Ho Đuc Dung -Volume 4 - Issue 4-2022, p.40-51.

Optimal design of geometrically nonlinear structures using

topology optimization
by Hồ Đức Dũng (Thu Dau Mot University)

Article Info: Received Jun 2nd, 2022, Accepted Oct. 9th,2022, Available online Dec.15th,2022
Corresponding author: [email protected]
https://doi.org/10.37550/tdmu.EJS/2022.04.359

ABSTRACT
This study deals with comparing the stiffness design of geometrically nonlinear
structures and linear structures using topology optimization. Bi-directional
Evolutionary Structures Optimization (BESO) is used for the design process. The
geometrically nonlinear behavior of the structures is analyzed using a total Lagrangian
finite element formulation and the equilibrium is achieved by Newton-Raphson
iterative scheme. The topology optimization of linear and nonlinear modeling is
performed. The sensitivity of the objective function is found with the adjoint method
and the optimization problem is solved using BESO’s update method. Objective
function of complementary work is evaluated. A special technique, the continuation
method, is applied to eliminate the instability of nonlinear structure optimization.
ANSYS APDL is also used to do FEA of optimal topology to verify the effectiveness of
geometrically nonlinear modelling. The results show that differences in stiffness of
structures optimized using linear and nonlinear modelling are generally small but it
can be large in some cases, especially structure highly involving buckling behaviour.

Keywords: BESO, linear and non-linear structure, topology optimization

1. Introduction
Topology optimization became one popular research subject in structure a few decades
ago. It finds not only structure design but also compliant mechanism design to meet the
best performance. The requirement for light-weight, low-cost and so on put topology
optimization in a high position. And, most of the work done is based on linearity behavior
which is not always valid for applications involving large deformation. That is the
motivation of this research.

40
 Thu Dau Mot University Journal of Science - Volume 4 - Issue 4-2022

Most of work done in topology optimization is based on linear behavior, assuming that
the structures with linear elastic materials undergo small displacement. Linear structures
can cover a large range of structural design problems. However, in many cases required
nonlinear solutions because of large deformation which energy absorption structures and
compliant mechanisms can be counted. And it can be called a geometrically nonlinear
structure.
There have been several prior works that considered geometrical nonlinearity in topology
optimization problems. It was introduced as a method for solving the topology
optimization problems of geometrically nonlinear structures and compliant mechanisms
(Brun & Tortorelli, 1998). The examples provided in the above-mentioned works were
not able to clearly show a significant difference in the converged topologies or values of
the objective function between linear and nonlinear modeling (Buhl, Pedersen &
Sigmund, 1999). With the examples provided, it showed that in many cases, the solutions
from the nonlinear modeling are only slightly different from the linear ones. However, if
snap-through effects are involved in the problems, the difference could be significant. It
proposed a microstructure-based design approach with a nonlinear FE formulation for the
topology optimization of structures with geometrical non-linearity (Gea & Luo, 2001). It
was considered topology optimization of non-linear compliant mechanisms represented
with frame elements (Pedersen, Buhl & Sigmund, 2001). An element removal and
reintroduction strategy for topology optimization problems with geometrical nonlinearity
were proposed (Bruns & Tortorelli, 2003). A level set-based topology optimization
method was developed for large deformation problems (Ha & Cho, 2008). BESO was
applied for topology optimization of geometrically nonlinear structures under both force
loading and displacement loading (Huang & Xie, 2008).

