You are on page 1of 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/326633268

Parametric design and optimization of shell structures using the Natural Force
Density Method

Conference Paper · January 2016

CITATIONS READS

2 2,121

2 authors:

Marcio Souza Ruy Marcelo Pauletti


University of São Paulo University of São Paulo
3 PUBLICATIONS   5 CITATIONS    120 PUBLICATIONS   331 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Membrane stuctures View project

fusion research View project

All content following this page was uploaded by Ruy Marcelo Pauletti on 28 July 2018.

The user has requested enhancement of the downloaded file.


Proceedings of the IASS Annual Symposium 2016
“Spatial Structures in the 21st Century”
26–30 September, 2016, Tokyo, Japan
K. Kawaguchi, M. Ohsaki, T. Takeuchi (eds.)

Parametric design and optimization of shell structures using the


Natural Force Density Method
Márcio S.V. SOUZA*, Ruy M.O. PAULETTI
Polytechnic School of the University of São Paulo
P.O. Box 61548, 05424-970 São Paulo, Brazil
marcio.sartorelli.souza@usp.br ; pauletti@usp.br

Abstract
In this paper, we present the development of a NFDM plugin for Rhino/Grasshopper environment, that
allows quick and robust parametric generation of bending-free shapes, either reticulated or continuous,
which we subsequently optimize using other Grasshopper plugins (Karamba and Galapagos). Even
though convergence of genetic algorithm is usually slower that of gradient optimization methods, we
found that the combination of NFDM shape generation and genetic algorithms offers a robust and overall
quick procedure to select efficient yet formally appealing shell geometries.
Keywords: form finding, parametric design, optimization, force densities, natural force densities, genetic
algorithms, grasshopper.

1. Introduction
Concrete shell structures had their apogee half a century ago, but manifestations declined thereafter, due
to increasing production costs and exhaustion of the formal repertoire. In subsequent decades, new
physical and numerical methods allowed exploration of a broader range of shapes, together with a
thorough assessment of their structural behavior. Recently, new methods for parametric design and
optimization finally stimulated a new architectural vanguard to re-explore shell structures, replacing the
previous, mostly opaque and geometrically limited concrete surfaces by translucent free-form grid-
shells, built from a variety of new materials. Nevertheless, even thou the current technology allows
construction of shells of very complex geometry, bending-free, funicular shapes still stand as the most
important class of shell structures, since they minimize the material required to withstand permanent
loads, and their typical double curvature shapes are also efficient to withstand variable loads mostly
through membrane-like stresses.
The Force Density Method (FDM) provides a convenient alternative for finding bending-free
geometries, approximating a continuous surface by a network of line elements that can also conveniently
represent the members of a grid-shell. By its turn, the Natural Force Density Method (NFDM) is an
extension of the FDM that preserves the linearity of the original method and overcomes some its
limitations to cope with irregular meshes that may arise from the discretization of irregular surface
geometries.
In this paper, we present the development of a NFDM plugin for Rhino/Grasshopper environment, that
allows quick and robust parametric generation of bending-free shapes, either reticulated or continuous,
which we subsequently optimize using other Grasshopper plugins (Karamba and Galapagos).

Copyright © 2016 by Márcio Sartorelli Venâncio de Souza and Ruy Marcelo de Oliveira Pauletti
Published by the International Association for Shell and Spatial Structures (IASS) with permission.
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

2. The Force Density Method


One of the first alternatives for form-finding was the definition of force densities, proposed by Linkwitz
[6] and Sheck [11], in the context of cable nets.
The Force Density Method (FDM) is based on the equilibrium of each node in a cable net. With
reference of the forces at the system in Fig.1, the resultant of internal forces acting on node i is
n n
Pi   Pij   N ij vij , (1)
j 1 j 1

Figure 1: System of central forces [Pauletti [9]]

Where N ij is the interaction force among the nodes i and j , while vij  lij / lij is the unit vector
oriented from i to j . Applying equilibrium conditions gives
n x j  xi
Fi   Nij 0 i  1, 2,..., n (2)
j 1 x j  xi
Relating nodes with nodal displacements results in a non-linear equation system. However, defining at
each element a force density nij

Nij
nij  (3)
x j  xi

a linear system with 3n equations is obtained


n

 n (x
j 1
ij j  xi ) Fi , i  1, 2,..., n (4)

With boundary conditions imposed, the system can be easily solved.


