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

net/publication/264177505

A Form-finding Technique for Three-dimensional Spatial Structures

Article in Architectural research · December 2013

DOI: 10.5659/AIKAR.2013.15.4.207

CITATIONS READS

5 6,287

1 author:

Sang Jin LEE

Gyeongsang National University
41 PUBLICATIONS 391 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sang Jin LEE on 20 March 2016.

The user has requested enhancement of the downloaded file.

 ARCHITECTURAL RESEARCH, Vol. 15, No. 4(December 2013). pp. 207-214 ISSN 1229-6163

A Form-finding Technique for Three-dimensional

Spatial Structures

Sang Jin LEE

ADOPT Research Group, Department of Architectural Engineering, Gyeongsang National University

http://dx.doi.org/10.5659/AIKAR.2013.15.4.207

Abstract A form-finding technique is proposed for three-dimensional spatial structures. Two-step discrete finite element (FE) mesh
generator based on computer aided geometric design (CAGD) is introduced and used to control the shape of three-dimensional spatial
structures. Mathematical programming technique is adopted to search new forms (or shapes) of spatial structures. For this purpose,
the strain energy is introduced as the objective function to be minimized and the initial volume (or the initial weight) is considered as
constraint function. Numerical examples are carried out to test the capability of the proposed form-finding techniques and provided as
benchmark tests.
Keywords: : From-finding, Spatial Structures, Design Optimization, Computer Aided Geometric Design

1. INTRODUCTION optimization of the spatial structures became important since it can

provide a way how to control the forms of spatial structures in more
The façade of tall buildings and the roof of the stadium structures general way. However, there have been a very few attempts on the use
(Culley and Pascoe, 2005) often appear to be free-forms in types of of CAGD in the form finding process for three-dimensional free-
spatial structure.1 With frequent appearances of free-form (Kim and form spatial structures although some successful uses of CAGD in
Lee, 2008), numerical analysis techniques are inevitably required in design optimization of continuum shells appeared (Bletzinger and
order to find its efficient form for the engineering problems where Ramm, 1993; Rao and Hinton, 1993; Lee, 2000).
large spatial structures are especially considered. In particular, the Therefore, we here propose a form-finding technique of three-
form-finding technique based on design optimization (Ramm et dimensional spatial structures based on CAGD. In particular, we
al., 1991; Bletzinger and Ramm, 1993; Rao and Hinton, 1993; Lee introduce new two-step FE mesh generator based on CAGD and
and Hinton, 2000; Lee, 2000; Lee and Bae, 2006; Bae and Lee, 2011) integrated it into the form-finding process. Finally, we provide a
is most desirable one to find the form of spatial structures which series of benchmark tests to show the performance of the proposed
meets some engineering requirements. form-finding technique.
Various research works on the design optimizations for spatial
structures (Ohsaki, 2010) have been carried out by many researchers. 2. GEOMETRIC DEFINITION
Size optimizations with weight minimization have been mainly
tackled in the early works on design optimization. The shape For successful form-finding process, it is necessary to prepare an
optimizations have been then gradually attempted to find efficient accurate geometric model of structures. The Coons patch
structural forms. In addition, it is found to be that the introduction representation (Coons, 1964; Anand, 1994) illustrated in Fig. 1 is one
of the CAGD (Computer-Aided Geometric Design) into the design of the most fundamental and effective parametric surface definitions.

Corresponding Author : Sang Jin LEE, Professor

Department of Architectural Engineering,
Gyeongsang National University
Tel : +82-55-772-1754 e-mail : [email protected]

The author is gratefully acknowledged for research grant from Gyeongsang

National University for his sabbatical leave to University of Cambridge
from September 2009 to August 2010.

This is an Open Access article distributed under the terms of the Creative
Commons Attribution Non-Commercial License (http://creativecommons.
org/licenses/by-nc/3.0/) which permits unrestricted non-commercial use,
distribution, and reproduction in any medium, provided the original work is
properly cited.
Figure 1. A parametric Coons patch representation
1
The term of spatial structure can be used for both continuum and discrete
structures. The Coons patch representation can be defined in a vector
©Copyright 2013 Architectural Institute of Korea.
208 Sang Jin LEE

form which has t hree s c a l ar comp onents and

:

where

where,

in which and are the values

of at the four corners of the Coons patch, is the The other tangent values and can also be
derivative of with respect to the point and determined by the same process. After calculating all the tangent
is the second derivatives at the and twist values using the above set of equations, all Coons patches
point . Therefore, the matrix can be divided into are now available to be used for interpolating the coordinates of the
four groups of sub-matrices as FE nodal points located within the Coons patches.