2. Geometrically Nonlinear Structures Optimization Method

All the methods for topology optimization used in this study are explained in detail here.
Firstly, a typical optimization method called BESO is clarified. Secondly, it is very
important to figure out the differences between geometrically linear analysis and
geometrically nonlinear analysis. Based on that issue, the finite element method affecting
the optimal topology using nonlinearity modelling is investigated (assuming that the
structures undergo large displacement with small strain).
2.1. Bi-directional topology optimization method
The original BESO appears to search for the minimum material volume subject to given
mean compliance or displacement. Topology optimization is often aimed at searching for
the stiffest structure with a given volume of material. In BESO methods, a structure is
optimized by removing and adding elements simultaneously. That is to say that, the
element itself, rather than its associated physical or material parameters, is treated as the
design variable (Huang & Xie, 2010). Thus, the optimization problem with the volume
constraint is stated as

41
Ho Đuc Dung -Volume 4 - Issue 4-2022, p.40-51.

1
M inimize : C =
T
F u
2
N (2. 1)
Subjec to : V −  V j x j = 0
*

j =1

0  x min  x j  1

where F and u are the applied load and displacement vectors and C is known as the mean
compliance. Vj is the volume of an individual element and V* is the prescribed total
structural volume N is the total number of elements in the analysis domain. The design
variable x j represents the density of individual elements limiting between prescribed x min
and 1. Notice that the material interpolation scheme (Bendsøe, 1989) has been applied.
2.2. Structural nonlinearity
2.2.1. Types of structural nonlinearity
Three main nonlinear behaviours are boundary nonlinearity, material nonlinearity and
geometrical nonlinearity.
Boundary nonlinearity is caused by independent displacement on the external boundary
condition. Contacts produce stresses and friction affected on changing in deformation.
The elastic material can generate nonlinearity behavior. The generated equation is
expressed as σ = E(ε)ε , in which Young’s Modulus is no more constant but proportional
with strains. Elastoplastic, viscoelastic and viscoplastic can be included in nonlinear
material behaviors.
In addition, geometrical nonlinearity is generated by nonlinear relationships in kinematic
quantities like strains and displacement. Large displacement, large rotation and large
strain are the roots of geometrical nonlinearity. In the linear model, displacements are
small so that the effect of geometrical changing can be ignored. But, when that change
becomes bigger, it has to be counted on the global stiffness matrix.
2.2.2. Incremental-iterative approach
Using the accumulated displacement, the resistant force (F) is obtained and the
unbalanced force ( t R − t F) , which is the difference between the applied and the resistant
forces, is determined. The iterative process at this load increment continues by calculating
a new tangent stiffness matrix, finding the displacement and the unbalanced force (Figure
). The equations used in the Newton-Raphson method can be stated as (Bathe, 2006).
t + t (it −1) t + t t + t it −1
u = R− (2. 2)
(it )
KT F
t + t t + t
 u (it ) = u it −1 +  u (it )
where Δt is a suitably chosen time increment and it denotes the iteration number of the
Newton-Raphson procedure in each time increment. The initial conditions at the start of
each time increment are:

42
 Thu Dau Mot University Journal of Science - Volume 4 - Issue 4-2022

t + t t + t t + t
= u; KT = KT; = F (2. 2)
z (0) t (0) t (0) t
u F

Convergence is achieved when both the errors, measured as the Euclidean norms of the
unbalanced forces and the residual displacements, are less than a minimum value. The
complete equilibrium path can be traced by finding the subsequent solution points at
higher load levels using the same approach.

Figure 1. Illustration of incremental Newton-Raphson approach (Abdi, 2015).

3. Results and discussion

Two numerical examples are given to perform the optimization procedure. BESO is
further developed to enable the topology optimization of geometrically nonlinear
structures undergoing large deformation. This is archived using total Lagrangian FE
formulation and an incremental iterative Newton-Raphson procedure to determine the
equilibrium solution at every evolutionary iteration. For example, the continuation
method and two-volume constraints are applied. The solutions for geometrically linear
modelling and geometrically nonlinear modelling are compared, and the necessity of
nonlinear modelling for the topology optimization of geometrically nonlinear structures
is investigated. Final optimal topologies are analyzed by ANSYS APDL.
3.1 Cantilever beam
Not many structures, which need to be optimized for stiffness, undergo large
displacements (Buhl, Pedersen & Sigmund, 1999). The first example considers designing
the stiffness of a slender cantilever beam. The initial dimension of a cantilever beam is
1*0.25*0.1m in length, width and thickness. Figure 2 shows the design domain of this
example. It is constrained at one end and free at the other, which is subjected to a
concentrated load in the middle. The covering volume of the final topology equals 50
percent of the initial domain volume. Material is linear elastic with Young’s modulus E=3
GPa and Poisson’s ratio υ = 0.4 . Also, an updated penalty method is employed in the
optimization process. BESO starts from the full design which is subdivided using a mesh
of 80*20 four-node plane stress elements. The BESO’s parameters are: evolution rate
ER=2 percent; rmin=75mm; convergence criterion: 0.1 percent.

