Development of a Beta Function Based Topology

Optimization Procedure
David G. Taggart1, Peter Dewhurst1, Lucian Dobrot1 and David D. Gill2
1

University of Rhode Island

College of Engineering
Kingston, RI 02881

Sandia National Laboratories

Albuquerque, NM 87185

Abstract: A novel finite element based topology optimization scheme has been developed and
implemented through the use of Abaqus user subroutines. The procedure is based on an iterative
material redistribution scheme in which the desired material distribution at each iteration is
imposed. A family of Beta probability density functions is utilized to provide a gradual transition
from an initial unimodal material density distribution to a bimodal distribution of fully dense and
essentially void regions. The efficiency and validity of the scheme is demonstrated through a
number of 2-D and 3-D test cases for which the optimal topology is known from analytic
optimality criteria. These test cases include classical minimum weight Michell structures as well
as newly derived optimal topologies for 3-D structures. The described method is the subject of
international patent application number PCT/US2006/062302.
To visualize the converged finite element results, procedures are developed to convert the final
bimodal material distribution data into contour surfaces of constant density. These contours are
then saved in a standardized CAD format (STL) that are imported into commercial CAD software.
The CAD models are used directly in rapid manufacturing equipment for the production of
prototype parts. Physical prototypes of optimized structures have been manufactured using Laser
Engineered Net-Shaping (LENS) and Dimension SST 3D-Printing. In addition to the test
cases for which optimal topologies are known from analytic optimality criteria, application of the
method to the design of an aerospace component will be presented.
Keywords: Design Optimization, Minimum-Weight Structures, Optimization, Postprocessing.

1. Introduction
Several finite element based optimization schemes have been demonstrated to determine minimum
weight structural topologies (1-4). Recent work at the University of Rhode Island (5,6) has led to
the development of schemes that have been implemented in Abaqus through the use of user
subroutines (7). In this paper, topology optimization using prescribed redistribution is shown to
provide a computationally efficient procedure. This scheme provides a robust optimization tool
that can be exercised within the Abaqus/CAE user interface.

2008 Abaqus Users Conference

2. Material Redistribution Scheme

In the topology optimization procedure, the desired final mass of the structure is specified at the
beginning of the analysis. This material mass is initially distributed uniformly throughout the
design domain resulting in a uniform, partially-dense material. All nodes are assigned an initial
relative density o = V f / V D , where Vf is the final structural volume and VD is the volume of
the partially-dense design domain. This initial distribution can be described by the probability
distribution function, fo, given by f o ( ) = ( 0 ) , where is the Dirac delta function and is
the relative material density (0 1). The corresponding cumulative distribution function, Fo,
is given by Fo ( ) = H ( 0 ) , where H is the Heaviside step function. The desired final material
distribution contains two distinct regions of fully dense material ( = 1) and regions that have zero
relative density (min <<1) with the fully dense regions representing the optimized topology. The
final material distribution can be described by the probability distribution function, ff, given by

f f ( ) = (1 o ) ( min ) + o ( 1)

(1)

and the corresponding final cumulative distribution is given by

F f ( ) = (1 o ) H ( min ) + o H ( 1)

(2)

A gradual transition from the initial distribution to the final distribution can be achieved through
the use of the beta function
f ( ) = ( , r , s ) =

r 1 (1 )s 1
B(r , s )

(3)

where r and s are adjustable parameters and

B (r , s ) =

(r ) (s )
(r + s )

(4)

where is the gamma function. The corresponding cumulative distribution function, also known
as the incomplete beta function, is given by
F ( ) = inc ( , r , s ) =

1
( ' ) r 1 (1 ' ) s 1 d '
B(r , s )

(5)

A non-dimensional time parameter, t, where 0t1, is introduced and appropriate functions r(t)
and s(t) are specified. The functions r(t) and s(t) are selected such that the total mass of material is
held constant and a smooth transition from the initial unimodal distribution to the final bimodal
distribution is achieved. One such family of distributions are shown in Figures 1 & 2.

2008 Abaqus Users Conference

10
9
8
7

f(x)

6
5
4
3
2
1
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 1. Transition from initial to final distribution

for t=0.1, 0.3, 0.5, 0.7 and 0.9.

1
0.9
0.8
0.7

F(x)

0.6
0.5
0.4
0.3
0.2
0.1
0

0.1

0.2

0.3

0.4

0.5
x

0.6

0.7

0.8

0.9

Figure 2. Transition from initial to final cumulative distribution

for t=0.0, 0.1, 0.3, 0.5, 0.7, 0.9 and 1.0.

2008 Abaqus Users Conference

At each finite element iteration, nodal densities are assigned based on the sorted nodal strain
energies computed from the previous iteration. Nodes with relatively low strain energy are
assigned reduced nodal densities and nodes with relatively high strain energy are assigned
increased nodal densities. Through direct assignment of nodal densities, the desired progression
of density distributions is enforced. In computing the element stiffness matrices, the nodal density
field is interpolated to give the Youngs modulus, E, at each Gauss point according to the relation
E = E d where Ed is the fully dense Youngs modulus. Convergence to the final topology can
be achieved in relatively few finite element iterations.

