You are on page 1of 97

politecnico di milano

Facoltà di Ingegneria
Scuola di Ingegneria Civile, Ambientale e Territoriale
Dipartimento di Ingegneria Civile e Ambientale

Master of Science in
CIVIL ENGINEERING - INGEGNERIA CIVILE

Form-Active Structures: Study of


Representative Examples

Supervisor:
prof. giorgio novati

Master Graduation Thesis by:


shuo wang
Student Id n. 841811

Academic Year 2016-2017


To my grandfather.
— Shuo Wang

致我的爷爷。
— 王朔
CONTENTS

Abstract xi
1 introduction 1
1.1 Form-Active Structures 1
1.2 Representative Examples for Tensile Form-Active Structures 3
1.3 First Examples of Geometrically Nonlinear Analysis of Cable Sys-
tems 6
1.3.1 Single Cable under Concentrated Force 6
1.3.2 Initially Flat Cable Net 7
2 the treatment of geometrically nonlinearity with ref-
erence to trusses 11
2.1 Linearization of the Joint Equilibrium Equations 12
2.2 The Geometric Stiffness Matrix 12
2.3 Newton-Raphson’s Iteration Algorithm 15
2.4 Examples 17
2.4.1 Shallow Space Truss 17
2.4.2 Shallow Geodesic Dome 18
2.5 Summary 22
3 form-finding of cable structures with force density method
25
3.1 Basic Theory of Force Density Method 26
3.2 Examples 29
3.2.1 Simple Cable Structure Without External Loads 29
3.2.2 Cable Net with a Upper Rigid Ring and Lower Anchorage
Points along a Square Base 30
3.3 Summary 30
4 form-finding of cable structures with dynamic relax-
ation 33
4.1 Basic Theory of Dynamic Relaxation 34
4.1.1 Velocity Tracing 34
4.1.2 Current Coordinates and Displacements 35
4.1.3 Calculation of Residuals 35
4.1.4 Boundary Conditions or Supports 36
4.2 Iteration Scheme 36
4.2.1 Initial Conditions 37
4.3 Stability of the Method 37
4.3.1 Fictitious Masses 38
4.3.2 Viscous Damping 39
4.3.3 Kinetic Damping 41
4.4 Calculation of Internal Forces for Trusses and Cables 42
4.5 Example: A Hyperboloid Structure 44
4.6 Summary 46

v
5 form-finding of inflatable dams 47
5.1 Literature Review 49
5.1.1 Dynamic Relaxation 49
5.1.2 Elastica Analysis 51
5.2 Basic Theory 54
5.2.1 Non-Dimensional Quantities 56
5.2.2 Governing Equations 57
5.2.3 The Jacobian Matrix 59
5.3 Form-Finding Procedure 62
5.4 Examples 64
5.4.1 Double-Anchor Inflatable Dams 64
5.4.2 Single-Anchor Inflatable Dams 75
5.4.3 Internal Air Pressure Update Procedure 77
5.5 Summary 80

bibliography 83

vi
LIST OF FIGURES

Figure 1.1 Tensile (a) and compressive (b) form-active shapes (Mac-
donald, 2007). 1
Figure 1.2 Heinz Isler: membranes under self-weight (a) and inverted
structures (b) used for the design of concrete shells. 2
Figure 1.3 Hybrid structural systems: integration of "bending- ac-
tive" linear elements and tensile membrane elements (Van
Mele et al., 2013). 2
Figure 1.4 A camping tent with flexible poles 3
Figure 1.5 The Dorton Arena at Raleigh, North Carolina, USA. (a)
Photo taken by a Pinterest user and (b) structural system
(Tibert, 1999). 3
Figure 1.6 The German pavilion (a) at the World’s Fair in Montreal
1967 and (b) its form-finding study model (Langdon, 2015). 4
Figure 1.7 The USA pavilion at the World’s Fair in Osaka 1970 5
Figure 1.8 Cable-net facade at Hilton Hotel, Munich 5
Figure 1.9 Cable-net facade at New Poly Plaza, Beijing 6
Figure 1.10 Initial configuration of the example. 6
Figure 1.11 Relationship between vertical displacement and vertical
load withoud prestress. 7
Figure 1.12 Initial configuration of the example. 8
Figure 1.13 Relationship between vertical displacement w and vertical
load P under different prestressing T0 . 9
Figure 2.1 Plan (a) and elevation (b) view for the shallow space truss.
A point force P = 100 kN is loaded at node A along the
direction of z-axis. 18
Figure 2.2 3D (a) and elevation (b) view of the truss under initial
and deformed configuration. 19
Figure 2.3 Global nonlinear behaviour in terms of load vs displace-
ment. 20
Figure 2.4 (a) Plan and (b) elevation view for the shallow space truss.
A force P = 300 kN is applied at node 1 along the direc-
tion of z-axis. 20
Figure 2.5 Plan (a) and zoomed 3d (b) view of the truss under initial
and deformed configuration. 22
Figure 2.6 Nonlinear behaviour in terms of load vs displacement at
the centred node. ninc presents for the number of incre-
mental load steps. 23
Figure 3.1 Connections in a cable structure. 26
Figure 3.2 A simple cable structure with zero external loads. The ar-
rows indicate the directions of the elements (Tibert, 1999). 27

vii
Figure 3.3 Different equilibrium configurations of the plane struc-
ture in Figure 3.2 (Tibert, 1999). 29
Figure 3.4 Topology of the cable net. Here m = 4 and n = 4. 30
Figure 3.5 Different equilibrium configurations of the cable net. 31
Figure 4.1 Initial conditions: velocity trace at t = 0. 37
Figure 4.2 System of nodal masses in a line. 38
Figure 4.3 SDOF time displacement trace. Effect of viscous damping
for (a) underdamped oscillations and (b) critically dampe-
doscillations. 40
Figure 4.4 A typical kinetic energy trace for a MDOF structure. 41
Figure 4.5 A trace for typical kinetic energy peak. 42
Figure 4.6 The link connecting joint i to joint k. 43
Figure 4.7 Initial (dotted) and equilibrium (solid) configuration when
the presstress ration is 2:3. (a) Plan view and (b) 3D view. 44
Figure 4.8 Initial (dotted) and equilibrium (solid) configuration when
the presstress ration is 1:1. (a) Plan view and (b) 3D view. 44
Figure 4.9 Initial (dotted) and equilibrium (solid) configuration when
the presstress ration is 2:1. (a) Plan view and (b) 3D view. 45
Figure 4.10 Kinetic energy trace for the structure when the presstress
ratio is 1:1. 45
Figure 5.1 A typical inflatable dam located in Mombaldone, Pied-
mont, Italy. 47
Figure 5.2 Typical composite structure of membrane (a) for inflat-
able dams and physical illustration of rubber sheet re-
inforced with aramid fabric (b) for the Ramspol dams
(Breukelen, 2013). 48
Figure 5.3 Illustration of the loading-shape interaction for an inflat-
able dam. 49
Figure 5.4 Illustration of the force-update extension in dynamic re-
laxation algorithm. 50
Figure 5.5 Possible element orientations in the cross-sectional profile
of the dam: 1. upstream with negative slope, 2. upstream
with positive slope, 3. downstream with negative slope
and 4. downstream with positive slope. 51
Figure 5.6 Geometry of different types of dams. (a) Folded dam,
(b) dam with fin and (c) spillway gate without (with)
fin. 52
Figure 5.7 Elastica element with tangential and normal force com-
ponents. 52
Figure 5.8 Illustration of a generalized element chain, sign conven-
tion and terminology. 54
Figure 5.9 Notations for a generic cable system. 55
Figure 5.10 Illustration for terminology for calculating the center po-
sition of the circular profile. 63
Figure 5.11 Finding the initial circular shape (curve with circles) start-
ing from flat configuration. 64

viii
Figure 5.12 Form-finding procedure (with external hydrostatic pres-
sure). The final equilibrium shape is shown in curve with
circles. 65
Figure 5.13 Final equilibrium shape of air-water inflated dams with
different upstream water head: (a) hu = 1.52 m, (b) hu =
5.8 m. Curves with circles present the final configura-
tion. 66
Figure 5.14 Membrane tension with different upstream water head. 67
Figure 5.15 Influence of different internal steps on N-R iterations when
m = 40. 69
Figure 5.16 Influence of different number of finite elements on N-R
iterations when (a) m = 20 and (b) m = 50. 70
Figure 5.17 Influence of different number of finite elements on equi-
librium shape when (a) m = 20 and (b) m = 50. 71
Figure 5.18 Values evaluated by Equation (5.47). 72
Figure 5.19 Illustration of N-R iteration steps for different initial con-
figurations. (a) Flat configuration and (b) circular config-
uration. 73
Figure 5.20 Illustration of (a) horizontal component and (b) vertical
component of membrane tension. 74
Figure 5.21 Form-finding procedure for single-anchor inflatable dams.
(a) Initial configuration (curve with circles) and (b) final
equilibrium configuration (bold curve with circles). 76
Figure 5.22 Cross-sectional equilibrium profile of 5 inflatable dams
with different anchorage-point separations. D is the dis-
tance between anchorage points. 77
Figure 5.23 Influence of anchorage-point separations on membrane
tension forces. 78
Figure 5.24 Approximated area for the dam cross-section with 12 el-
ements. Areas that fall outside of the profile boundaries
are defined as negative values. 79
Figure 5.25 Final equilibrium shapes of an inflatable dam assuming
constant internal air pressure and an internal air pressure
update. 81
Figure 5.26 Membrane tension values obtained with and without the
internal air pressure update when the upstream head changes. 82

L I S T O F TA B L E S

Table 2.1 Coordinates of each node for shallow space truss. 17


Table 2.2 Comparison among solutions. 18

ix
Table 2.3 Coordinates of each node for shallow geodesic dome. Unit:
[cm]. 21
Table 2.4 Displacements of each node for shallow q geodesic dome,
where u is the total displacement i. e. u = u2x + u2y + u2z . 21
Table 5.1 Membrane tension for various methods. 65
Table 5.2 Membrane tension for various methods. "RE" and "ESE"
represents rigid element and extensible straight element
respectively. 68
Table 5.3 Minimum internal steps for selected element number. m
is element number, nstep is internal step number. 68
Table 5.4 Minimum load increments for different initial configura-
tions when m = 40. nstep is load increment number. 72
Table 5.5 Basic data for single-anchor dam. 75

x
ABSTRACT

The present thesis is devoted to study representative examples of "form-active"


structures and to present computational approaches and procedures which can
be adopted to determine their initial configuration and their response to the
applied loads. An introduction to the "form-active" structures and their applica-
tions are presented in Chapter 1.
Chapter 2 is devoted to the derivation and the discussion of the tangent stiff-
ness matrix, composed of the elastic stiffness matrix and the geometric stiffness
matrix, in the convert of 3D trusses or cable systems, under the assumption of
small strains and linear elasticity but large displacements. A combination be-
tween Finite Element method and Newton-Raphson’s iteration is generated in
the MATLAB® R2017a environment. The important role of geometric stiffness
matrix in the nonlinear analysis is introduced. The results are compared to com-
mercial software (Midas GEN 2017 ) and are found to be in excellent agreement.
The effects for the accuracy of this method are discussed.
Cable structures are mainly discussed in Chapter 3 and Chapter 4, where the
Force Density Method and Dynamic Relaxation Iteration are explained respec-
tively. Examples show the advantages of these methods and further improve-
ments are elaborated.
Chapter 5 presents an original study of form-finding procedure for inflatable
dams. In this chapter, it is illustrated that an alternative tool with computational
advantage for the form-finding and analysis of inflatable dams through the dis-
cretized finite element method. This method has various applications in the area
of inflatable dams such as single-anchor and double-anchor dams, as well as the
application of updated internal air pressure scheme for large external loads. It
is also flexible to deal with different initial configurations such as flat or circular
configuration. This flexibility gives a possibility to analyze different inflatable
dams with short or no anchorage separation (single anchorage systems).

xi
INTRODUCTION
1
1.1 form-active structures

The problem of "form-finding" is relevant to structures in which the internal


forces are axial forces or in-plane membrane forces of tensile nature, with van-
ishing bending actions and shear actions. Such structures have no rigidity and
therefore their shape is dictated by the external forces (and prestressing forces,
if any) acting on them; for a structure of this type, the equilibrium configuration
is uniquely linked to the applied forces and is called “form-active” shape for the
given load (see Figure 1.1). The term "form-active" was introduced by Engel in
his book Structure Systems, first edition published in 1967.
Tensile form active shapes.
shapes
bending stiffness, a cable
cable m
shape –the ‘form-active’ shsh
it to equilibrate the applie
applie
purely tensile internal forc
forc
(a)
arrangements produce diffdiff
shapes.

Compressive form-active sh
s
(b)

Figure 1.1: Tensile (a) and compressive (b) form-active shapes (Macdonald, 2007).
H. Isler:
H. Isler: membranes
membranes under
under self-weight
self-weight andand inverted
inverted structu
struct
used
used
Tensile form-active for the
for
structuresthe design
design
include of concrete
of
cables,concrete shells
cable-nets,shells
prestressed mem-
branes and air-supported membrane structures. The present thesis studies some
examples of such form-active structures and the computational approaches and
procedures which can be adopted to determine their initial configuration and
their response to the applied service loads.
With reference to a cable and to its form-active shape corresponding to a
given applied loading, one can note that if a rigid structure is constructed whose
longitudinal axis is the mirror image of such form-active shape, then the rigid
structure too will be subjected exclusively to axial internal forces when the same
load is applied. And if the load is reversed, in the mirror-image rigid structure
all the axial internal forces are compressive (Figure 1.1 (b)). This "principle"
carries over to membranes and corresponding mirror-image rigid counterparts,
and has been used by the Swiss engineer Heinz Isler (1926-2009) to design some
of his thin-shell concrete structures (see Figure 1.2).
Coming back to tensile membranes, it is worth mentioning that in recent years
there has been a growing interest in hybrid systems which integrate membranes
with flexible bending elements. Although the topic is not dealt with in the

1
Compressive
Compressive form-active
form-active shapes
shapes

2 introduction
H. Isler:
H. Isler: membranes
membranes under
under self-weight
self-weight and and inverted
inverted structures
structures
usedused for design
for the the design of concrete
of concrete shells
shells

(a) (b)

Figure 1.2: Heinz Isler: membranes under self-weight (a) and inverted structures (b)
used for the design of concrete shells.

present thesis, an example of such structural system is represented in Figure


1.3, taken from Van Mele et al. (2013), for illustrative purposes.
Hybrid structural sy
Hybrid structural systems:
integration of ‘bend
integration of ‘bending-
active’ linear eleme
active’ linear elements and
tensile membrane elements
(a) 3D view
tensile membrane e

camping tent
with flexible (fiberglass) poles

Example(*): 2 elastically bent arch beams camping ten


(restrained to the ground) + 2 analogous
suspended bending elements + membrane;
with flexible
the suspended elements create ‛high points’
without additional, external structural
(b) Plan view
elements.
The bending-active (initially straight)
Example(*): 2 elastically bent arch beams
Figure 1.3: Hybrid structural systems: integration of "bending- active" linear elements
elements can be made of Fibre Reinforced
(restrained to the ground) + 2 analogous
and tensile membrane elements
Polymer (FRP) (Van
composite
bending stiffness)
Mele
(high et al.,low
strength, 2013).

suspended
The system bending
(*) Van
consists of two
actively elastically
bent
elements
Mele et al., "Shaping
bent
linear elements", Int J arches
+with
tension structures membrane;
(externally
Space Struct, 28, 2013. restrained to the
ground) and the suspended
of two other suspended elements bending create
elements, ‛high points’ with ten-
in interaction
without
sile membrane additional,
elements. The flexibleexternalelastic beams, structural
initially straight, are called
elements.
"bending-active" beams and function both as support and as shape-defining
system for the membrane surface. Bending-active textile hybrids of this type are
particularlyThe bending-active
applicable for temporary (initiallyand mobile straight)constructions. One application
elements structures
of such bending-active can be made of Fibretent
is the camping Reinforced
with flexible (fiberglass)
Polymer
poles (Figure 1.4). The (FRP) composite
bending-active (initially (high strength,
straight) elementslow
can be made
bending
of Fibre Reinforced stiffness)
Polymer (FRP) composite which have the properties of high
strength, low(*)density and
Van Mele etlow
al., bending
"Shapingstiffness.
tension structures with
It is evident that the form-finding of such
actively bent linear elements", Int hybrid structures
J Space is more
Struct, 28, 2013.complex
than for standard active-form structures; and several research studies can be
1.2 representative examples for tensile form-active structures 3

Chapter 2

LiteratureFigure
review
1.4: A camping tent with flexible poles

found in the recent literature focusing on numerical form-finding methods for


form- and bending-active structures.
2.1 The Historical review
next section will introduce some representative examples of tensile form-
active structures, especially the cable structures. Chapter 3 and Chapter 4 will
The focus
first structures regarded
on the study ascable-net
of these cable roofs are four pavilions with hanging roofs
structures.
built byPneumatic
the Russian engineerorV.inflatable
structures G. Shookhov
damsatcan
an also
exhibition in Nizjny-Novgorod
be regarded as form-active
in 1896. Duringthe
structures, thediscussion
1930’s a small number of
of inflatable roofwill
dams structures of moderate
be presented sizes were
in Chapter 5.
built in the U.S.A. and Europe, but none of major importance [88].
A big1.2step
representative examples
in the development for tensile
of suspended form-active
roofs came in 1950 when structures
Matthew
Nowicki designed the State Fair Arena, Figure 2.1, at Raleigh, North Carolina,
USA.The firstNowicki
Sadly, exampledied
for that
tensile
sameform-active
year in a structures is the
plane crash, but J.S.
his Dorton Arena or
work continued
the State
through Fair Arena,
the architect Figure
William 1.5, at
Henry Raleigh,
Deitrick andNorth
civil Carolina, USA.Severud
engineer Fred This is and
a big
step in the development of
in 1953 the arena was completed [88]. suspended roofs and came in 1950 designed by
Matthew Nowicki (1910-1950); the stadium was the first permanent structure in
the world to use a cable-supported roof.

