.3922 -o.0673 0.9190 o•.35.37 -3.2957 2.7086
29 0.9154 0.3800 .0.02.3$ 0.9247 0.)$20 .3.0095 2.s201
)) 0.9):)0 o.m1 0.0104 0.9261 0.3494 -2.6.310 2.s~
31 0.9411 0.3762 0.0408 0.9264 0.3487 ~.47$0 2.3200
32 o.9S6h o.J643 0.0776 0.9307 0.3400 -2.38$6 2.1aos
JJ 0.9704 o.JS18 0.1112 o.~.3Sl 0.))18 -2.)70$ 1.9694
.;:. o.nm. o•.%3 o.i:.74 0.9367 o.~19 -2.0764 1.910?
l:) o.9860 0.3369 0.1789 0.,392 0.3220 -2.o673 1.6087
)S 0.9937 o.32e6 0.2061 0.9h17 0.3169 -2.0000 1.7069
31 0.9949 0.3240 0.2.300 0.9426 o.Jlla -2.2hS2 1.63$1
JS 0.9975 o.Jlb2 o.2s20 0.9455 0.307) -2.)188 1.s400
39 o.m1 o•.)J6S o.210s 0.9466 o.))!6 ~.S186 1.b802
1'0 0.9997 o.)066 o.2a34 0.9h71 o.JQJa -2.6999 1.~o6
bl 0.9999 0.))17 0.2~ 0.9483 o•.J>Q8 -2.9398 1.1:000
42 1.0000 0.))00 0.2977 0.9 86 o.):)03 .3.2216 1.3500
1 1.0000 0.2222 0.l)OO o.211 I o.mo ·l·~~zz i.~zsz
41
Tabl1 l• (C 1.)
Tb
iol
22 -o. 19 .2 • -1.
-el IX
LO
-2.311 - • 7
2
-2.)400 ..3 • .3
23 -o.s1e9 1.249S ).2.)65 -1.221 -2.1752 -J.26U -2.2131 .3.3)6.)
24 .o.S91'8 0.9947 3.08?.3 -1.2699 -1.99b9 -2.8121 -2.0453 -2.8800
2s -o.S7kS o.?OSO 3.1166 .1.26S6 -1.$950 -2.sseo -t.6 99 .2.6W)2
26 -0.6000 o.S862 3.os99 -1.2915 -1.2GS7 -2.,3Sb6 -1.).38$ -2.4h?S
27 -o.Sll&S o.Jl!i 3.1700 -1.2387 -1.40.$1 -2.21'61 -1.47SO -2.3(,04
2 -o.Shli6 0.2226 J.10.i; -1.2.389 -1.1347 -2.0872 -1.4,X>l -2.2186
29 -o.siss 0.0766 3.ol.sOO -1.2s11 -1.0046 -1.8649 -1.0712 -2.oooo
.YJ -o.So.$6 -0.0263 3.099b -1.21'6.) -o.6222 -1.5268 -o.61SS -1.9 00
.31 -o.S2SJ -0.1011 2.9032 -1.2937 -o.68o) .J..So64 ..0.7476 -1.6;2
32 -o.L9$6 -0.1602 2.646J -1.2613 ..0.1001 .J..1;048 .o.1eoo -1.;618
)J -0.4717 -o.27o6 2.12s1 -1.2)16 -o.5728 -1.2229 .0.9849 -1.3791
34 -o.4600 -0.):)99 2.7096 -1.2320 -o.62$S -1.1677 -o.111s -1.3J8S
35 -o.4h2S -0.3809 2.6400 -1.2111 -0.6211 -1.0450 -0.7219 -1.3.300
.36 -0.4232 -0.4322 2.s1so -1.1e1s -0.7163 -o.961) -o.eL;o -1.1322
-o.t,106
~
-o.SJ18 2.~000 -1.1600 -1.0209 -o.B052 -1.2199 -0.9665
.0.3910 .o.62)1 2. 00 -1.1312 .1.1696 -0.6937 .J..~us .o.8J7S
39 -o.3528 -o.7J8S 2.4000 .1.0335 -1.3912 -o.6280 -1.1023 -0.7700
LO -0.)780 -o.8)7S 2.3500 -1.1188 -1.6373 -0.3937 -2.0280 -o.456$
h1 -0.)5;9 -0.92L9 2.3199 -1.0916 -1.90S7 -0.3034 -2.)800 -0.3790
b2 -o.)468 -1.06S4 2.JCX)O -1.ob20 -2.1864 -0.1926 .7396 -0.240
1 -o.J2z2 2.22 -2. 062 -o.oz6a -2.2600 -o.o~
2
1h••• real interval• ere in col h of Ta 1• • alu11 of Ft'b
•nd b arc sh ln col S and 6 of Table 4.

St lla
Forcc1 b ar. acting •long th• parabola 22~2-1. 1hu•!cre, their vtrtl•
cal c 1 en be al td by th• .Coraul•
rsb •lnI · rnb

•t•
dnD• .z..
1·_z...1_+_
· _.1.
I
I 11 11 dtfl In atep 10.

The hotbantal C°"10M My be
~ coa'f • ffiSb
ere
Yalu•• ot lino , col( , , and FVSb &r. in col 7, 8, 20 21 of

Table b, r pec:tively. Mb 11 conllderc p !Uva en dinct• dowmm'd,
ffiSb ls pcdUve en direct tr 1 to 22.
t 12•
1 pl n t the corn..,>cndlng ;>olntl
bei a ~n• tc:cu, th 1 · t bl cont&ln 1
th• plane tang t to th• hyper at th• points. ThcrU'ore, they n t .-:ti •1
th• lntera t Ion ot th• pl • nor 1 and t• , t to the pu1bol• 22-la•l at
Ch point.
the UH of CllCul lt Cl ba sh that th dlr Uon coalne of th•
nor 1 to• pl to 1 cutain 1urt1 • u iv n b,y
43
Te 1• ii. r~c •
1+122
FSi, k.
-.i
•
Point ' 1 ~ 6
i. 9;, J.2 2. • ·.3· -5· 1895 o•
2) 1.7614 3.102s 2.9b11 2.9529 -6.SJSO .9.e;23 o.S999
24 1.7246 2.9742 2.92o6 2.9))6 -?.994h -6.4407 0.$885
2S 1.seoJ 2.4973 2.0373 2.s1~9 -'..i· 749S -7.6009 o.ssse
6 1.s1YJ 2.4643 2.8320 2.el!£6 •J.7941 <,.9377 o.$SS1
2.10ss o.L90S
27
28
2g
1.3291
1.3233
1.2ose
1.16SS
1.7600
1.h$86
2.7037
2.6h76
2.768
2.70h6
2.67S6
-~.0840
.3.8678
-2.8661
,.mb
-6.S.35S
-s.3)12
0.4900
o.hSb
)) 1.1241 1.26SS 2.611S 2.629) -1.'/S62 -!).206U 0.431
31 1.06a2 1.1400 2.sea!i 2.sm -1.9h,;8 -4.2962 o.us2
.32 o.9S91 0.9187 2.$415 2.$660 -2.oois .0072 osns
33 o.6SlS 0.7286 2.$600 2.$)18 -2.s133 .3.;192 o.3333
JU 0.777) o.ShSl 2.b69 2.Slli9 -1.6044 -J.)662 o.314e
3S 0.6825 o.i.t.) 2.hS.36 2.!i617 -1.7771 -3.2741 o.2a90
~ o.ssco o.Jt..34 2.L2B9 .~1.&12 .0600 -2.S6S6 0.2417
37 o.scn1 o.2s11 2.~lOl.J 2.4196 ~.8134 ~.242S 0.21~
.)6 0.4126 0.1704 2.3917 2.!tOlO ..3.3900 -2.0213 0.1727
39 O.)H6 0.10.34 2.3790 2.3853 ....-l.0496 -1.sm O.lJSO
ha 0.2,aa o.os12 2.~s 2.3737 ....!,.5007 -1.1521 0.0967
h1 o.11.os 0.0196 2 •.)601 2.3643 -?.622S -o.89$S 0.0831
L2 o.Ol6J 0.0021 2.3579 2.3S90 -6.1616 -o.s68o 0.0892
1 0.0000 0.0000 2-~1.2 1.178.$ -J.1879 -0.1130 0.0000
Table L. (c~.)
?olnt
ill·&!·
1.0 --
llEc. 11ln.
G-
22 o. • ) •J -1. • 0.7300
23 o.eooo -0.6986 134•19 -1.0241 0.619 Sl-42 0.764
24 o.aoss -o.71S8 13$-4.3 -o.91S3 0.5747 S4-SS 0.8183
2S o.8))9 -0.7.lSO 137·16 -0.9228 o.s1oa S9·17 o.8S91
26 0.8320 -0.75~7 138-00 -o.6693 o.4822 61-10 o.8760
27 o.s1os -o.11So lh0-48 -o.6156 0.4000 ~s o.916S
28 o.a?OS -o.194s 11.2-37 -o.76U1 0.3740 68-02 0.9274
29 o.689S -o.8lli 144·.34 -0.711) 0.3226 71·12 0.91'66
)) 0.9006 .o.8.)21 146-19 -o.666S o.2s79 73-16 0.9577
.31 0.9091 -o.6Sho 148-39 -o.6092 0.2534 7.)-19 0.9673
.32 0.92.so -0.812> lS0-16 -o.S600 0.2119 77-U6 0.9773
33 0.9200 -o.e 152-59 -0.5099 0.1700 80-13 o.96SS
J4 o.923s -o.90he J.Sh-1.&6 -o.1;706 0.1401 81·29 0.9890
JS 0.9600 -0.9250 157-40 -0.1'106 0.1166 83-11 0.9929
;;6 0.9696 -o.9t-OS 16o-OS -o.;6ll 0.0872 64-00 o.99hS
31 0·977S -o.9SS6 162-52 -0.308; o.0646 66·16 0.9979
~ o.9aso .o.968S 165-35 -0.2571 O.C444 87-27 0.9990
Y) 0.9 90 ·<>-91~ 16S-2S -o.,oso 0.0261 c-28 0.9996
hO o.9a94 -0.9899 171-?1 -o.1432 0.0133 89·12 0.9999
41 o.99as -o.99h.) 173~ -o.1o;o 0.0001 B~.30 0.9999
h2 1.00)0 -o.99n 179•11 -o.olhJ 0.0001 90-00 1.0000
l l.OCOO -1.0000 180-00 -0.0000 0.0000 ~-\>O 1.0000
Table h• (Cent.)
-.
~olnt
22 o. • -2. 0 -i. 71 • l • 0
2.3 o.11ss 1&.68$0 .Ji.SS69 .J.S7SO -2.620) 7.89SO s.9000
24 0.6982 h.1692 -4.J:)l.&8 .3.S208 -2.4779 6.1ess h·9S13
2s 0.6782 3.22o6 -3.4965 •J.006h -1.7821 6.))56 1i.22so
26 0.6$61 2.$043 -2.6.)99 -2.$087 -1.3723 S.7SS2 3.8471&
27 0.6320 2.ses2 •).1647 -2.9000 -1.26$9 S.69$1 3.2011
28 o.6o71 2 • .3478 •J.068S •2.&M -1.11s9e s.2019 2.e.;-,s
29 o.S798 1.6.$88 -2.3542 -2.2274 -0.7)67 4.7SOS 2.4296
)) o.SSh6 0.9746 -1.~611 -1.boS6 -o.ti2ss h.6SS6 2.2401
.31 0~203 1.01os -1.6~ -1.6022 -o.4150 3.9000 1.781.S
32 o BM 0.9772 -1.7 -1.70S3 -o.)662 J.9000 1.3100
33 0.1543 1.1140 -2.2400 -2.22s2 -o.)810 .).2.;86 1.1710
.34 o.!i2S8 o.16e1 •1.629.3 -1.6091 -0.2416 J.1008 1.060)
35 0.3800 o.67SS -1.6422 -1.6324 -0.19:$0 J.138.3 0.945)
J6 0.3398 0.7000 -1.9m -1.9262 -0.1690 2.1aoa 0.6899
31 0.2946 o.a300 -2.66o6 -2.S100 -0.1662 2.1974 o.472S
J8 0.2490 o.w.o .3.2800 •J.27SO -O.J.168 1.9892 0.348J
39 o.2008 0.8099 -J.96S3 .3.96JS -o.1osa 1.811.3 o.2478
40 0.141 o.6816 -h.7498 -b.7498 -o.06SS 1.lli17 0.1113
41 o.10liS o.s66S .t;.60S7 -S.60)7 -O.OS60 o.ee97 0.07~
b2 O.Olh.3 0.092$ -6.4516 -6.l.S16 -o.006h o.S6eo 0.0496
1 0.0000 0.0000 -2.1ez2 -~·1812 -o.OCXX> 0.11~ 0.0000
6
Tabl• • (Ceft:l.)
PH FV
• •
6~;
Point 72
22 5.1"1
2) s.0141 lo.saso
24 b•.X>76 9.1405
25 J,.52l'J 1.W66
26 h.3529 6.:!)17
27 4.4292 s.1e..s3
28 h.0$21 ~.18Sl
29 3.99.)6 4. 84
.)J h.26ol 3.2147
Jl J.h5SO 2.7920
32 J.SJ18 2.2872
33 2.6576 2.2aso
34 2.8590 1.7290
l) 2.9433 l.6210
j6 2.6118 1.3699
37 2.0312 i.~2s
JS 1.5434 1.1923
39 1.10.ss l.OS77
~ l.0762 0.7929
41 o.a337 o.66o9
b2 0.5616 0.11&21
l o.11p 0.0000
h?
ere
f 11 th• • 1• we the llOl"•l • th• oaltlv• dlrcctl t th• z-
axle.
1hl• r•l to th• t• lftt 1• 1110 •1 to th• dlrteU of th• tore•
f'Tb "1ic 11 contal ed In th• ta tnt lane. Tlltretore, the torcu Frb •k•
•n angle 180• ( vlth th• XY-pl•n••
Tb• vutlc11 component• ot Ffb can bl obtained, thcrcfot• 1 !ff
Mb• PTb •ln (180-/ ) • FTb atnf
th• projected COlllpon•nt• of rrh on th• XY•;:>lan• are equal to
FRI']) • F1'b OI (180•/ ) • -n'b co.f