Still, it is convenient for large meshes to adopt a matrix notation instead of the presented. The Cartesian
coordinates of the system, as the external and internal forces, can be expressed by 3 global vectors X ,
F and P , respectively defined by
 x1   F1   P1 
x   F  P 
X   2  ;F   2  ;P   2  (5)
     
     
 xn  3nx1 Fn  3nx1  Pn  3nx1

2
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

Where xi  [ xi ]3 x1 , Fi  [ Fi ]3 x1 and Pi  [ Pi ]3 x1 are local coordinate and force vectors of each node. It
is also convenient to substitute the sum inside the definition of P from the n nodes to b elements
connecting those nodes. The vector of nodal coordinates and the vector of internal forces of the eth
element, as show in Fig. 2, are related by
b
xe = A e x ; P =  A eTpe (6)
e=1

T T T
Where xe =  x1eT xe2 T  and p = p1 pe2T  = N e -v eT v eT  , with v e  vij  as the
e eT
3 x1

director cosine of the element inside the global Cartesian coordinate system. Ae is defined as the
Boolean incidence matrix of the element.

Figure 2: Linear element with local and global indexes (Pauletti [9]).

Once these definitions are given, the equation (4) can be rewritten as
Kd X = F (7)
e
Where K d  A
b1
eT
k ed Ae is the force density stiffness matrix of the system, and

I -I 3 
k ed  ne  3
I 3 
(8)
-I 3
is the element local stiffness matrix, with n e as the element force density and I 3 as an order 3 identity
matrix.
3. The Natural Force Density Method
The Natural Force Density Method (NFDM) preserves the linearity of the original FDM meanwhile
issues related to irregular triangular meshes are overcome. The natural force density derives from the
natural forces defined at the natural membrane finite element, first proposed by Argyris [3]. Pauletti [8]
redefines the formulation of previous developments into a more concise notation.
The nodes and faces at Fig. 3 are numbered anti-clockwise, where each node index is coincident with
the face index in front of it. Node coordinates are referenced at both global and local Cartesian system,
with the local representation expressed by ‘^’.

3
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

Figure 3: Natural Membrane Finite Element (a) unit vectors, (b) internal forces and (c) N 3 definition (Pauletti
[9]).

To define the internal vector forces, it is convenient to define natural forces N i , which are parallel to its
faces. For the face 3 (Fig 4.c), it is defined as
N 3  12  3th3  1
3  3 At   3V
1
3 (9)

Where  3 is the natural stress parallel to the face 3, A the element area in the actual configuration, t
its thickness and V  At its volume. Analogue expressions give a natural force vector
N   N1 N 3  , and then the natural force densities vector is given by n   n1 n3   NL1
T
N2 n2
, with L as an order 3 diagonal matrix, where L ii  i . Thereafter, it can be shown that, for a prescribed
natural force densities vector n , there is a linear relationship between the natural force vector Pn and
element nodal coordinates vector x , according to Pn  k n x , with k n as the element stiffness matrix
defined by

 n2  n3  I 3 n3I 3 n2 I 3 
 
k en   n3I 3  n1  n3  I 3 n1I 3  (10)
 n2 I 3 n1I 3  n1  n2  I 3 
Assemble of the natural and linear force density elements leads the solution of the linear system
described by Eq. 7. However, the definition of natural force densities are not intuitive, requiring previous
definition of an initial stress field σˆ 0  Tσ n0 , where σˆ 0 is the element stress field at its local coordinate
system, σ n the stress field at the global coordinate system and T is a transformation matrix (Pauletti
0