(a) (b)
(a) (b)
Figure 1.5: The Dorton Arena at Raleigh, North Carolina, USA. (a) Photo taken by a
Pinterest user and (b) structural system (Tibert, 1999).
Figure 2.1: The State Fair Arena at Raleigh, North Carolina, U.S.A., (a) Repro-
duced
Its design from [10],
features (b) cable
a steel Structural system,saddle-shaped
supported reproduced from [16].
roof in tension, held
up by parabolic concrete arches in compression. The structure is based on two
parabolic concrete arches which lean over to the point that they are closer to
On an exchange visit to the U.S.A. in 1950 a German student in architecture, named
being parallel to the ground than they are to being vertical. The arches lean
Frei Otto, previewed the drawings for the Raleigh Arena in the New York office of
Fred Severud. Otto saw that the project embodied many of his own ideas about how

5
4 introduction

toward and beyond each other such that they cross each other 8 meters above
ground. These arches, approaching horizontal in plane, thus serve as the outer
edges of the structure, which when viewed from above appears almost elliptical.
The arches are supported by slender columns around the building perimeter.
Cables are strung between the two opposing arch structures providing support
for the saddle-shaped roof.
The next example for tensile cable structures is the German pavilion at the
World’s Fair in Montreal 1967 (Figure 1.6), designed by Frei Otto (1925-2015).
This pavilion is the first large cable net structure with fabric cladding, this kind
of roof makes the weight of the pavilion lighter: its steel and plastic roof weighed
only 150 tons; one third to one fifth the weight of normal roofing materials. This
system used a roof of steel cable net suspended from eight slender steel masts
of varied height, situated at irregularly intervals and supported by steel cables
anchored outside the structure, covered an area the size of a city block.

(a) (b)

Figure 1.6: The German pavilion (a) at the World’s Fair in Montreal 1967 and (b) its
form-finding study model (Langdon, 2015).

Note that the form-finding model in Figure 1.6 (b) is not a numerical model
but an experimental soap film one developed by Otto, while the present thesis
is devoted to study the numerical models for form-finding problems of some
representative examples of form-active structures.
Another example for the form-active structure is United States pavilion at the
World’s Fair in Osaka 1970 (Figure 1.7). This pavilion uses a large low-profile
super elliptic air-supported roof, with a membrane attached to a diagonal ca-
ble net, which is a pioneering structure at 1970s. The designer of the pavilion
is David Geiger (1935-1989), who is also the inventor of this kind of the air-
supported fabric roof system. This system was welcomed at that time because
it could provide the economically best alternative to span large distances.
The last two examples in this section are devoted to cable-net facades. The
first cable-net facede is located at the Hilton Hotel (formerly Kempinski Hotel,
Figure 1.8) in Munich, Germany. This facade is 40 m wide by 25 m tall and
its longer-span horizontal direction was installed with the pretension cables
with a diameter of 22 mm. These cables are spaced at 1.5 m on center, and
the pretension in the cables can limit their deflections to 0.9 m (Mazeika and
Kelly-Sneed, 2007).
1.2 representative examples for tensile form-active structures 5

Figure 1.7: The USA pavilion at the World’s Fair in Osaka 1970

Figure 1.8: Cable-net facade at Hilton Hotel, Munich

A more recent example of such facade is the New Poly Plaza (Figure 1.9)
in Beijing, China. It is a building with a glass wall 90 m high by 60 m wide
designed by Skidmore, Owings & Merrill LLP. The facade is supported by an
orthogonal cable net with stainless steel cables spaced at 1333 mm horizontally
and 1375 mm vertically. Panes of glass of a slightly smaller size (to accommodate
the joints) are clamped in place by stainless steel fittings attached to the cables
at their intersection.
6 introduction

6
fA

R A
4

Figure 1.9:θ Cable-net facade at New Poly Plaza, Beijing


3
C(x0,y0) T(s)

2
1.3 first examples
R of geometrically nonlinear analysis of ca-
ble systems
1

0
-1 0 1 2 3 4 5 6 7

Since the form-finding problem of form-active structures is a kind of geometri-


D = anchorage-point separations
cally nonlinear problem, here are two examples to show how the treatment for
geometrically nonlinearity will be done.

1.3.1 Single Cable under Concentrated Force

Consider a single cable with two fixed end (Figure 1.10), the total initial length
of the cable is 2L0 , and the concentrated load P is applied in the middle of the
cable. The stiffness of half of the cable can be represented as k = EA/L0 .

β v
N N

Figure 1.10: Initial configuration of the example.

Then the vertical equilibrium equation for the mid-point of the cable in the
deformed configuration can be written as:
P
N cos β = (1.1)
2
where
j
v
q
N = k ( L − L0 ) , cos β = , L = L20 + v2 (1.2)
L
Then, Equation (1.1) can be rewritten in dimensionless form:
     k
v 1 1 P
1− q = (1.3)
L0 1 + ( Lv0 )2 2 EA i
m
l

z
1.3 first examples of geometrically nonlinear analysis of cable systems 7

which can also be interpreted as:

Fint (v) = Fext (1.4)

Equation (1.3) illustrates a strong nonlinear relationship between the non-


dimensional displacement v/L0 and the non-dimensional external load P/EA.
Figure 1.11 shows the plot of Equation (1.3); the horizontal axis is the the non-
dimensional displacement while the vertical axis is the non-dimensional ex-
ternal load as a function of v/L0 . Practically, Equation (1.3) can be solved by
Newton-Raphson’s technique which will be introduced in the next section.

0.3

0.25

0.2
P/EA

0.15

0.1

0.05

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
v/L 0

Figure 1.11: Relationship between vertical displacement and vertical load withoud pre-
stress.

1.3.2 Initially Flat Cable Net

Consider a structure consisting of 4 prestressed cables, as it is shown in Fig-


ure 1.12. The length L0 is fixed and vertical loads P are concentrated on the 4
central nodes. The other nodes are fixed to the ground by hinges. In this way
the problem consists of a double-symmetric structure loaded with symmetrical
loads; this symmetry can reduce reduce the unknowns of the problem; only 2
independent nodal displacement components u and w.
The analysis is carried out by writing two equilibrium equations for node G 0
(in deformed configuration) in terms of the unknown displacements u and w,
and solving the nonlinear equations by Newton-Raphson’s technique.
8 introduction

I K
Prestressed cable net Elongations of the
P P cables converging
in node G:
D
A e1 = e 2 = 2u
B C
P P 1 e3 = e 4 = L 3 − L 0

F 2 G 4 H L3 = L 4 = u 2 + (u − L0 ) 2 +
E
3 w N1
u G′M = (− u i + (L0 − u) j − w k) /
u N4
N2
J G′H = ( (L0 − u) i − u j − w k ) / L
M N3 G′
x, i G′M , G′H being unit vectors
y, j
z, k
N1 = N 2 N3 = N 4
Equilibrium equations
Figure of node
1.12: Initial G’ (in deformed
configuration config):
of the example.
In horizontal direction x: N 2 + N 3 ( u L3 ) − N 3 ( L 0 − u ) / L3 =0
Let T0 be the pretension and ei be the elongation for each cable from a flat
In vertical
configuration direction
to the 2N3 w L3 Then
z: configuration.
prestressed ( )
= P the internal force in each
cable can be written as:
N 2 = TEA
0 + EA ( L0 ) 2u N3 = T0 + ( EA L0 )( L3 − L0 )
Ni = T0+ e (1.5)
L0 i
G. Novati - Politecnico di Milano
At the node G0 , the elongation can be written as:

e1 = e2 = 2u, e3 = e4 = L 3 − L 0 (1.6)

where
q
L3 = L4 = u2 + (u − L20 ) + w2 (1.7)

Since N1 = N2 and N3 = N4 , then the equilibrium equations of node G 0 in


deformed configuration are:
u L −u
N2 + N3 − N3 0 =0
L3 L3 (1.8)
w
2N3 =P
L3
With all the equations above, Equation (1.8) can be rewritten in the terms of
displacement:
 2u − L0
( T0 + 2ku) + T0 + k( L3 − L0 ) =0
L3 (1.9)
w
2 T0 + k ( L3 − L0 ) =P
L3
1.3 first examples of geometrically nonlinear analysis of cable systems 9

where k is the stiffness and k = EA/L0 .


Equation (1.9) is in the form:

Fint (u, w) = Fext (1.10)

where Fint (u, w) is nonlinear.


Now, the residual or unbalanced forces can be defined as:

R(u) = Fint (u) − Fext (1.11)

where u is the set of unknown displacements (u, w).


Then the system R(u) = 0 can be solved by Newton-Raphson’s iteration.
This iterative algorithm will be introduced in later chapters. The results can be
found in Figure 1.13, which illustrates a nonlinear behaviour: as the vertical
concentrated load P increases, the vertical displacement w increases nonlinearly,
while enlarging the prestress T0 in the cables, the vertical displacement w is
reduced. This is due to the introduction of geometric stiffness and this particular
phenomenon will be discussed in the next chapter.

50

45

40

35
Vertical loads [KN]

30

25

20

15

10
T 0 = 100 KN
5 T 0 = 140 KN
T 0 = 180 KN

0
0 20 40 60 80 100 120 140 160 180 200
Vertical displacements [mm]

Figure 1.13: Relationship between vertical displacement w and vertical load P under
different prestressing T0 .
T H E T R E AT M E N T O F G E O M E T R I C A L LY N O N L I N E A R I T Y
WITH REFERENCE TO TRUSSES
2
As it is illustrated that the last computational example of the cable net in Chap-
ter 1, the form-finding problem may be solved by stiffness matrix method, as
the Jacobian matrix plays the role of the "stiffness" matrix. In this chapter, such
method is discussed with the respect to truss structures with the property of
geometrically nonlinearity.
A structure must satisfy the deformed equilibrium conditions in the nonlin-
ear case, the pieces of the structure must fit together in their loaded state and
a constitutive equation must be satisfied. Thus, the truss must meet all these
conditions with assumptions that the structures still remain in the elastic range
and no bulking and unstable cases.
The geometrically nonlinear problem will be solved by Newton-Raphson’s
iteration as well as Finite Element method when dealing with multi-degree-of-
freedom cases.
A typical step of this analysis which is reviewed from the book by Levy and
Spillers (2013) and Meek (1991), can be described as fellows. Given a fixed joint
load matrix P and a starting configuration which is not in equilibrium with this
joint load matrix, the following sequence of actions must be taken:

• Compute the unbalanced load R. Since the member forces F are not in
equilibrium with the given load P, the unbalanced load can be computed
as R = P − NT F, whereN is the geometric matrix composed of cosines
between member local axes and global axes.

• Solve for the incremental displacements. Under the unbalanced load R the
structure will still displace. This computation involves solving the system
(KE + KG )δu = R for the node displacement δu, where KE is the classical
elastic stiffness matrix and KG is the geometric stiffness matrix which will
be discussed below.

• Updating the coordinates. The coordinate matrix X is updated to describe


the deformed configuration as X = X + δu.

• Compute new member forces. The new coordinates of the structure im-
ply new member lengths which in turn imply new member forces. The
updated length change will be used here.

• Repeat this sequence of calculations. Computation stops when the unbal-


anced force R → 0. Practically, this procedure is controlled by a relatively
small tolerance. It will be argued below that these calculations comprise
the application of Newton-Raphson’s method to the equations of node
equilibrium.

11
12 the treatment of geometrically nonlinearity with reference to trusses

2.1 linearization of the joint equilibrium equations

In nonlinear analysis, the equilibrium P = NT F holds at deformed configuration.


Under a perturbation this equation takes the form,

dP = dNT F + NT dF (2.1)

It is assumed that all variables can be described in terms of the joint coordinate
matrix R and its perturbation, the node displacement matrix δu. In Equation
(2.1) the term NT dF describes the change in member forces with the matrix N
fixed which is simply linear elastic analysis. That is,

NT dF → KE δu (2.2)

It is noted here that the variation of the member forces, dF, with the equilibrium
"operator", N, fixed simply returns linear elastic theory. It is then the term dNT F
of Equation (2.1) which gives rise to the geometric stiffness matrix KG , i.e.

d(NT F) F f ixed → KG δu (2.3)

The way to compute matrix KG as a gradient will now be explained. This ar-
gument is based on the fact that a small variation dF of a function F can be
represented as:

dF ∼ ∇F · dx (2.4)

∇F is of course the gradient matrix of the function F. A vector F of n functions


with m variables has an n × m gradient matrix as indicated in Equation (2.5)
below.
 ∂f ∂f ∂ f1

∂x1 ∂x2 · · · ∂xm
    1 1
f1 x1  
 
 f2 
 
 x1   ∂ f2 ∂ f2 ∂ f2 
· · ·
F=  ..  ; x =  ..  ; ∇F =  (2.5)
    ∂x1 ∂x2 ∂xm 
.  .   .. .. .. 
 . . . 

fn xm ∂ fn ∂ f2 ∂ fn
∂x1 ∂x2 · · · ∂xm

2.2 the geometric stiffness matrix

In order to complete details of the derivation of the geometric stiffness matrix as


a gradient, it is convenient to work with (NT F)i , the contribution of bar i to the
2.2 the geometric stiffness matrix 13

equilibrium equations. More explicitly, this term is shown below for one typical
bar with two node as:
 . 
..
 
 ( ni ) x 
 
( n )
 
 i y 
 
 ( ni ) z 
 
(NT F)i =  ...  Fi (2.6)
 
 
−(n ) 
 i x
 
 −(ni )y 
 
 −(ni )z 
 
..
.

The term [(ni ) x , (ni )y , (ni )z ] T from Equation (2.6) contributes to the start node
(node A) while the term [−(ni ) x , −(ni )y , −(ni )z ] T does to the end node (node
C)for the typical bar.
It is the gradient of Equation (2.6) which gives the contribution of bar i to the
geometric stiffness matrix. Referring back to the original discussion of Newton-
Raphson’s method, for the case of the equilibrium equations discussed here,

f →(NT F)i = NiT dFi ,


x →R, (2.7)
dx →δu

and

∇f → ∇(NT F)i = (∇NiT ) Fi (2.8)

Now since NiT is a function of the coordinates [ x A , y A , z A ] T and [ xC , yC , zC ] T ,


chain rule of differentiation yields can be used:

d(NiT F)i = (∇NiT · dx ) Fi =



 T
∂Ni ∂NiT ∂NiT
= dx A + dy A + dz A +
∂x A ∂y A ∂z A (2.9)
∂NiT ∂NiT ∂NiT

+ dxC + dyC + dzC Fi
∂xC ∂yC ∂zC

With dx A = (δu A ) x , dy A = (δu A )y , dz A = (δu A )z , dxC = (δuC ) x , dyC = (δuC )y


and dz A = (δuC )z , i. e. the change in coordinates are interpreted as nodal dis-
placements, the contribution of member i to the geometric stiffness matrix be-
comes:
" #
T AA T AC
∇(Ni ) ∇(Ni )
(KG )i = ∇(NiT Fi ) = T CA
Fi (2.10)
∇(Ni ) ∇(NiT )CC
14 the treatment of geometrically nonlinearity with reference to trusses

where
 ∂ (n ) ∂ ( ni ) x ∂ ( ni ) x

i x
∂x A ∂y A ∂z A
 
 ∂ (n ) ∂ ( ni ) y ∂ ( ni ) y 
∇(NiT ) AA =  ∂xi y ∂y A ∂z A 
 A 
∂ ( ni ) z ∂ ( ni ) z ∂ ( ni ) z
∂x A ∂y A ∂z A
 ∂ (n ) ∂ ( ni ) x ∂ ( ni ) x

i x
∂xC ∂yC ∂zC
 
 ∂ (n ) ∂ ( ni ) y ∂ ( ni ) y 
∇(NiT ) AC =  ∂xi y ∂yC ∂zC 
 C 
∂ ( ni ) z ∂ ( ni ) z ∂ ( ni ) z
∂xC ∂yC ∂zC
(2.11)
− ∂(∂xniC)x − ∂(∂yniC)x − ∂(∂zniC)x
 
 
 ∂ (n ) ∂ (n ) ∂ (n ) 
∇(NiT )CC =  − ∂xi y − ∂yiC y − ∂ziC y 
 C 
∂ ( ni ) z
− ∂xC − ∂(∂yniC)z − ∂(∂znCi )z
 ∂ (n )
− ∂(∂yniA)x − ∂(∂zniA)x

− ix
 ∂x A 
 ∂ (n ) ∂ (n ) ∂ (n ) 
∇(NiT )CA =  − ∂xi y − ∂yiA y − ∂z iA y 
 A 
∂ ( ni ) z ( ni ) z
− ∂x A
− ∂∂y A
− ∂(∂znAi )z

The derivatives indicated in Equation (2.10) can now be computed completing


the description o the geometric stiffness matrix. Symbolically,
" #
(KG )iAA −(KG )iAA
(K G ) i = (2.12)
−(KG )iAA (KG )iAA

with
 
1 − (ni )2x −(ni ) x (ni )y −(ni ) x (ni )z
Fi 
(KG )iAA =

 (n ) (n ) 1 − (ni )2y −(ni )y (ni )z  (2.13)
Li  i y i x 
−(ni )z (ni ) x −(ni )z (ni )y 1 − (ni )2z

or
Fi
(KG )iAA = (I − ni niT ) (2.14)
Li

where

ni = [(ni ) x , (ni )y , (ni )z ] T (2.15)

Equation (2.13) follows directly from differentiation as will be shown below.