Forcff FRTb ate not parpendlc:ular to the proJtc:Uon ot th• gro1 22-42•1.
The angle ot ln:llnltlon of thtH torcu vlth the projtetlcn o! the groin can
bl obt•ln.t<l troa Plg. 9. P'l'ca thll tlgure lt lhbs lhcwn thtt
12 + bJ. • L2
•
•110,
~erctore,
t
48
·.
flg.9 - ANGLE €. ..
-2 • c2 + b'J. •2 Cb COi 6-e
Th•t•t re,
a2 +~-Cc t•nt- t {)2 • c2 + t:J. -2 c COi 6·
•ca obt•ln
cos~• -tan 1tt f
ire
1J lrdf Ir• the • le• •fine ln 1tep1 11 Ind 12.
en th• 1 111 ~ are cbt11ned, th• fore 1 f'RTb can ruolved. lnto tvo
ca11p nt1. 1h•H arc noraal to the proj«:Uan ct th• groin, t.e., M'b •
th• c •nt• t• ntlal to th• project! Cf th• roln,
t.1., Tb • f'Rfb x c e: • Tbtse nor•l c:e111:>cr21nt1 are lance .by th• or•l
c ~cnt1 t th• Jlcent ~tr.
ta f , c e- , ~. 11 €:, M'J>, mb, mb •nd ftlTb are

in 11, 12 1 131 1.ij, 16, 17, 18 • 19 of Table !&, r ectlv1Jy.
St l)s
1b• Uve 1 d 11 U11ly • 1mtd to be lt01'aly dlltrlbut9d a horl•
z tal proJ t I • 1beretort I 1n Cll ul•t lng th• ltfelHI du• to thl• type or
. 1 din I • JJ, Jl .32 •1 be ed. 1hb hpl111 thlt 1 thod 1lallar to
that uaed ln st • 1 to 11 l c:alc l•t the 1tre1111 d to ad lced ould b •
tolloved. Hevner, th• 1p 11'1 Uve loed t roor1 a hotlz tal proJtctlOl'l
c1 be r ..ce to • aurtllce le1d !nee an curate c1lculmtlon ot th• curved
1urt1e1 b p dbl•. lhl1 surflc• 1re1 can obt•l~ dividing th• total
vuttcal forcu c trlbute<t to th• groin, ln col 2) 1 by th• d•d
lced ptr cnlt tr a, t.c.,
zs.112
1urt1e1 are • o,os • 15lla.25 1q. ft.
50
the L. L. to be equal to .)'.) lb. per aq. ft., th• ratio ot tct 1
1urt1C• to th• bork t•l projtctlon 11
~ • 2.472.
1ht L. L. pa t 1urt111 ere• 1• then
?ht tuUo o! L. L. to D. 1.. i8
~-~·02U3
rr;i:; so.oo • •
To lrolud• • e.ttect L. L., th• w1 • t 1tHtH1 ind tote•• 1lntd
in the prevt ould be ultlplled

Step Jlia
Yalu t th• hotbmt•l •nd vuUctl cmipcx.-ntt o! toa:a •t .acb polnt ot
the orcln tre own tn eel 22 . 23 Table Ji. U thc•c Vil S Ire
Uplicd by 2 x 1.243 (in order to 1"llud• tbc forcq c~trU:iuttd to th• groin
fr the adjacent p•nel th• tt!ac:t ot llv• lead), th• total torce1 acting
en tb• groin vlll be t1lned.
Atter 111 the tore act! on the ,roln are abt.dn 1 lt can be est
u a thtcceblng ct a tv ln cd arch. 'nl• efflCtlvc vldt.h ICU 11 an arch,
•• recoaei~•d by th• Portl• Cta AH 1auon8, is l to 1.s2fa, tn
t h t I• the annge ndl or th• 1'Ch at th• p int ot lnt1r11et1on o! th•

adjlc t h,ypcr. Hr.w'v t, ~'lt duls; ct tht areh 1• caitt-4 line• th• •in pur-
pose ot Is r t 11 to d l;n the 11 p rt ot th• atr~tur..
st.-,, isr
OlrtcUQ Of prL-¥:1;>•1 ltHIHS can be detua!ned by l•tU s ln Eq. 40
be 1 to ztro. 'there.tore,
Sl
2T lln iX._ + Ty tt In 2 C)(

tan 2 J?•
'.? fx + 2T COi IX + Ty CCI 2 6(
1ihtr• j3 • 90•fi ln tht. c•H •t• the ang1H ot pr1":1pa1 stresses vith the
podtlv• rt ot t.h• X-gmentors p11•lnG by the point. Value ot prlnc:lpal
1tru1 s arc, therefore,
2
T2 •••x ci'2p . ~ C$n! JJ C!! (£..P,) T era CB~l.
• n °' + iea aln P<.. + :I •ln O<..
F point 22 it Is t that
Tx • -s.3623, Ty • 2.e1s1, end T· 4.15.

th•nfo:e,
•
jJ. 66 ~J·.
sln j • 0.9162, cos fl • o.4006

sin (j..ol) • o.2~01, coe ~~ • 0.9682.
Fr thl• • can obt.aln
T1 • 6.68 h/tt.. ~r salon
T2 • o.36 k/l.t. tension •
• 0.36 )( 1.243
ft' • J x 12 x 1000 • 12. 4.l psi.
S2
TbeH valuu ara vtll below th• allowable liJllt1 spsclflld by th•
Jlalldlng Code CACI .318-6J).
By lmpectlon or otb.•r v~lu•s e! T, Tx and r1 , thown In oolu.n 3l ot Table
1 •nd aoluea 1 •m 8 Of Tlble 3, rupectlvdy, it !1 evident. that no further

checking 11 required.
1h• •bc71• 1tr11se1 do not require 1ny relnf or.:eaent. However, ACI COll-
alttH J.3411 st1tcd that "Kinlaum rtlntorcuent shall be prwided •• requited
in th• Building Cede (ACI 318-63) wen where not ttquit~d l:I;)' 1na~sl1. • This
relnforaeiaent ii usUlllly prwided. to aoco:aodata unsyaaetrical load.I •nd 1tre1se1

due to vol\1Htr1c changes.
The 11inlaua rctntorccacnt spec lfitd. in crUcle 807 JC,I 318-63 may be uted
ln this case. Therefore,
Min .Al • 0.002 x 12 x .3 • 0.072 1q. ln. per ft.
Ullng #3 •t 12 in. c. to c. provldts 0.11 1q. ln. p.r ft.
lt 11 deslrabla to place th• rdntorccaent •long the genenton, so that

aw:h rclntOl'C:eaent tuna •long 1tralght lln••·
S3
Discussion ot Results
As antlclpat•d, lt Is evident trca th• results obt•lned ln step 1$ ot th•

dtslgn fer the grolnm vault that th• stresses ln concrete are well bclcv the
•llowble stresses spec !!led by the >Cl :CUUdlng Code .318-6). However, th•
thiaknHs ot th• shell ls not usually dictated by tha strength rtqulraents but
by the cover onr the reln!orcemcnt, better ln1ul1Uon, •nd fire proofing.
Th•nfou, lt 1c.• that th.r• ls no netd. to reduce the th1cknu1 of th• shell.
An approxl•t• check ot th• accuHcy of the result• can be obh!ned by
C:Olllpltlng the ttal surface ar• of the tr11ngl9 1-B-42• 1 to Its projection on
tht XY-plant 1 as 1ugge1ted b;y C1nd1116 1 l .e.,
1
•
*~:~s 2.472.
Thls ratio Ht• to be r•sonable, e.peclall;y 1t lt ls COllpH'ed to the
ratio 2.8 obtained by Candd• ln his exanpl• of • groined v1ult 20 •· sq. ln
plan and with 1 rls• ot 10 ••
Sh
Coaclualon
1h• Mllbtan1 theory provn tot>. 1 11tls1'1etory method to glv• • re11oaabl1

e1tl111ta o! the stresse1 in h)'perbolic paraboloid shells. With th• provl•lon
that they are subjected to • Wllforlllly dlstrlbut.ed losd, th• bound.lr;y condlUCllS
satisfy th~ conditions or equilibrium, •nd th• rtse-to-span ot shell r•tlo 11
1
within the 1pecified llait o! >·
'Ihe hyperbolic p•nboloid lhells have bee • very pcpular end vldely
t1vortd bec1use they ara vcll ..dlpt•d to th• use o! ceff:rcte. They show a
l'eurkablc rcd1ta11u to any fora ot vibtatiOflll lhcck. They ire 11eon011lc1l
•nd. 1110 give the erc:hitcct an q>portunity to dtvdcp aore l•gln11Uvc 1nd
gl'lc•tul snap•• Of •tructures than th• c:c..ventlonal l!taa and colu. bsildlngs.
SS
Acknowltdgacnt
Gntei\&1 1ck.novlcdgasnt ls 9clc to Protu1or V. H. RoHbl'•ugh tor
•fficicnt direction of th• vork, 9n1...rag..cnt throughout th• preparation

of this report, lad tar aid in organizing th• •tcr111.
1HE STRESSES IN HYPERBOLIC P.ARAOOWID
s-IELLS USIN:J 1HE MEMBRANE 1HEORY
by
Yousif Kellow Dawood

B.S., University of Baghdad, 1959
AN AffiTRACT OF
P.. MASTER•S REPORT
submitted in partial fulfillment of the
M!>STER OF SCIENCE
KANSAS STATE UNIVER:>ITY

Manhattan, Kansas
1965
AI~ .AB3TRACT
The .me:'!lbrane theory in :nany c;:;ses £;ives ci reasonable estimate of the stress
distribution of dou!Jly curved shells. Hi th t!"le prov is ion that they are sub-
jected to a unifor;nly distributed loe.d , the boundary conditions s~tisf'y the
conditions of equilibri~~ , and the rise-to-span of shell ratio is not less than
1/5. It is a gocd :by-pass method to avoid the elaborate calculations encount-

ered in using the bendin£ theory.
The hJ;perbolic par2boloid surface can be defined as a surface of trans-
lation for~ed by 2oving a vertical par8bol2, opened upward, over another
vertical ~zrabola opened downward. It can also be defined as a warped paral-
lelogra;il formed by r.: cving a straisht lir.e along two nonparallel, nonir.tersecting
lines in sp:::ce.
T'ne equation Z•1XY represents the sinplest possible equation of the second
desree ~nich describes the surf8ce of the h~erbolic ,araboloid.
T:te seneral equations for membrane stresses for doubly curved shells c2n
be obt2 ined :by writing the eJ..?ress ions for the eqt;.ilibr iu;11 of a:i ele;ner.t for med
by four intersectins lines contained in the surface. Tne expressions for
e~uilibri~~ can be sin;ilified by transfcrminn the actual stresses acting on the
ele;Je:it into fict it i0us stresses zcting on the ;;>reject ion of the ele:n.ent on the
XY-plcne. Tne resulting expressions 8re in the form of three partial differ -
ential equations of the second order. 1be use of the stress fu.4ction will
reduce these three equztions to one. This will facilitate the solution of the
differential equations. However, in the case of the h~erbolic paraboloid shells,
subjected to a unifor:c. surface load or t.mifornly distributed load on a horizon-

t8l proJection, direct inteGration of the partial differential equations is
rel~tively si:nple.
L1e hyperbolic paraboloid shells have become very popular 8nd widely
f::::vored because they are well-adapted to the use of concrete. They show a
re:;.ar~-..able l'esista:tce to 2rry form. of vibrational shock, are more economical,
a:1d also sive the architect an opportunity to develop ~r.ore imaginative and grace-
ful shapes of structures than the c onve;it ional beam and c olur.m buildings.
Tne sroined vault is taken as a design example. Tne basic stresses in the
structures are found by using the equations of stresses previously derived.
However, readjustment of stresses is required i!'l this case due to the existence
o! a free edge ·which cannot withsta!'ld any types of stresses, since there is no
structt:.ral ele~ent which is sble to resist the;n. This readjustnent includes
cancelling out the stresses calculated at the free edge by introducinG a set of
stresses equal and C?posite to those obtained at this edge. TI1.is involves the
introduction of the same additional stresses at the sroin. Further~ore, the
horizcnt~l and vertic~l components of forces contributed to the groin by the
shell are calculated, 2nd a suggested procedure for desi£n of the groin is pre-
sented. TI1.e principal stresses are also obtained which are used as the b2sis
o; ~esisn for the concrete shell section.