[9]). At last, the natural force densities vector can be easily defined by n  VL T σˆ 0 .
-2 -1

The viable configuration defined by the pair  σˆ , X  can be used directly or utilized as a first estimative
inside iteration procedures.
4. Application to parametric design
Using the parametric tool Grasshopper (Davidson [4]), plug-in of Rhinoceros (McNeel [7]), custom
components were developed to apply the FDM and NFDM procedures. First, series of components are
used to transform the problem geometry into necessary information for method solving, defining linear
force density elements by lines, triangular natural force density elements by meshes and gravity,
supports and local loads by points and vectors.

4
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

Figure 4: Method data flow inside custom Grasshopper components.

FD elements also receive information about their own force density, section area and specific weight,
while NFD elements receive its initial stress, thickness and also specific weight. The elements geometric
and physical properties are for self-weight definition by the initial geometry, referred to a vector defining
gravity. However, the initial geometry differs of the resultant one, requiring an iterative procedure to
find the real weight of the shape (See 5.1).
Supports receives besides point information, also displacement vector, acting plane and d.o.f. restriction.
An intermediate component takes all these information to define node indexing, build the stiffness
matrix, the external forces vector and grab all these information inside a single data structure. The model
then goes to the final component for the linear equation solving, model coordinate vector update and
new geometry generation. The geometry then can be can be exported to a finite element analysis
program for structural analysis, or it can simply be done inside Grasshopper, with the FEA component
library Karamba (Preisinge [10]). The self-weight iteration procedure can be activated by a Boolean
parameter, and iterations limit number and maximum error needs to be defined.

Figure 5: (a) Resultant shape and shape with (b) increased NFD elements stress, (c) increased FD elements
stress, (d) support positions changed and (e,f) initial geometry change.

The methods applications inside parametric space provides a robust and interactive shell and membrane
design tool, where both initial shape and method parameters can be changed in real-time (Fig. 5).

5
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

Parametric design also allows the definition of very complex surfaces to define initial geometry, and
referencing Rhino3D objects inside Grasshopper allows parameters changes inside the 3D modelling
software. One example is in Fig. 6, where NFDM is applied to a mesh generated by Metaballs as
boundary curves, which are organic-looking objects that are inside Grasshopper library.

Figure 6: Form-finding of shell with a MetaBall generated initial mesh. MetaBalls center points can be
controlled inside Rhino3D environment.

5. NFDM Form-Finding for shells


Both FDM and NFDM were developed for the form-finding of taut structures (membrane and cables),
though FDM has already been used in some cases for timber grid-shell form-finding as the Multihaeim
Hall (1974) and the Solemar Therme (1987). While for membranes the procedures works fine defining
supports coordinates and managing the force densities for more or less tensioned surfaces while
checking their displacements, shells requires some specific considerations when looking for viable
shapes.
5.1. Self-weight
Considering self-weight is essential for shell design, as it rules its behavior. When designing a shell it is
generally desirable to get the closest to a funicular shape as possible.
Funicular shapes are those which shows only compressive behavior, without bending, maximizing
material efficiency. A simple way to achieve these shapes were discovered long time ago by Robert
Hooke (1676), testifying in your third law: “As hangs the flexible line, so but inverted will stand the
rigid arch”. The basic idea is that, as chain do not resist to bending, it search for a new geometric
configuration in a way that only tension forces are acting in. So, inverting the shape will provide a only
compression form if made from a rigid material, as compression stress are more likely to lead to 2º order
effects. Using this principle, it is only necessary to consider the surface self-weight inside NFDM
external forces, to the method simulate the hanging chain procedure and provide a resultant shape for a
shell, if inverted.
Nevertheless, inside NFDM the shape of the initial configuration and the resultant configuration differs.
So, computing the self-weight of the initial shape will not result exactly in the equilibrium shape,
requiring an iterative procedure to obtain the real external forces vector.
For a linear element, we compute the self-weight of the initial configuration
 F1e   F01e   S 1
 e   e 
2 1
(11)
 F2   F02 
e e
Where Fi are the external force and F0i prescribed loads acting on node i of the element at gravity
direction,  the element density, S the section area and its length. For triangular elements the
definition is similar