Consider the x-component of the unit vector ni when written in terms of nodal
coordinates,
x A − xC x A − xC
( ni ) x = p = (2.16)
( x A − x C )2 + ( y A − y C )2 + ( z A − z C )2 Li
2.3 newton-raphson’s iteration algorithm 15

Differentiation yields,

∂ ( ni ) x 1 1 x A − xC 1
= − · 2( x A − xC ) = (1 − (ni )2x )
∂x A Li 2 Li Li
∂ ( ni ) x 1 1 x A − xC 1
= − · 2 ( y A − y C ) = − ( ni ) x ( ni ) y (2.17)
∂y A Li 2 Li Li
∂ ( ni ) x 1 1 x A − xC 1
= − · 2 ( z A − z C ) = − ( ni ) x ( ni ) z
∂z A Li 2 Li Li

The derivatives of (ni )y and (ni )z with respect to ∂x A , ∂y A and ∂z A follow in


a similar fashion given the form of Equation (2.16). Derivatives with respect to
the coordinates at the end node C are the negative of derivatives with respect to
the coordinates at node A.
One can recall the formulation of the elastic stiffness matrix and the Equation
(2.18) and Equation (2.19) is obtained.
" #
(KE )iAA −(KE )iAA
(K E ) i = (2.18)
−(KE )iAA (KE )iAA

with,

( EA)i
(KE )iAA = (ni niT ) (2.19)
L0i

Finally, the tangent stiffness matrix is formulated in Equation (2.20) as the


sum of both stiffness matrices mentioned before.

(Kt ) i = (K E ) i + (K G ) i (2.20)

2.3 newton-raphson’s iteration algorithm

The following is a Newton-Raphson’s iteration algorithm for performing a geo-


metrically nonlinear truss analysis. This is an implicit formulation which uses
Newton-Raphson’s iterations at the global level to achieve equilibrium during
each incremental load step. Material nonlinearities are not presently included
in the algorithm. A program implementing this algorithm has been written in
MATLAB and some representative results are provided at Section 2.3 The algo-
rithm proceeds as follows:

a. Define/initialize variables
• P = the total vector of externally applied global nodal forces
• Pn+1 = the current externally applied global nodal force vector
• F = the vector of truss axial forces vector
• u = the vector of global nodal displacements, initially u = 0
• X = the vector of nodal x coordinates in the undeformed configura-
tion
16 the treatment of geometrically nonlinearity with reference to trusses

• Y = the vector of nodal y coordinates in the undeformed configura-


tion
• Z = the vector of nodal z coordinates in the undeformed configura-
tion
• L = the vector of truss element lengths based on current u, the initial
lengths are saved in a vector L0
• Kt = KE + KG , the assembled global tangent stiffness matrix

b. Start Loop over load increments


1 Calculate load factor λ = 1/nIncrements and incremental force vector
dP = λP
2 Calculate global stiffness matrix Kt based on current values of L, L0 and
F.
3 Solve for the incremental global nodal displacements du = K− 1
t dP
4 Update global nodal displacements un+1 = un + du
5 Update the global nodal forces Pn+1 = Pn + dP and current truss ele-
ment lengths L
6 Calculate the vector of new internal truss element axial forces Fn+1 . For
trusselement i the axial force is shown in Equation (2.21). Using this
expression is for avoiding ill conditioning during the computation
(Yaw, 2011).

( EA)i L2i − L20i


Fin+1 = · (2.21)
L0i Li + L0i

n +1
7 Construct the vector of internal global forces Nint based on Fn+1
n +1
8 Calculate the residual R = Pn+1 − Nint
kRk2
9 Calculate the convenience conv = kPk2
. An L2-norm is used here:kRk2 =

RT R
10 Iterate for equilibrium if necessary. Set up iteration variables.
• Iteration variable k = 0
• Tolerance tol = 0.001
• Maximum iteration maxIter = 100
• δu = 0
• Ftemp = Fn+1
11 Start Iterations while convergence > tolerance and k < maxIter
i Calculate the new global stiffness Kt
+1
ii Calculate the correction to un+1 : δunk+ n +1
1 = δuk + K− 1
t R.
+1
iii Update L based on current un+1 + δunk+ 1
2.4 examples 17

Table 2.1: Coordinates of each node for shallow space truss.


node x[m] y[m] z[m]
A 0 0 0

B -10 -5 -1

C -10 5 -1

D 10 5 -1

E 10 -5 -1

iv Calculate the vector of new internal truss element axial forces


+1
Fktemp
n +1 +1
v Construct the vector of internal global forces Nint based on Fktemp
n +1
vi Calculate the residual R = Pn+1 − Nint
kRk2
vii Calculate the convenience conv = kPk2
viii Update iterations counter k = k + 1
12 End of while loop iterations

c. Update variables to their final value for the current increment


• Fn+1 = Ftemp
• un+1 = un+1 + δu

d. End Loop over load increments

2.4 examples

2.4.1 Shallow Space Truss

This example is adopted from the book by Ghali, Neville, and Brown (2003).
This example also plays the role of a benchmark to check the MATLAB codes
described in the present thesis.
A shallow space truss with 4 bars is shown in Figure 2.1. Node B to E are
fixed while node A is loaded by a point load P = 100 kN downwards. The axial
rigidity for each bar is EA = 40000 kN and the coordinates for each node are
shown in Table 2.1.
The vertical displacement u A of node A the axial force N in each bar, which
should be identical since the symmetry, are calculated. Table 2.2 lists the solution
from the book writed by Ghali, Neville, and Brown (2003) as well as the solution
from commercial software Midas GEN 2017 and also the one from MATLAB.
The result shows that the accuracy of the MATLAB code mentioned in Section
2.3 which gives a relative error of 0.2% in vertical displacement and 0.1% in
axial force.
18 the treatment of geometrically nonlinearity with reference to trusses

B E

A
x
z

y
C D A

(a) (b)

Figure 2.1: Plan (a) and elevation (b) view for the shallow space truss. A point force
P = 100 kN is loaded at node A along the direction of z-axis.

Table 2.2: Comparison among solutions.


uA N
Ghali’s 0.483 190.1

Midas Nonlinear 0.483 190.1

MATLAB 0.484 189.9

Midas Linear 0.884 280.6

It can be noted that the influence of geometric stiffness matrix from Table
2.2, as one can find the displacement from linear analysis is almost double as
the one from nonlinear analysis. This scenario is due to account of geometric
stiffness matrix that actually increases the total stiffness matrix as which is the
sum of both geometric and elastic stiffness matrix, while the latter is the only
one that is used in linear analysis.
A deformed shape and a typical nonlinear behaviour of load-displacement
relationship can be seen in Figure 2.2 and Figure 2.3 respectively.
In Figure 2.3, as the load increments are set to 20, one can notice that the
strong nonlinear behaviour between external load and vertical displacement at
node A, which is the natural consequence of the geometric property of shallow
truss.

2.4.2 Shallow Geodesic Dome

The example of shallow geodesic dome is adopted from the paper published by
Tanaka, Kondoh, and Atluri (1985).
The geometry is shown in figure Figure 2.4. The outer nodes, i. e. the nodes
from 8 to 13, are fixed while the central node 1 is loaded by a force directed
along z-axis. The elastic stiffness for each bar is EA = 106 N. Based on the
paper by Tanaka, Kondoh, and Atluri (1985), the first buckling load is around
3.15 × 10−4 EA N. Thus, a external force P = 300 N at node 1 is applied to avoid
2.4 examples 19

-1

-5

-1 undeformed
0
-10 deformed
-5
0
5 5
10
(a)

-1

undeformed
deformed
-1

-10 -8 -6 -4 -2 0 2 4 6 8 10
(b)

Figure 2.2: 3D (a) and elevation (b) view of the truss under initial and deformed con-
figuration.

instability. The initial configuration of the structure as specified in Table 2.3 is


considered.
20 the treatment of geometrically nonlinearity with reference to trusses

100

90

80

70

60
Load [N]

50

40

30

20

10

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Displacement [cm]

Figure 2.3: Global nonlinear behaviour in terms of load vs displacement.

10 4 3 8

5 1 2 x

11 6 7 12
y

13 z

(a) (b)

Figure 2.4: (a) Plan and (b) elevation view for the shallow space truss. A force P = 300
kN is applied at node 1 along the direction of z-axis.

The displacements for each node from MATLAB code are listed at Table 2.4.
These results are computed by using 20 incremental loads. Only free nodes, i. e.
node 1 to node 7 are considered, as the node 8 to node 13 are fixed so that their
displacements should be null.
Because of the symmetry of the structure, the horizontal displacements of
node 1 should be zero and the code meets this fact since actually u x and uy are
numerically zero. Another consequence of symmetry is that the displacements
2.4 examples 21

Table 2.3: Coordinates of each node for shallow geodesic dome. Unit: [cm].
node x y z node x y z
1 0 0 0 8 43.3 -25 8.216

2 25 0 2 9 0 -50 8.216

3 12.5 -21.65 2 10 -43.3 -25 8.216

4 -12.5 -21.65 2 11 -43.3 25 8.216

5 -25 0 2 12 0 50 8.216

6 -12.5 21.65 2 13 43.3 25 8.216

7 12.5 21.65 2

Table 2.4: Displacements of eachq


node for shallow geodesic dome, where u is the total
displacement i. e. u = u2x + u2y + u2z .

node ux [cm] uy [cm] uz [cm] u[cm]


1 -2.1e-18 2.3e-15 0.581 0.581

2 0.019 -1.0e-15 -0.035 0.040

3 0.009 -0.016 -0.035 0.040

4 -0.009 -0.016 -0.035 0.040

5 -0.019 -1.0e-15 -0.035 0.040

6 -0.009 0.016 -0.035 0.040

7 0.009 0.016 -0.035 0.040

of node 2 and node 5 should also be symmetric and this consequence can also
be seen in the Table 2.4. It is also noted here that one can find the total dis-
placements of node 2 to 7 are identical; this phenomenon is also the result of
symmetry.
The last part as last subsection is the listing of figures for deformed configu-
ration (Figure 2.5) and the nonlinear behaviour (Figure 2.6)
As one can find in the Figure 2.5, the symmetry property also holds for dis-
placement, and this fact is also coincided with the previous conclusion.
In the Figure 2.6, the dotted line presents the result from a number of incre-
mental loop steps equal to 5 while stared line presents the one equal to 20. It
can be mentioned that the accuracy of Newton-Raphson’s method is also influ-
enced by the number of the load steps. Since the larger load step is chosen, the
more inner iterations for displacement correction δu will be performed and the
efficiency of the method is more reduced. If the number of inner iterations is
beyond the threshold of iteration steps (which can be set manually), the loop
will be jumped out and thus the accuracy is declined.
22 the treatment of geometrically nonlinearity with reference to trusses

50

40

30

20

10

-10

-20

-30

-40
undeformed
deformed
-50
-60 -40 -20 0 20 40 60
(a)

-4 20

-2 0
0
-20
-40 -30 -20 -10 0 10 20 30 40

undeformed
deformed

(b)

Figure 2.5: Plan (a) and zoomed 3d (b) view of the truss under initial and deformed
configuration.

2.5 summary

A derivation and explanation of the ingredients of a geometric stiffness matrix


for nonlinear analysis, in a small strain assumption, is provided. A combination
between Finite Element method and Newton-Raphson’s iteration is coded and
run in the MATLAB® R2017a environment. The important role of geometric
stiffness matrix in nonlinear analysis is introduced. The results of the formula-
tion are compared to representative benchmarks and are found to be in excellent
agreement. The effects for the accuracy of this method are discussed.
2.5 summary 23

300

250

200
Load [N]

150

100

50

0
0 0.1 0.2 0.3 0.4 0.5 0.6
Displacement [cm]

Figure 2.6: Nonlinear behaviour in terms of load vs displacement at the centred node.
ninc presents for the number of incremental load steps.
FORM-FINDING OF CABLE STRUCTURES WITH FORCE
DENSITY METHOD
3
The form-finding methods for the cable structures may be classified into two
different groups, one is the linear methods and the other is nonlinear methods;
Force Density Method is the one of those linear methods. This chapter discusses
the fundamental theory of Force Density Method as applied to cable structures.
The Force Density Method was initially presented by Schek (1974). This method
is based on the force-length ratios, as it is called force densities, which are de-
fined for each branch of the cable net structure. Force Density Method is a
simple linear system of equations for a possible initial configuration. It is a big
improvement over the previously proposed Grid Method, which is one of the
simplest methods of form finding. The Grid Method is also a linear solution to
the initial equilibrium problem for orthogonal cable nets. However, because of
the restrictions placed on the structure in that method, the resulting shapes are
few (Tibert, 1999), while Force Density Method is a more general method; Schek
(1974) illustrated that the force densities are fairly suitable for the description of
the equilibrium configuration of any general cable nets.
The Force Density Method is a strategy to solve the equilibrium equations for
a cable net without demanding any initial configuration of the structure; it is
like a mathematical trick. Here is a simple example for descprition.
For a general cable network in Figure 3.1, the equilibrium equations in the x-,
y-, and z-directions at that node can be expressed as:
x j − xi x − xi x − xi x m − xi
Tij + Tik k + Til l + Tim + Fxi = 0
Lij Lik Lil Lim
y j − yi y − yi y − yi y m − yi
Tij + Tik k + Til l + Tim + Fyi = 0 (3.1)
Lij Lik Lil Lim
z j − zi z − zi z − zi z m − zi
Tij + Tik k + Til l + Tim + Fzi = 0
Lij Lik Lil Lim

Consider the equilibrium equations above. These equations are nonlinear


since the element length L is a function of the node coordinates. However, if
instead of the element forces, the force-to-length ratios (denoted by q) for each
element are specified, then Equation (3.1) can be written as:

qij ( x j − xi ) + qik ( xk − xi ) + qil ( xl − xi ) + qim ( xm − xi ) + Fxi = 0


qij (y j − yi ) + qik (yk − yi ) + qil (yl − yi ) + qim (ym − yi ) + Fyi = 0 (3.2)
qij (z j − zi ) + qik (zk − zi ) + qil (zl − zi ) + qim (zm − zi ) + Fzi = 0

As one can find, obviously the main advantage of using the force densities
q as description parameters for a cable structure is that any state of equilib-
rium can be obtained by the solution of one system of linear equations. Hence

25
26 form-finding of cable structures with force density method

i
m
l

y x

Figure 3.1: Connections in a cable structure.

the obtained equilibrium configuration has the prescribed force density in each
element without any other conditions. The detailed theory of Force Density
Method will be introduced in the next section.

3.1 basic theory of force density method

The following contents are adopted from the papers by Schek (1974) and Hernández-
Montes, Jurado-Piña, and Bayo (2006) as well as the PhD thesis by Tibert (1999).
In this chapter, only Linear Force Density Method is considered.
One assumption in the Force Density Method is that the cables are straight
and pin-jointed to each other or to the supporting structure. One can start with a
graph of a network as Figure 3.2. Then all nodes are numbered from 1 to ns , and
all elements from 1 to m. The fixed nodes are taken at the end of the sequence
with a number of n f . All the other n nodes are free. Thus, the total node number
is ns = n + n f . Afterwards, the connectivity matrix Cs can be constructed with
the aid of the graph (Figure 3.2). Each branch/element j has the node numbers
k and l (from k to l). The connectivity matrix Cs for the structure is define by
(i = 1, 2,..., ns ):


 +1, for i = k

cs ( j, i ) = −1, for i = l (3.3)


0, in the other case

Since the free and fixed nodes are separated, the connectivity matrix can be
divided into two matrices:

Cs = [C Cf ] (3.4)
Pds

3.1 basic theory of force density method 27

6
1 2

1 2

3 4 5 6

3
7 8
8 9
7 10

4 5
y
11 12
9
z x

Figure 3.2: A simple cable structure with zero external loads. The arrows indicate the
directions of the elements (Tibert, 1999).

where C and C f is the matrix contained with free and fixed nodes respectively.
m Denoting the vectors containing the coordinates of the n free nodes x, y, z, and
similarly for the n f fixed nodes x f , y f , z f , the coordinate differences u, v and w
for each element can be written as:

u = Cs xs = [C C f ][x x f ]T = Cx + C f x f
v = Cs ys = [C C f ][y y f ] T = Cy + C f y f (3.5)
w = Cs zs = [C C f ][z z f ]T = Cz + C f z f

The length l j of the element and the element forces s j form the m-vector l and
s. The load vectors are f x , fy , and fz . The vector elements in u, v, w and l can be
rebuilt as diagonal matrices U, V, W and L. It is easily shown that the sum of
the forces in each node is zero and therefore the network is in the configuration
of equilibrium if the following equilibrium equations are implemented:

C T UL−1 s = f x
C T VL−1 s = fy (3.6)
C T WL−1 s = fz
28 form-finding of cable structures with force density method

where the Jacobian matrices are demonstrated as:


∂l ∂l ∂l
= C T UL−1 , = C T VL−1 , = C T WL−1 (3.7)
∂x ∂y ∂z
By using the force-to-length ratios for the elements, i. e. the force densities, as
description parameters, Equation (3.6) are written as:
C T Uq = f x
C T Vq = fy (3.8)
T
C Wq = fz
where the vector q is described as:
q = L −1 s (3.9)
With Equation (3.11) and the identities:
Uq = Qu, Vq = Qv, wq = Qw (3.10)
where Q is the diagonal matrix belonging to q, Equation (3.8) can rewritten as:
C T QCx + C T QC f x f = f x
C T QCy + C T QC f y f = fy (3.11)
T T
C QCz + C QC f z f = fz
For simplicity by setting D = C T QC and D f = C T QC f , Equation (3.11) can
be expressed as:
Dx = f x − D f x f
Dy = fy − D f y f (3.12)
Dz = fz − D f z f
Then the equation system can be solved:
x = D −1 ( f x − D f x f )
y = D −1 ( f y − D f y f ) (3.13)
z = D −1 ( f z − D f z f )
Since Equation (3.12) is solved by elementary algebra, two cases for matrix D
may take place:
1 Determinant of D is not zero
The matrix D is full ranked and the form of the structure is governed
by the values chosen for the force densities. In the case of a prestressed
cable net with a given connectivity (i.e. fixed C and C f ) the number of
equilibrium shapes is identical to the number of vectors q. This justifies
the use of the force densities as description parameters for a cable net
(Tibert, 1999).
2 Determinant of D is zero
The system can be solved for x only when vectors D f x f is in the space
spanned by the linearly independent vectors of matrix D (if the external
loads are zero). Similarly for y and z.
3.2 examples 29