Concrete Shells Derived from Experimental Shapes
Heinz Isler
Dr Eng.. Burgdorf. S\,itzerland
Summary and the ame. l11ey arc merely two dif-

ferent aspect of the ame challenge.
Thi paper presents the author· experience in the design and con rruction of
1n non-shell huilding design. the archi-
more than one thousand concrete -;hell tructure over a 40 year period. The shell
tect defines the structure·!> hape and
shapes pre ented are not based on geometric concepts. but result from shape-
the engineer define the tatic de ign.
finding experiment l11e hapes are created automarically by natural la\\S.
In shell de ign. both aspect arc unit-
ed. Preferably. de ign of a hell is the
work of one indh idual. If more than
Introduction re ist gra\'il) and other load . Both the
one designer is involved. they must be
pace enclo ure and load-bearing
willing lo subordinate thcmscl\'CS to
A shell i a double-curved heet-like function are accomplished by one ele-
the emerging logic of the shell's form
roof made of rigid material. lf correct- ment. the hell.
ly shaped and con tructed. a hell can as it evolves through experimentation.
be elr-supporting. needing no girders.
tru e or beam to carry it. Further- Structural Simplicity Experime ntal Shapes
more. a hell need no roofing.. since it
A shell structure is inherenLly simple.
is the roof it elf. A rich repertory of hell hapes has
having the same internal and external
The two function of roof and roof- been developed. Thc e non~geometric
shape. There is only one element be-
upport are accomplished by one hape re ult from the experimental
tween the structure's in ide and its
structural clement. This dual function manipulation of a heet or membrane
ouL-;idc: the thin layer of the hell it-
expo ed to a variety of force and han-
of a shell is a great advantage and an self.
dling technique :
even greater challenge.
This shape has to folio\\ strict law of
Shell structure are economical. A sin- - 1lle pneumatic, or inflated hill
tatic in order to be both safe and eco-
g.le. imple element made of common. method, based on ob cnation of a
nomical. De ign of a shell Lructure
easily handled materials fulfils the two bed pillow by the author in 1954
cannot be split into the architect' ta k
(figs. /- -I).
main task of a building: the enclosure and the engineer's. The La k of archi-
of pace and the tructural strength to tectural and tructural de ign arc one - The hanging membrane method.
where th,e hape deriYcd is frozen
and inverted (Figs. 5-1 I). Thi
method of shape generation \\as
·discovered' by the author after ob-
cn ing a piece of cloth hanging in a
reinforcement mesh in 1955.
- Spanning of elastic membrane in
different edge frame.. "ith or "ith-
out ·stamping po. I · in the urface.
The flow method, whereby an cx-
Fi~. J:P11e11mmic f/wpe tferi1·ed from square pandfog foam pas ed through a tubt:
frame. preliminary model - quare or otherwi e - produces a
Fig. 3: AC'curate shape-fimling model, pneu- hape with a double cun ature (Figs.
matic method: Accurate elastic membrane is 12. /]).
inflated by air pressure. The coordi11ntes of
e1•ery cun·e are de1ermi11ed with 1111 e1cc11rt1- - and other technique or combina-
cy of0.02 mm. tion of techniques (Figs. f.I. 15).
Remarkably. each of these methods
can lead to a limitlc number of new
shape . Fig. 16 gives a hint as to the va-
riety of such shapes.
Shap e Characteristics
Each method of generating shell
hape imparn. specific and character-
Fig.1: Rectangular ~hells 20X 14 m. shape Fig. .J: P11eummic shape derfred from an ir- istic qualitie · to the shape it create:..
found by pne1111wtic method. 1955 regular frame For instance. the pneumatic method
142 Shell and, patial Structures Structural Engineering International 3/94

leads lo hapes \ 'Cr) similar Lo a . oap
bubble. Pncumaticall) generated hell
shapes have the same amoun t of medi-
um curvature a t every point of their
surface. a lthough the curvature varie
from point 10 point. The buckling sta-
bility of pneumatically formed she ll .
therefore. is constant at all point - a
rather ideal static behaviour.
The hanging method leads to mem-
brane that have o nl) tension tre ·se
When the hanging shape is frozen and Fig. 5: lfcmgin!! 111e111bm11e. prelimilwry model
in,cncd. the stre se change to com-
pres ·ion. ll1is characteristic i very
u eful for creating concrete hell
without tension. "hich thu re i t
cracking and the o the r well-kno" n ef-
fects of deterioration.
On the o ther hand. because of the
compression stresses.. these she lls are
a lso subject to buckling instability.
Shells of this type have clear limit as
10 span. thickness and free edge . They
must be handled by appropriate
mean . such as the ·bending· (up or
down) of the free edge to enhance
their !ability and provide sufficient
overall double cur\'ature.
Solutions for point- uppon of the Fig. 6: Shapc-meas11ri11g modt•l witli rm accuracy vf 1 '••1 111111 m t•1·ery poi111. prc>ducc>d from "
shells are rcadil) suggested by the fcmn·crea1i11g lw11gi111: experi111e111
forms themselves. The oblique reac-
tions al the supports points have to be
carefully considered. however. U ual-
1~. hori7ontal thru l i taken by a pre-
stressed tie. For corrosion protection.
tht! tie can be embedded in a concrete
beam. hs O\t.!r-pre tressing l!liminates
cracking. of the protecting tie beam.
a suring the truclure· durability.
Design Methods and Models

U ually engineer develop a tructural
design incremental!). starting from a
fixed idea a nd adapting it to each de-
tail requircmem untiJ a complete
scheme is achieved. Thi approach cer-
tainly can be followed in shell design: Fig. 7: Open-air theater, Gro1:.i111:e11. (icr111m11·, shape fowui by hanging 111ethod, fh·e·po111t·
howc' er. its ill\ er e can al o apply: a supponed shell 011 slope. 1977
numl1er of ha pc can be proc.luced and
compared in terms of the structure·s
sp1.:cific requirement . In any ca e. pre-
es narrO\\S. leading in mo t cases to a n Nevertheless, physical models are in-
liminary physical model. are recom-
optimal solution. o r to a compromise dispensable for insights into the be-
mended.
without too many deficiencies. haviour and appearance of new shells
Models can be a · e ed according to and absolutely should not be ignored.
the functional requiremem of the The ir creatio n i nol only enjoyable -
Physical i ersus Computer Models
1
building. including pccial con traints especially in the preliminary pha e of
such as viewing. lines in a stadi um or Today. most structures can be simulat- rough models: they al o sho" unmis-
acoustic properties in a theater. Static ed on a computer. Static and stability takably how the real structure reacts.
behaviour and erectability must be investigation of shell lructures can Overall or local instabilitie or "eak
evaluated and questions of formwork also be done by computer. ll1i ' Urcl~ point arc revealed "hen the model
and. of course. price must al o be con- i a great help and can accelerate the are made really thin. Even elementary
sidered. GraduaU). the range of choic- de ign proce s. load tests can be \'ery instructive.
Structural En ~in cc ring lntem alional 3/94 Shell and Spatial Structures l.J:>
helpful in the creative process of shell
design. but only when the structure
the) propose have been built and mon·
itored over a period of time.
An accurate physical model on the
other hand i the Lil..eness of the real
building. and contain all of the struc-
ture ·s problems and questions. also the
one not yet known. Unre tricted ex-
ploration of the unknown i. a most im-
portant tep in innovation. '111erefore.
phy ical models are likely to remain a
necessary design tool for new "ihdl
form in the future.
Fig. 8: /11tloor swimming pool. Brugg. Swir:.erla11d. Four-pnim-!tupporrecl sht!ll. 35 x 35 111. Shell Construction
1981
Form work
A crucial step in the con truction of a
shell i the de ign of the formwork.
Curved gl ue-laminated timber beam
or flexible metal trus e upportcd by
light caffolding are favoured methods
(Fig. 17). The curved beam cll.!fine Lhc
shape and upport the light timber
strips over which wood fibre . lab
[iJled with rockfibre insulation arc
placed (Fig. 18).
The curved beams can be reused when
Fif;I. 9: fl1ree-poi11Huppom•d sliells 11'irlt free edge.~. ltigltway serl'lce swrion. Sll'ir:.erland.
the same shell form is re peated. Dif-
1968 ferent projects can share fonmvork el-
emenL in order to keep co ts IO\\. 111c
resulting designs need not look al all
the same. nor even very imilar (Figs.
12. 13).
Cladding
Cladding i an even more important
cost factor in hell constructio n. Nor-
mally. double-curved and especially
spherical surface cannot be clad with
quare or rectangular panels. as can
single-cun·ed cylindrical surface-.. To
cover a sphere. every panel has to be
cut to differing angles. making the
work very expensive.
Pig. JO: Site/I:. for sports cewn te1111i~ lwll w1d indoor \11·i111ming pool. lleimberR. S11'ir:.er· Pneumatic. hanging and llo\\ -generat-
land. !97R ed hell . however. although doublc-
cun·ed. can largd) be clad by square
or rectangular panels. without cu tting.
For statically accurate physical model For shape-finding. accurate phy ical There i no direct explanation for this
test . a great deal of knowledge and modelling remain invaluable. The au- surpri ing fact. yet it ha functioned in
experience, a well a patience. i nec- thor ha done all shape-finding for new practice for decade .
e sary. Such a time-consuming process shells merely by physical experiment
can only be undertaken in exceptional This contribute con iderahly to the
or by design. This is possible when
ca cs due to its inherent expense. For economy of shells. Indeed. without thi
careful and very accurate techniques
ta tic or stability a sessmcnt. computer advamage. shell construction would be
are applied. lbis natural way of find-
programs arc undeniably quicker and substantially more expensive.
ing hapes leads to the answer directly.
cheaper. 'oneth~lcl>s. a healthy critical without computers or other aids. To- There 1s still a son of secret to the cffi.
ense hould be employed when inter- day. some computer programs aspire cient cladding of hell : the operation
preting the result derived from com- tu simulate the c direct experimental must tart a t particular lines on the
puter. processes. Somcda) computers may be shetrs surface. otherwise all the panels
I+! Shell and Spatial Structures Structural Engineeri ng International 3/94

follo" ing the first ro" must be cut.
The slarl line \'aries with every shape
and has to be carefully identified be-
fore 1he site work starts, a nother a rgu-
ment for physical models.
l nsula1io11 Panels
The cladding panels arc at the same
time insulation slabs which remain a
part of the building. Again. cost is re-
duced by u ing one element for two
purpo es. The inf>ulation labi. arc ex-
Fig. 11: Remmg11lt1r shell~ for ll!lllli' 11,11/. \farin. S11'i1::t·rltmd. 1983
po cd to the inside of the shell. On the
upper side. their rough surface adheres
them 10 the concrete. which penetrates
it when poured.
About one panel out of 1en thousand
tends to fall after the formwork i re-
moved. Therefore. additional hooks
are placed on the panels to guarantee
. ccure attachment. The hooks must be
non-corroding and must not allow heat
loss through the insulation panels.
In ulation panel with noi e-absorbing
Fig. 12: Fo11r-poi111-mppor1ed slrell 11•itli [rel! edges for a garden center. Ca111ori110. Swa:;er-
surface arc recommended. This leads
land. 1971
to excellent acoustic properties in the
building. eliminating echoes. Perforat-
ed metallic heets are rather ineffec-
tive for this purpo c.
R ei11forcemen1
Experience has shown that shell thick-
ness should not be le s than 8-9 cm.
Two layers of reinforcement arc rec-
ommended (Fig. 19). Jn this way. defor-
mation and cracking under local
bending are kept to a minimum.
Ideally. reinforcement bar should be