6
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

 F1e   F01e  1


 e   e   At  
 F2    F02   3 1 (12)
 F3e   F03e  1
   
Where A is the element area and t its thickness.
Then FDM is applied and residual forces vector r is given
r  Fi 1  Fi
Where Fi is the global force vector at iteration i . At last, the self-weight is iteratively updated, as the
shape modifies, until r is bellow a stipulated error. Similar approach was used with linear elements
by Block [1] inside his Thrust Network Analysis (TNA) Method.

Figure 7: 1º principal stress of shell (a) without iterations and (b) after 5 iterations.

5.2. Buckling
Shell structures failures occurs mostly by second order effects, principally buckling. Form-finding
techniques do not take this issue in account, requiring always a preliminary linear buckling analysis
during design and validation of the structure from precise non-linear analysis. With help of Grasshopper
plugin Karamba, it’s possible to obtain the shape buckling factor due to self-weight and many external
loads as necessary during design process. When a final shape is obtained, the shape can be exported to
be analyzed into FEA software’s with more precise analysis, as ANSYS, Abacus and OASYS.

Figure 8: (a) Triangular gridshell from FDM, with 3cm box section cross and wood as material. (b) Karamba
linear buckling analysis and ANSYS [2] non-linear analysis, with (c) 1º order and (d) 2º order buckling.

7
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

Furthermore, simple genetic algorithms as Galapagos provides an interesting manner to maximize or


even achieve a pre-determined buckling factor providing linear and natural force densities as
optimization parameters (See 6.2).
5.3. Constructive patterns
Shell shapes are generally very complex, which can make construction a challenge. Continuous shells
requires a very precise topographical mapping of the building area and its shape, as discrete shells can
bring problems if elements dimensions varies much. Even with recent digital fabrication technologies,
which can overcome many of these issues, it can be still desirable in project to define patterns that turns
construction an easier step.
With linear elements of the FDM it is possible to apply an iterative procedure to find prescribed lengths,
taking advantage of the presence of the element length inside force densities definition.
We first compute the self-weight of each linear element as it have its prescribed length
 F1e   F01e   S 1
 e   e 
p
1 (13)
 F2   F02  2 
Where p is the prescribed length. Then, the FDM procedure is applied and each element force density
is updated
nie
nie1  p (14)
i

With i as the element length at step i . The procedure continues until

Lp  L   (15)

Where Lp is the prescribed lengths vector and L is the resultant lengths vector. Similar approach were
also used by Block [1] inside TNA.

Figure 9: (a) 2.6m (131 iterations) and (b) 2.7m (128 iterations) constant length bars resultant grid from an initial
2.5m grid and its first principal stress fields.

6. Optimization inside parametric space


Optimization problems can be solved by different manners: calculus, enumerative and random based
methods. Calculus based methods are very important and highly used for structural optimization, but
they search domain can lead to local solutions and not consider the whole search space. Genetic
algorithms, by other side, are random based methods that can have larger search space. They are oriented
inside a logical search based on the survival of the fittest, analogue to evolution theory where the
population are objective functions (Goldberg [5]).