3. Two dimensional case studies

The 2-D implementation utilizes 4 node quadrilateral elements with full integration. A design
domain that extends beyond the expected optimal topology is selected. Four test cases as depicted
in Figure 3 are considered. The corresponding optimal topologies for each case are shown in
Figure 4. Case 1 is the well known center fan topology first obtained by Michell (8) for the case
of a simply supported beam with a single central load and a half-space design domain. Changing
the right hand support from a roller to a pin support (Case 2) produces a different topology. This
topology, with a mirrored structure about the horizontal axis, is identical to the simply supported
beam with a full design domain (Case 3); this solution also appears in Michells work (8). Case 4
is the well known Chan cantilever (9). All of these results are consistent with topologies obtained
through analytic considerations.
Case 2

Case 1

Case 3
Case 4

1.5

Figure 3. Two Dimensional Test Cases

4

2008 Abaqus Users Conference

Case 1

Case 2

Case 3

Case 4

Figure 4. Minimum weight topologies for 2-D test cases

4. Extension to three dimensions

The implementation of the topology optimization scheme in 3-D utilizes 8 node linear hexahedral
(brick) elements with full integration. Visualization of the final density field with the
Abaqus/CAE environment is achieved using View Cut and Display Group tools. For
improved visualization of the minimum weight topology and for the manufacture of prototype
components, a special purpose post-processing code was developed. This code reads model
definition and nodal density data from a file created by WRITE statements in the Abaqus user
subroutine, URDFIL. The post-processing code reads this data file and, using the marching cube
algorithm (10), generates a triangulated surface corresponding to threshold density, typically
2008 Abaqus Users Conference

=0.5, input by the user. The triangulated surface data is saved in STL format and is available for
import to standard CAD software, STL viewers, and rapid manufacturing equipment.
As a demonstration, a 3-D generalization of the 2-D Michell center fan topology (Case 1 above) is
investigated. In the 3-D case, the semi-infinite design domain is subjected to a concentrated
normal force with equally spaced radial roller supports along a circular region whose center is the
load application point. For the case of eight support points, the optimal topology can readily be
shown to comprise four Michell 180 arch structures intersecting at the pole. To model this
problem, the design domain is taken to be a cylinder whose radius and height exceed the radius to
the support locations. Using symmetry, the design domain can be taken as a 1/16 sector of this
cylinder (see Figure 5a). An axial concentrated force is applied at the corner point that represents
the center of the full cylinder domain. To model each of the 8 support locations, zero axial and
circumferential displacement conditions are imposed at a single point along a radial edge of the
domain. Finally, symmetry displacement boundary conditions are imposed along the two
symmetry planes such that the model represents one sixteenth of the actual full cylinder design
domain. The region was meshed using hexahedral elements as shown in Figure 5b.
The optimized topology as represented by the triangulated surfaces generated during postprocessing is shown in Figure 6. The results for the 1/16 sector are mirrored to provide the full
model topology. Good correlation with the expected topology is observed.

(a)

(b)

Figure 5. 3-D test case, a) loads and boundary conditions, b) finite element mesh.

2008 Abaqus Users Conference

Figure 6. 3-D test case numerically generated topology.

A second family of 3-D test cases is minimum weight tubular truss structures subjected to
combined axial and torsion load. Minimum weight topologies have been shown to consist of
members oriented along orthogonal helices (11). For pure torsion loading, the helical truss
members are oriented 45 with the tube axis. For pure axial loading, truss members oriented
parallel to the tube direction provide the minimum weight design. For combined loading with
axial load F and torque T, the optimal cylindrical structure layout comprises orthogonal families of
helices intersecting the axial direction at angles and (/2-), respectively, where

= cot 1 ((Fr / T ) / 2 ) / 2

(6)

The topology of such a structure is illustrated in Figure 7.

2008 Abaqus Users Conference

Figure 7. Helical spirals for combined axial and torsion loads (=/6).

For this problem, the design domain is taken to be a thick walled cylinder subjected to
concentrated forces at the ends of the cylinder given by fn in the axial and ft in the tangential
directions. The finite element model along with the optimized results for the case of pure torsion
(fn=0, ft=1) are shown in Figure 8. This figure shows the domain and boundary conditions, the
final density field and an image of the STL model created during post-processing. The STL
models for varying combinations of axial and torsion loadings are shown in Figure 9. The helix
angle, , was estimated from these models and compared to the theoretical optimum given in
Equation 6. As shown in Figure 10, the numerical scheme accurately predicts the optimum helix
angle for all cases.

Figure 8. Pure torsion, cylindrical domain numerically generated topology.

2008 Abaqus Users Conference

Figure 9. Results for combined axial and torsion loads (fn/ft=, 3.464, 2.383, 1.678,
0.727, 0.352 and 0)
45

Theoretical
theoretical

ATO obtained
Numerical
40

35

(o)

30

25

20

15

10
0

0.5

1.5

2
fn/ft

2.5

3.5

Figure 10. Comparison of theoretical and numerical optimum helix angle, .