as long as pos ible and join t should be
avoided. However. labour costs com-
pel the u e of reinforcement mesh.
Fig. 13: F11·e-poi111·s11pported .~Ire/I with camile1·er edge!>. 20()() mY. garden n'lllt'r. Parfa.
which is quickly placed but has 1he
Frnnce, 1969
grea1 disadvan1age of needing fre-
quent joints and overlapping. Overlap-
ping of 1he reinforcing me. h can ub-
~tantially reduce rigidi1y and force a
from the atmo phere, as well as allow- porosity at the surface. There are no
1hicker struc1ure. It is impos ible. for
ing the pene1ration of acid rain. The pores. or very small ones on the sur-
example. 10 overlap four nets in the
pores and channels in the concrete de- face of a well trowcllcd concrete shell.
lower layer and four in the upper layer
crease in size as the depth below the Thus. capillarity is re\•erscd. BelO\\ the
at the same poinl in a shell 8 cm thick.
surface increases. having a capillary ef- surface. the pores are larger than at the
The overlapping points have to be dis-
fect. The acid is transported deep into urface. so any moi Lure is 1ra11 ported
1ributed according to a clear concept.
the concrete. altering its basic chem- outward and atmospheric anJ \\;llcr-
istry and making it acidic. Instead of borne contaminant are pre,cntcd
Concrele Coi1er protecting the reinforcement. the con- from entering. the concrclc.
crete promote it corrosion.
There i considerable controversy In decade. -old shell structu1e. . car-
aboul how much concrete cover is Fonuna1ely. indeed ome" hat miracu- bona1ion is only 1- 3 mm deep and ha"
necessary for a successful shell s1ruc- lou ly. the phenomenon of carbona- slopped. Shells can succes'>full) ha\'C a
ture. ln traditional concrete construction does not occur in the author's con- concrete cover of 15-20 mm onl). Thi~
tion, carbonation can penetrate up to crete shells. 111e trowclling of the con- benefit wa not foreseen "hen trow-
30 mm or more below the surface. The crelc 's surface. which gives it a smooth elling the wet concrete surface 30 )Cars
surface draws in destmctive gases and clean appearance. also eliminates ago.
StTUctural Engineering lnlcm ational 3/9-' Shdl and Spatial Structu r~:.- 1-15
Therefore. the workers have to be fit
and \\ell 1raincd.
Monitoring
After a hell ha been poured. the con-
crete hardl'ned and ripened. the pre·
stres ing set and the formwork re-
moved. the real lifetime of the building
begins. The hell has tow ith tand grav-
itational loads. temperature changes
and all the innuence of ''ind and
weather and atmo pht!ric aggrc,...ion.
Shell h<l\C <,tatic propenie... quit~ dif-
ferent from ordinal) building Their
behaviour can lead to a long sen ice
life. but al o to premature collapse.
llle pecific peculiarity of '>hell., i'
their lo'" change of shape. From the
Fn:.. I .J. Pril'dtt rt»Htlence. 511 i1:.erlu11d. 1986 moment a hell stand on its O\\ n.
light hape denection · commence.
Denections C\Cn in Lhc ft~t fc,, day'>
after completion can be quite con 1d-
erable. and they continue to mcrca..,c.
Such dencctions are harmlcs-; for some
hape : but for other.; they can be latal.
rr. for instance. a deflection leads to a
flattening or a curved part. the rndius
of cun a tun.! around it increa. cs. \\ hich
results m mcreased tre es and de-
creascd -.1ability.
Inadequate or weak shell shape'> arc
characteri<,etl b~ their inability to limit
deflection before they reach a critical
tagc. For ..uch hells. collap c b onl) a
mailer of time.
u ~ g
fig.15. 1\1r Forn• \Jmewn. Diibemlor.f. 511'11:.erlwul. 1987
A good hell i poured in one 1ep.

0 u w
1/0
Concrete
The hell ha.., to be rrong. wcather-re-
i 1ant anti durable. A concrete mix
without interruption and without
joints. The etting of the concrete
~ ~
''ilh additives uch a plasticizer'\ can
accompli!.h this. Since the stresses in a
therefore has 10 be rcrnrded. In thi
way. concre te at temporary joints can
.
~ t!!j ~
\\ell designed and \\ell baped shell arc
not \Cf} high. there is no need for high
trcngth or special concrete. \Vorkabil-
accept new material. Tile duration of
the setting lime ha lo be carefully e-
lected in consideration of \\Cather con-
dition . Seriou weather forecasts are
e.
~ (3
~
i1~ i., the most imponant factor.
Un1formit~ of the mixture is 'nal. A

mdispen able. Pouring is to be a' oided
in rain) weather. a the ne\\ I) poured
Cl 0
handful of bad concrete can produce a
fi ·t-:.11ctl hole through the hell. Mo t
of the -.hell is inclined. The concrete
concrete \\OUld be \\ashed awa~. Shells
hould not be poured at temperatures
under+ 4 ·c. The thin layer or concrete
? e e
~ ffi
has to have sufficient tiffne By cools too quickly. etc.
adding tixotropic ingredienc . stiffne s
and \\Ori.ability arc combined. The The concreting \\Ork is not easy. a the
amount of cement should not be 100 workers must mo,·e about on teep Fig. 16: 'Sl111pt• ft11111lil'' •protlun·d hr p11t•11
high. in order to limit creep and lopes. Furthermore. the finished parts mutit:. hu111:111g or f1011 experimellt!> /I/.
hrinkage. normal!~ 325 kg. m·1• cannot be touched after tro"clling. 1969
146 Shdl Jnd patial tructure-. tructur::il E ng ineering Jnlc rn::ilio nul 3/9.t
Shell ... tructure.., should be monitored
pl!riodicall) after completion to gain
information about the character and
degree of <li.,placements. 'ormal hell
bcha\ iour indicate that di placement
arri' cat an end 'aluc after some years.
thus auaining a degree of st!curity. But
monitoring should continue.
There an! definite tabilit) limits for
-hell de igns. EH!f) hm compres. ion Fi~. 17: Formwork. for .\lu•ll /8X 49 m · g/11e-lm11i11att•ti be"ms un "light _\ct1ffoldim:. 19"'8
member ha., a critical load which may
not be exceeded. Theof). tests on-site
and ml!asuremcnt ha,·e hcd ome
light on the stability limits for hells.
hut the subject is not yet full) under-
stood. E pccially the combination of
instabilit) modes and long-term be-
ha' iour still need., a grca1 deal of sci-
entific investigation.
Future Development
The shell as a -.tructural form i o effi-
cient that it i" as..urcd of hright future.
Sooner or later. ecological concerns Fig. 18: Tu11her stripJ ""d im11/u1io11 1/ah.1 011 /1t•\ible mewl trllS!>e.i;.. Sic/1 Bldi:. Gem•rn.
"ill compel their \\ idcr U'ic. Dc\'elop- S11 it:erland. 1969
ing countrie., arc rich in manpower.
but poor in re ource Pro\·ided the
craft men and engineer.; in these coun-
tric., have the neccs af) skills. shells
should be \\Clcomcd and \\ idely ap-
There is a general trend toda) in the 121 ISLER. H. \ e11 Shapes for Slrcll1·
de ig.n of building. to brt!ak out of the Tin!lll\ Yean Aftt'r. JASS Bulletin 'r.
plied. 71 Tl. \ladrid. J9"'9.
boring uniformit\ of cubic architec-
Even in the richer countries. hell ture. Many new building are decorat- 13) ISLER. H. Typologie 1111d Ted111if.. dt•r
ha\'e their place. not only becau e of ed "ith ome form of curn!d elements. 111otleme11 Sclwlen. \\'erk. Bauen und \\oh-
their long-term economy. hut bccau...e impo ed for aesthetic reasons to inter- nen. Di:c. 1983.
of their pleasing 'ihapes a well as their rupt the monoton) of the hox-hke
[4] BILLI GTO~. D. P. The Tower 1111d
great varict). structural forms. Shell rorm · can pro-
tire Bmlgt•. Princeton Uni\'. Press. 1985. pp.
vide a remedy for design monoton). 210-232.
Grid . frame-. and other geometric spa-
offering a rt!al altemati\e.
tial 'itructure arc \'cry frcque111ly u'ied (51 RAMt--1. E.: CHL'1 CK. E. //<'i11:
toe.lay to co,er la1 ge areas. Indecd. for /\ll!r. Sdwle11. Karl Kramer Verlag. Stull-
large spans in excess of. '>a). 90 m. Education of the tudent "ho "ill be
the architect and engineer-; of the fu - gan. 1986.
shelb arc not appropriah.:. Ho"c'er. ture ha to find ne\\ (or rather. old) (6] BAL/.. ~I .Arc/1irecmral .A~pect~ oj Or-
for medium '>pan.... shells present con-
way . Student must be mo,ed a\\3}. at ~1111ic Forms 111ul Desil{11 of Concrete Sl1t•ll.1.
siderabk adn1ntages. IA Bulletin ;-..o.10632.19<'9.
least occasionall}. from the computer
monitor. The hands-on adventure of (7] ISLER. H Ele~mue .\.lc>delle. Deut'Ch\.
hapt!-finding is part of the proccs Bau1eitung. 7 1990.
Creation demand discipline. i...no" l-
edge. and opennes · Without a\\ 111 for (8) ISLER. H. Ge111•ra1i11e Shell Sh11pe.{ b'
l'hnirnl F•.xperm1e111.1. IASS Bulletin '\r
implicit~. shell hapt!s in ma<,scs
111 J.t Copenhagen. 1991.
would become tedious. \\Or c than
\\hat we complain about toda). (9] ISLER. 11. .\.fotlemer Sclwle11bt111. Ar-
~- cus 18. 1992.
Perhap the architecture of tomorro" I IOJ
COPCLAXD. J. A. Str11c111ral Art.
\\ill disco\'er interesting nc" territOr) ! Concrete Quarterly. Winter 1992.
[Ill ILER. H. The \\ay 10 Slr"pe. Int.
d flu? lt'.\I. i/ltotrtl//llllJ and :.lid/ :.IW/11'\ 111 tlit:. llrtl
ymp. on Innovative World of Concrete.
de 11rt' the illft•llec11111/ pmpt•rt1 of II /1/t•r
Bangalore. India. 1993.
[ 12] ISLFR. H. Longterm 8t•ltm·io11r of
C( • ~ Related Literature Slre/lr,, Seiken IASS-S)mp.. Tokyo. Oct.
c..
1993.
Fig. 19: Dt'for111<11io11 t111J.:le with one layt'r [I] ISLER. H. .\'e11 Sltape1 f<n '>lie/1.1. (13) ISLER. H. Concretl' Sire/ls Toti".' ·
of rt'111fonc111c111 and ll'itlr 111·0 (helc>11 > !ASS Bulktin :\r. ~ C-3. ~ladrill. 19:W. IASS-A CE ymp.. Atlanta. L"SA. 1994.
tructurul Engineering International 3/9.t Shell and Spatial Structure,. t.i7

Engineering Structures 75 (2014) 39–50
Contents lists available at ScienceDirect
Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
Design process for prototype concrete shells using a hybrid cable-net

and fabric formwork
Diederik Veenendaal ⇑, Philippe Block
BLOCK Research Group, Institute of Technology in Architecture, ETH Zurich, Stefano-Franscini-Platz 5, 8093, Switzerland
a r t i c l e i n f o a b s t r a c t
Article history: This paper sets out to explore the potential of combining a cable net with a fabric, in particular to scale
Received 3 November 2013 the concept of flexible formworks to the size of large-span roofs and bridges, especially when applying a
Revised 20 May 2014 thin coat of concrete or mortar to form a shell structure. By carefully designing the cable net and its
Accepted 21 May 2014
topology, and calculating and controlling the prestressing forces, it is possible to form a wide range of
anticlastic shapes, beyond those of the hyperbolic paraboloid.
A complete workflow for the computational design of a shell shape and its corresponding flexible
Keywords:
formwork are presented as a proof-of-concept for future work. Two prototype shell structures were built
Shell structure
Cable net
based on this workflow to validate the overall approach, to compare the built geometry with that of the
Fabric formwork design model and to identify further challenges when developing and scaling up the concept. In addition,
Flexible formwork a comprehensive overview of flexible formworks for anticlastic shells is presented to frame the present
Form finding research.
Shape optimization Ó 2014 Elsevier Ltd. All rights reserved.
1. Introduction between the combined cable-net and fabric formwork discussed

here and flexible formworks in precedent work. In addition, it is
Concrete shell structures, if properly designed and constructed, used to frame some of the design choices made for two built
are able to cover large spaces at minimal material cost through prototypes. Section 3 presents a complete workflow for the design,
efficient membrane action. optimization and form finding of these cable-net and fabric-
However, they are challenging to construct, traditionally formed, anticlastic shell prototypes, before discussing their actual
requiring full and generally rigid formworks, which are both mate- construction in Section 4.
rial- and labor-intensive. The materials are often used only once,
since they are customized for a specific doubly curved geometry.
Due to the amount of work involved, these structures are generally 1.1. Context of research project
not competitive in a contemporary building environment where
labor is expensive. The workflow and prototype structures, presented in this paper,
It is possible to reduce the amount of material, especially of the are intended to further inform and develop the design of the HiLo
falsework, by introducing a flexible formwork. In this case, the roof. HiLo is a research and innovation unit for NEST demonstrating
shuttering is replaced by a fabric, and the falsework by a cable ultra-lightweight construction. It is planned as a 16 m 9 m
net, supported by an external frame at its boundaries. The duplex penthouse apartment for visiting faculty of Empa and
challenge is then to design the flexible formwork such that the Eawag. NEST is a flagship project of Empa and Eawag in collabora-
resulting shape matches the designed geometry. tion with the ETH Domain. It is a dynamic, modular research and
After summarizing the wider context and the specific objectives demonstration platform for advanced and innovative building
of this research project, Section 2 presents a review of historical technologies on the Empa–Eawag campus in Dübendorf, Switzer-
and recent construction methods for anticlastic (negatively doubly land, to be completed in 2015 (Fig. 1). As a ‘‘future living and work-
curved) shell structures with an emphasis on flexible formworks. ing lab’’, NEST consists of a central backbone and a basic grid to
This review is intended to illustrate some of the differences accommodate exchangeable living and office modules, such as
HiLo, allowing novel materials and components, and innovative
systems to be tested, demonstrated and optimized under
⇑ Corresponding author. real-world conditions. HiLo is a collaborative effort of the BLOCK
E-mail addresses: dveenend@ethz.ch (D. Veenendaal), pblock@ethz.ch (P. Block). Research Group and the Assistant Professorship of Architecture
http://dx.doi.org/10.1016/j.engstruct.2014.05.036
0141-0296/Ó 2014 Elsevier Ltd. All rights reserved.
40 D. Veenendaal, P. Block / Engineering Structures 75 (2014) 39–50
timber, catenary in profile, and placed in parallel. A slightly