3. AUTOMATIC MESH GENERATION

The FE mesh generator is implemented by using the surface

definition described in the previous section. There are three major
steps in the present type of FE mesh generation.
Eq. (3) indicates that a Coons patch representation requires the
coordinate data and appropriate derivatives of the coordinates at 3.1 Forming the Coons patch surface
the four corners of a patch as shown in Fig. 1. The free-form surface has to be represented by the coordinate
A surface is usually expressed by an assembly of Coons patches data of the key points. The Coons patch can be formed as described
which is formed by cubic splines running in the and direction in Table 1.
as shown in Fig. 2. Table 1. The procedure how to form Coons patch

Figure 2. Assembly of Coons patch having (k+1) curves and (l+1)

curves in u- and v- directions respectively. The patch number
has been assigned under the assumption that i=j=2 and k=l=4.

Therefore, the corners of patches can be treated as the key points

of the splines and the derivatives can be evaluated in a manner The grid of Coons patches is prepared in a structured
similar to that used in splines where appropriate key points satisfy mesh data format. For example, a two-dimensional array
continuity thus avoids the possibility of having any kinks. If we can be used to store the coordinates
consider a -curve having a natural end condition at which of the key points in a two-dimensional grid of Coons patches.
passes through points the tangent values Therefore, the grid data of the Coons patches will have the lengths
can be determined by the matrix calculation and for the u- and v-directions respectively.
A Form-finding Technique for Three-dimensional Spatial Structures 209

3.2 Base grid pattern in parametric plane The generation procedure of double layer grid pattern in
The present automatic mesh generator uses a base grid pattern to parametric space is summarized in Table 2.
produce three-dimensional discrete spatial structures. Therefore,
the preparation of the base grid pattern in the parametric (u,v) Table 2. Double layer grid pattern generation
plane is required. Any form of base grid pattern can be used in the
present FE mesh generation process. However, a double layer of
grid patterns are used in numerical examples and explained here. In
this study, we use a three dimensional unit grid which has 8
members including 4 diagonal members as illustrated in Fig. 3. To
control the shape of double-layered grid structure in effective way,
we introduce a virtual mid-surface.2 In this study, as shown in Fig. 3,
all imaginary nodal points will be prepared in the mid-surface.

Note that all the information will be first prepared in

parametric plane and then mapped into the target surface.
Therefore, the grid information should be the subset of the
geometric definition prepared by using Coons patch.

3.3 Mapping the grid pattern into the real surface

Since the base grid pattern is a subset of the parametric grid of the
Coons surface, an identification process has to be carried out to
find a right Coons patch to which the nodal points belong. As
illustrated in Fig. 5, each nodal point of the base grid pattern is then
Figure 3. Unit grid for double layer grid pattern mapped onto the real surface defined by the Coons patch
which is already identified. Therefore, the coordinates of the nodes
Then, the corner nodes of unit square will be mapped into the top on virtual mid-surface can be calculated.
surface and the nodes at the center of unit square will be mapped
into the bottom surface by using normal vector of the mid-surface.
Then, we connect the nodes on the top and bottom surfaces with
diagonal elements and complete the double-layer unit grid with the
adding 4 elements on the top surface as shown in Fig. 3. The base
grid pattern for double layer spatial structure illustrated in Fig. 4
can be also created by the same way used in the unit grid.

Figure 4. Virtual mid-surface for controlling the shape

of double layer spatial structure
Figure 5. Mapping process (Lee, 2010)
2
The mid-surface will be used for the control of shapes of spatial structures
during the design optimization. In addition, top and bottom surfaces can be
Once, we get the nodal coordinates of virtual mid-surface, we
used instead of mid-surface.
have to map the node onto the top surface and other nodes
210 Sang Jin LEE

will be mapped onto the bottom-surface as explained in Fig. 4. Note small perturbation in the generating of the target surface or the
that we do not have to use the mid-surface strictly and therefore distance between top and bottom surfaces of double layer grid
the top and bottom surfaces can be also used to control the form of represented by Coons patches, the design sensitivities are evaluated
target spatial structures by finite difference (FD) method. The design sensitivities are
calculated in the following manner:
4. DESIGN OPTIMIZATION PROCESS
(a) Calculate the objective function and constraint
4.1 Problem definition functions associated with the initial set of
The optimization problem (Lee and Hinton, 2000) can be design variable using the process described in step 3
generally defined as: (b) Evaluate the design sensitivities with respect to design
variables

The Pseudo-code of the design sensitivity calculation is

summarized in Table 4.

