National Autonomous University

of Mexico

Faculty of Engineering

“Evolutionary Structural Optimization:

Development
of a Computing Application for the Optimal
Design of Elements Under Plane Stress”

THESIS

To obtain the title of:Mechanical

Engineer

Presents:
Fernando Lopez Osorio

Thesis Director:
Dr. Fernando Velázquez Villegas
University City February 2011.
Assigned Jury:

President: Dr. HERIBERTO DE JESÚS AGUILAR JUAREZ

Vocal: Dr. FERNADO VELÁZQUEZ VILLEGAS

Secretary: Dr. LEOPOLDO ADRIÁN GONZÁLEZ GONZÁLEZ

1st. Alternate: Dr. Jesus Manuel Dorador Gonzalez

2nd. Alternate: MY ANTONIO ZEPEDA SANCHEZ

Place where the thesis was carried out:

University City, DF Mexico.

THESIS TUTOR:

Dr. FERNADO VELÁZQUEZ VILLEGAS

Acknowledgements

It is difficult to express the deep gratitude I feel towards all those people without whose
collaboration it would never have been possible to carry out this work, research and
development, on a fascinating subject in engineering.

I owe my parents, among many things, their constant effort and contribution to my
education and well-being.

To our beloved National Autonomous University of Mexico for the support provided for
the realization of this work through project IN108909 “Computational Tools for Optimal
Design in Engineering and Technological Innovation Projects” of the Support Program for
Research and Technological Innovation Projects (PAPIIT) of the General Directorate of
Academic Personnel Affairs (DGAPA). I would also like to thank our alma mater for
providing me with a space in the Faculty of Engineering, where I have learned solid
knowledge over the years.

To the Optimal Mechanical Design Group (GDMO), belonging to the CDMIT of the
Faculty of Engineering of the UNAM, directed by Dr. Fernando Velázquez Villegas and MI
Antonio Zepeda Sánchez, I owe among many other things, the opportunity they gave me to
work with them in structural optimization, this difficult discipline based on numerical
methods on the one hand and structural calculations on the other, areas of knowledge in
which both are great specialists, as well as the supervision and guidance of the work.

To my friends and colleagues in the optimization group who, in addition to offering me

their trust and friendship, provided me with diverse information, some calculation
subroutines, ideas and large doses of enthusiasm for the results obtained.

iv
Index
Assigned Jury:.....................................................................................................................3
Acknowledgements............................................................................................................iv
1.1 Justification of the proposal.....................................................................................7
1.2 Objectives of the thesis..........................................................................................10
Chapter 3...............................................................................................................................37
Implementation of the Algorithm.........................................................................................37
Chapter 4...............................................................................................................................50
Application and Results........................................................................................................50
Conclusions...........................................................................................................................74
References.............................................................................................................................76
4.1

Conclusions………………………………………………………………………………
Future work………………………………………………………………………….
Bibliographic References…………………………………………………………….

5
Introduction and objectives
“We are almost dwarves, sitting on the backs of giants. We
see more things than the ancients, and at a greater distance,
not because of the sharpness of our own sight or the height of
our stature, but because they support us and elevate us with
their gigantic stature.
Bernard of Chartres

Structural design should be seen as a creative approach to the constructive solution of an

engineering problem posed by the society to which it belongs. Throughout history, solutions
have been modified and, through interconnected or generational communication, improved.

Until recently, in the design of mechanical structures, only the limitations imposed by
materials had been discussed. However, a major limitation in the development of recent
structures is the difficulty in carrying out structural analyses. An event that marked the
breaking of this last limitation was the appearance of computers. These have become the
drivers of structural analysis, increasing speed and certainty in obtaining solutions.

Structural optimization aims to establish a methodology to systematically carry out the

structural design process, using current materials, computer tool technology and the
creativity of the engineer.

In the so-called “Traditional” design methods, the experience of the designer who uses
them is fundamental from the idealization stage to the manufacturing phase. The designer
must essentially make the following decisions [24]:

• Define the specifications of the problem.

• Choose appropriate design criteria.
• Model and evaluate.
• Perform calculations using techniques such as FEM.
• Interpret and evaluate the results of the FEM.
• Choose the best solutions and even propose improvements.

This design process fundamentally requires a great deal of experience on the part of the
designer, who must use it in almost all stages of the design process.

6
 The concept of optimization is at the basis of engineering, since the classic function of the
engineer is to design new, better, more efficient, and less expensive systems. The ability of
optimization methods to determine the best design without exploring all possible designs is
based on the use of a relatively modest level of mathematics and the task of performing
iterative numerical calculations, using clearly defined logical procedures or computationally
implemented algorithms.