2008 Abaqus Users Conference

Another interesting 3-D test case involves pure torsion loading with the design domain taken to be
of infinite extent. For modeling purposes, the infinite domain is approximated by a thick wall
cylinder (Figure 12) in which the inner and outer diameter extend well beyond the expected
minimum weight topology. As shown in Figure 11, circumferential forces are applied along a
circular path near the inner diameter of the cylinder. At the opposite end of the cylinder, nodes
along the corresponding circular path are constrained such that motion in the axial and
circumferential directions is prevented. The minimum weight topology was identified by Michell
(8) to consist of loxodromes, or spherical spirals, oriented at 45 with longitudinal lines as shown
in Figure 12. The numerical optimization scheme accurately predicts this topology (see Figure
13). Using the STL file generated by the post-processing code, a physical model was fabricated
on an Dimension SST 3-D printer and is shown in Figure 14.

Figure 11. Pure torsion, infinite domain - loads and

boundary conditions

10

2008 Abaqus Users Conference

Figure 12. Michells loxodrome for pure torsion.

Figure 13. Pure torsion, infinite domain - numerically generated topology.

2008 Abaqus Users Conference

11

Figure 14. Prototype component manufactured from STL file.

5. Application to Component Design

To illustrate application to an aerospace component design, the numerical optimization scheme
was applied to the design of an optical lens assembly shown in Figure 15. The particular
component selected for optimization is the housing. The housing is supported through three
flexures which are bonded to three mounting brackets. The loads on the mounting bracket consist
of six axial forces equally spaced around the perimeter of the lens support region and are
associated with contact with a wave spring. For simplicity, details of the support region design are
not considered and the optimization is applied to the cylindrical region of the housing.
For this analysis, the mass of the housing was held equal to that of the original design. Hence, the
optimized structure (Figure 16) is both stronger and stiffer than the original design. Using Laser
Engineered Net Shaping (LENS ), a prototype optimized component was manufactured from
the STL file as shown in Figure 17.

Laser Engineered Net Shaping and LENS are registered trademarks of Sandia National
Laboratories
12

2008 Abaqus Users Conference

Mounting
Bracket
Wave Spring
Housing

Figure 15. Optical lens assembly.

Figure 16. Optimized lens housing.

2008 Abaqus Users Conference

13

Figure 17. Prototype component manufactured from STL file.

6. Conclusions
A novel finite element topology optimization procedure is presented. This procedure is shown to
identify minimum-weight topologies for several 2-D and 3-D problems for which the optimal
topology is known from classical theoretical solutions. It is believed that this scheme provides an
efficient method for identifying optimal topologies for complex design problems. Since the
scheme has been implemented using user subroutines in Abaqus, application to general design
problems is simplified due to the robust user interface tools for both problem definition and
visualization of final results. Finally, the generation of STL format models of the optimized
structure provides visualization and rapid manufacturing capability.

14

2008 Abaqus Users Conference

7. References
1. Bendsoe, M. P., and Kikuchi, N., Generating optimal topologies in structural design using a
homogenization method, Computer Methods in Applied Mechanics and Engineering, 71,
197-224, 1988.
2. Bendsoe, M. P. and Sigmund, O., Topology Optimization, Springer-Verlag, 2004.
3. Xie, Y. M. and Steven, G. P., A simple evolutionary procedure for structural optimization,
Computers and Structures, 49, 885-896, 1993.
4. Xie, Y. M., and Steven, G. P., Evolutionary structural optimization, Springer, London,
1997.
5. Nair, A., Demircubuk, M., Dewhurst, P. and Taggart, D., Evolutionary techniques for
identifying minimum weight topologies and for the suppression of global instabilities,
ABAQUS Users Conference, May 2005.
6. Nair, A., Taggart, D. and Dewhurst, P. A novel evolutionary topology optimization
algorithm, in preparation.
7. Abaqus Analysis User Subroutine Reference Manual, version 6.6, Abaqus, Inc., Providence,
RI, 2006.
8. Michell, A. G. M., Limits of economy of material in frame-structures, Philosophical
Magazine, 6, 589-597, 1904.
9. Chan, A. S. L., The design of Michell optimum structures, Report No. 142, College of
Aeronautics, Cranfield UK, December 1960.
10. Lorensen, W. and Cline, H. E., "Marching Cubes: A High Resolution 3D Surface
Construction Algorithm", Computer Graphics - Proceedings of SIGGRAPH '87, Vol. 21, No.
4, pp. 163-169, 1987.
11. Dewhurst, P. and Taggart, D. G., Analysis of minimum weight symmetrical structures for
combined axial and torsional loading, in preparation for submission to Structural and
Multidisciplinary Optimization, 2008.

8. Acknowledgement
The authors would like to appreciation to Sandia National Laboratories, Albuquerque, New
Mexico, for support of this work. Sandia is a multiprogram laboratory operated by Sandia
Corporation, a Lockheed Martin Company, for the United States Department of Energy's National
Nuclear Security Administration under Contract DE-AC04-94AL85000.

2008 Abaqus Users Conference

15