5. Carry out the optimization: Having evaluated the design

sensitivities in step 4, the search direction can be determined and
where is the vector of design variables, then using the SLP, MDFM, SQP algorithm, a new design can be
is the objective function, are the inequality constraints, calculated. If the new design satisfies the convergence criteria then
are the equality constraints and and are lower and stop. Otherwise, update the design and goto step 2.
upper bounds of the design variable respectively. In addition,
is the number of inequality constraints, is the number of equality
constraints and is the number of design variable. Table 3. FE analysis for the objective and constraint functions
In this study, strain energy is used as the objective function
with a volume constraint. The program DoT (1999) is used as
the optimizer. In DoT, three methods are available for constraint
optimization: the modified feasible direction method (MFDM),
sequential linear programming (SLP) and sequential quadratic
programming (SQP) methods. In this study, all three methods are
used.

4.2 Optimization Algorithm

The basic algorithm of the design optimization based on Table 4. Pseudo-code for the design sensitivity calculation
mathematical programming (Vanderplaats, 1984) is described in
this section. Five main steps are involved:

1. Define the problem: The initial configuration and thickness

of free-form structure can be defined by using a Coons patch
representation. The movement direction of design variables must be
specified at this stage. For the FE analysis, the material properties,
loading and boundary conditions have to be prepared. Therefore,
the movements of the design variables will be determined by the
positions of key points.

2. Generate FE mesh: The two-step mesh generator described

earlier is used in this study. The FE mesh has to be prepared in the
parametric plane in either a single layer or double layer pattern.
In this study, the in-house code is implemented to generate the
parametric double layer grid pattern. The parametric grid pattern is
then mapped onto the surface representations which is prepared in
step 1.
4.3 Design Parameterization
3. Carry out FE analysis: A space truss FE analysis is carried We now consider the parameterization of the three-dimensional
out by using the standard truss FE. The objective and constraint spatial structure for the design optimization process. Recalls that
functions are calculated in this step. The linear static FE analysis is we are focusing on shape optimization and therefore the CAGD
carried out as summarized in Table 3. described earlier is of prime significance. The surface definition
described in the previous section can be utilized in various ways. In
4. Evaluate the design sensitivities: With the introduction of a shape optimization for spatial structure, the surface definition can
A Form-finding Technique for Three-dimensional Spatial Structures 211

be used to: whereas the design variables allow movement in the z-direction in
Fig. 7. Optimal shapes resulting from such restriction will vary.
• express the surface geometry as accurately as possible, Note that the design variables are linked in the longitudinal
• generate a discrete FE surface mesh with a two-step mesh direction in Fig. 6.
generator, and
• control the shape of spatial structure during the design
optimization iteration

However, the usage of surface definitions in shape optimization

requires careful consideration to achieve desirable optimal shape.
In the next sections, the selection of suitable design variables
is discussed in conjunction with CAGD and the linking, move
directions and perturbations of the design variables are discussed.

4.3.1 Selection of the design variables

The selection of the design variables is crucial in shape
optimization since the optimal shapes are highly dependent on it.
For the shape optimization, various quantities have been used as
design variables such as nodal coordinates, control points of splines
and key points of splines. In this study, the key point coordinates
of the Coons patches serve as the design variables in design
optimization. The mid-surface of free-form is represented by
Coons patches and the key points pass through the free-form mid-
surface and continuity is maintained at all stages of the shape Figure 7. Possible movement directions of the design variables.
optimization.
4.3.4 Perturbation
4.3.2 Linking of the design variables It is necessary to perturb the design variables by tiny amounts to
During the design optimization process, the linking of design calculate design sensitivities. Note that the key point coordinates
variables is necessary for several reasons: (a) it can preserve initial of the Coons patches serve as design variables and these values
types of structure such as dome and cylinder and (b) it guarantees a will be perturbed in turn during the design sensitivity calculation.
reduction in the number of design variables. Fig. 6(a) shows the After the perturbation of the coordinates, the Coons patches must
cylindrical shape of roof. There are a total of 49 key points at the be redefined. Re-meshing follows to update the FE mesh for spatial
junctions of the splines in the u and v directions or corners of the structure. Note that the parametric mesh will not be changed
Coons patch. In order to retain a prismatic shape during the design during the perturbation process so that the mapping of the initial
optimization process, linking rules are required. Only seven design parametric grid FE mesh data onto the redefined surface will take
variables to (the radial movements shown in Fig. 6(b)) are place during every perturbation. The effect of perturbation of a
adopted. Typically, design variable is the radial movement of all design variable is illustrated in Table 5.
key points in the associated longitudinal straight line shown in Fig.
Table 5. Effect of perturbation of a design variable on the cylindrical shell roof
6(b). This is achieved by variable linking.
with constant thickness.