Essentially, optimization methods consist of searching for certain values for certain
variables, so that, by fulfilling a set of requirements usually represented by algebraic
equations, they provide us with the best possible value for a function that is used to evaluate
the design.

1.1 Justification of the proposal.

Structural optimization is currently a field in constant evolution, in which significant

progress has been made in recent years, partly thanks to the increase in technological
capacity in the computational field, being today recognized as a practical design tool.

One of the main objectives in the industry is to reduce manufacturing costs without
affecting the response to the demands of the structure; that is why the search for optimal
designs becomes necessary in order to take full advantage of the properties of the materials.
To find these better designs, simulation tools are used to predict, modify and optimize
mechanical behavior.

The application of structural optimization in the different areas of engineering provides an

advantage to the industry in general, through these techniques it is possible to reduce the
number of analyses that are traditionally required, while obtaining an optimal design for the
imposed conditions. Below are some cases of structural design by applying optimization
methods.

In naval structures, the hull shape has been optimized in a preliminary stage to optimize
both the power required in calm waters and the behavior at sea using an evolutionary
algorithm that considers the change in the weight of the vessel, caused by the variation in
the shape of the hull [5]. The distribution of the holds and the master section, tanks, frame
clearance and navigation area of the vessel to be designed is also optimized [20].

7
 The application of optimization in aeronautical components is also currently used [6], an
example is the static and vibrational structural analysis of an aeronautical component that is
part of the structure of the vertical tail rudder of a commercial aircraft, as shown in Figure
I.1.

Fig. I.1. Aircraft breakdown, initial starting model (the red area is the surface
to be optimized) and optimized model [6].

In aerospace engineering, it can be found that in 1996, using genetic algorithms and
evolutionary computing, a new design was produced for a cargo-carrying boom that could
be mounted in orbit and used with satellites, space stations, and other aerospace construction
projects. The result, a twisted, organic-looking structure that has been compared to a human
femur (see Figure I.2), uses no more material than the standard arm design, is light, strong,
and far superior at dampening vibrations [8].

Fig.I.2 Initial information geometry, genetically optimized design (diagonal elements

omitted for clarity) [8].

8
 Fig. I.3 Complete optimized design and testing of the final product [8].

Due to the advantages of structural optimization, the UNAM Center for Mechanical
Design and Technological Innovation is creating the foundations and pillars of a new project
that investigates and implements optimization techniques. The creation of the Optimal
Mechanical Design Group (GDMO) allows the development of algorithms and computing
applications in order to provide the Faculty of Engineering with a new support tool in the
design process. As part of these bases, this thesis addresses an optimization method for
planar structures.

Throughout this work, a series of preliminary concepts of the optimization methods and
their resolution using finite elements as the analysis tool chosen for the optimal structural
design problem are presented.

1.2 Objectives of the thesis.

The main objective of this thesis is the development of a computer program for optimal
structural design under plane stress conditions. The optimization process is driven by the
Bidirectional Evolutionary Structural Optimization (BESO) algorithm, while the structure
analysis is performed with a fixed finite element mesh.

The main interest of this thesis is not focused on the study of optimization algorithms, nor
on the particular use of the finite element method, but rather on delving into the integration,
results, and perspectives of a computer program for the optimal design of continuous
structures under plane stress conditions.

The associated tasks contemplated in this thesis are the following:

• Create an algorithm that generates a mesh of triangular finite elements over an

9
established domain.

• Implement an algorithm that allows the analysis of two-dimensional continuous

structures using the finite element method.

• Implement the Bidirectional Evolutionary Structural Optimization algorithm and relate

it to the generated mesh.

• Apply the algorithm for topology optimization with Bidirectional Evolutionary

Structural Optimization, and fixed mesh in 3 case studies.

1.3 Personal goal.

Develop solid knowledge and generate capabilities for the application of concepts
associated with structural optimization and element analysis, as well as reinforced attitudes to
interact professionally.

1.4 Organization of the thesis.

The content of the work carried out in the thesis presented is briefly discussed below:

The first chapter presents the historical development of structural optimization from the
first approaches to the current state. The general ideas that lead to the concept of optimal
design are introduced, as well as the different existing methods for solving said concept.
The description of the state of plane stresses is also carried out.

Chapter two covers the topics of meshing types, the case of finite elements for plane stress
and details the importance and operation of the algorithm proposed for the structural
optimization used in this thesis.

Later in chapter three, three essential modules in the mathematical programming algorithm
for structural optimization based on fixed mesh, finite element analysis and optimization
concept are described in general terms.

In the fourth chapter, the validity of the approaches presented in the previous chapters is
contrasted by solving examples found in the basic literature on structural optimization.