43
Ho Đuc Dung -Volume 4 - Issue 4-2022, p.40-51.

Figure 2. Design domain and boundary conditions of the geometrically nonlinear

cantilever beam

Figure 2. Evolutionary history of complementary work and volume fraction using

BESO’s update scheme and linear modelling subject to load case of 60 kN

Figure 4. Evolutionary history of complementary work and volume fraction using

BESO’s update scheme and nonlinear modelling subject to load case of 60 kN

44
 Thu Dau Mot University Journal of Science - Volume 4 - Issue 4-2022

Several observations can be drawn from all these processes. In some initial iterations,
because the effect of ER, elemental densities are not classified clearly with BESO. that
effect can be seen in topology. It is noticed that the “optimal” topologies obtained for
linear modelling are symmetric. With nonlinear modelling, topology is not symmetric.
The purpose of optimal design is that the displacement of topology is as small as possible.
A comparison among four topologies is given in TABLE 1. Nonlinear displacement is
the most important value which is supposed to minimize.
TABLE 1. Comparison of the optimization results of linear and nonlinear modelling using
MMA and BESO at load case of 60 kN.
modelling Update Objective Linear FEA Nonlinear FEA Computation iters
scheme value (J) displacement (m) displacement (m) time (s)
Linear BESO 2327 -0.0778 -0.0776 25 83
Nonlinear BESO 2279 -0.0763 -0.0758 2504 124
For all the topologies both the complementary work and displacement (linear and nonlinear)
are listed. Generally, there is not much difference in this case. Nonlinear modelling can give
a bit better result compared with linear. The biggest error is lower than 0.5 percent.
Because of the slightly lower complementary work of nonlinear modelling compared with
linear, it could be argued that the difference is insufficient to justify the efficiency of
nonlinear analysis. The next example, which can figure out the important role of nonlinear
modelling essential, will be implemented.
3.2 Clamped beam
It can be seen that in the last example, the difference between linear and nonlinear modelling
is less than 5 percent. In order to emphasize the differences between linear and nonlinear
modelling, the example of the clamped beam which contains buckling behavior is given.
In addition, one more structural optimization model is considered. Figure 5. Design domain
and boundary condition of clamped beam problem subject to centre load show a beam
clamped at both ends. Concentration force is applied to the centre point of the top edge. The
full topology is 8 m long, 1 m in width and 0.1 m in thickness. The final topology covers 20
percent of the volume. Material is linear elastic with Young’s modulus E=3 GPa and
Poisson’s ratio υ = 0.4 . Here, the updated penalty method is not employed in the
optimization process. Volume constrains is 20 percent. BESO starts from the design which is
subdivided using a mesh of 320*40 four-node plane stress elements. The BESO parameters
are: evolution rate ER=2 percent; rmin= 75 mm; convergence criterion 0.1 percent.

Figure 5. Design domain and boundary condition of clamped beam problem subject to
centre load

45
Ho Đuc Dung -Volume 4 - Issue 4-2022, p.40-51.

Figure 6. Evolutionary histories of complementary work and volume fraction using

BESO’s update scheme and linear modelling
The main difference between linear and nonlinear topology is two horizontal beams along
the top edge of the domain. With a small load, it may be effective to resist the
displacement. It is easy to collapse with a bigger load. Figure6 and Error! Reference
source not found.7 show the evolutions of complementary work and volume fraction of
topologies.