prestressed hessian fabric was tacked to the arches and, under
the weight of the applied cement mortar, sagged in between the
falsework arches to form corrugations, acting as a lost formwork.
The thickness of the first thin coat of cement, the prestress in the
fabric and the spacing between the arches would determine
the depth of the corrugations, and thus the stiffness of the shell.
The system was competitive as it reduced the cost of molds and
scaffolding, and required no skilled labor [4].
Waller patented a specific system in 1955 for spans of up to
150 m using prefabricated, external trussed arches from which to
suspend the fabric. By the end of the 1970s, the Ctesiphon system
had been used around the world for the construction of over 500
shell structures. Two of Waller’s last and largest were the Chivas
Distillery Warehouses in Paisley, Scotland. The two structures,
Fig. 1. Visualization of the preliminary design for NEST, with HiLo constructed at
100 ft (30.5 m) and 150 ft (45.7 m) long, each featuring three
the top corner. ÓEMPA and Gramazio & Kohler. 100 ft (30.5 m) spans, have a thickness of 2.5 in. (6.4 cm), with
the fabric spanning 2.54 m between the arches [5] (Fig. 2). The suc-
and Sustainable Technologies (SuAT), both at the Institute of Tech- cess of the system was partially attributed to the rising demand for
nology in Architecture, ETH Zurich, joined by Supermanoeuvre in unobstructed covered spaces with increased clearances in addition
Sydney as well as Zwarts & Jansma Architects (ZJA) in Amsterdam. to global shortages at the time of the ‘‘modern wonder material,
HiLo introduces several innovations, and this paper relates in steel’’, as well as timber [6].
particular to the development of a reusable and lightweight The renowned shell builder Félix Candela Outeriño (1910–
cable-net and fabric formwork system to construct the anticlastic, 1997) used the Ctesiphon system for his first shell, an experimental
thin shell roof, with no internal falsework. Van Mele and Block
[1,2] presented a method for finding the distribution of forces
required in such a cable-net or membrane formwork to obtain a
particular shape, after it has been loaded with concrete. This con-
trol allows a range of pre-defined, non-analytical, anticlastic
shapes to be designed and constructed, thus offering room for
shape optimization. In 2010, the first author provided consultancy
to ZJA for a competition to design an interstate wildlife crossing. In
this context, ZJA proposed to use a cable-net supported fabric to
push the concept of a flexible formwork to the scale of long-span
bridges emphasizing its qualities and constructional advantages
[3]. The present paper is a continuation of these earlier ideas.
1.2. Objectives
The objectives of this paper are twofold. First, to establish, as a

proof of concept, a complete workflow for the structural design of
an anticlastic thin concrete shell taking into account the fabrica-
tion constraints of a hybrid cable-net and fabric formwork. Second,
to construct prototype shells based on this workflow in order to
identify challenges in both computational and constructional
aspects. The results of this paper are then used to inform future
research and development of the workflow and construction
method for full-scale structures in general, and the roof structure
of HiLo in particular.
2. Historical overview of related work
The innovations in this and future work are rooted in the longer
history of using (prestressed) fabrics as a flexible formwork to
construct thin-shell concrete structures. This section serves as a
first comprehensive overview of flexible formworks for anticlastic
shells. The overview is meant to frame the present research and
highlight some of the similarities and differences with our partic-
ular approach.
2.1. Fabric-formed shells
The British engineer, James Hardress de Warenne Waller

Fig. 2. Formwork [5] (top) and current state in 2013 (bottom) of the shell of the
(1884–1968), was the first to apply fabrics to the construction of Chivas Distillery Warehouse, now Chivas Central Bottling Hall and despatch
thin shells. He developed the ‘Ctesiphon’ system, which started warehouse, Scotland, ca. 1959, by James Waller et al., with 30.5 m spans and
from reusable, lightweight falsework arches, made of steel or thickness of 6.4 cm. Photo (bottom) courtesy of Chivas Brother Ltd.
D. Veenendaal, P. Block / Engineering Structures 75 (2014) 39–50 41
vault with a 6m span in San Bartolo, Mexico City and again for a although it is doubly curved, it can be described by straight
rural school near Victoria, Tamaulipas in 1951 [7]. elements. This presented a uniquely efficient way of constructing
shuttering made from straight, reusable timber planks (Fig. 3).
Although both this and the Ctesiphon system were ingenious
2.2. Hyperbolic paraboloid shells construction systems at the time, their disappearance from prac-
tice is most likely related to the general decline of shell building.
After that, Candela moved on to geometries based on the hyper- The rising cost of labor and declining price of steel, made shell
bolic paraboloid (hypar). This is a ruled shape, meaning that building less attractive.
Alternative structural systems to cover large spaces became
available through the development of steel and timber gridshells,
and more importantly of tensioned membrane structures. In the
latter case, the membrane is the structure itself, requiring no
formwork and falsework. These systems became possible through
the increasing quality, strength and affordability of coated polymer
fabrics. At the same time, these fabrics gave rise to new opportuni-
ties for fabric-formed concrete [8] and the present method.
2.3. Fabric-formed hypars
An alternative to Candela’s system of forming the surface of a

hypar is to prestress a fabric, similar to Waller’s system. In this
way, no heavy timber shuttering and no falsework at all are needed
to carry the surface of the shell, as all loads from the fresh concrete
are transferred to the boundaries. Such a system was pioneered
and patented by Joseph A. Kersavage in the 1970s at the University
of Tennessee. Using strips of insect, plastic or metal screen, and
later fabric, a hypar surface was formed, which was then bonded
by applying a semi-rigid material such as acrylic plastic (Fig. 4a)
[9], or by brushing or spraying a mix of latex, cement and sand
to a thickness of only 1 cm [10,11]. The formwork system was
developed further in collaboration with George Nez and Albert
Knott, starting with a 1977 storage building roof consisting of four
hypars for the Rocky Mountain Park, and resulting in several other
roof structures in the USA. This ultimately led to the successful
application of the system in about 20 projects in developed and
mostly developing countries, most of which feature arrangements
of several hypar shells (Figs. 4b and c). Like Waller’s Ctesiphon sys-
tem, their solution requires low- to unskilled labor with little or no
supervision. These roofs have attracted some recent academic
interest, investigating the seismic behavior of these structures
[12], and the properties of the latex-modified concrete [13].
Fig. 3. Falsework (top) and current state in 2008 (bottom) of the shell of the Chapel
2.4. Recent fabric-formed anticlastic shells
Lomas de Cuernavaca, Mexico, 1958, by Félix Candela et al. It has a minimum 18 m
span and thickness of 4 cm. Photos (top) ÓPrinceton University Library and In 2006, several experiments at Eindhoven University of
(bottom) ÓEduardo Alarcón. Technology were intended to demonstrate a proposed construction
Fig. 4. Latex-modified concrete hypar roofs by TSC Global, Port-au-Prince, Haiti, 2011. Photos (b and c) by Stephen Riley, TSC Global (Creative Commons License CC-BY-NC-SA
2.0).
method using fabric and shotcrete. A 7 m high, 2.5 m wide proto- above 80% of the applied load. The wires, spaced 12 in. (30.5 cm),
type with a thickness of 7 cm was constructed (Fig. 5a). The project and prestressed up to 600 lb (2.7 kN) using ‘‘standard prestressing
concluded that the construction method would be feasible in a equipment’’ [23].
developed country, but that the surface could have deviations as As a result of these experiments, two 64 ft (19.5 m) square
a result of the application of shotcrete, up to several centimeters buildings were soon successfully completed; the Bay Refining
[14]. A follow-up feasibility report showed an additional experi- Gas Station and Carwash in Midland, Michigan [24], and the club-
ment replacing the fabric by a stiffer metal mesh to resolve the house at the Purdue Golf Course in West Lafayette, Indiana [25]
issue of form control, and mentioned the use of a cable net with (demolished in the mid-1990s), both an assembly of four hypars
rebar for the actual proposed construction method [15]. (Fig. 6). The Purdue shell had two layers of 0.135 in. (3.4 mm)
By 2009, the Centre for Architectural Structures and Technol- wires, 6 ft (1.8 m) offset from the straight line generators. The
ogy, or CAST, at the University of Manitoba, Canada, had under- bottom wires were spaced 12 in. (30 cm) apart, the top layer
taken several experiments in small- to fullscale fabric formed 24 in. (60 cm) apart. The shell used 3 in. thick EPS foam, a 0.5 in.
shell elements; single curved, synclastic and anticlastic shapes, stiffening mortar, a 3 in. concrete cover and traditional rebar,
many with local wrinkles corrugating the surface [16]. 16.5 cm in total.
At the Vrije Universiteit Brussel, a series of ten prototypes were
constructed from a prestressed fabric using shotcrete [17,18]. The 3. Methodology
final shell had a thickness of 5 cm and a 2 m span (Fig. 5b). After
applying the shotcrete, the deformed shape was calculated and To fully realize the structural efficiency of a flexibly formed
compared to a numerical model, but for the ten shells, deviations shell, it is crucial to both design an optimal shell within the pro-
from the expected deflections were measured ranging between ject’s constraints and to control the cable forces such that its form,
5% and 58%, to more than 100% for non-coated fabrics. These errors despite the formwork’s flexibility and the weight of the wet con-
were attributed to ‘‘slip at the fixation points, and the dynamic crete, is in the end exactly as required. A computational approach
effects of the shotcreting’’. to realize this goal was developed and is explained in this section
Yet another independent physical experiment was carried out in more detail. The procedure is shown in Fig. 7, and consists of
at the University of Edinburgh, as part of a student project, to eight steps:
investigate constructional aspects of the fabric forming of a hypar
shell [19]. The work is unpublished at present, but briefly shown by 1. establishing the boundary conditions (Section 3.1) and tar-
Brennan et al. [20] (Fig. 5c). get shape of the shell (Section 3.2);
2. generating a cable-net topology and mapping it onto the
2.5. Discrete flexible formworks surface (Section 3.3);
3. patterning of the surface for the fabric (Section 3.4);
An alternative tensile formwork system is to replace the fabric 4. computing the nodal target loads based on the target shape
by discrete elements, such as the system briefly shown by De and thickness (Section 3.5);
Bolster et al. [21] in the context of hypar shells. This reusable 5. best-fit optimization of the cable forces under load in the
and reconfigurable ‘cable net’ was proposed by Mollaert and final state (Section 3.6);
Hebbelinck [22] and was in fact constructed as a network of edge 6. materialization of the cables (Section 3.7);
chains and adjustable belts. The network was covered with EPS 7. determination of the cable forces prior to casting through
tiles before applying a fiber-reinforced mineral polymer on either static analysis of the unloaded cable net (Section 3.8); and
side (unpublished work by Sven Hebbelinck). 8. design of the external frame that ensures the chosen
A similar, older system is the ‘‘offset wire method’’ (OWM), boundary conditions (Section 4.1).
developed by Waling and Greszczuk [23] at Purdue University,
Indiana, USA. The method used two layers of wires to sandwich a Calculations for loading cases other than the design loading
layer of EPS tiles on top of which concrete could be cast. This gen- (self-weight of the formwork and concrete) need to be carried
erated a hypar shell, whilst avoiding the need for ‘‘a forest of out afterwards to further check the design, although they could
falsework’’. potentially be integrated in step 1 for the shell and step 7 for the
After building a 37 7/8 in. (1 m) square small-scale model, they cable net.
proceeded with a large-scale 20 ft (6.1 m) square laboratory model
with a rise of 7 ft (2.1 m). The shell had 3 in. (7.6 cm) thick EPS 3.1. Boundary conditions and initial shape
designed for a 2 in. (5.1 cm) thick concrete cover, deviating by
2.7 in. (69 mm) from a true hypar at its center. A coating of mortar As a starting point, the boundary conditions and initial shape of
reduced deflections to 0.43 in. (11 mm), but showed cracking our shell are based on a ‘simple’ hyperbolic paraboloid (hypar)
Fig. 5. Recent hypar shells using flexible formwork.