3.2 examples
3.3. FORCE
3.3. THE THE FORCE DENSITY
DENSITY METHOD
METHOD
3.2.1 Simple Cable Structure Without External Loads
3.3. THE FORCE DENSITY METHOD
(a) Elements
This example is a plan cable structure and its layout can be1–12 have in
found q =Figure
1 3.2. (b) Element
The results are adopted from Tibert (1999). element 4 ha

(a) Elements
(a) Elements 1–12 qhave
1–12 have = 1q = 1 (b) Elements
(b) Elements 1–3, have
1–3, 5–12 5–12 qhave
= 1q and
= 1 and
(a) q = 1 for all elements element
element 4 has 4q has
= 10q = 10
(b) q = 10 for element 4 and q = 1 for
the rest elements
(a) Elements 1–12 have q = 1 (c) Elements 1–3, 5–12
(b) Elements 1–3, have
5–12 qhave
= 1q and
= 1 and (d) Interior
element
element = −0.1
4 has 4q has q = 10 elements ha

Figure 3.6: Different equilibrium configurations for the p

3.3.2 The non-linear force density met

Multiple equilibrium shapes can be obtained with the


However, these shapes may be unsatisfactory from a s
example, the mesh may be irregular and the force dis
fore, it is necessary to find a configuration which is in
(c) q = −0.1 for element 4 and q = 1
satisfies additional conditions. These conditions are ge
(d) q = 1 for inner elements and q = 5
for the rest elements the extended force elements
for boundary density method. It is preferred to st
(c) Elements
(c) Elements 1–3, 5–12
1–3, 5–12 have qhave
= 1q and
= 1 and (d) Interior elements
tions using the shapehave
(d) Interior elements qhave
found 1 qand
= with = 1the
andlinear
edge edge method. I
element
element
Figure 43.3: q has q = −0.1
= −0.1
has 4Different elements
elements
equilibrium configurations of qhave
haveplane
the q=5
= 5 structure in Figure 3.2
(Tibert, 1999).
(c) Elements
Figure
Figure 3.6: 1–3,equilibrium
5–12equilibrium
3.6: Different
Different have qconfigurations
= 1configurations
and for(d) Interior
for
the elements
the plane
plane have
structure
structure qin=Figure
in Figure 13.5.
and edge
3.5.
element 4 has q = −0.1 elements have q = 5 61
Figure 3.3 illustrates the final equilibrium configurations of this structure
among different ratios of force density. The results show that for a single ele-
Figure 3.6: Different
the equilibrium configurations for the planewill
structure
resultininFigure 3.5.
3.3.2
ment
3.3.2 aThe The
decrease non-linear
in
non-linear force force
density
force density
relative
density to themethod
method others a elon-
gation of that element. The decrease in the force density, even with negative
values,
Multiple canequilibrium
Multiple also beshapes
equilibrium applied tobe
shapes
can theobtained
can structure.
be obtained
with with the linear
the linear force force density
density method.
method.
3.3.2
However,
However,
The non-linear force density method
thesethese shapes
shapes may may be unsatisfactory
be unsatisfactory from from a structural
a structural pointpoint of view.
of view. For For
example,
example, the meshthe mesh
may may be irregular
be irregular and force
and the the force distribution
distribution unsmooth.
unsmooth. There-
There-
Multiple
fore, fore, equilibrium
it is necessary
it is necessary shapes
to afind
to find can be obtained
a configuration
configuration with
whichwhich the
is in is linear force
in equilibrium
equilibrium density method.
and which
and which also also
However,
satisfies
satisfies these shapes
additional
additional may
conditions.
conditions. be unsatisfactory
TheseThese from
conditions
conditions a structural
are generally
are generally point of
non-linear
non-linear view.
and soandis soFor
is
example,
the extended
the extended the mesh
force force may
density
density be
method. irregular
method. and the force
It is preferred
It is preferred distribution
to start
to start unsmooth.
the non-linear
the non-linear There-
computa-
computa-
tionsfore,
usingitusing
tions is necessary
the the shape
shape tofound
found find
withawith
configuration
the the linear
linear which Inis contrast
method.
method. inInequilibrium
contrast
to thetoand
the which also
non-linear
non-linear
satisfies additional conditions. These conditions are generally non-linear and so is
the extended force density method. It is preferred to start the non-linear computa-
30 form-finding of cable structures with force density method

3.2.2 Cable Net with a Upper Rigid Ring and Lower Anchorage Points along a Square
Base

This example is a cable net with simple topology, i. e. 4 cable element converging
at each internal node, and all the nodes at the base and those of the rigid upper
ring are fixed (Figure 3.4). The cable net is consist with m radial cable elements
and 4n circumferential cable elements. Here by "cable element", it means a por-
tion of cable limited by two nodes. This example is analyzed by changing the
force density ratio between radial and circumferential cable elements via MAT-
LAB codes generated from the Force Density Method discussed previously.

-1

-2

-3

-4

-5
-6 -4 -2 0 2 4 6

Figure 3.4: Topology of the cable net. Here m = 4 and n = 4.

This example is parametric, i. e. one possibility is changing the number of


radial or circumferential cable elements (m, n) under the given topology. In this
case, only one fixed set (m = 4, n = 4) is studied.
Figure 3.5 shows the final equilibrium shapes of the cable net. The results
illustrate that a change of force density ratio can alter the final configuration,
which is coincided with the previous example. When enlarging the force den-
sity in the circumferential cables, the structure looks tighter; while if the force
density in radial cables increases, the structure behaves more straight or linear.

3.3 summary

The Force Density Method can be used to generate solutions of discrete net-
works, through linear systems of equations, which are in an exact state of equi-
3.3 summary 31

10

15
4

10
2

5
-2

0 -4
10
-6
5 10
5
0 -8
0
-5
-5
-10
-10 -10 -10 -8 -6 -4 -2 0 2 4 6 8 10

(a) q = 1 for all elements (3D view) (b) q = 1 for all elements (plan view)

10

15
4

2
10

5
-2

-4
0
10
-6
5 10
5
0 -8
0
-5
-5
-10
-10 -10 -10 -8 -6 -4 -2 0 2 4 6 8 10

(c) q = 1 for radial elements and q = (d) q = 1 for radial elements and q =
5 for the circumferential elements 5 for the circumferential elements
(3D view) (plan view)
10

15
4

10 2

5
-2

0 -4
10
-6
5 10
5
0 -8
0
-5
-5
-10
-10 -10 -10 -8 -6 -4 -2 0 2 4 6 8 10

(e) q = 5 for radial elements and q = (f) q = 5 for radial elements and q =
1 for the circumferential elements 1 for the circumferential elements
(3D view) (plan view)

Figure 3.5: Different equilibrium configurations of the cable net.


32 form-finding of cable structures with force density method

librium without requiring iterations and thus some kind of convergence crite-
rion. It is able to be generated quickly and easy to solve for a given topology, by
varying the force densities and external loads. Since the Force Density Method is
independent from material properties, there are two interesting consequences:

• First, resulting designs can be materialized arbitrarily, given the initial


lengths of the cable net in undeformed configuration and this operation
does not change the final equilibrium configuration.

• Second, the loads can be simply multiplied to any realistic value, and then
the internal force distribution can be calculated; again this modification
does not affect the geometry.

The Force Density Method may be summarized as follows (Tibert, 1999). The
variables specified by the designer are:

• structural topology, and

• kinematic boundary conditions.

• force density prescribed for each element.

The problem unknowns are:

• equilibrium configuration, and

• internal force distribution.


FORM-FINDING OF CABLE STRUCTURES WITH DYNAMIC
R E L A X AT I O N
4
This chapter describes the fundamental theory of Dynamic Relaxation as ap-
plied to structures using finite elements. The method is described in detail
and recent developments is discussed. The use of Dynamic Relaxation for form-
finding and analysis of cable structures is described.
Dynamic Relaxation is a step by step method for tracing the motion of a
structure from the time of loading to when it reaches a position of equilibrium
arising from the effects of damping. Since the method of Dynamic Relaxation is
for static analysis in which the motion of degree of freedom is not concerned,
the mass at each degree of freedom is fictitious. The analyst will however be
concerned that the path to the solution is as rapid and efficient as possible.
Dynamic Relaxation technique was developed in 1965 by A.S. Day for the fi-
nite differential analysis of concrete pressure vessels (Topping and Iványi, 2008).
It is a direct or vector method of nonlinear static structural analysis. The concept
of the method was known much earlier by Otter, Cassell, and Hobbs (1966) with
the technique by making velocities vanish, the solution of static problems will
be extracted by vibration analysis. According to Veenendaal and Block (2012),
Dynamic Relaxation is favourable in the case of form-finding for cable networks
compared to stiffness matrix methods. The important features of this method
when used with a finite element idealisation are:

• The method is not for dynamic problems but by using a fictitious damped
dynamic analysis, the solution of a static problem is determined.

• The method does not utilise an assembled structural stiffness matrix and
hence it is particularly suitable for highly nonlinear problems, i. e. geomet-
ric nonlinear problems.

• The method is always expressed in terms of the current coordinates of


the structure hence the method automatically permits analysis of the large
displacements in geometric nonlinear problems.

Since 1977, the method has been extensively developed by Barnes (1977) in
his PhD thesis for the analysis and design of tension structures. The method
is particularly easy to program as the stiffness matrix need not to be formed.
The method of Dynamic Relaxation is a direct application of Newton’s second
law (equation (4.1)), but it is a static analysis method where the motion should
be as rapid and efficient as possible, therefore, suitable fictitious masses can be
assigned to each node.

F = Ma = M v̇ (4.1)

33
34 form-finding of cable structures with dynamic relaxation

In certain cases, for the most rapid path to determining the solution, different
masses may be assigned in each of the three coordinates directions. The method
may be described as pseudo-dynamincs.
It is best to use simple finite elements with Dynamic Relaxation rather than
fewer complex elements. The element types usually implemented with this
method are:

• For form-finding
a Truss elements of constant force or force density
b Cable elements of constant tension force
c Beam elements of constant moment
d Plane stress/strain triangular elements of constant stress with possible
warp and weft specification
e Geodesic strings for control of seam lines

• For analysis
a Truss elements
b Cable elements (i. e. no compression elements)
c Beam elements
d Plane stress/strain constant triangular elements
e Constant moment plane elements

4.1 basic theory of dynamic relaxation

The following contents are adopted from the papers by Lewis, Jones, and Rush-
ton (1984), Barnes (1999) and his PhD thesis Barnes (1977), the books by Topping
and Iványi (2008), Adriaenssens et al. (2014) and Lewis (2003).
The basic equation for Dynamic Relaxation is Equation (4.2). This equation
t in the
expresses the scene for any time t the out of balance or residual force Rix
x coordinate direction at joint or node i.
t t t
Rix = Mix v̇ix + Cix vix (4.2)

where Mix and Cix are the fictitious mass and viscous damping factor at joint i
in the x coordinate direction, and vixt and v̇t are the velocity and acceleration at
ix
time t for the joint i in the x coordinate direction.
It is noted that the viscous damping term Cix vixt is proportional to the velocity,

but in an opposite direction.

4.1.1 Velocity Tracing

The analysis traces the behaviour of the structures at a series of points in time t,
t + ∆t, t + 2∆t, t + 3∆t,...etc. Over any time step ∆t, the velocity is assumed to
4.1 basic theory of dynamic relaxation 35

vary linearly with time t. Hence the average velocity over time step ∆t is given
by:
t+∆t/2 t−∆t/2
t vix + vix
vix = (4.3)
2
and the acceleration is assumed constant over the time step hence
t+∆t/2 t−∆t/2
t vix − vix
v̇ix = (4.4)
2
Substituting the terms for v and v̇ from Equation (4.3) and Equation (4.4) into
Equation (4.2), the result is
Mix t+∆t/2 t−∆t/2 C t+∆t/2 t−∆t/2
t
Rix = (vix − vix ) + ix (vix + vix ) (4.5)
∆t ∆t
which shows that the integration scheme is overlapping because the residuals
are calculated at the end of each time step and the velocities and accelerations
are calculated at the half time step.
The rearrangement of Equation (4.5) enables the calculation of the velocities at
the new time step (t + ∆t/2) from the those of the previous time step (t − ∆t/2).

Mix /∆t − Cix /2


   
t+∆t/2 t−∆t/2 t 1
vix = vix + Rix (4.6)
Mix /∆t + Cix /2 Mix /∆t + Cix /2

4.1.2 Current Coordinates and Displacements

Using Equation (4.6) the displacement of joint i in the x direction during time
interval from t to (t + ∆t) is given by:
t+∆t/2
∆xit+∆t = ∆tvix (4.7)

and the current coordinates may be expressed as:

xit+∆t = xit + ∆xit+∆t (4.8)

Similar equations can be written for the y and z coordinates.

4.1.3 Calculation of Residuals

Once the current coordinates have been determined by using Equation (4.8),
the internal forces (Ti ) should be determined for each node. The calculation
of internal forces is discussed in Section 4.4. The internal forces are calculated
at the the joints where the residuals are determined. The contribution of each
element connected to joint i is summed with the applied loading Fi to give the
residual force at time (t + ∆t). For example in the x direction,
 t+∆t
t+∆t T
Rix = Fix + ∑ ( x j − xi )tm+∆t (4.9)
m l m
36 form-finding of cable structures with dynamic relaxation

where m represents the indices of all elements connected to joint i. Similar equa-
tions may be written for the y and z directions. Note that the current geometry
should be used to calculate the components of the residual forces since the struc-
ture is geometrically nonlinear.

4.1.4 Boundary Conditions or Supports

As it is suggested by Bagrianski and Halpern (2014), boundary conditions may


be imposed by setting zero residual forces or velocities explicitly in the direc-
tions of fixity at the fixed nodes. One alternative way, which is used in this
chapter, is assigning large masses to fixed nodes. Hence the assigned mass for a
fixed degree of freedom x at joint i is shown in Equation (4.10).
Mix = 1048 (4.10)

4.2 iteration scheme

The method of dynamic relaxation is an iterative algorithm which consists of


two main procedures:
• Calculation of the out of balance forces at each node of the structure us-
ing Equation (4.9). The residuals are initially equal to the applied loading
unless there is some prestress in the structure.
• Calculation of the nodal velocities using Equation (4.6) then using Equa-
tion (4.8) for updating the coordinates of the nodes.
The general cycle of the iterative process is shown below:
1 Set initial conditions, and zero kinetic energy;
2 For each element, determine residuals from stresses and loads;
3 For each node, determine velocities and update coordinates;
4 Determine current kinetic energy;
5 If the kinetic energy greater than a specified tolerance, then goto step 2; else
6 End the loop.
The convergence of the method is monitored through the level of kinetic energy
of the structure. The kinetic energy Uk is calculated by:
n m
Uk = ∑ ∑ Mij v2ij (4.11)
i j

where n is the number of nodes and m is the number of dimensions. Conver-


gence is generally assumed when the kinetic energy is smaller than a threshold
(Uk < Uε ) and the residuals of all the degrees of freedom are less than a specific
tolerance. This is achieved when an appropriate viscous damping factor is used
or kinetic damping scheme is employed.
4.3 stability of the method 37

4.2.1 Initial Conditions

To ensure the initial conditions (v0ix = 0 and R0ix = Fix ) the velocity at time ∆t/2
must be given by:
∆t/2 −∆t/2
vix = −vix (4.12)

This formula may be confirmed by defining an imaginary velocity at time −∆t/2


∆t/2
that is equal in magnitude but of opposite sign to vix such that the velocity at
t = 0 must be zero since the velocity is assumed to vary linearly with time.

Vix

-∆t/2 t
∆t/2

-Vix

Figure 4.1: Initial conditions: velocity trace at t = 0.

Figure 4.1 illustrates this concept. For time t = ∆t/2, Equation (4.6) can be
rearranged as follows:
∆t/2 −∆t/2 t
vix ( Mix /∆t + Cix /2) = vix ( Mix /∆t − Cix /2) + Rix (4.13)

Noting that the residual a time t = 0 is equal to the applied loading Fi and
−∆t/2
substituting for vix from Equation (4.12) gives an expression for the initial
velocity at time t = ∆t/2,

∆t/2 ∆t
vix = F (4.14)
2Mix ix
where Fix is the initial external force at joint i in the x direction.

4.3 stability of the method

The controlling parameters for the stability of the method are:


• the nodal mass components which may be fictitious;

• the method of damping and the associated damping factor, and


-∆t/2 t
38 ∆t/2
form-finding of cable structures with dynamic relaxation

• the time interval or step. -Vix


Usually the time step is a fixed value and the other two factors are varies until
stability of the iteration is achieved.

4.3.1 Fictitious Masses

If the time step ∆t is too large or the masses are too small then instability of
the iteration may occur and the analysis will not converge to an equilibrium
state. Generally convergence may be achieved by reducing the time interval or
increasing the fictitious masses. It has been shown by Barnes (1999) that for any
∆t convergence usually may be assured by using fictitious masses define by the
following equation:
∆t2
Mix = S + (a term) (4.15)
2 ix
where Six is the largest direct stiffness of the ith joint in the x direction. The
proof of Equation (4.15) can be derived from a simple system, where the nodes
are in line connected by truss elements which is shown in Figure 4.2. The proof
given here was developed using the original study by Barnes (1999) as a basis.
The greatest, primary stiffness falls in the direction of the line. Masses are as-

g h i j k

Figure 4.2: System of nodal masses in a line.

signed to the nodes and it is assumed that they move only along the line. This
arrangement reflects the most critical condition. In this undamped system, us-
ing Equation (4.6), the velocity at node i at time step (t + ∆t/2) can be written
as:
∆t t
vit+∆t/2 = vit−∆t/2 + R (4.16)
Mi i
For the next time interval the velocity can be calculated by:
∆t t
vit+3∆t/2 = vit+∆t/2 + Ri − Sij ∆t(vi − v j )t+∆t/2

(4.17)
Mi
where S is the stiffness in the form of EA/lij and ∆t(vi − v j )t+∆t/2 is the elon-
gation of the truss between joint i and j. The bracketed term is therefore equal
to Rit+∆t . By substituting Equation (4.16) into Equation (4.17), Rit term can be
eliminated and the result is:
∆t2
−vit+3∆t/2 + 2vit+∆t/2 − vit−∆t/2 = S (v − v j )t+∆t/2 (4.18)
Mi ij i
4.3 stability of the method 39

A similar equation can be written for node j. Subtraction of the two equations
from each other will enable the velocity of node i relative to node j to be ex-
pressed:

2∆t2
−vijt+3∆t/2 + 2vijt+∆t/2 − vijt−∆t/2 = Sij vijt+∆t/2 (4.19)
Mi

Instability occurs when the relative velocity at the current time step is equal to
or greater than the relative at the previous time step in magnitude but it points
to an opposite direction. This can be expressed as:

vijt+3∆t/2 = −vijt+∆t/2 = vijt−∆t/2 (4.20)

Substituting this into Equation (4.19) then a bound on the nodal mass for stabil-
ity is given:

∆t2
Mi = S (4.21)
2 ij
This expression is only valid when the principal stiffness directions coincide
with the global coordinate system. Hence to ensure stability of the iteration the
express in Equation (4.15) may be used.
The stiffness mentioned above is expressed as Equation (4.22) for cable or
truss members,
 
Ei Ai Ti
Si = + t (4.22)
li li

where Ti is the tension in the cable member and lit is the current length of
the member. For cables subject to elastic straining only the elastic stiffness will
dominate but for cables where the tension is constant as in the case of form-
finding the geometric stiffness will condition the iteration.