Figure 7. Evolutionary histories of complementary work and volume fraction using

BESO’s update scheme and nonlinear modelling
TABLE 2. Comparison of the optimization results of linear and nonlinear modelling
using BESO
Modelling Update Objective Linear FEA Nonlinear FEA Computation iters
scheme value (J) displacement (m) displacement (m) time (s)
Linear BESO 45072 -0.2254 buckling 2576 129
Nonlinear BESO 38328 -0.2156 -0.1925 23184 42

46
 Thu Dau Mot University Journal of Science - Volume 4 - Issue 4-2022

Not like the last example, totally different designs are obtained from nonlinear topology
design comparing with linear topology design. Complementary work, which is the
objective function, for nonlinear designs is much lower than linear designs. It means that
the topology obtained using nonlinear modelling has a higher stiffness at the design load.
Nonlinear results with BESO consist of a triangle in the middle of the structure under
compression. With linear results, it also includes two horizontal bars on the upper edge.
The computational time required to obtain the nonlinear design is approximately seven
hours. It is necessary in order to arrive at the optimal design. This cost of time is caused
by a large number of elements and taking buckling behaviour into account.
A topology optimization procedure for the stiffness design of structures undergoing
geometrically nonlinear deformations has been proposed. In many cases, computational
time-consuming in geometrically nonlinear modelling is a big issue compared with the
difference in the objective function. But in some case like a clamped beam, the difference
in objective value may be large.
3.4 2D topology analysis using ANSYS APDL
Geometrically linear FEA and geometrically nonlinear FEA are implemented by using
ANSYS APDL. Because geometrically nonlinear FEA is more complicated than a linear
one, it is necessary to evaluate the reliability of the FEA solution using Matlab code. One
of the most popular commercial
topology software
FEA using chosenAPDL.
ANSYS is ANSYS APDL.
In order to evaluate solutions, all Save the optimal design with void and
the character of the modelling in solid elements.

ANSYS APDL is set similarly to its

modelling in Matlab code. Figure 8 Use matlab to create ANSYS
command, use elemental densities
shows the procedure of optimal input to define Young’s modulus.
topology FEA using ANSYS
APDL.
Open ANSYS APDL and execute
ANSYS command

Save and display output data of

ANSYS APDL

Figure 8. The procedure of8.

Figure optimal topologyofFEA
The procedure using
optimal ANSYS
topology APDL
FEA
using ANSYS APDL
The optimal topology of the cantilever beam is used as a test case to validate the developed
ANSYS simulation. All the parameter is set similarly to the designed cantilever beam. The
deflection of the beam is obtained from linear and nonlinear FEA using the developed
ANSYS command and the results are compared with those obtained from analyzing the
same problem using Matlab code. Figure .8 illustrates the optimal cantilever beam
deflection using ANSYS FEA for both linear and nonlinear geometrical modelling.

47
Ho Đuc Dung -Volume 4 - Issue 4-2022, p.40-51.

TABLE 3. Comparison of the linear and nonlinear response of the optimal cantilever beam
in Figure 9 using the Matlab FEA modelling and ANSYS FE modelling.
Linear Nonlinear
modelling modelling
Linear FEA Matlab FE -0.0778 -0.0763
modelling
ANSYS FE -0.0778 -0.0763
modelling
Nonlinear FEA Matlab FE -0.0776 -0.0758
modelling
ANSYS FE -0.0777 -0.0759
modelling

(a)

(b)
Figure 9. Illustration for the large deformation of cantilever beam subject to a designed
load of 60 kN with (a) linear modelling with BESO’s update method; (b) nonlinear
modelling with BESO’s update method

48
 Thu Dau Mot University Journal of Science - Volume 4 - Issue 4-2022

The is no difference in linear results. With nonlinear results, it can be seen the small gap.
The error between FE modelling using ANSYS APDL and Matlab is lower than 1 percent.