y2 x2
z¼c 2

b a2
where a; b, and c are parameters determining the shape. For a

symmetric hypar with straight edges within our bounding box,
a ¼ b ¼ 1 and c ¼ h=w2 ¼ 1:2=1:82 ¼ 10=27. In the rest of our
design process, these straight boundaries are kept fixed, for simplic-
ity of our initial prototypes.
Eventually, when the shell has cured and the formwork is
removed, the lower supports each have four constraints; three
translations and one rotation (Fig. 9). The shell has no edge beams
and (approximately) uniform thickness. For single or arrangements
of four hypars, reducing or entirely removing any edge beam (pos-
sibly thickening the shell at the supports) decreases overall shell
bending [26,27]. Although maximum displacements may increase,
they are not significant compared to serviceability limits.
3.2. Target shape
Although some of the thinnest known shell structures are

hypars, very slight improvements to their geometry can drastically
improve their structural behavior [28]. However, their formwork
can then no longer be described by straight elements, whereas a
Fig. 6. Laboratory model, formwork and completed state of the Purdue Golf flexible formwork can accommodate these changes as long as the
Clubhouse, Indiana, USA, ca. 1962, by J.L. Waling et al., has a minimum 13.8 m span
and 8.9 cm (structural) thickness. Reprinted with permission from [25] by the
shape remains anticlastic. Subject to this condition, the target
National Academy of Sciences. Courtesy of National Academies Press, Washington, shape for the cable net can be any given shape. In this case, as part
D.C. of our conceptual design process, structural shape optimization is
applied. Our objective is to minimize the maximum deflection
and thus optimize the stiffness.
Research at the Vrije Universiteit Brussel, subsequent to that of
Cauberg et al. [17], described the form finding of such flexibly
formed, anticlastic shells using FDM [29] and DR [30], both
1.2m
1.8m
1.8m
Fig. 8. Boundary conditions and NURBS control points, allowed to move vertically
within the bounding box while maintaining anticlasticity.
Fig. 7. Outline of the computational procedure (* indicates iterative procedure; the

target shape can be iteratively determined or given by the designer).
with a square plan, with the bounding box measuring

w b h ¼ 1:8 m 1:8 m 1:2 m (Fig. 8) and a thickness of
between 1.5 and 2.5 cm (to be determined). For a set of given
x- and y-coordinates, the heightfield of the hypar is described by Fig. 9. FE mesh with TRIC-elements and support restraints.
modelling them as discrete networks under a given selfweight of load-bearing and shape-defining part of our flexible formwork sys-
the concrete. A set of two force densities (corresponding to the tem. Several criteria govern the design of the cable-net topology:
two orthogonal directions) or a (fictitious) elastic stiffness (identi-
cal for all links) were varied to obtain a shape that fitted some spa- The valency of the cable net should generally be even to allow,
tial requirements of the designer. Guldentops et al. [29] also for practical reasons, continuous cables up to the boundaries.
compared such a form to that of a hypar of similar (but not identi- Continuous cables should terminate at the boundaries to allow
cal) size and boundary conditions, showing deflections to be lower. for more convenient prestressing (and thus control) throughout,
However, to reduce deflections even more, we apply shape optimi- rather than loop within the mesh.
zation instead of form finding. Continuous cables should follow principal curvatures of the
Bletzinger et al. [31] suggested to use a form-finding method as surface to reduce the amount of prestressing required, as the
a shape generator for structural optimization by varying the loads cables’ load capacity is proportional to prestress and curvature.
(distribution of selfweight). However, compared to varying Continuous cables should have as low geodesic curvature as
geometric parameters (e.g. control points of the surface), they possible, if they are to serve as guides or seam lines for second-
demonstrate that results are less optimal (for uniform thickness), ary fabric strips or cutting patterns (thus reducing material
since the degrees of freedom are reduced. This leads us to choose waste).
a geometric parameterization, using a NURBS surface, for the time The density of the mesh should be fine enough such that the
being (Fig. 8). The control points of the NURBS surface are allowed demands on the secondary fabric are low (in terms of strength,
to move vertically within the bounding box of our boundary condi- prestressing and patterning requirements).
tions, such that the resulting geometry never becomes synclastic. The density of the mesh should be coarse enough, such that the
The shape is optimized using a genetic algorithm by varying the total length of cable, the number of intersections and the
control points [32]. amount of prestressing work (thus material and labor cost)
When optimizing a shape to achieve the lowest maximum are reduced.
deflection (serviceability limit state), with self-weight as design
loading, we expect the material to stay within its elastic range. Given these considerations, we start by roughly orienting a
Therefore, as long as no stress limits are exceeded, we can apply a quad mesh along the lines of principal curvature (Fig. 10a). The
linear elastic model. Since the present project uses a custom end points are used to plot geodesics along the surface of the shell,
concrete mix design, and does not include material testing, optimi- corresponding well with principal curvature (Fig. 10b). Then, the
zation is carried out for the probable limits of linear elastic stiffness. end points are moved to optimize the mesh width (Fig. 10c). The
Based on Eurocode strength classes, from regular C25/30 to high- objective is to reduce the standard deviation of the mesh width.
strength C90/105, the linear elastic stiffness E of the concrete might By moving the end points, the total number of nodes along the
vary between 31 GPa and 44 GPa. The results show that the stiff- boundary increases, but in terms of construction, this simplifies
ness, within this range, does not significantly influence the shape. matters, since at each point, only one cable now terminates. We
The model’s thickness is designed to be in the range of 1.5– set our mesh width at approximately 20 cm with a 30 cm upper
2.5 cm, so we optimize for both of these limits as well (Table 1). limit. Table 2 shows the result of the manual optimization.
In summary, we minimize the maximum deflection by varying
the vertical position of the four control points within the bounding
box of the boundary conditions. The problem is not subject to any 3.4. Cutting patterns
other constraints, since displacements are very low and additional
load cases are not considered at this stage. As discussed, aligning the seam lines of cutting patterns along
The FE software used (see Section 3.9) provides one type of shell geodesics minimizes waste. The cables, already geodesic, are cho-
element, that is described as being similar to the TRIC-element sen as seam lines, and the surface in between is unrolled using
[33], which already produces accurate results for doubly curved an area-preserving approach. The surface is subdivided, triangu-
geometries at coarse meshes, avoids any shear locking [34] and lated and flattened. The degree of subdivision is increased until
neglects transverse shear deformation [35] meaning that it applies the difference between surface areas of the flattened pattern and
to thin shells. the original doubly curved surface are within a certain tolerance,
Table 1 shows the results from optimization compared to the in our case 0.01 m2 (Fig. 11).
initial hypar. We observe that the relative reduction in deflections
increases with shell thickness, and thus the influence of flexural
behavior. The shape for the lower bound uniform thickness of
1.5cm is chosen for further development (Fig. 9).
3.3. Cable-net topology design
The next step, having settled on the required shape of the shell,
is to map a cable net onto the surface that serves as the main
Table 1
Results (height at midspan) and fitness (max. deflection at the tips) of shape Fig. 10. Optimization of the cable-net geometry.
optimization, identical for C25/30 and C90/105.
Thickness Height at midspan Deflection at tips

(cm) (cm) (mm) Table 2
Mean, minimum and maximum mesh width, and standard deviation, in (cm).
Hypar 1.5 60 0.079
Optimized 1.5 62 0.078 Mean Min. Max. St. dev.
Hypar 2.5 60 0.040
Optimized 2.5 78 0.037 Initial 21 15 28 4
Optimized 23 18 26 2
the force densities q or forces f required after casting of the

concrete.
Using a branch-node matrix C to describe the topology of the
cable-net of m cable segments and n nodes, the coordinate differ-
ences for all m cable links are u ¼ Cx; v ¼ Cy, and w ¼ Cz [38].
The branch-node matrix and the coordinate vectors are split in
two based on the n ¼ nN þ nF free and fixed nodes:
C ¼ ½ CN CF ; x ¼ ½ xN xF , y ¼ ½ yN yF and z ¼ ½ zN zF . We
can define static equilibrium in each direction as
CTN Uq ¼ DN xN þ DF xF ¼ 0 ð1Þ
CTN Vq ¼ DN yN þ DF yF ¼ 0 ð2Þ
CTN Wq p ¼ DN zN þ DF zF p ¼ 0 ð3Þ
where, to simplify notation, DN ¼ CTN QCN and DF ¼ CTN QCF . The

matrices U; V; W and Q are the diagonal matrices of vectors
u; v ; w and q.
This is rewritten as a single system
¼ DN x
Aq p N þ DF x ¼0
F p ð4Þ
Fig. 11. Cutting patterns, cut along cable lines, used for the fabric shuttering and
the textile reinforcement of the second prototype.
where DN and DF are block diagonal matrices, and remaining vec-
tors and matrices are vertically stacked.
Of the two directions in the net, the standing cables are chosen N and
We linearize Eq. (4) with respect to the unknowns q and x
for the seams, such that the fabric strips can simply be placed over obtain [39]:
the cable net without the possibility of sliding down. The patterns
DxN
are designed to skip every other cable, to compare the effect on the DN kA A0 q0
¼p ð5Þ
concrete imprint. For engineered membrane structures, the cutting Dq
patterns are shrunk to compensate for the prestress, such that the
where weighting factor k was set to 10 in our case. This is an under-
resulting surface is properly tensioned and smooth. In our case, for
determined system of equations, since the left-hand side matrix is
simplicity of construction and aesthetic reasons, the patterns are
3nN ð3nN þ mÞ. Such a system can be solved with a least-squares
not compensated, such that the unstrained fabric sags between
solver such as GNA. In each iteration i, one sets qiþ1 ¼ qi þ Dq, then
the cables once the concrete is applied.
updates DN as well as the residuals, and repeats the iteration until
convergence is reached, in our case if kDx N k < ¼ 0:001, occurring
3.5. Target loads after 707 iterations (22 s). Upon convergence, the required pre-
stresses under the load of the fresh concrete can be calculated by
To determine the forces in the cable net, we first calculate the simply multiplying the force densities with the lengths of the cable
target loads on the nodes of the net as an approximation of our links.
target shape. Each load is assumed to be the tributary area of a cor- Fig. 12 shows the results for an orthogonal net, one based on
responding node in the net, extruded by the average thickness of OWM (see Section 2.5) and one aligning with principal curvatures,
the shell. Although the shape is optimized for 1.5 cm, the thickness each mapped onto our target surface. Note that the first cable net
for the load calculation is changed to 2.5 cm to accommodate the would have infinite forces on a hypar, but for the optimized shape
expected sag of the fabric. This means the additional thickness is has a slight curvature, allowing us to find a solution.
assumed to act as dead load, but not contribute structurally to
the shell. First, the Voronoi diagram of the projected cable-net
3.7. Materialization
nodes is determined. The cells are projected onto the target sur-
face, planarized and then extruded normal to the surface by the
Having determined the forces after casting in Section 3.6, we
assumed thickness. The density is set to 24 kN/m2.
wish to know the forces prior to casting in order to construct and
prestress our formwork. To do so, we materialize and dimension
3.6. Best-fit optimization the cable net. The smallest steel cable available to us is 2 mm INOX
with a tensile strength of 3.7 kN. This is well above the maximum
We wish to find the required cable-net forces such that, under tensile stress after casting (factor of 13.5). The stiffness of our cable
given loads of the wet concrete, the resulting concrete shell takes
the form of the target shape. As mentioned, this problem has
already been addressed and solved by the second author through
a non-linear extension of thrust network analysis (TNA) [2]. This
approach used the Levenberg–Marquardt algorithm (LMA) as a
least-squares solver, which interpolates between the Gauss–
Newton algorithm (GNA) and the method of gradient descent. For
the sake of comparison (see Section 5), non-linear extension of the
force density method (FDM) [36,37] is implemented and adapted
to our problem. The approach uses GNA as a least-squares solver.
302 806 1214 [N] 117 216 364 [N] 95 193 274 [N]
The goal is to minimize changes in coordinates DxN ; DyN ; DzN
and force densities Dq, such that the coordinates x; y; z approach Fig. 12. Cable forces after casting from best-fit optimization, where aligning with
the target shape, while satisfying static equilibrium. The results are principal curvatures reduces the required prestress.
is not specified by the manufacturer, but presumably it is matrices are not implemented in IronPython, the ALGLIB library
E ¼ 195 N=mm2 based on the relevant German DIN-norm. for IronPython was later used to solve sparse least squares prob-
So, having chosen a material and cross-section, with an axial lems with xalglib.linlsqrsolvesparse, based on the LSQR
stiffness EA ¼ 195p, the initial lengths l0 can directly be computed algorithm. Unrolling of the surface to create cutting patterns was
from the forces f and lengths l in the final state [40]: achieved by adapting a Grasshopper definition by Andrew
1 Heumann.
l0 ¼ I þ ðEAÞ1 F l ð6Þ
4. Results
where I is an identity matrix of size m, and EA and F are diagonal
matrices of all stiffnesses EA and the forces f respectively. Based on the computational approach presented in the previous
section, two prototype shells were built as a proof-of-concept.
3.8. Static analysis
4.1. External frame
From the final state geometry, and the newly given stiffnesses
and resulting initial lengths, it is now possible to compute the A timber frame was built to follow the boundaries of the shell
intermediate geometry and forces prior to casting. In that situation, and resist the applied prestresses, using 9 cm square fir elements
the fresh concrete has not been applied yet, so assuming the (Fig. 14). The frame was designed such that the upper part would
selfweight of the formwork to be negligible compared to the pre- be removable for demolding, whilst the lower part would support
stresses, we simply remove the loads when computing the residual the two bottom corners of the hypar. A tension tie connected the
forces and attempt to find a new equilibrium shape. This FE two bottom corners to resist the horizontal thrust from the shell.
analysis is performed using dynamic relaxation [41] (this method Because the frame edges might deflect due to bending in the order
is well-known as a form-finding method, but here we used real of 1 mm (based on calculations), it was decided during construc-
material values, making it equivalent to FE analysis with a tion to add an additional timber cross, connecting the midpoints
particular type of element and solver). of the frame edges (see Fig. 14, right) and a top member for the sec-
After dynamic relaxation, we obtain new lengths and recom- ond prototype (to allow access for measurements).
pute the forces. The cables were dimensioned for the forces after
casting and Fig. 13 shows that, as expected, the range of forces 4.2. Cable net and prestress
has decreased for the unloaded state.
The cable net was made from 2 mm stainless steel cable. Cables
3.9. Implementation were guided through the timber frame along cringles, terminating
at eyebolts using crimp sleeves. For the first prototype, nodes were
The procedure, outlined in Fig. 7, was implemented in fixed by winding a piece of wire around the intersecting cables. For
Grasshopper for Rhinoceros, using its built-in Galapagos the second, cross clamps were used instead (Fig. 15), which also
components for the genetic algorithm, plug-ins Karamba to served as measuring points for photogrammetry.
evaluate deflections, and Kangaroo for the dynamic relaxation. In At one end of each of the twenty cables, a turnbuckle was used
addition, custom Python components were written in IronPython, to introduce prestress. The prestressing was controlled by measur-
using the NumPy library. NumPy functions used were ing the lengths of the cable segments during prestressing, attempt-
numpy.linalg.inv (which uses Lapack’s subroutine DGESV, based ing to approach the digital model as closely as possible.
on LU factorization), to solve linear systems such as in Eq. (6), and Prestressing was then further refined by measuring and checking
numpy.linalg.lstsq (which uses Lapack’s subroutine DGELSD) the forces in several cable segments. In the first prototype, this
to solve the least squares problem in Eq. (5) using singular value was done by measuring the elongation of springs at twenty loca-
decomposition (SVD). Because NumPy algorithms for sparse tions inside the net. For the second prototype, a portable cable ten-
sion meter was used instead, allowing more measurements of
higher accuracy.
4.3. Fabric
The first fabric was a PP geotextile, Propex 60-7041, with a ten-