8
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

With NFDM inside parametric design, it is possible to achieve satisfying optimization results with built-
in Grasshopper plugin Galapagos, a genetic algorithm which take sliders as parameters to search along
an objective function.
Defining natural and linear force densities (also boundary conditions, if necessary) as parameters,
Galapagos can do geometric and structural optimization search leading to many issues as volume, area,
displacement and buckling problems. Also, Karamba provides structural information to the genetic
algorithm and has its own optimization components.
6.1. Covered area and deformation energy
For design constraints, maximum covered area ( A ) is preferred, but also, for structural efficiency,
minimum internal elastic energy ( Wint ). To achieve that for Galapagos, which allows only one objective
function, an optimization coefficient    A / Wint is defined, with  as a weighing factor, and the
genetic algorithm is set up to search for a maximum  . As parameters, border linear elements force
densities are defined at each internal and external edges, and triangular elements receives its initial stress
field.

Figure 10: Design options obtained with Galapagos search with its covered area, elastic energy and   0,1 .

By result, many design options can be found, as genetic algorithms works with populations of solutions,
and the final shape can be decided by the designer, choosing a solution that corresponds to its
expectations.
6.2. Buckling
The gridshell shown in Fig. 8 were found defining the same force densities along all linear elements.
For buckling optimization, each element have its individual force density parameter, which are
manipulated by Galapagos. Then, setting it up to maximize the load multiplier for buckling will start a
search for a more stable structure.

Figure 11: Optimized shape and its first buckling mode (a) linear analysis and (b) non-linear analysis.

Also, the definition of force densities for each linear element responded to slightly irregular patterns on
the grid, which can be aesthetically interesting.
6.3. Volume
For minimum volume optimization, simple genetic algorithms do not respond well as it is a highly
constrained problem, and simply setting up Galapagos for minimum weight will result in very inefficient

9
Proceedings of the IASS Annual Symposium 2016
Spatial Structures in the 21st Century

shells. However, Karamba has a component for cross section optimization which finds minimum cross
sections with structural safety respected, which can be applied for desired shapes found with NFDM.

Figure 12: Wind load cases and resultant shape under self-load after cross section optimization, limited to
maximum 80% of structure utilization and maximum displacement of 3cm.

In case of the triangular gridshell at Fig. 11, the structure force densities were optimized for buckling,
but only for its dead load. However, wind loads needs to be considered, and at more than one direction.
A quick procedure can be done in Karamba dividing these loads into load cases, and cross section
optimization search for a solution which satisfies all of them.
7. Acknowledgments
The first author acknowledges the support by São Paulo Research Foundation (FAPESP) under the grant
2015/24924-2.

References
[1] Adriaenssens S., Block P., Veenendaal D. and Williams C. Shell structures for architecture: form
finding and optimization. (1nd ed.), New York: Routledge, 2014.
[2] ANSYS Inc. ANSYS® Academic Research, Release 16.2. http://www.ansys.com/
[3] Argyris, J.H. Dunne, P.C. Angelopoulos, T. Bichat, B. Large natural strains and some special
difficulties due to non-linearity incompressibility in finite elements. Computer Methods in Applied
Mechanics and Engineering, 1974, p. 219–278.
[4] Davidson S. Grasshopper. http://www.grasshopper3d.com.
[5] Goldberg, D.E. Genetic Algorithms in Search, Optimization, and Machine Learning. Longman,
1989.
[6] Linkwitz, K. About formfinding of double-curved structures. Engineering Structures, 1999, p.
709–718.
[7] Mcnell. Rhino3D. http://www.rhino3d.com.
[8] Pauletti R. M. O. and Pimenta, P. M. The natural force density method for the shape finding of taut
structures. Computer Methods in Applied Mechanics and Engineering, 2008.
[9] Pauletti, R. M. O. An extension of the force density procedure to membrane structures. IASS
Symposium, 2006, v. 1, p. 490-491.
[10] Preisinge C. Linking structure and parametric geometry. Archit Design, 2013, 83, 110-113.
[11] Scheck, H.J. The force densities method for form finding and computation of general networks in
computer methods. Applied Mechanics and Engineering, 1974, p. 115-134.

10

View publication stats

You might also like