Figure 6. Cylindrical shell roof: (a) geometry, (b) 6x6 Coons patches

4.3.3 Move directions

The move directions of the design variables have to be prescribed
for the optimization tasks. This is crucial data in shape optimization
process since the prescribed movements of the design variables
restrict the design space. Consider the cylinder shown in Fig. 6. The
design variables allow movements in the radial direction in Fig. 7
212 Sang Jin LEE

5. NUMERICAL EXAMPLES Table 6. The result of shape optimization for cantilever space truss

Ndv MDFM SLP SQP

In this study, numerical examples are carried out for the
verification of the form-finding technique described in previous 1 -0.6 0.6 0.0 -3.02e-2 -2.83e-1 -3.60e-1
sections. 2 -0.6 0.6 0.0 5.35e-1 9.05e-2 -2.40e-2

3 -0.6 0.6 0.0 6.00e-1 5.98e-1 4.53e-1

5.1 Cantilever space truss
A double-layer cantilever space truss subjected to a line load - - 2.13e-1 3.65e-1 4.83e-1
is optimized. The geometry of the cantilever space - the strain energy of initial configuration
truss is illustrated in Fig. 8. The spans of the space truss are
and its depth is . The following
material properties are used: elastic modulus and cross section area
.

Figure 8. Geometry of cantilever space truss

Figure 10. Optimum shape of cantilever space truss (x-z plan view)
A double grid pattern with the size of and is
used for automatic FE mesh generation. A Coons patch is The numerical results for the first test are presented in Table 7
used to represent the mid-surface of space truss. To control the and the optimum thickness profile of space truss in x-z plan is
shape or thickness profile of the space truss, the shape and thickness illustrated in Fig. 11.
design variables are assigned into the key point of Coons patch. The
position of design variables are described in Fig. 9. Table 7. The result of thickness optimization for cantilever space truss

Ndv MDFM SLP SQP

1 0.2 1.0 0.6 0.950 1.0 1.0

2 0.2 1.0 0.6 0.493 0.633 0.340

3 0.2 1.0 0.6 0.450 0.200 0.242

4 0.2 1.0 0.6 0.800 1.0 0.996

- - 1.0 0.615 0.516 0.604

Figure 9. A 3x3 Coons patches for cantilever space truss: (a) shape design
variables ( ) and (b) thickness design variables ( )

In this example, the effect of shape and thickness design variables

is separately tested with different cases. In the first case, three shape
design variables ( ) are tested and in the second case, four
thickness design variables ( ) are tested without shape design
variables. All design variables is allowed to move in the z-direction.
The strain energy of the space truss serves as the objective functions
with the initial volume constraint. Three different optimization
techniques MFDM, SLP and SQP are used to search an optimum
form of this structure.
The numerical results for the first case are presented in Table 6
and the optimum shape of space truss in x-z plan is illustrated in Figure 11. Optimum thickness of cantilever space truss (x-z plan view)
Fig. 10.
A Form-finding Technique for Three-dimensional Spatial Structures 213

5.2 Four corner support (FCS) space truss against the point loads. The initial geometry of the space truss, the
The FCS space truss subjected to point loads is side constraints and the optimum solution are described in Table 8.
optimized. The geometry of FCS space truss is illustrated in Fig. 12
where the locations of point loads are marked on the top chord. The Table 8. The result of shape optimization for FCS space truss
spans of FCS space truss are and its depth is Ndv SQP
. The following material properties are used: elastic
modulus and cross section area 1 0.0 1.2 0.0 -0.536
. 2 0.0 1.2 0.0 1.2

3 1.2 1.5 1.2 0.9

4 1.2 1.5 1.2 1.48

5 1.2 1.5 1.2 0.9

- - 1.0 0.8

The optimum shape of FCS space truss is illustrated in Fig. 14.