(a)

(b)
Figure 10. Illustration for the large deformation of a clamped beam subject to a designed
load of 400 kN with (a) linear modelling with BESO’s update method; (b) nonlinear
modelling with BESO’s update method

49
Ho Đuc Dung -Volume 4 - Issue 4-2022, p.40-51.

TABLE 4. Comparison of the linear and nonlinear response of the optimal clamped beam
in Figure obtained from the Matlab FE modelling and ANSYS FE modelling
Linear Nonlinear
modelling modelling
BESO BESO
Linear FEA Matlab FE -0.2254 -0.2156
modelling
ANSYS FE -0.2254 -0.2156
modelling
Nonlinear FEA Matlab FE buckling -0.1925
modelling
ANSYS FE buckling -0.1934
modelling

Because of the results for optimal topology in Figure 9(a), there is no result for nonlinear
FEA. With others, the result from ANSYS APDL is similar to Matlab. The error is less
than 2 percent. Thus, the approach introduced is reliable for the FEA of optimal
topology. There are many special behaviors of nonlinear modelling like locking,
buckling and so on.

4. Conclusion
In those implemented example with topology optimization, the stiffness of structure is
reduced slower than volume. The topology optimization results achieved from linear and
nonlinear modelling showed that, for the presented examples, the solutions achieved from
the optimization using non-linear modeling have a higher performance than those with
linear modeling. Although there is not a significant difference between the solutions
achieved from linear and nonlinear modeling in the first example of this study, nonlinear
modeling consumes much more computational time than linear modeling. It is almost 50
times higher. The displacement of topologies using nonlinear modeling is lower than 3
percent compared with linear modeling. The results from the second example case which
involves buckling (snap-through) effects, showed the importance of implementing
nonlinear modeling in large displacement problems. The topology using linear modeling
goes buckling when the applied force is lower 17 percent of the design load. And, the
topology using nonlinear modeling goes buckling when the applied force is higher 40
percent than the design load.
Nonlinear FEA using ANSYS APDL is necessary to evaluate optimal topology. And, it
is possible to use ANSYS APDL to build a finite element modeling. the difference
between ANSYS FEA results and Matlab FEA results is less than 1 percent.

50
 Thu Dau Mot University Journal of Science - Volume 4 - Issue 4-2022

References
Abdi, M. (2015). Evolutionary topology optimization of continuum structures using X-FEM and
isovalues of structural performance (PhD thesis). University of Nottingham, Nottingham,
United Kingdom.
Bruns, T. E., & Tortorelli, D. A. (1998). Topology optimization of geometrically nonlinear
structures and compliant mechanisms. Mechanical Science and Engineering, 1874-1882.
Buhl, T., Pedersen, C., & Sigmund, O. (1999). Stiffness design of geometrically nonlinear
structures using topology optimization. Structural and Multidisciplinary Optimization, 93-
104.
Bendsøe, M. P. (1989). Optimal shape design as a material distribution problem. Structural
Optimization, 193-202.
Bathe, K.-J. (2006). Finite element procedures. Prentice Hall, Pearson Education, Inc.
Bruns, T. E., & Tortorelli, D. A. (2003). An element removal and reintroduction strategy for the
topology optimization of structures and compliant mechanisms. Mechanical Science and
Engineering, 1413-1430.
Gea, H., & Luo, J. (2001). Topology optimization of structures with geometrical nonlinearities.
Computers & Structures, 1977-1985.
Ha, S., & Cho, S. (2008). Level set based topological shape optimization of geometrically
nonlinear structures using unstructured mesh. Computers & structures, 1447–1455.
Huang, X., & Xie, Y. (2008). Topology optimization of nonlinear structures under displacement
loading. Engineering Structures, 2057–2068.
Huang, X., & Xie, M. (2010). Evolutionary Topology Optimization of Continuum Structures:
Methods and Applications. Chichester, United Kingdom: John Wiley & Sons, Ltd.
Luo, Z., & Tong, L. (2007). Shape and topology optimization of compliant mechanisms using a
parameterization level set method. Computational Physics, 680-705.
Pedersen, C. B., Buhl, T., & Sigmund, O. (2001). Topology synthesis of large-displacement
compliant mechanisms. Numerical Methods in Engineering, 2683–2705.

51