4.3.2 Viscous Damping

The damping factor that causes the structure to approach the static position
most rapidly should be considered for the analysis. The factor is called the
critical damping factor. In Figure 4.3 (Adriaenssens et al., 2014) the trace for a
single degree of freedom (SDOF) problem with a number of different damping
factors are illustrated.
The underdamped oscillations is slow to reach convergence and since the
trace does not pass through the static solution, this kind of analysis does not
give any bounds on the accuracy of the analysis. The critically damped trace
represents the most efficient path to the solution. In multi-degree of freedom
(MDOF) problems the trace will not be ideal but it is still similar to that shown
in Figure 4.3.
The critical damping factor may be estimated by undertaking an undamped
run to obtain an estimate of the highest frequency and by using the expression
40 form-finding of cable structures with dynamic relaxation

Figure 4.3: SDOF time displacement trace. Effect of viscous damping for (a) under-
damped oscillations and (b) critically dampedoscillations.

derived for the critical damping factor for a SDOF problem, the ith degree of
freedom for example (Lewis, 2003). The critical damping factor is
p
Cic = 2 Si Mi (4.23)

The frequency is
s
1 Si 1
f = = (4.24)
2π Mi T

hence,
p p p p
Cic = 2 Si Mi = (2 f )(2π ) Mi Mi = 4π f Mi (4.25)

where the highest frequency obtained from a MDOF problem is used to estimate
the critical damping factor. Next a damping constant for the complete structure
is defined as:
 
Cic
k = ∆t (4.26)
Mi

The damping constant k can be applied for all joints in the structure and it
indicates that the ratio of damping force per unit mass is constant - such that
joints of larger mass will be more heavily damped. In this case Equation (4.6)
may be rewritten as:

t−∆t/2 1 − k/2 ∆t
    
t+∆t/2 t 1
vix = vix + Rix (4.27)
1 + k/2 Mix 1 + k/2

which is usually written as:


t+∆t/2 t−∆t/2 t
vix = Avix + Bix Rix (4.28)

where
1 − k/2 ∆t
 
1
A= Bix = (4.29)
1 + k/2 Mix 1 + k/2

This means that A is constant for the whole structure while Bix differs for each
joint and each coordinate direction.
4.3 stability of the method 41

Figure 4.4: A typical kinetic energy trace for a MDOF structure.

It is clear for a successful dynamic relaxation procedure the time step, ficti-
tious masses and viscous damping terms must be used. The viscous damping
term is usually determined by the use of an undamped trial analysis where the
frequency is used to estimate a suitable damping factor from Equation (4.25).
This undamped trial analysis is sometimes referred as the "trial run". This esti-
mation procedure and the trial run may be avoided by use of kinetic damping
which is discussed in the next section.

4.3.3 Kinetic Damping

There are many parameters to fix for the efficient solution of any problem. The
number of parameters may be reduced by the use of kinetic damping which
does not require the determination of a viscous damping trial run. Therefore
only the time step and the fictitious nodal masses are required. In this way the
time interval may be fixed and the masses estimated from Equation (4.15). In
the case of instability the masses can be increased by a term or alternatively the
time interval may be reduced.
Kinetic damping is an alternative to viscous damping that was suggested
by Cundall for application to unstable rock mechanics problem (Barnes, 1999).
According to Hüttner, Máca, and Fajman (2015), the method of kinetic damping
has been found to be very stable and rapidly convergent when dealing with
large displacements. In this case no damping factor is used hence,

∆t
A=1 B= (4.30)
Mix

and the kinetic energy of the complete structure is traced as the undamped os-
cillations proceed and all current nodal velocities are set to zero whenever an
energy peak is detected. For a linear elastic system oscillating in one mode, the
first kinetic energy peak achieved would represent the static equilibrium posi-
tion. For practical problems, however, the process must be continued through
further peaks, until the required degree of convergence is achieved. Figure 4.4
(Adriaenssens et al., 2014) shows the kinetic energy trace for a typical structure.
After detecting an energy peak, coordinates will have been projected to time
(t + ∆t). But the "true" kinematic energy peak will have occurred at some ear-
lier time t∗ as shown in Figure 4.5. To determine the coordinates at time t∗
42 form-finding of cable structures with dynamic relaxation

KE
KE peak
peak at t = t*

t a b
∆t/2 c

t t-∆t t-∆t/2 t
t-3∆t/2 t-∆t t-∆t/2 t t+∆t/2
Rt-∆t Rt
∆t*
xt-∆t vt-∆t/2 xt
Figure 4.5: A trace for typical kinetic energy peak.

a quadratic can be fitted through the current kinetic energy value (c) at time
(t + ∆t/2) and two previous kinetic energy value (a and b) in Figure 4.5.
j k
It is convenient for computation to keep records of the difference between the
previous and current kinetic energies. In terms of these differences ∆t∗ is given
by:

b−c
∆t∗ = ∆t = ∆tq (4.31)
(b − c) − ( a − b)
dis.
Since coordinates have been updated using average velocities (at mid-points
of time intervals), they should be reset according to the same scheme. Thus:
∗ t+∆t/2 t−∆t/2
xit = xit+∆t − ∆tvix − ∆t∗ vix (4.32)

Hence, using Equation (4.28), Equation (4.30), Equation (4.7) and Equation (4.31):

t
∗ t+∆t/2 ∆t2 Rix
xit = xit+∆t − ∆t(1 + q)vix + q KE (4.33)
2 Mi

An alternative is to assume the peak occurs at (t + ∆t/2) and thus q = 1/2


in Equation (4.33). This is the same as taking the velocities during the interval
(t − ∆t/2) → t and t → (t + ∆t/2) as constants equal to the mid-point values.

4.4 calculation of internal forces for trusses and cables

Once the current coordinates have been determined using Equation (4.7), the
new (extended) length my be calculated at time (t + ∆t). In that case the current
internal force in link element m, such as a truss or cable, can be determined as
fellows:
EAm t+∆t
Tmt+∆t = 0
0
( lm − lm ) + Tm0 (4.34)
lm
g h i j k
4.4 calculation of internal forces for trusses and cables 43

k
Tm

Tm

n i
y j
z
x

Figure 4.6: The link connecting joint i to joint k.

where lm 0 and l t+∆t is the initial and the current length of link at time ( t + ∆t ),
m
EAm is the elastic modulus multiplied by the cross sectional area of the link m,
and Tm0 is the initial prestress in link (if not specified by an initial slack length).
Pay attention that if the link element is a cable and Tmt+∆t < 0, which means
the force is compressive, then Tmt+∆t must be set equal to zero.
If the link connects joints i to k as it is shown in Figure 4.6, then force in the x
direction at joint i from linked element m is given by:
 t+∆t
t+∆t xk − xit+∆t

t+∆t
Tixm = Tm t+∆t
(4.35)
lm
Similarly the force at node k in the x direction is:
t+∆t t+∆t
Tkxm = − Tixm (4.36)
Note that current geometry need only be used to calculate the components
of forces if the structure is geometrically nonlinear. The contribution, in the
x direction, of each member connected to the ith joint are summed with the
applied loading Fix to give the residual force at time (t + ∆t):
t+∆t t+∆t
Rix = Fix + ∑ Tixm (4.37)
m
where the m is the number of elements at joint i. And similar equations can be
written for the y and z directions.
When dynamic relaxation is used to calculate the coordinate positions of cable
structures subject to prestress then Tmt+∆t is set equal to a prestress value Tm0 and
the coordinate positions of joints calculated. This is sometimes referred to as
form-finding since the geometry of the structure due to the prestress in each
member and under the action of dead load is determined. The final current
length of each member when the structure is in equilibrium as a result of the
prestress may be calculated as lm t+∆t using the current coordinates of the joints

at the end of the calculation. These tensioned lengths under the prestress may
used to determined the slack or initial lengths of the members.
44 form-finding of cable structures with dynamic relaxation

4.5 example: a hyperboloid structure

Using the MATLAB program, a Dynamic Relaxation code is implemented and


an example is presented. A cable net structure with a rectangular plan view
(two lower and two higher anchored points) is considered as an example. The
input data are performed by "Gmesh" as a graphical aid to MATLAB.
This structure is a hyperboloid structure, i. e. at each point the main curva-
ture is negative. The following figures show the different equilibrium shapes
by changing the presstress ratio between boundary and inner cables. Note that
the prestress in each boundary/inner cable is constant while the prestress in
boundary cables may be different from the one in inner cables.

1.3
3
1.2

2.5
1.1

1 2

0.9 1.5

0.8
1

0.7
0.5
0.6

0
0.5 1.2
1
0.8
0.4 0.6 0.2 0.4 0.6
-0.4 -0.2 0
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.4 -1 -0.8 -0.6
-1.2

(a) (b)

Figure 4.7: Initial (dotted) and equilibrium (solid) configuration when the presstress
ration is 2:3. (a) Plan view and (b) 3D view.

1.3
3

1.2

2.5
1.1

2
1

0.9 1.5

0.8
1

0.7
0.5

0.6

0
0.5 1.2
1
0.8
0.6 0.2 0.4 0.6
0.4 -0.4 -0.2 0
0.4 -1 -0.8 -0.6
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 -1.2

(a) (b)

Figure 4.8: Initial (dotted) and equilibrium (solid) configuration when the presstress
ration is 1:1. (a) Plan view and (b) 3D view.

Figure 4.7 to Figure 4.9 show that changing the prestress ratio between bound-
ary cables and inner cables can influence the final equilibrium configuration of
the structure. As the prestress in boundary cables increases, the equilibrium
4.5 example: a hyperboloid structure 45

1.3

3
1.2

2.5
1.1

1 2

0.9 1.5

0.8
1

0.7
0.5

0.6

0
0.5 1.2
1
0.8
0.6 0.2 0.4 0.6
0.4 -0.4 -0.2 0
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.4 -1 -0.8 -0.6
-1.2

(a) (b)

Figure 4.9: Initial (dotted) and equilibrium (solid) configuration when the presstress
ration is 2:1. (a) Plan view and (b) 3D view.

configuration for this structure looks stiffer; this phenomenon is consistent with
the description of Tibert (1999) and the conclusion of Section 3.2.2.
10

7
Kinetic Energy [kJ]

0
1000 2000 3000 4000 5000 6000 7000 8000
Iterations

Figure 4.10: Kinetic energy trace for the structure when the presstress ratio is 1:1.

Figure 4.10 illustrates the kinetic energy trace in the case with a constant
prestress in all the cables. After several rise and fall, the kinetic energy finally
converges to zero, and thus this structure meets the equilibrium configuration.
This behaviour is also consistent with the illustration in Figure 4.4.
46 form-finding of cable structures with dynamic relaxation

4.6 summary

The Dynamic Relaxation method provides a practical design tool for the anal-
ysis and design of structures. In this chapter, a Dynamic Relaxation method
is introduced to the general form-finding problems with cable nets. The exam-
ple shows the advantages of this method which suitable for highly nonlinear
problems. The initial convergence is fast which permits quick, economical pre-
liminary studies, while the required number of iterations are usually large until
convergence but the operations are extremely simple. Hence, one possible im-
provement for Dynamic Relaxation method is using an initial state which is
close to an admissible solution; this trick may reduce the number of iterations
and thus the code can be converged faster.
F O R M - F I N D I N G O F I N F L ATA B L E D A M S
5
Inflatable dams or rubber dams are membrane structures inflated with air and/or
water which are considered as attractive alternatives for water barriers. These
structures can be also regarded as the form-active structures because of the air-
supported property.
Since the mid-1950s, the first inflatable dam was constructed in Los Angeles
by an American engineer (Tam, 1997), over 2000 inflatable dams have been con-
structed by 2011 (Chu, Guo, and Yan, 2011). One of the largest in Europe so far is
the rubber dam used for the Ramspol storm surge barrier in Netherlands. The
barrier uses three identical inflatable rubber dams each of whose dimensions
are 75 m long, 13 m wide and a design height of 8.35 m. A typical inflatable
dams as shown in Figure 5.1 are usually made of polymer with nylon or aramid
reinforced(Figure 5.2); according to literature, the height of inflatable dams may
range from 3 to 8 meters while their length can reach 120 meters (Watson, Suher-
man, and Plaut, 1999). The thickness of the membrane can vary from 7 mm to
16 mm and the elastic modulus may fluctuate from 0.1 GPa to 2 GPa (Mysore,
Liapis, and Plaut (1998) and Parbery (1976)).

Figure 5.1: A typical inflatable dam located in Mombaldone, Piedmont, Italy.

Inflatable dams are used in various situations, such as creating groundwater


supply, recreational basins, preventing contamination, diverting water for irri-
gation, hydroelectricity, tidal or flood control, and raising the height of existing
dams to increase reservoir capacity. The two main ways to fill the rubber dams
are:

47
48 form-finding of inflatable dams

(a)

(b)

Figure 5.2: Typical composite structure of membrane (a) for inflatable dams and phys-
ical illustration of rubber sheet reinforced with aramid fabric (b) for the
Ramspol dams (Breukelen, 2013).

• Air filled rubber dam. This kind of dams are inflated by pumping air
inside the rubber body to form the shape of circle with designed height
and pressure. It can be placed across channels, streams to store water or
divert water for irrigation or other purposes. The advantages of air filled
dams are:
1 Quick to inflate and deflate by air blowers;
2 Lower cost;
3 Narrower concrete sill demand for installation;
4 Deflectors are available to reduce the vibrations.

• Water filled rubber dam. This kind of dam is a tear-shaped rubber dam
filled with stabilizing water. Compared to air filled rubber dams, normally
it is considered to be expensive and slow to fill but more stable to provide
optimal control over upstream water levels compared with air-filled ones.
The advantages of water filled dams are:
1 Stabilizing water-filling with heavier weight to minimize vibrations;
2 Ideal for applications with tail water elevation or broad watercourses.
This kind of dams are capable to withstand a higher overtopping up
to 50% of the dam height, while air fill dams can only capable as 20%
as its height;
3 Lower maintenance cost;
4 Optional water circulation system to eliminate the possibility of freezing
in winter.
5.1 literature review 49

5.1 literature review

Inflatable dams were invented six decades ago, however, predicting the equi-
librium shape of inflatable dams is still challenging. Previous studies of the
behaviour and analytical solutions of the inflatable dams are usually based on
material idealization and shape approximation. Anwar (1967) approximated the
hydro-static equilibrium profile, the free downstream membrane portion is ana-
lyzed based on the equation of a circle, and for the loaded upstream membrane
elliptic integrals is used. Binnie (1973) developed a closed form solution for wa-
ter filled dams accounting for physical restrictions of the base and perimeter
lengths. However, both solutions are based on an idealized weightless, inexten-
sible membrane as well as linked to specific load cases.
Since inflatable dams are force-modeled structures, their equilibrium shape
must be determined by either experimental or numerical form-finding meth-
ods. Streeter, Rhode-Barbarigos, and Adriaenssens (2015) introduced one possi-
ble form-finding algorithm based on Dynamic Relaxation method, where they
neglect the bending stiffness of the inflatable membrane. Meanwhile, Watson,
Suherman, and Plaut (1999) solved this form-finding problem using a combi-
nation of quasi-Newton methods and globally convergent homotopy methods,
while they assume the bending stiffness of the membrane is not negligible.

5.1.1 Dynamic Relaxation

Streeter, Rhode-Barbarigos, and Adriaenssens (2015) have introduced Dynamic


Relaxation for form-finding of inflatable dams. This is a kind of 2D form-finding
problem similar to the one which characterizes cable-net systems or tensile mem-
brane structures. The Dynamic Relaxation is employed for the cross-sectional
analysis under this circumstance. Therefore, only tensile elements exist. Further-
more, "kinetic" damping is used to obtain static equilibrium by this paper. As
it has been mentioned at Section 4.3.3, the motion of the structure is traced
and when a local peak in the total kinetic energy of the system is detected, all
velocity components are set to zero. The process is then restarted from the cur-
rent geometry and repeated until the energy of all modes of vibration has been
dissipated and a static equilibrium has been achieved.

initial profile
pressure In
hu inside 1 iteration pr
Fu0 Fu1

hc0
hc1

Figure 5.3: Illustration of the loading-shape interaction for an inflatable dam.