sile strength of 42 kN/m, and a 5.2 m roll width. The geotextile was
chosen purely for its similarity to the North American Propex
315ST, used by Prof. Mark West in many of his experiments at
C.A.S.T. It is very similar in weight, strength and hydraulic
properties.
95 274 [N] 151 211 [N]
Fig. 13. Cable forces after and before casting. The range of prestress is smaller in the Fig. 14. Design and photo of the first prototype, with a cross added in the as-built
latter state. frame.
at our concrete lab. The mix design was not the focus of our present
research, so the only criteria were to obtain a mixture with high
slump, resistance to shrinkage cracking and tensile capacity. The
proportions by weight were:
1 kg cement (Holcim Normo 5R, CEM I 52.5).

0.1 kg microsilica (Elkem Grade 971-U).
0.7 kg fine sand/aggregate (0/4 mm).
0.015 kg PVA fibers (Kuralon K-II 6/12 mm).
0.24 l water.
0.010–0.015 l plasticizer (BASF Glenium ACE 30).
0.015 kg stabilizer (Sika 4R).
4.5. Construction
Construction of the first prototype took place in April, 2013

(Fig. 17). The second prototype was cast in February, 2014. Apart
from the initial woodwork, it involved only unskilled labor. The
various activities are summarized in Table 3.
4.6. Cost estimation
Table 4 shows the materials and their costs for the first proto-
type formwork. The cost of the formwork is CHF 527.79, or CHF
162.90 per plan square meter with nearly 60% due to the timber
frame. We expect that the relative cost will go down at larger
Fig. 15. Construction of the second prototype (clockwise from top left): turnbuck-
les, stitched seam, surface after casting, final result, corner support detail, and node
with cross clamp.
The second fabric was a PP Proserve F0899 with a tensile

strength of 54–60 kN/m, and 3.6 m roll width, used for underwater
fabric formworks. The fabrics were tacked onto the timber frame.
The second fabric was also clamped to the frame with an additional
timber profile to control the edge geometry.
4.4. Concrete
After prestressing the net, the fabric was applied and a 0.9–
2.9 cm (average ca. 2.4 cm, see Fig. 16) thick layer PVA fiber rein-
forced cement mortar was hand-rendered on top. The second shell
was cast while continuously measuring the thickness for better Fig. 17. Construction of the first formwork and shell.
control, by distributing the concrete accordingly. It was also fitted
with an additional layer of AR-glass textile reinforcement. The mix
design and choice of fiber reinforcement was adapted from Máca
et al. [42] based on further discussion and availability of materials Table 3
Labor involved in constructing , and adapting and reconstructing the formwork.
Activity Prototype 1 (hours) Prototype 2 (hours)

Woodwork 8 1
Carpentry 39 15
Installation 34 32
Prestressing 12 14
Patterning 8 5
Concreting 22 14
Fig. 16. Diagram of varying shell thickness for the first prototype due to
undulations between cables (not to scale). Total 123 81
Table 4
List of materials, quantities and cost in CHF for the first formwork.
Component Type Qty. Cost per unit Total excl. Total cost
Timber 100 100 fir 25 m 11.50 287.50 310.50
Fabric Propex 60-7041 9 m2 0.71 6.39 7.60
Tacks 1.8 20 mm 0.5 kg 26.10 13.05
Cable INOX V4a 2.0 mm 100 m 0.52 52.29 58.10
Wire Steel wire 1 mm 12.00 12.96
Cringle 6 mm 60 0.62 37.26 40.24
Turnbuckle M5 20 2.00 40.00 43.20
Crimp sleeve 2 mm 60 0.09 5.49 5.93
Quick link 4 mm 40 0.56 22.32 24.11
Eye screw M6 40 20 0.56 11.20 12.10
Total 527.79
spans, where the ratio of surface to edge is higher, and for multiple average deviation from the target surface l ¼ 2:0 mm with a
use, since much of the first prototype was reused for the second. standard deviation of r ¼ 1:5 mm.
Table 5 and Fig. 18 show an overview of our two as well as
4.7. Measurements and accuracy precedent prototypes for which data was published. Here the total
difference between design model and final loaded state are
After the first shell had hardened, thirteen measurements were compared, where the difference is the sum of deformations due
taken at nodes of the cable net to see how much the actual proto- to loading and deviations between theoretical and physical model.
type deviated from the computer model. Measurements using The proposed design procedure accounts for the final loaded state
ruler and laser level revealed an average deviation l ¼ 22 mm of the formwork, so the second prototype is particularly accurate
with a standard deviation of r ¼ 1 mm. The second shell was made by comparison. In general, good results were obtained in each
to improve the method of measuring forces and geometry as the study, considering the inherent flexibility of these formworks.
main source of error. Forces were measured using a Tensitron
ACX-250-M portable cable tension meter, while photogrammetry 5. Discussion
was used to register nodal positions. The cable net was loaded
using weights of sand equivalent to the nodal loads in the com- A historical investigation yielded a large number of precedents
puter model. This also meant that the cables remained accessible of flexibly formed, anticlastic shell structures. A recurring topic
in the loaded state, allowing measurements of forces. The resulting among these works is the lack of control of deviations. Based on
Table 5
Comparison of deformations and deviations, relative to the span, in flexible formworks.
Ref. Surface Eq. thickness in Span s (m) Plan (m) Deformation d due Deviation D from Rel. difference
concrete (cm) to loading (mm) theoretical (mm) dþD
s (‰)
[23] Cable + EPS-foam 3.5 1.4 11 18 5 16