Figure 12. Geometry of FCS space truss

In the automatic FE mesh generation, the double-layer grid

pattern having the size of and is used. A
Coons patch is used to represent the mid-surface of entire space
truss. Both shape and thickness design variables are simultaneously
used in this form-finding process. The positions of the shape and
thickness design variables are presented in Fig. 13. Two shape
design variables ( ) are assigned into the key points except the
corners and three thickness design variables ( ) are assigned
to all key points. All design variables is allowed to move in the
z-direction.

Figure 14. Optimum shape of FCS space truss: (top-left) plane view,
(top-right) side view and (bottom) perspective view

6. CONCULSIONS

A form-finding technique based on design optimization

process with CAGD is proposed for three dimensional discrete
spatial structures. A new two-step mesh generator is successfully
Figure 13. A 3x3 Coons patches for FCS space truss: (a) shape design
variables ( ) and (b) thickness design variables ( )
integrated in the present form-finding process. The performance
of the present technique is tested by using numerical tests. From
numerical results, the present technique is successful to find new
The SQP is used to search an optimum form of FCS space truss forms of spatial structures with double-layer grid pattern and it can
214 Sang Jin LEE

be possibly applicable to the spatial structures with appropriate grid

patterns in the parametric plane.

REFERENCES

Anand, V.B. (1993) Computer graphics and geometric modelling

for engineers. John Wiley & Sons, New York.
Bletzinger, K.U. and Ramm, E. (1993) Form finiding of shells by
structural optimization, Engineering with Computers, 9, 27-35.
Bae, J.E. and Lee, S.J. (2011) A study on the shape control technique
for force density method, Proceeding of Symposium of
Architectural Institute of Korea, 31(2).
Coons, S.A. (1964) Surfaces for computer-aided design of space
figures. Mech. Eng. Dept., MIT.
Culley, P. and Pascoe, J. (2005) Stadium engineering. Thomas
Telford Ltd.
DOT design optimization tools user’s manual (1999) Vanderplaats
Research & Development, Inc.
Kim, J.S. and Lee, S.J. (2008) Structural design of free formed
structure. Review of Architecture and Building Science, 52, 4,
48-54.
Lee, S.J. and Bae, J.E. (2006) Shell design optimization technique
considering the appearance of close frequencies in optimization
process, The Proceeding of Symposium of the Korean Society for
Noise and Vibration Engineering.
Lee, S.J. and Kanok-Nuchulchai, W. (1998) A nine-node
assumed strain finite element for large deformation analysis.
International Journal for Numerical Methods in Engineering,
42, 777-798.
Lee, S.J. (2000) Shape and thickness optimizations of prismatic
shells using a simple sweep geometric model. Journal of
Computational structural engineering institute of Korea, 13(2),
221-230.
Lee, S.J. and Hinton, E. (2000) Dangers inherited in shells optimized
with linear assumptions. Computers and Structures, 78, 478-486.
Lee, S.J. (2010) Automatic FE mesh generation technique using
computer aided geometric design for free-form discrete spatial
structures. Journal of Korea Association for Spatial Structure,
10(2), 77-86.
Ohsaki, M. (2010) Optimization of finite dimensional structures,
CRC Press.
Ramm, E. and Mehlhorn, G. (1991) On shape finding methods and
ultimate load analyses of reinforced concrete shells. Engineering
Structures, 13(2), 178-198.
Ramm, E, Bletzinger, K.U. and Kimmich, S. (1991) Strategies in
shape optimisation of free-form shells, in Festschrift Erwin Stein,
Nonlinear computational mechanics - a state of the art, edited by
P. Wriggers and W. Wagner, Berlin, Springer.
Rao, N.V.R. and Hinton, E. (1993) Structural optimization of
variable thickness plates and free form shell structures, Structural
Engineering Review, 5, 1-21.
Vanderplaats G. (1984) Numerical optimization techniques for
engineering design. McGraw-Hill Book Company
(Received June 9, 2013/Accepted December 3, 2013)

View publication stats