2 int. ext. 2 1 ext. int. 1


50 form-finding of inflatable dams

The basic scheme for Dynamic Relaxation established by Streeter, Rhode-


Barbarigos, and Adriaenssens (2015) follows the same as it has been employed
in Chapter 4. However, a "force update" procedure is introduced as well. Since
changes in the form of the membrane can also induce variations in the loading
and vice versa, the shape of an inflatable membrane dam and the applied hy-
drostatic loading are coupled. The applied loading includes both the hydrostatic
force and the internal pressure (originating inside the membrane). Additionally,
the circular initial cross-sectional profile of the dam is adopted for simplicity, de-
fined by the total membrane length and an anchorage-point separation. Figure
5.3 illustrates the loading-shape interaction for an inflatable dam, where hu is
the upstream height, hc0 and hc1 correspond to the initial height and the height
after the first iteration for element 1 of the membrane and Fu0 and Fu1 are the
corresponding resulting forces on the same element (Streeter, Rhode-Barbarigos,
and Adriaenssens, 2015).
The force update extension is based on an equilibrium shape which is identi-
fied by the Dynamic Relaxation from an initial dam cross-sectional profile and
its corresponding hydrostatic loading and internal pressure. Once the first the
equilibrium shape is calculated, the profile of the hydrostatic loading is reevalu-
ated, and if necessary, updated. The equilibrium shape along with its updated
loading are adopted as inputs for the next form-finding and analysis iteration.
The procedure continues until there is no significant change in the loading and
the equilibrium cross-sectional profile of the dam. Figure 5.4 illustrates the this
extension in the Dynamic Relaxation algorithm (Streeter, Rhode-Barbarigos, and
Adriaenssens, 2015).

Force estimate/update
Initial Equilibrium
profile Form-finding/analysis profile
using DR method

Figure 5.4: Illustration of the force-update extension in dynamic relaxation algorithm.

Another modification compared to a general Dynamic Relaxation scheme is


the position of applied loads. In a classic Dynamic Relaxation scheme, the mem-
branous cross-sectional profile is composed of uniaxial tensile elements, and
forces are applied on the nodes of the elements. However, for the analysis of
inflatable dams, forces are estimated applied on the centroid of the elements.
5.1 literature review 51

The internal or external hydrostatic force applied on an element is estimated


initial profile
based on the difference in elevation between the centroid of the
pressure element and
the hhydrostatic
u
inside 1 iteration
head multiplied by the specific weight of water. The resulting
force is assumed as Fu0 a uniform
Fu1 pressure along the length of the element and
thus evenly distributed between the two nodes of the element. This processing
hc0
strategy may also be applied to airhc1pressures (if any), which are also evenly dis-
tributed to the nodes of the element. It is important to keep track of the sign
of each force with respect to the global axes since elements can have four pos-
sible orientations as they are shown in Figure 5.5 (Streeter, Rhode-Barbarigos,
and Adriaenssens, 2015). Once the nodal forces have been found, dynamic relax-
ation is applied along with the corresponding dam profile, and an equilibrium
profile is identified.

2 int. ext. 2 1 ext. int. 1

ext. 1 1 int. int. 2 2 ext.

Orientation 1 Orientation 2 Orientation 3 Orientation 4

Figure 5.5: Possible element orientations in the cross-sectional profile of the dam: 1. up-
stream with negative slope, 2. upstream with positive slope, 3. downstream
with negative slope and 4. downstream with positive slope.

As for boundary conditions, the nodes at boundary are explored for the base
introduction, which means nodes on the cross-sectional profile of the dam in
contact with the ground are identified based on their coordinates and defined
as boundary nodes.
The paper by Streeter, Rhode-Barbarigos, and Adriaenssens (2015) presents a
computationally advantageous tool for the form-finding problems of inflatable
dams. However, there may be some disadvantages, for instance, the force update
procedure for calculating the residual forces is comparably unfriendly to gener-
ate computer codes since one may keep focus on the position of elements where
both external and internal pressures applied. The introduction of boundary con-
ditions is also not easy to achieve for single anchorage system as the contact
between membrane and base may change step-by-step during iterations.

5.1.2 Elastica Analysis

Watson, Suherman, and Plaut (1999) presented an alternative way to compute


the equilibrium shape of inflatable dams. In this paper, several types of inflatable
dams (with bending stiffness) have been studied (Figure 5.6). For the folded
dam, the material is unstrained when it is horizontal, and one end is lifted,
folded back, and clamped to the other end and the foundation (point C in Figure
5.6(a)) This problem is related to the folding of flexible strips or sheets. For the
dam with fin, it is to be treated as a thick sheet of material is again unstrained
Fig[ 3[ In~uence of internal pressure on height of dam with no external water[
0289 L[ T[ Watson et al[ : International Journal of Solids and Structures 25 "0888# 0272 0287

52 form-finding of inflatable dams


nondimensional internal pressures p 0999\ 1999\ and 2999[ Naturally\ the contact length of the
elastica with the foundation decreases as p increases[ The nondimensional dam height hd is plotted
as a function of p in Fig[ 3[ When the internal pressure is zero\ the similarity solution given by
Wang "0870\ 0876# when it isfrom
is applicable\ flat.which
It ishslit
d intosCtwo9[685\
9[193\ sheets
f"9# except
08[8\ andnear
`"9# one
32[6[edge. The opposite edge is
clamped, and pressure is applied between the two sheets. The edge that was not
1[1[ With external water
slit acts as a fin, which is useful in cases of overflow. For the dams with spillway
Now consider thegate,
foldedadamtypewithof
external water ofgate
spillway Hw and speci_c
heightraised by anweight G on the bladder is examined, and
inflatable
upstream side\ as depicted in Fig[ 4[ In this case the origin of the coordinate system is placed at
point A\ located on the bladder
the dam at 0283 may
the water be folded
surface\ sinceetthe
L[ T[ Watson
back orequations
al[ : governing
may
International Journal
have
areand
of Solids
a fin
di}erent
Structures
as described above. In this
over
25 "0888# 0272 0287
segments CA and AB[ Also\ instead
chapter, of using
only F anddams
folded G\ the tension per unit
are concerned. width T and shear force
Fig[ 7[ In~uence of external water height on height of folded dam for g 29\999 and p 2999[

Fig[ 03[ In~uence of external water height on height of dam with _n for g 29\999 and p 2999[
Fig[ 8[ Geometry of dam with _n[
Fig[ 4[ Geometry of folded
(a)dam with external water[ (b)

decrease the height hd of the dam[ The variation of hd with the water height hw is presented in Fig[
7 for the same values of g and p as used in Fig[ 6[

2[ Dam with _n

2[0[ Without external water

The pressurized cross section of this type of dam is sketched in Fig[ 8\ in the absence of external
water[ The _n at D is at the midpoint of the perimeter from the clamped end C\ and the origin is
taken at the lift!o} point B[ Equations "3a c# and "7a\b# govern on segment BD "9 ⇡ s ⇡ sD#

Fig[ 04[ Geometry of spillway gate without _n[


(c)

Figure 5.6: Geometry of different types of dams. (a) Folded dam, (b) dam with fin and
From A to B in Fig[ 04\ the governing equations are "3a f# except that p is replaced by ⌧p[ For
(c) spillway
simpli_cation gate"3a\b#
in this case\ without (with)
are used fin.
in "3d\e# and the resulting equations are integrated to
give
0277 L[ T[ Watson et al[ : International Journal of Solids and Structures 25 "0888# 0272 0287
f fA py\ ` `A px[ "00a\b#
Then the equations to be used are
dx dy du dm
cos u\ sin u\ m\ " fA py# sin u "`A px# cos u[ "01a d#
ds ds ds ds
At A "s 9#\ x y m 9[

Fig[ 5[ Element of elastica with tangential and normal force components[


Figure 5.7: Elastica element with tangential and normal force components.

This method is based on a set of equilibrium equations directly written on an


per unit width
elastica V areas
element utilized\ as de_ned
a portion of in themembrane,
the free body diagram of an element
as shown in Fig[ 5[5.7
in Figure Equilibrium
(Watson,
provides the relations
Suherman, and Plaut, 1999). Consider a folded dam as in Figure 5.6(a) water of
Hw anddT
heightdM specificdu dV Γ on
weight du the upstream side. The coordinate origin is
V\ ⌧V \ T P[ "4a\b\c#
dS dS dS dS dS

With the use of "0c# and "4a#\ one can integrate "4b# and obtain

0 M1
T T9 ⌧ \ "5#
1 D
5.1 literature review 53

chosen at point A, located on the dam at the surface between water and air, since
the governing equations are different over segments CA and AB. The tension
per unit width T and shear force per unit width V are utilized, as defined in
the free body diagram of an element in Figure 5.7. Under this terminology, equi-
librium equations can be written and the governing equations from A (where
s = 0) to B (where s = s B > 0) are given:

dm
=v (5.1)
ds

dv 1
= t0 m − m3 + p (5.2)
ds 2
where the non-dimensional quantities are used along with:

L2 V L2 T L2 T0 Hw L4 Γ
v= , t= , t0 = , hw = , γ= (5.3)
D D D L D
and
1 M2 1 Eh3
T = T0 − , D= (5.4)
2 D 12 (1 − ν2 )

From C (where s = sC < 0) to A, the equations are the same except that p is
replaced by p + γy (where y < 0) in Equation (5.2). At s = 0, x = y = 0. Hence,
the unknown for this case are:

z = ( θ (0) , m (0) , v (0) , t c , s B , s C ) (5.5)

and the nonlinear system to be solved for z is:


 
y(s B ; z) + hw
 

 θ (s B ; z) + π 

m(s B ; z)
 
J (z) =  =0 (5.6)
 

 y (sC ; z ) + hw 

θ (sC ; z )
 
 
x (s B ; z ) − x (sC ; z ) + s B − sC − 1

This nonlinear system is solved by a combination of quasi-Newton meth-


ods and homotopy methods. The method presented by Watson, Suherman, and
Plaut (1999) is easy to evaluate as it is simply equilibrium equations written at a
portion of membrane, but it is not efficient since it is very sensitive to choose the
unknowns and initial conditions. Due to the highly nonlinear governing equa-
tions, despite the advantages of collocation over shooting, a quintic spline collo-
cation formulation was not successful in this case. Both quasi-Newton methods
and homotopy methods failed on the spline collocation formulation. In general,
this kind of method is still challenging in analysis of inflatable dams. Simple
shooting and quintic spline collocation do not always succeed, and multiple
hc0
hc1
54 form-finding of inflatable dams

shooting generally is even less robust than collocation so it should not be tried
at all.
As one can seen the advantages and disadvantages of previous studies, a dis-
cretized equilibrium method will be introduced in following sections, which is
more robust, efficient and easy-generated compared among the previous meth-
ods.
2 int. ext. 2 1 ext. int. 1
5.2 basic theory

In this section, basic equations of the discretized equilibrium method will be


proposed.
ext. This1method is 1an upgraded int.version of Harrison
int. (1970),
2 where they
2 ext.
did the same discretization as what will be dealt with in this chapter but dif-
ferent numerical
Orientation 1 procedure. Harrison 2(1970) uses aOrientation
Orientation trial and error
3 procedure
Orientation 4
starting from the first node to the last node, so unless the guessed shape is cor-
rect, the solution would not satisfy equilibrium at the last node. In this chapter,
Newton-Raphson method is employed to solve the nonlinear problem.
This method is carried out on the basis of two assumptions:

• It is assumed that the behaviour of the 3D membrane structure can be


represented by behaviour as a 2D cross section of unit width;

• The perimeter of the section may be considered as a finite number of small


elements which would be treated as rigid bodies or extensible straight
elements.

y
gj+1

i j
3
fj+1
2

(a)
1

f1 pi
gi+1
g1 pi
i fi+1

fi
(b)
gi

x
Figure 5.8: Illustration of a generalized element chain, sign convention and terminol-
ogy.
5.2 basic theory 55

With these assumptions, for a generic element chain in Figure 5.8, it will have
an analogous behaviour as a cable system. In a continuous cable system, the
tension in the cable will satisfy the following equation (Finzi, 1976):
dT
P+ =0 (5.7)
ds
where P is the external load for the cable, T is the tension in the cable and ds
the local coordinate (Figure 5.9).

fA fB

A B

T(s) T(s + ds)


ds

C C’

Pds

Figure 5.9: Notations for a generic cable system.

Since P is a vector with a generic direction, it can be decomposed into tangent


component Pt and normal component Pn , both of which are scalars. Then the
Equation (5.7) can be rewritten as:
dT T
Pt + = 0, Pn + =0 (5.8)
ds r
where r is the curvature of the small portion ds.
The second equation in Equation (5.8) presents that a constant membrane
tension along the whole cross section in the case of rubber dams, since both the
hydrostatic pressure and air pressure are perpendicular to the membrane. This
fact can be examined in later examples.
Figure 5.8 shows a chain of elements and its unknown forces. Note that the
applied pressures are not drawn in Figure 5.8 (a) for simplification but they
are actually acting on this element chain as in Figure 5.8 (b). In those figures,
f i and gi are horizontal and vertical nodal forces respectively, pi is the applied
pressure which is positive when applied towards the right side with respect to
the element axis.
With a discretization of m elements, the cross section profile has (m + 1) nodes.
The number of unknowns for this profile is consist of two contributions: one is
from the (m − 1) internal nodes and the other from the 2 boundary nodes. In
each internal node, the nodal coordinates ( xi , yi ) and vertical force gi (while
horizontal force f i could be computed by equilibrium) are unknown; in the
boundary nodes, the coordinates are fixed, unknown variables are nodal forces
f 1 , g1 and gm+1 . Finally, the total number of unknowns for this problem is 3(m −
1) + 3 = 3m. To solve these unknowns, 3m equations are required. However, this
56 form-finding of inflatable dams

demand can be achieved by the 2 equilibrium equations (vertical and rotational)


for each element chain and 1 geometric constraint each single element.
The above equations constitute an equation system or residuals R3m×1 , where
m is the total number of elements. The unknown forces and coordinates form a
vector of unknown X3m×1 , the problem can be solved by incremental nonlinear
methods such as Newton-Raphson method.
Equation (5.9) represents the governing equations, then it can be expanded
by Taylor expansion at a neighborhood of an approximated solution (or initial
guess) Xk as it is shown in Equation (5.10).

R(X) = 0 (5.9)

 
∂R
R L (X) = R(Xk ) + (X − Xk ) = 0 (5.10)
∂X Xk

where R L is the linearized term of the nonlinear term R.


Equation (5.10) can be written as:
 
∂R
∆X = −R(Xk ) (5.11)
∂X X
k

where ∆X = (X − Xk ), and it can be solved by:


  −1
∂R
∆X = − R(Xk ) (5.12)
∂X Xk

Then a new set of unknowns can be reached:


  −1
∂R
Xk+1 = Xk + ∆X = Xk − R(Xk ) (5.13)
∂X X
k

The partial derivatives in Equation (5.10) to Equation (5.12) present the Jacobian
matrix which can be denoted by Equation (5.14) in compact notation.
 
∂R
J(Xk ) = (5.14)
∂X X
k

It is noted again that the equilibrium equations are written at the element
chains instead of single elements (but the geometric constraints are written at
single elements), since the latter scheme will make the condition number of
the Jacobian matrix significantly large and thus leads to singularity in terms of
numerical computation.

5.2.1 Non-Dimensional Quantities

As the residual vector R contains equilibrium equations for forces and mo-
ment, as well as geometric equations, it is very important to introduce non-
dimensional quantities by rewriting theses equations in a non-dimensional form.
5.2 basic theory 57

Denoting with hat the original variables, the non-dimensional quantities are
defined without hat. Let ( x̂i , ŷi ) be the coordinates for ith node, and fˆi , ĝi be the
nodal forces; p̂i be the applied pressure and lˆi be the length for a single element.
The total length of the cross-sectional profile is donated to be the reference
length L and atmosphere pressure p0 to be the reference pressure. Then the non-
dimensional parameters which are presented in below sections can be defined
as:

lˆi 1 x̂ ŷ fˆi ĝ p̂
li = = , xi = i , yi = i , fi = , gi = i , p i = i (5.15)
L m L L p0 L p0 L p0

As for the applied pressures, according to the sign convention in Figure 5.8,
the resultant pressure for a element i suffered by both external hydrostatic pres-
sure ph and internal air pressure pint can be written as in Equation (5.16), which
will be discussed in Section 5.3

p̂i ( p + ph ) − ( p0 + pint )
pi = = 0 (5.16)
p0 p0

5.2.2 Governing Equations

The governing equations are:


 
R1 ( X )
 
 R2 ( X ) 
 
 R (X) 
 3 

 .
..


 
 R3i−2 (X) 
 
 
R(X) =  R
 3i−1 ( X ) 
 =0 (5.17)
 R3i (X) 
 
 .. 
.
 
 
 
 R3m−2 (X)
 
 R3m−1 (X)
 

R3m (X) (3m×1)


58 form-finding of inflatable dams

where m is the total number of elements and X is global vector of unknowns:


 
f1
 g1 
 
 
 
 x2 
 
 y2 
 
 
 g 
 2 
 .. 
 . 
 
 xi 
 
X=  (5.18)
 yi 
 
 gi 
 
 . 
 .. 
 
 
 xm 
 
 y 
 m 
 
 gm 
 
g m +1 (3m×1)

In each subvector of governing equation system, R3i−2 , R3i−1 and R3i repre-
sent the vertical equilibrium, rotational equilibrium and geometric constraint
respectively. As it is shown in Figure 5.8, for a element chain (1 − j), the govern-
ing equations can be written as:
j
R3j−2 = Vj = ∑ Vji + gj+1 − g1 = 0 (5.19)
i =1

j
R3j−1 = M j = ∑ Mji + g1 (x j+1 − x1 ) − f1 (y j+1 − y1 ) = 0 (5.20)
i =1

R3j = Cj = m2 ( x j+1 − x j )2 + m2 (y j+1 − y j )2 − 1 = 0 (5.21)

where m in Equation (5.21) is the total number of elements, Vji in Equation (5.19)
and M ji in Equation (5.20) are the vertical and rotational equilibrium contribu-
tion of single element i to the whole chain (1 − j) respectively, which can be
written as:

Vji = − pi ( xi+1 − xi ) (5.22)

1 
M ji = pi ( xi+1 − xi ) x j+1 − ( xi + xi+1 ) +
2 (5.23)
1 
+ p i ( y i +1 − y i ) y j +1 − ( y i + y i +1 )
2
5.2 basic theory 59

Note that Equation (5.21) may be rewritten as Equation (5.24), however, in


the sense of numerical computation, Equation (5.24) may introduce errors since
when the number of elements goes more, the last term will close to zero.