Cable + EPS-foam + mortar 4.0 8.6 6.1 6.1 69 13 8.0
4.6 8.6 6.1 6.1 11 13 2.8
[17] Coated PP, PE, PVC 3.6–5.0 2 22 15 0.75–8.7 (5–58%) 7.9–11.9
Uncoated PP, PE, PVC P15 (P100%) P15
[Present paper] Cable + uncoated PP 2.4 2.55 1.8 1.8 N/A 22 8.7
Cable + uncoated PP 1.5 2 0.8
Fig. 18. Sequence of differences from design model to final built state for Table 5.
this historical overview and our own study, possible strategies to the concept of calculating prestresses in the final state, and
improve accuracy are: materializing before finally calculating prestresses prior to
casting;
to reduce the applied loading by layering and curing the shell in the design and construction of two prototypes with a discussion
stages, thereby creating a partially or entirely self-supporting on their accuracy.
structure as soon as possible;
to increase the stiffness of the material (e.g. coated instead of It is clear that much work needs to be done for the further
uncoated fabrics, cable nets) and/or the prestress such that development of this construction method. However, the present
the formwork is less sensitive to inaccuracies of the applied research has enabled us to identify which particular aspects war-
load; rant our immediate attention. Future work will focus on improving
to design for the loaded state to exclude deformations from the the optimization procedure and tackling the challenges associated
final comparison; and with scaling up the construction method.
to accurately measure and correct the required prestresses.
Acknowledgements
Our approach is based on the latter three: a cable net for which
we calculate the prestresses required in the final state, then mea- The authors would like to thank Regine van Limmeren, Masoud
sure and control the prestresses required in the initial state. In this Akbarzadeh and Ramon Weber as well as Mile Bezbradica, Jonas
paper, we adapted a non-linear FDM for the calculation of the final Sundberg and David López López for the construction of the two
state. Its convergence was slow, but satisfactory for the present prototypes respectively. David Novák and Prof. Konrad Schindler
model. However, some preliminary numerical experiments with assisted with the photogrammetry. Further support at ETH Zurich
more complicated shapes have shown that non-linear FDM tends was provided by Oliver Zgraggen, Paul Fischlin, Heinz Richner and
to diverge, making it unsuitable for future work in its current form. Thomas Jaggi. We acknowledge our former students Paul Mayen-
In the method by Van Mele and Block [2] convergence is vastly court and Matthias Amstad, who worked on a previous formwork.
improved by reducing and controlling the amount of unknowns: We especially thank Manuela Tan, and Noah Nichols from Pro-
first by solving vertical equilibrium separately, and second by iden- pex as well as Kaloyana Kostova and John Orr from the University
tifying mi independent sets of the m unknown force densities. In of Bath for sponsoring the two fabrics; Jennifer Stevenson, Chivas
other words, the amount of unknowns decreases from 3n þ m to Brothers Ltd., for providing current photos of the Paisley ware-
n þ mi , where mi 6 m, and n are the number of free nodes. houses; Charlotte Erdmann, Engineering Library of Purdue Univer-
In further work, the objective will be to scale up the construc- sity, for identifying the Bay Service Station and securing additional
tion method. copyright permissions; and George Nez and Brad Wells, TSC Global,
In terms of computation, thickness in addition to shape optimi- for information regarding their hypar shell roofs.
zation will be included as well as stress constraints when checking
for multiple load cases based on Swiss building code and wind References
modelling. The material model will be updated based on results
from material testing of the textile and fiber reinforced concrete. [1] Van Mele T, Block P. A novel form finding method for fabric formwork for
concrete shells. In: Proceedings of the international association for shell and
In terms of construction, the timber frame and turnbuckles will spatial structures symposium 2010. Shanghai, China; 2010.
be replaced by steel and jacks. Development of the steel details and [2] Van Mele T, Block P. Novel form finding method for fabric formwork for
prestressing strategy will be an important focus. Partially leaving concrete shells. J Int Assoc Shell Spatial Struct 2011;52(4):217–24.
[3] Torsing R, Bakker J, Jansma R, Veenendaal D. Large-scale designs for mixed
the frame as a permanent edge beam will be evaluated, depending fabric and cable-net formed structures. In: Orr J, editor. Proceedings of the 2nd
on results from shape and thickness optimization, as well as based international conference on flexible formwork. Bath, UK; 2012. p. 346–55.
on non-structural considerations. [4] Billig K. Concrete shell roofs with flexible moulds. J ICE 1946;25(3):228–31.
[5] Anon. Curved roofs of large span. Concr Constr Eng 1959;54(5):171–4.
Finally, the full-scale project will serve as an enclosure, and [6] Waller JdW, Aston A. Corrugated shell roofs. ICE Proc – Eng Div
therefore integrating the structural shell with building systems 1953;2(4):153–82.
and insulation while retaining its thinness is another important [7] Faber C. Candela: the shell builder. Reinhold Publishing Corporation; 1963.
[8] Veenendaal D, West M, Block P. History and overview of fabric formwork:
aspect of the overall design.
using fabrics for concrete casting. Struct Concrete – J Fib 2011;12(3):164–77.
[9] Kersavage JA. Method for constructing a tensile-stress structure and resultant
6. Conclusions structures, patent us 3,927,496; 1975.
[10] Stepler R. Hypar structures, light, easy to build, cheap – and permanent. Pop
Sci 1980;216(2):73,74–77,168.
The field of fabric formwork represents over a century of [11] Knott A, Nez G. Latex concrete habitat. Canada: Trafford Publishing Company;
exciting inventions and discoveries, yet provides many opportuni- 2005.
[12] Balding D. The experimental seismic testing of hypar shells. Fourth-year
ties still. A historical overview of flexibly formed shell structures undergraduate project. Cambridge University; 2013.
and prototypes, often hypars, has been presented, many of which [13] Carlton WS. Material behavior of latex-modified concrete in thin hyperbolic
were constructed with comparable goals to the present work. Our paraboloid shells. Master thesis. University of Oklahoma; 2013.
[14] Pronk A, Houtman R, Afink M. The reconstruction of the philips pavilion
proposed solution has been developed to allow large-span form-
volume 1. In: Sources of architectural form, theory and practice. Kuwait City,
works with little or no falsework and enable a wider range of anti- Kuwait; 2007.
clastic shapes than those possible through traditional means. [15] Pronk A, Dominicus M. Philips-paviljoen; 2012.
[16] West M, Araya R. Fabric formwork for concrete structures and architecture. In:
The first prototype presented here is the first cable-net and
Krõplin B, Oñate E, editors. Proceedings of the international conference on
fabric formwork, and an important step towards a new cable-net textile composites and inflatable structures, structural membranes 2009.
and fabric formed roof to be built in Switzerland. Our main Barcelona; 2009.
contribution is a complete computational design workflow for [17] Cauberg N, Parmentier B, Vanneste M, Mollaert M. Fabric formwork for
flexible, architectural concrete. In: Oehlers D, Griffith M, Seracino R, editors.
the prototype shell and its required formwork. Proceedings of the international symposium on fibre-reinforced polymer
This includes reinforcement for concrete structures, FRPRCS-9. Sydney, Australia; 2009.
[18] Cauberg N, Tysmans T, Adriaenssens S, Wastiels J, Mollaert M, Belkassem B.
Shell elements of textile reinforced concrete using fabric formwork: a case
the formulation of design criteria for the cable net and its study. Adv Struct Eng 2012;15(4):677–89.
topology; [19] Ghaui B. My interests & thoughts as a part ii student; 2012. <http://
ghaui.wordpress.com/tag/hyperbolic-paraboloid/>.
[20] Brennan J, Walker P, Pedreschi R, Ansell M. The potential of advanced textiles [32] Pugnale A, Sassone M. Morphogenesis and structural optimization of shell
for fabric formwork. Construct Mater 2013;166(4):229–37. structures with the aid of a genetic algorithm. J Int Assoc Shell Spatial Struct
[21] De Bolster E, Cuypers H, Van Itterbeek P, Wastiels J, De Wilde W. Use of hypar- 2007;48(3):161–6.
shell structures with textile reinforced cement matrix composites in [33] Argyris J, Tenek L, Olofsson L. Tric: a simple but sophisticated 3-node
lightweight constructions. Compos Sci Technol 2009;69:1341–7. triangular element based on 6 rigid body and 12 straining modes for fast
[22] Mollaert M, Hebbelinck S. Adaptent: a generating system for cable net computational simulations of arbitrary isotropic and laminated composite
structures. In: Fifth international conference on space structures. Guildford, shells. Comput Methods Appl Mech Eng 1997;145:11–85.
Surrey, UK; 2002. [34] Argyris J, Papadrakakis M, Apostolopoulou C, Koutsourelakis S. The tric shell
[23] Waling J, Greszczuk L. Experiments with thin-shell structural models. In: element: theoretical and numerical investigation. Comput Methods Appl Mech
Proceedings, annual meeting, American Concrete Institute, vol. 57; 1960. p. Eng 2000;182:217–45.
413–31. [35] Preisinger C. Karamba User Manual for Version 1.0.3; 2013.
[24] Maddex D, Dow Alden B. Midwestern modern. Archetype Press; 2007. [36] Linkwitz K, Schek H-J. Einige Bemerkungen zur Berechnung von
[25] Waling J, Ziegler E, Kemmer H. Hy-par shell construction by offset wire vorgespannten Seilnetzkonstruktionen. Ing Arch 1971;40:145–58.
method. In: Proceedings, world conference on shell structures. San Francisco, [37] Gründig L, Schek HJ. Analytical form finding and analysis of prestressed cable
CA, US; 1962. p. 199–200. Washington, D.C., National Academy of Sciences- networks. In: International conference on tension roof structures. London, UK;
National Research Council Publication. p. 1187–453. 1974.
[26] Ortega NF, Robles SI. The design of hyperbolic paraboloids on the basis of their [38] Schek H-J. The force density method for form finding and computation of
mechanical behaviour. Thin-Wall Struct 2003;41:769–84. general networks. Comput Methods Appl Mech Eng 1974;3:115–34.
[27] Jadik T, Billington DP. Gabled hyperbolic paraboloid roofs without edge beams. [39] Linkwitz K, Veenendaal D. Chapter 12: nonlinear force density method:
J Struct Eng 1995;121:328–35. constraints on force and geometry. In: Adriaenssens S, Block P, Veenendaal D,
[28] Tomás A, Martí P. Optimality of candela’s concrete shells, a study of his Williams C, editors. Shell Structures for Architecture. Routledge; 2014. p. 143–
posthumous design. J Int Assoc Shell Spatial Struct 2010;51(1):67–77. 55.
[29] Guldentops L, Mollaert M, Adriaenssens, De Laet L, De Temmerman N. Textile [40] Linkwitz K. Formfinding by the ‘‘direct approach’’ and pertinent strategies for
formwork for concrete shells. In: Domingo A, Lazaro C, editors. Proceedings of the conceptual design of prestressed and hanging structures. Int J Space Struct
the international association for shell and spatial structures (IASS) symposium 1999;14(2):73–87.
2009. Valencia; 2009. p. 1743–54. [41] Barnes MR. Form finding and analysis of tension structures by dynamic
[30] Tysmans T, Adriaenssens S, Wastiels J. Form finding methodology for force- relaxation. Int J Space Struct 1999;14(2):89–104.
modelled anticlastic shells in glass fibre textile reinforced cement composites. [42] Máca P, Zatloukal J, Konvalinka P. Development of ultra high performance fiber
Eng Struct 2011;33:2603–11. reinforced concrete mixture. In: IEEE symposium on business, engineering and
[31] Bletzinger KU, Wüchner R, Daoud F, Camprubí N. Computational methods for industrial applications; 2012.
form finding and optimization of shells and membranes. Comput Methods
Appl Mech Eng 2005;194:3438–52.
1
I
UNUt:Hi->1'ANlllNC TllE HYPERBOLIC P1\HAHOLOI D
P I
HE ..\Tl:'\G A:'\[) COOLL'\G
A SHOPl'f'\r, n::'\TER
._.,....,_
-
~
-
__ __
.........,____
....... ._ _
,........,. ... ,..,.......,._..,,,,, ,.....
H'Tltt:-..' .\S!\l.hl"' h>H "'V Hnunor)f- t'A.RAROf.(111)
.,
. .,.
,

Felix Candela Concrete Shell Structures

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Felix Candela Concrete Shell Structures

Uploaded by

Copyright:

Available Formats

Finite-Element Analysis of Felix Candela’s Chapel of Lomas

Introduction tures in order to appreciate and understand his works.

JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008 / 47

J. Archit. Eng. 2008.14:47-52.

Fig. 3. Hyperbolic paraboloid

his engineering knowledge and expertise as a builder to select

48 / JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008

J. Archit. Eng. 2008.14:47-52.

he meant the form could distribute stresses without excessive

JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008 / 49

J. Archit. Eng. 2008.14:47-52.

Fig. 7. Section cut on model of Cuernavaca Chapel

Finite-Element Analysis Results

50 / JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008

J. Archit. Eng. 2008.14:47-52.

Felix Candela’s Chapel of Lomas de Cuernavaca proves an excel-

The writers gratefully acknowledge Mrs. Dorothy Candela’s do-

JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008 / 51

J. Archit. Eng. 2008.14:47-52.

SMAX 关kPa 共psi兲兴 41 共6兲 76 共11兲 14 共2兲

52 / JOURNAL OF ARCHITECTURAL ENGINEERING © ASCE / JUNE 2008

J. Archit. Eng. 2008.14:47-52.

Geometry and Construction Optimization: An example

Zlata Tošić1 · Sonja Krasić1 · Naomi Ando2 · Milos Milić1

© Kim Williams Books, Turin 2019

Keywords Construction optimization · Geometry · Ruled surfaces · Digital

Geometry and Construction Optimization

A level of complexity arising from the application of topology optimization

Fig. 1 Workflow for designing and optimization of a new model

Fig. 4 Grasshopper drawing of the position of all elements of the new model

Table 1 Geometry and construction properties of the new HP in function of number “N”

Mass 19,449 kg 27,810 kg 36,173 kg 44,538 kg

Fig. 5 Interdependence of number of division and maximum deformation in the new HP model

Calculation of Structural Properties of Both Models Using Femap

Structural Properties of the Existing Building

Candela produced structures of astonishing elegance and efficiency thanks to his

Fig. 9 Stress distribution in the existing object—max: 12.64 MPa (FEMAP with NX Nastran)

Fig. 10 Total translation in the existing object—max: 0.244 m (FEMAP with NX Nastran)

Structural Properties of the New Model

Fig. 11 Elements and loads of the new model (FEMAP with NX Nastran)

Fig. 12 Stress distribution of the new model—max: 62.72 MPa (FEMAP with NX Nastran)

Fig. 13 Total translation of the new model—max: 0.164 m (FEMAP with NX Nastran)

Comparison of Construction Properties of the New and the Existing

Digital Fabrication of Elements

Fig. 14 Percentage of mass and

Fig. 16 Grasshopper drawing of five groups of elements divided by length

Table 2 Group’s specification of rod elements of the new model

Length of the rods (m) L = 1.64 − 2.15 L = 2.36 − 3.57 L = 3.74 − 4.65 L = 5.08 − 6.22 L = 6. 32− 6.90

Table 3 Group’s specification of the panels of the new model

Area of the panels ­(m2) A = 1.54 − 1.73 A = 1.80 − 2.18 A = 2.20 − 2.57 A = 2.63 − 3.01 A = 3.26

Fig. 17 Grasshopper drawing of five groups of panels divided by area

Fig. 18 Final model of the new building

FELIX CANDELA, ELEGANCE AND ENDURANCE:

NOAH BURGER1 AND DAVID P. BILLINGTON2

Figure 1. Restaurant Los Manantiales [5]

1. INTRODUCTION Today, the romantic setting has faded. The canals

of 1-5/8” (40mm). Candela was taking a risk when

Prior to Candela’s free edge shells, most shells used

To Candela, the edge rib seemed cumbersome; the

Normally, a shell’s dead load causes bending in its

3.1 Groin (V beam)

The groin was examined first, using membrane

V-beam S11 Stresses [(+) Tension]

-4 analyzed next. The maximum S11 compression

Fig. 1 Workflow for designing and optimization of a new model

Fig. 4 Grasshopper drawing of the position of all elements of the new model

Table 1 Geometry and construction properties of the new HP in function of number “N”

Fig. 5 Interdependence of number of division and maximum deformation in the new HP model

Fig. 9 Stress distribution in the existing object—max: 12.64 MPa (FEMAP with NX Nastran)

Fig. 10 Total translation in the existing object—max: 0.244 m (FEMAP with NX Nastran)

Fig. 11 Elements and loads of the new model (FEMAP with NX Nastran)

Fig. 12 Stress distribution of the new model—max: 62.72 MPa (FEMAP with NX Nastran)

Fig. 13 Total translation of the new model—max: 0.164 m (FEMAP with NX Nastran)

Fig. 14 Percentage of mass and

Fig. 16 Grasshopper drawing of five groups of elements divided by length

Table 2 Group’s specification of rod elements of the new model

Table 3 Group’s specification of the panels of the new model

Area of the panels (m2) A = 1.54 − 1.73 A = 1.80 − 2.18 A = 2.20 − 2.57 A = 2.63 − 3.01 A = 3.26

Fig. 17 Grasshopper drawing of five groups of panels divided by area

Fig. 18 Final model of the new building