1
R3j = Cj = ( xi+1 − xi )2 + (yi+1 − yi )2 − =0 (5.24)
m2
Another remark is that the unknowns stored in vector X do not comprise the
horizontal nodal force f i (except f 1 ) which is physically presents at each node.
However, these fores can be computed by exploiting horizontal equilibrium once
the vector X is solved.
The horizontal equilibrium for the chain (1 − j) is:
j
Hj = ∑ Hji + f j+1 − f1 = 0 (5.25)
i =1

where Hji is similarly the horizontal equilibrium contribution of single element


i to the whole chain (1 − j):

Hji = pi (yi+1 − yi ) (5.26)

From Equation (5.25), the unknown f j+1 can be solved by:

j
f j+1 = f 1 − ∑ Hji (5.27)
i =1

If one reorders the notations in Equation (5.27), the final expression for un-
known f 1 is shown in Equation (5.28), and the membrane axial forces at position
( xi , yi ) can be computed as the resultants of f i and gi .
i −1
f i = f 1 − ∑ Hij , i = 2, 3, 4, · · · (5.28)
j =1

Once the nodal forces f i and gi are computed, the membrane tension ti can
be solved by vector summation:

~ti = ~f i + ~gi (5.29)

5.2.3 The Jacobian Matrix

This section deals with the partial derivatives in Equation (5.10) to Equation
(5.12), i. e. the terms in the Jacobian matrix.
First of all, the derivatives of Vj respect to X are concerned.

j
∂Vj ∂Vji ∂g j+1 ∂g1
=∑ + − (5.30)
∂X i =1
∂X ∂X ∂X
60 form-finding of inflatable dams

For the term of ∂Vji /∂X, since only xi and xi+1 will contribute to the deriva-
tives, this term can be evaluated as:
∂Vji ∂Vji
= pi , = − pi (5.31)
∂xi ∂xi+1

For the remaining terms, the results are:

∂g j+1 ∂(− g1 )
= 1, = −1 (5.32)
∂g j+1 ∂g1

Then, the derivatives of M j respect to X are evaluated.

j
 
∂M j ∂M ji ∂ g1 ( x j+1 − x1 ) ∂ f 1 ( y j +1 − y 1 )
=∑ + − (5.33)
∂X i =1
∂X ∂X ∂X

For the term of ∂M ji /∂X, since only xi , yi , xi+1 , yi+1 , x j+1 and y j+1 will con-
tribute to the derivatives, this term can be evaluated as:
∂M ji
= − p i ( x j +1 − x i ) ,
∂xi
∂M ji
= − p i ( y j +1 − y i ) ,
∂yi
∂M ji
= p i ( x j +1 − x i +1 ) ,
∂xi+1
∂M ji (5.34)
= p i ( y j +1 − y i +1 ) ,
∂yi+1
∂M ji
= p i ( x i +1 − x i ) ,
∂x j+1
∂M ji
= p i ( y i +1 − y i )
∂y j+1

For the remaining terms, the results are:



∂ g1 ( x j + 1 − x 1 )
=( x j+1 − x1 ),
∂g1
 (5.35)
∂ − f 1 ( y j +1 − y 1 )
= − ( y j +1 − y 1 )
∂ f1

Pay attention that the remaining terms are also contributed by x j+1 and y j+1 ,
and thus the full derivatives of M j with respect to x j+1 and y j+1 are:

∂M j
= p i ( x i + 1 − x i ) + g1 ,
∂x j+1
(5.36)
∂M j
= p i ( y i +1 − y i ) − f 1
∂y j+1
5.2 basic theory 61

The next thing is to evaluate the derivatives of geometric constraints Cj with


respect to X. Since only x j , y j , x j+1 and y j+1 contribute to the derivatives, the
final results are:
∂Cj
= − 2m2 ( x j+1 − x j ),
∂x j
∂Cj
= − 2m2 (y j+1 − y j ),
∂y j
(5.37)
∂Cj 2
=2m ( x j+1 − x j ),
∂x j+1
∂Cj
=2m2 (y j+1 − y j )
∂y j+1

Finally, the Jacobian matrix looks like:


 
∂V1 ∂V1 ∂V1 ∂V1
∂ f1 ∂g1 ∂x2 ··· ∂gm+1 

 ∂M1 ∂M1 ∂M1 ∂M1 

 ∂ f1 ∂g1 ∂x2 ··· 
∂gm+1 
 
 ∂C1 ∂C1 ∂C1
··· ∂C1 
   ∂ f1 ∂g1 ∂x2 ∂gm+1 
∂R  . .. .. .. 
J(X) = =  .. . . .  (5.38)
∂X X

 ∂Vm

∂Vm 

 ∂ f1
∂Vm
∂g1
∂Vm
∂X3 · · · ∂gm+1 

 
 ∂Mm
 ∂ f1
∂Mm
∂g1
∂Mm
∂x2 · · · ∂gm+1 
∂Mm 
 
∂Cm
∂ f1
∂Cm
∂g1
∂Cm
∂x2 · · · ∂g
∂Cm
m +1 3m×3m

where, the elements in X are in the order of what is shown in Equation (5.18)
However, in order to generate the Jacobian matrix easier, one may consider
an augmented version for both unknown vector X b (3m+4)×1 (Equation (5.39)) and
Jacobian matrix bJ(X
b )3m×(3m+4) . This scheme makes the unknown vector and
Jacobian matrix more regular and easier to generate by computer codes. Once
the augmented Jacobian matrix bJ(X b ) is generated, 4 columns can be deleted
corresponding to nodal coordinates (x̂1 , ŷ1 , x̂m+1 and ŷm+1 ) which are given
62 form-finding of inflatable dams

data, then the augmented Jacobian matrix bJ(X


b ) becomes a square matrix now
which can be used in Newton-Raphson iteration.
 
f
 1 
 x̂ 
 1 
 ŷ1 
 
 
 g 
 1 
 
 x2 
 
 
 y2 
 
 g2 
 
 . 
 . 
 . 
 
 xi 
X=
b   (5.39)
 yi 

 
 gi 
 .. 
 
 . 
 
 x 
 m 
 
 ym 
 
 g 
 m 
 
 
 x̂m+1 
 
 ŷ 
 m +1 
g m +1 
(3m+4)×1

5.3 form-finding procedure

The form-finding procedure basically contains two phases:


• Phase 1: starting from an initial circular cross-section with only air pres-
sure inflated and calculated each nodal coordinates by the given data
(cross-sectional total length L and distance between anchorage points D).
This initial configuration is the consequence of the equations studied by
Parbery (1976). The physical reason is without any hydrostatic pressure
applied, only air pressure inflated, which is perpendicular to the elements,
the curvature of the equilibrium shape should be constant: this means the
initial equilibrium shape should be (or a portion of) a circle.
• Phase 2: find the final equilibrium shape with both internal air pressure
and external hydrostatic pressure applied incrementally to the membrane,
as it is shown in Figure 5.12. In this figure, broken curve represents the
initial shape obtained from Phase 1.
In order to calculate the required parameters in Phase 1, one may consider
the equation system by Equation (5.40), which is also nonlinear. The notations
in Equation (5.40) are shown in Figure 5.10.
5.3 form-finding procedure 63

R
4

3
θ
C(x0,y0)

2
R

0
-1 0 1 2 3 4 5 6 7

D = anchorage-point separations

Figure 5.10: Illustration for terminology for calculating the center position of the circu-
lar profile.

The radius R and center ( x̂0 , ŷ0 ) of the circle can be found by Equation (5.40)
and Equation (5.41) :

2Rθ = L, 2R sin(θ ) = D (5.40)

where L is total length of the cross-section, D is the distance between two an-
chorage points, θ is the angle shown in Figure 5.10.
Once the nonlinear system of Equation (5.40) is solved, center ( x̂0 , ŷ0 ) of the
circle can be computed by:

D
x̂0 = ; ŷ0 = − R cos(θ ) (5.41)
2
Then, each node coordinates are determined by Equation (5.42).

x̂ = − R sin(θi ) + x̂0 ; ŷ = − R cos(θi ) + ŷ0 ; (5.42)

where θi is the angle corresponding to ith node and varies from (π − θ ) to


( π + θ ).
Note that the terms in Equation (5.40) to Equation (5.42) are dimensional,
non-dimensional terms can be computed by divided L for both side in Equation
(5.40) to Equation (5.42).
As the initial configuration is set as a circular shape, the exact solution for
membrane tension can be computed by the second equation in Equation (5.8),
and rewritten as:
pint R
t i +1 = (5.43)
p0 L
64 form-finding of inflatable dams

Then the horizontal and vertical components for membrane tension can be
expressed non-dimensionally as:

f i+1 = mti+1 ( xi+1 − xi ), gi+1 = mti+1 (yi+1 − yi ) (5.44)

In fact, Phase 1 can also be achieved by assuming a flat configuration. In this


case, the internal pressures are increasing progressively and simultaneously one
"fixed" anchorage point (point B0 in Figure 5.11) is moving along the base axis
until its "final" fixed position (point B in Figure 5.11). This procedure can be
verified in the following example.

A B B’

Figure 5.11: Finding the initial circular shape (curve with circles) starting from flat
configuration.

In Phase 2, the upstream water level hu , downstream water level hd and inter-
nal water level hi (if any) should be given, then the elements which are subjected
by hydrostatic pressures may be found, simply comparing the nodal positions
of elements with the given water level(s). Once the elements subjected to both
external and internal pressures are determined, the pressure load vector is mod-
ified for these elements by Equation (5.16). Note that the hydrostatic pressure
is assumed applied at the mid point of each elements with average pressure
because of the trapezoidal distribution of hydrostatic pressure. Then the same
procedure introduced in Section 5.2 can be applied in order to find the final
equilibrium shape.

5.4 examples

5.4.1 Double-Anchor Inflatable Dams

This example is adopted from the paper by Harrison (1970). Two inflatable dams
with initial flat configuration (as in Figure 5.8) having 18.3 m perimeters and
anchorage-point separations of 6.1 m are considered. Figure 5.13 shows the
cross-sectional equilibrium profiles of the two dams with upstream heads of
1.52 m and 5.8 m respectively. In both cases, the downstream head is fixed at
1.52 m, and the internal pressures are generated by a internal water head of 1.52
m and an air overpressure of 27.58 kPa.
5.4 examples 65

upstream

internal water level downstream

A B

Figure 5.12: Form-finding procedure (with external hydrostatic pressure). The final
equilibrium shape is shown in curve with circles.

5.4.1.1 Rigid elements


Firstly, rigid elements are employed to solve this problem. In this case, the cross-
section profile is discretized into 100 elements, i. e. m = 100.
Figure 5.14 shows the results of membrane tension, which is almost constant
along the local element direction ("dS" in Figure 5.9); this result matches with the
analysis in Section 5.3. Table 5.1 lists the average membrane tension in Figure
5.14 along with previous studies by Harrison (1970) and Parbery (1976).

Table 5.1: Membrane tension for various methods.


upstream tension by tension tension by tension by
head Harrison by Parbery flat con- circular
(1970) (1976) figuration config.
1.52 m 111.059 kN/m 110.562 kN/m 109.475 kN/m 109.712 kN/m

5.8 m 86.431 kN/m 83.877 kN/m 78.867 kN/m 79.061 kN/m

Table 5.1 shows the numerical method with rigid elements have a good agree
with previous studies in the case of lower upstream water level (hu = 1.52m).
But in the higher upstream water level (hu = 5.8m), the difference occurs. In
terms of the initial configurations in Phase 1, it is indicated that the membrane
tension calculated using the algorithm correctly predicts the forces starting from
both flat and circular configuration. Table 5.1 also reveals that a increased exter-
nal water pressure results in an reduction in membrane tension. The internal
pressure can thus be seen as a pre-stressing force for the membrane structure.

5.4.1.2 Extensible straight elements


In order to improve the accuracy with higher upstream water level, extensible
straight elements will be used. An elastic modulus of 2,068,500 kN/m2 is em-
ployed for each element with a material thickness of 7.62 mm and longitudinal
66 form-finding of inflatable dams

5
Vertical position [m]

0
-1 0 1 2 3 4 5 6 7
Horizontal position [m]
(a)

5
Vertical position [m]

0 1 2 3 4 5 6 7 8
Horizontal position [m]
(b)

Figure 5.13: Final equilibrium shape of air-water inflated dams with different upstream
water head: (a) hu = 1.52 m, (b) hu = 5.8 m. Curves with circles present
the final configuration.
5.4 examples 67

120

115

110
Membrane tension [kN/m]

105

100

95
hu = 1.52m
hu = 5.8m
90

85

80

75
10 20 30 40 50 60 70 80 90 100
Element position

Figure 5.14: Membrane tension with different upstream water head.

depth of 1 m. The element number m is 100 and the initial configuration is


circular.
Since the elasticity is introduced to the problem, the equation system must
be modified. One possibility is replacing Equation (5.21) by Equation (5.45) and
the other parts remain unchanged.

ti 2
R3j = Cj = m2 ( x j+1 − x j )2 + m2 (y j+1 − y j )2 − (1 + ) =0 (5.45)
EA
where ti is the membrane tension.
In fact, the membrane tension ti is also a function of nodal coordinates and
forces. Hence the Jacobian matrix J(X) should also be modified. However, in
terms of application, this change may be neglected.
Table 5.2 reveals the improvement by introducing tensile stiffness. The nu-
merical method with extensible straight elements have an excellent agree with
the referencs by Harrison (1970) and Parbery (1976) in the both cases of lower
(hu = 1.52m) and higher upstream water level (hu = 5.8m).

5.4.1.3 Efficiency of the methods


As this discretized method can achieve the final equilibrium shape even begin-
ning with a flat configuration, it is necessary to discuss the efficiency for this
situation.
In Phase 1 at Section 5.3, if one "anchorage" point is moving from an end posi-
tion of the flat configuration to the real anchorage position, there will be several
parameters may have influence on the result. One is the element number and
68 form-finding of inflatable dams

Table 5.2: Membrane tension for various methods. "RE" and "ESE" represents rigid
element and extensible straight element respectively.
upstream tension by tension tension by tension by
head Harrison by Parbery re ese
(1970) (1976)
1.52 m 111.059 kN/m 110.562 kN/m 109.712 kN/m 111.141 kN/m

5.8 m 86.431 kN/m 83.877 kN/m 79.061 kN/m 85.446 kN/m

the other may be the internal steps of moving the point. The physical meaning
for internal steps is the subdivisions of the distance between the flat "anchor-
age" position to the real anchorage point, as in Equation (5.46), where nstep is
the internal steps and xm i
+1 is the x-coordinate for the "anchorage" point in ith
internal step. For each selected element number (m), the required minimum in-
ternal steps (nstep) are shown in Table 5.3. Note that in this case, the applied
load is internal air pressure only.

i L−D i
xm +1 = 1 − (5.46)
L nstep

Table 5.3: Minimum internal steps for selected element number. m is element number,
nstep is internal step number.
m nste p tension
[k N / m]
20 8 110.8971

30 14 110.8086

40 19 110.7778
(40) (30) (110.7778)

50 27 110.7635

accu. 110.7381

Table 5.3 reveals that the more elements are introduced during the computa-
tion, the more internal steps are needed to converge, and the better accuracy is
reached in the sense of membrane tension. For a fixed number of finite elements
(m = 40), an increase of internal steps (nstep from 19 to 30) provides the same
result of membrane tension, which means the minimum internal step is suffi-
cient to reach the same level of convergence; this fact can also be seen in Figure
5.15.
Figure 5.15 shows the number of Newton-Raphson’ iterations in each internal
steps for moving the "anchorage" point. For a given number of internal step, the
first step costs the most number of N-R iterations in order to get convergence;
the following steps costs almost the same number of N-R iterations, i. e. the
5.4 examples 69

25
nstep=19
nstep=30

20

15
Iterations

10

0
5 10 15 20 25 30
Steps

Figure 5.15: Influence of different internal steps on N-R iterations when m = 40.

convergence speed is almost constant. Increasing internal steps dose not change
convergence speed much, especially in later steps. However, an increase of in-
ternal steps results the decline of iterations in the first step, which may make
numerical computation stable.
Figure 5.16 shows that although the number of finite elements is growing, the
convergence rate remains a same level in each internal step. However, a benefit
of a higher of finite element number is the circular configuration is more regular
(Figure 5.17), which will make contribution to the Phase 2 with better accuracy.
As it has been discussed in Section 5.3, the configuration with air pressure
only should be a circular shape. To verify whether the equilibrium shape in
Figure 5.17 (curve with small circles) is circular or not, one possibility is to
check the Equation (5.47) is holden or not.
 2
2 2 R
( x i − x0 ) + ( y i − y0 ) = (5.47)
L

where ( x0 , y0 ) and R/L are the non-dimensional coordinates of the center of


and the radius the circle respectively. They can be computed by Equation (5.40)
and Equation (5.41).
Figure 5.18 shows the difference and error of each node evaluated by Equation
(5.47). It is illustrated the errors between the shape found from a flat configura-
tion to a real circular configuration is at the level of 10−5 , which is very small.
This check also proves that the discretized method is capable to deal with the
form-finding problem starting from a flat configuration.
70 form-finding of inflatable dams

25

20

15
Iterations

10

0
1 2 3 4 5 6 7 8
Steps

(a)

25

20

15
Iterations

10

0
5 10 15 20 25
Steps

(b)

Figure 5.16: Influence of different number of finite elements on N-R iterations when (a)
m = 20 and (b) m = 50.
5.4 examples 71

10

6
Vertical position [m]

-2

-4
0 2 4 6 8 10 12 14 16 18
Horizontal position [m]
(a)

10

6
Vertical position [m]

-2

-4
0 2 4 6 8 10 12 14 16 18
Horizontal position [m]
(b)

Figure 5.17: Influence of different number of finite elements on equilibrium shape


when (a) m = 20 and (b) m = 50.

After the initial equilibrium shape is generated whether from a flat configu-
ration or circular configuration in Phase 1, Phase 2 is entered. Following para-
72 form-finding of inflatable dams

-5
10
0.04823 9

(x-x 0)2+(y-y 0)2


0.04822 (R/L)2
8
Errors

0.04821 7

0.0482 6
(x-x 0 )2 +(y-y0 )2 and (R/L)2

0.04819 5

Errors
0.04818 4

0.04817 3

0.04816 2

0.04815 1

0.04814 0

0.04813 -1
0 5 10 15 20 25 30 35 40 45
Nodes

Figure 5.18: Values evaluated by Equation (5.47).

graphs will compare the efficiency for both initial configurations in the same
case with hu = 5.8m (Figure 5.13 (b)). Table 5.4 shows the minimum internal
steps required in this form-finding procedure; Figure 5.19 shows the N-R iter-
ation steps in each load increments for each methods. Note that in this case,
the physical meaning of "internal steps" is the load increments composed by
Newton-Raphson’s iteration.

Table 5.4: Minimum load increments for different initial configurations when m = 40.
nstep is load increment number.
initial nste p tension
config. [k N / m]
Flat 3 84.049

Circular 4 85.910

Table 5.4 shows that in the case with upstream water head at 5.8 m, form-
finding procedure starting from a circular configuration requires one more load
increment than the one from a flat configuration; this disadvantage may be
neglected practically. Figure 5.19 also illustrates that for both method, the con-
vergence rates are at the same level.
Note that one may check the plots for membrane tension components ( f i and
gi ) in order to make sure the results are correct. A typical plots for these forces
are shown in Figure 5.20. Since it is similar for the final equilibrium shapes that
computed by different number of incremental load, and the membrane tension
5.4 examples 73

10

6
Iterations

0
1 2 3
Steps

(a)

10

6
Iterations

0
1 2 3 4
Steps

(b)

Figure 5.19: Illustration of N-R iteration steps for different initial configurations. (a)
Flat configuration and (b) circular configuration.

is almost constant along the circular axis, the check for its component is very
74 form-finding of inflatable dams

important. If the plot for any of them is not smooth, the result is not correct and
the load increments need to be modified.
100

80
Horizontal component of membrane tension [kN/m]

60

40

20

-20

-40

-60

-80
0 5 10 15 20 25 30 35 40 45
Element position

(a)

80

60
Vertical component of membrane tension [kN/m]

40

20

-20

-40

-60

-80

-100
0 5 10 15 20 25 30 35 40 45
Element position

(b)

Figure 5.20: Illustration of (a) horizontal component and (b) vertical component of
membrane tension.
5.4 examples 75

5.4.2 Single-Anchor Inflatable Dams

This discretized method can also deal with single-anchor inflatable dams, but in
a slightly different way. In the case of single-anchor dams, the numerical method
is a kind of design-driven codes. The basic procedures are:

1 Assign a suitable value for initial perimeter L and a fictitious anchorage dis-
tance D. Use the form-finding procedures in Section 5.3 to generate a
initial configuration which is shown in Figure 5.21 (a). At this stage, all
applied loads, including hydrostatic loads and internal air pressure are
subjected to the membrane.

2 Move the fictitious anchorage point (usually the right node) towards the real
anchorage position (usually the left fixed node) along the base step by step.
At each step, update the fictitious anchorage distance D; track the angle
of each element, when the tangent of the last element (the mth element) is
near zero, stop the loop.

3 The final perimeter of the membrane can be calculated by adding up the


updated fictitious anchorage distance to the initial length L.

Table 5.5: Basic data for single-anchor dam.


Initial perimeter 20 m

Fictitious anchorage distance 7m

Upstream water level 5.8 m

Downstream water level 1.52 m

Internal water level 1.52 m

Internal air pressure 27.58 kPa

Table 5.5 lists the data for a single-anchor inflatable dam. Here 50 rigid ele-
ments are employed.
After running the code, the final equilibrium configuration for the single-
anchor dam is shown in Figure 5.21 (b). The final perimeter for the membrane
is 23.5 m.
To investigate the effect of the number of anchorage points on the behaviour
of inflatable dams, the example of this single-anchorage dam is further explored.
Figure 5.22 shows the final equilibrium shapes for several inflatable dams with
different anchorage-point distances. In these cases, the cross section perimeter
L is fixed at 23.5 m and the anchorage-point distance varies from 7 m to 4
m; other parameters keep the same as in Table 5.5. Figure 5.22 illustrates the
single anchorage structure presents a 20% to 25% decrease in the overall dam
height, which agrees with the research by Streeter, Rhode-Barbarigos, and Adri-
aenssens (2015). Another fact is for those double-anchor dams, although these
76 form-finding of inflatable dams

5
Vertical position [m]

0 1 2 3 4 5 6 7 8 9
Horizontal position [m]
(a)

5
Vertical position [m]

0
0 1 2 3 4 5 6 7 8 9
Horizontal position [m]
(b)

Figure 5.21: Form-finding procedure for single-anchor inflatable dams. (a) Initial con-
figuration (curve with circles) and (b) final equilibrium configuration (bold
curve with circles).
5.4 examples 77

9
D=7m
D=6m
8 D=5m
D=4m
Single-anchored
7

6
Vertical position [m]

0
0 1 2 3 4 5 6 7 8 9 10
Horizontal position [m]

Figure 5.22: Cross-sectional equilibrium profile of 5 inflatable dams with different


anchorage-point separations. D is the distance between anchorage points.

cross-sectional equilibrium profiles differ significantly in size, they have an over-


all similar shape.
Figure 5.23 shows the variation in membrane tension as the anchorage dis-
tance decrease from 7 m to 0 (single-anchored) for the same structure under a
downstream head of 1.52 m, an upstream head of 5.8 m and an internal pres-
sure generated by a 1.52 m hydrostatic pressure and a 27.58 kPa air pressure.
It is a particular behaviour for the influence of anchorage-point separations
on membrane tension that the linearly reduction of membrane tension as an-
chorage points get closer. One explanation for this behaviour is for dams with
lager anchorage points, the angle between the mth element and the base ground
("downstream angle") increases, and consequently the enlargement may result
in lager membrane tension. This phenomenon has been researched by Parbery
(1976).

5.4.3 Internal Air Pressure Update Procedure

The internal air pressure update procedure is introduced by Streeter, Rhode-


Barbarigos, and Adriaenssens (2015) for analyzing inflatable dams sustained
lager external loads, since the application of a large external loading can lead to
78 form-finding of inflatable dams

125

120

115
Membrane tension [kN/m]

110

105

100

95

90

85

80
0 1 2 3 4 5 6 7
Anchorage distance D [m]

Figure 5.23: Influence of anchorage-point separations on membrane tension forces.

significant variations in the internal pressure of the structure. The basic proce-
dure for this problem is similar to the form-finding procedure which has been
introduced in Section 5.3; the only difference is the internal pressure is mod-
ified at each load increment due to the change of membrane volume. Hence,
the assumption for this problem is the membrane is a perfectly sealed struc-
ture inflated by an ideal gas under constant temperature. This implies that, for
the cross-sectional analysis of the dam, as longitudinal length of the dam is
assumed as 1 m, changes in internal air pressure correlate to changes in the
internal area (instead of the whole volume of the dam). The internal area of the
dam cross-sectional profile A is approximated based on the sum of a series of
discrete rectangles:
m
A= ∑ Ai (5.48)
i =1

where Ai is the area of the rectangle corresponding to the ith element and m is
the total number of elements.
The width of each rectangle is defined by the horizontal difference in the
position between the two nodes of the elements, while the height of the rectangle
is given by the elevation of the centroid of the element. Figure 5.24 illustrates
the area approximation for the cross-sectional profile of a dam with 12 elements.
5.4 examples 79

Elements whose areas fall outside the membrane (areas A10 to A12 on Figure
5.24) are defined with a negative value.
7

5
Vertical position [m]

A9
3 A6 A7
A5 A8

A4
2

A3 A10
1
A2 A11
A1 A12
0
0 1 2 3 4 5 6 7 8 9
Horizontal position [m]

Figure 5.24: Approximated area for the dam cross-section with 12 elements. Areas that
fall outside of the profile boundaries are defined as negative values.

The original internal air pressure is used as the initial air pressure for the
first cycle, i. e. the first load increment of the discretized algorithm. During this
load increment, transversal area of the membrane is computed as the initial
area. The difference, ∆A, between the transversal area of the equilibrium shape
in later cycles and the initial area is then estimated and linked to the internal
pressure change:
∆A
∆p = − p (5.49)
A k k −1
where ∆p is the variation of the internal pressure at the kth load increment, Ak
is the transversal area at the kth load increment and pk−1 is the internal pressure
at the (k − 1)th load increment.
This allows the estimation of the updated pressure value which is employed
in the next cycle of the finite element algorithm in conjunction with the new
equilibrium profile. The updated pressure at kth load increment is thus pk =
pk−1 + ∆p. The process is repeated for each load increment until total load is
applied.
This updated internal air pressure procedure is employed to study the effects
of changes in the internal air pressure on membrane tension based on the ex-
80 form-finding of inflatable dams

ample presented by Harrison (1970) and Parbery (1976). The specific data the
inflatable dam is the same as in Section 5.4.1 and Figure 5.13 (b). A circular
inflatable membrane dam under a downstream head of 1.52 m and an internal
pressure generated by a 1.52 m hydrostatic head and a 27.58 kPa air pressure
with an anchorage separation of 6.1 m is thus considered. All parameters are
held constant while the upstream head increases from 1.52 m to 5.8 m. Figure
5.25 shows the corresponding cross-sectional equilibrium profiles revealing a
significant change in the shape of the inflatable structure. This figure reveals
that with updated internal air pressure, the final configuration of the inflatable
dam is stiffer than the one with constant internal air pressure. Figure 5.26 shows
the tension in the membrane obtained with and without the internal air pres-
sure update as the upstream head increases. For upstream water level higher
than 3 m, As the external loading increases, the membrane tension of dams
with updated air pressure has a trend of increase while the tension in the mem-
brane with constant air pressure decreases. For upstream head values lower
than 3 m, the change in membrane tension is similar for both cases. This fact
illustrates that the updated air pressure scheme can sustain larger external load,
as the dam with updated air pressure looks differ than the one with constant
air pressure in terms of membrane height (Figure 5.25) and membrane tension.
Consequently, assuming a constant pressure will underestimate membrane ten-
sion and should be avoided when a large external loading is applied.

5.5 summary

This chapter presents an alternative tool with computational advantage for the
form-finding and analysis of inflatable dams through the discretized finite el-
ement method. This method has various applications in the area of inflatable
dams such as single-anchor and double-anchor dams, as well as the application
of updated internal air pressure scheme for large external loads. This method
is also flexible to deal with different initial configuration such as flat or circular
configuration. This flexibility gives a possibility to analyze different inflatable
dams with short or no anchorage separation (single anchorage systems). Finally,
the updated pressure scheme allows the method to properly model perfectly
sealed systems. Furthermore, the study of the updated pressure shows that the
constant pressure assumption which is commonly used in the research, may not
be employed under large external loads since it makes the membrane less stiff
compared with the one using updated pressure scheme.
5.5 summary 81

5
Vertical position [m]

1
constant internal air pressure
updated internal air pressure
0
0 1 2 3 4 5 6 7 8 9
Horizontal position [m]

Figure 5.25: Final equilibrium shapes of an inflatable dam assuming constant internal
air pressure and an internal air pressure update.
82 form-finding of inflatable dams

125 35
constant internal air pressure
updated internal air pressure
120 34

115 33

Maximum air pressure [kPa]


Membrane tension [kN/m]

110 32

105 31

100 30

95 29

90 28

85 27
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Upstream head [m]

Figure 5.26: Membrane tension values obtained with and without the internal air pres-
sure update when the upstream head changes.
BIBLIOGRAPHY

Adriaenssens, Sigrid, Philippe Block, Diederik Veenendaal, and Chris Williams


(2014). Shell Structures For Architecture: Form Finding And Optimization. Rout-
ledge (cit. on pp. 34, 39, 41).
Anwar, Habib O (1967). “Inflatable dams.” In: Journal of the Hydraulics Division
93.3, pp. 99–119 (cit. on p. 49).
Bagrianski, Serguei and Allison B Halpern (2014). “Form-finding of compressive
structures using Prescriptive Dynamic Relaxation.” In: Computers & Struc-
tures 132, pp. 65–74 (cit. on p. 36).
Barnes, Michael R (1999). “Form finding and analysis of tension structures by
dynamic relaxation.” In: International journal of space structures 14.2, pp. 89–
104 (cit. on pp. 34, 38, 41).
Barnes, MR (1977). “Form finding and analysis of tension space structures by
dynamic relaxation.” PhD thesis. City University London (cit. on pp. 33, 34).
Binnie, AM (1973). “The theory of flexible dams inflated by water pressure.” In:
journal of Hydraulic Research 11.1, pp. 61–68 (cit. on p. 49).
Breukelen, Marjolein van (2013). “Improvement and scale enlargement of the
inflatable rubber barrier concept.” In: Maste Thesis, Delft University of Tech-
nology (cit. on p. 48).
Chu, Jian, Wei Guo, and SW Yan (2011). “Geosynthetic tubes and geosynthetic
mats: analyses and applications.” In: Geotechnical Engineering 42.1, p. 57 (cit.
on p. 47).
Engel, Heino (1967). Structure Systems. Hatje Cantz Pub (cit. on p. 1).
Finzi, Bruno (1976). Meccanica Razionale. Zanichelli (cit. on p. 55).
Ghali, Amin, Adam Neville, and Tom G Brown (2003). Structural Analysis: A
Unified Classical And Matrix Approach. Crc Press (cit. on p. 17).
Harrison, HB (1970). “The analysis and behaviour of inflatable membrane dams
under static loading.” In: Proceedings of the Institution of Civil Engineers 45.4,
pp. 661–676 (cit. on pp. 54, 64, 65, 67, 68, 80).
Hernández-Montes, E, R Jurado-Piña, and E Bayo (2006). “Topological mapping
for tension structures.” In: Journal of Structural Engineering 132.6, pp. 970–977
(cit. on p. 26).
Hüttner, M, J Máca, and P Fajman (2015). “The efficiency of dynamic relaxation
methods in static analysis of cable structures.” In: Advances in Engineering
Software 89, pp. 28–35 (cit. on p. 41).
Langdon, David (2015). AD Classics: German Pavilion, Expo ’67 / Frei Otto and Rolf
Gutbrod. Online Documentation (cit. on p. 4).
Levy, Robert and William R Spillers (2013). Analysis Of Geometrically Nonlinear
Structures. Springer Science & Business Media (cit. on p. 11).
Lewis, Wanda J (2003). Tension Structures: Form And Behaviour. Thomas Telford
(cit. on pp. 34, 40).

83
84 Bibliography

Lewis, WJ, MS Jones, and KR Rushton (1984). “Dynamic relaxation analysis of


the non-linear static response of pretensioned cable roofs.” In: Computers &
Structures 18.6, pp. 989–997 (cit. on p. 34).
Macdonald, Angus J (2007). Structure And Architecture. Routledge (cit. on p. 1).
Mazeika, Aaron and Kieran Kelly-Sneed (2007). “NASCC preview-Getting Started
with Cable-Net Walls-While cable-net-supported glass walls remain an atyp-
ical design solution for exterior wall systems, their popularity is growing.”
In: Modern Steel Construction 47.4, p. 55 (cit. on p. 4).
Meek, John L (1991). Computer Methods In Structural Analysis. CRC Press (cit. on
p. 11).
Mysore, GV, SI Liapis, and RH Plaut (1998). “Dynamic analysis of single-anchor
inflatable dams.” In: Journal of Sound and Vibration 215.2, pp. 251–272 (cit. on
p. 47).
Otter, Joseph RH, Alfred C Cassell, and Roger E Hobbs (1966). “Dynamic re-
laxation.” In: Proceedings of the Institution of Civil Engineers 35.4, pp. 633–656
(cit. on p. 33).
Parbery, RD (1976). “A continuous method of analysis for the inflatable dam.”
In: Proceedings of the Institution of Civil Engineers 61.4, pp. 725–736 (cit. on
pp. 47, 62, 65, 67, 68, 77, 80).
Schek, H-J (1974). “The force density method for form finding and computation
of general networks.” In: Computer methods in applied mechanics and engineer-
ing 3.1, pp. 115–134 (cit. on pp. 25, 26).
Streeter, M, Landolf Rhode-Barbarigos, and Sigrid Adriaenssens (2015). “Form
finding and analysis of inflatable dams using dynamic relaxation.” In: Ap-
plied Mathematics and Computation 267, pp. 742–749 (cit. on pp. 49–51, 75, 77).
Tam, Paul Wing Ming (1997). “Use of rubber dams for flood mitigation in Hong
Kong.” In: Journal of irrigation and drainage engineering 123.2, pp. 73–78 (cit.
on p. 47).
Tanaka, K, K Kondoh, and SN Atluri (1985). “Instability analysis of space trusses
using exact tangent-stiffness matrices.” In: Finite elements in analysis and de-
sign 1.4, pp. 291–311 (cit. on p. 18).
Tibert, Gunnar (1999). “Numerical analyses of cable roof structures.” PhD thesis.
KTH (cit. on pp. 3, 25–29, 32, 45).
Topping, Barry HV and Peter Iványi (2008). Computer Aided Design Of Cable
Membrane Structures. Saxe-Coburg Publications (cit. on pp. 33, 34).
Van Mele, Tom, Lars De Laet, Diederik Veenendaal, Marijke Mollaert, and Philippe
Block (2013). “Shaping tension structures with actively bent linear elements.”
In: International Journal of Space Structures 28.3-4, pp. 127–136 (cit. on p. 2).
Veenendaal, D and P Block (2012). “An overview and comparison of structural
form finding methods for general networks.” In: International Journal of Solids
and Structures 49.26, pp. 3741–3753 (cit. on p. 33).
Watson, Layne T, Surjani Suherman, and Raymond H Plaut (1999). “Two-dimensional
elastica analysis of equilibrium shapes of single-anchor inflatable dams.” In:
International Journal of Solids and Structures 36.9, pp. 1383–1398 (cit. on pp. 47,
49, 51–53).
Bibliography 85

Yaw, Louie L. (2011). 3D Co-rotational Truss Formulation. Online Documentation


(cit. on p. 16).

You might also like