Finally, the conclusions of this thesis and possible future lines of research in the field of
structural optimization are presented.

10
 Chapter 1

Chapter 1
Background
“Works are not built to last. They are built for a purpose or
function that leads, as an essential consequence, to the
construction being maintained in form and condition over time.
Its resistance is a fundamental condition; but it is not the only
purpose, nor even the primary purpose.”
[Torroja Miret Eduardo]

Since ancient times, builders or designers have optimized, to a certain extent, based
essentially on empirical knowledge, but their actions have been strongly limited by the lack
of a well-established methodology for structural optimization.

Historically we can find in the literature several precursors of optimal structural design
and some of them will be mentioned below chronologically:

Galileo's analysis of traction distribution in beams subjected to bending in the 17th

century, although incorrect, allows a rational approach to obtaining the optimal shape in
beams of variable thickness [14].

Half a century later, Johan Bernoulli applied the hypothesis of plane deformation of the
section and Hooke's law, postulating a linear distribution for the state of traction in the
orthogonal sections of the guideline, his theory allowed him to address the problem of the
design of beams of uniform resistance [14].

Parent in the first decade of the 18th century, discovered the concept of neutral fiber of the
section and solved the problem of designing beams of uniform resistance subjected to the
action of moving loads [14].

In the last third of the 18th century, Lagrange analyzed the elastic design of axisymmetric
columns of minimum weight subjected to axial loads [14]. His conclusions, although
incorrect, since he does not consider the action of the dead weight and, therefore, obtains a
cylindrical solution of constant section, opened the way that allowed Clausen in 1851 to
correctly solve the problem, obtaining for the combined action of centered loads and dead
weight, the exponential distribution of the area of the section as a function of the height
[14].

Maxwell makes the first major contribution to the construction of a theory of optimal
structural design. It demonstrates an important theorem relating to articulated structures

Page | 1
 Chapter 1

under a single load state, relating the lengths of the bars with the maximum tractions and
compressions allowed in each one [14].

Levy subsequently carried out an in-depth study on the design of trusses (lattice structures)
and arches of uniform resistance. His demonstration, which proves the necessarily isostatic
nature of the optimal configuration of a truss (openwork deck) subjected to a constant load
state, is of particular interest and subsequently generates numerous controversies analyzing
the possible extension of this result to other types of structural typologies, load states and
optimization criteria [14].

Michell presents a revolutionary contribution to structural optimal design techniques. His

theory, based on Maxwell's realizations, develops a powerful methodology applicable to
two-dimensional articulated structures subjected to a single load state, which allows finding
the configuration of minimum weight with restrictions on the maximum admissible values
of the tensions in the bars and, therefore, optimal in this sense. Michell's methodology
involves great operational complexity and, except in very simple cases, its application was
carried out until the appearance of the first computers [13].

The principles of optimal design were formulated in the period 1955-1960, and are mainly
due to Klein, Pearson and Schmit. Their respective fundamental contributions can be
summarized in the following terms:

• The formulation of the generalized structural optimization problem

• The integration of structural analysis and optimization into a single, coherent

design scheme.

• The use of general structural analysis methods (FEM) and the subsequent
application of structural optimization to problems with continuous two-
dimensional and three-dimensional definition domains.

During the 1970s Francavilla, Ramakrishnan, and Zienkiewicz proposed to characterize

the optimal shape, with the aim of minimizing the concentration of tractions, through
predefined geometric parameters.

In 1977 Oda presented a study where the optimal shapes corresponding to two planar
problems are obtained by introducing changes in some preselected finite elements [15].

Rodríguez and Sereig in 1985 introduced an algorithm based on FEM where the optimal
shape is achieved by maximizing the use of material [19].

By 1990, Mattheck and Burkhardt proposed an optimization method based on the analogy

Page | 2
 Chapter 1

between the geometry of the structure and the tree growth mechanism with the aim of
minimizing stress concentrations [12].

Xie and Steven presented a method called ESO (Evolutionary Structural Optimization) in
1993, which through a simple iterative process removes the least efficient material from the
design [27].

Bendsoe and Kikuchi also in 1993 developed the homogenization method in which a
material model with small cavities is introduced into the design, solving the optimal design
problem by determining the ideal porosity [2].

Currently, algorithms based on the process of natural selection and biological evolution
(evolutionary algorithms) are confirmed as the most powerful and robust methodology for
optimal design [26].

This chapter describes the optimization algorithms for the optimal design problem in
continuous structures, these algorithms are fundamental for the completion of this thesis.

1.1 Structural optimization.

Evolutionary optimization techniques constitute a set of emerging heuristics, successfully

used to solve a wide range of problems in the areas of optimization, artifact design,
information search, device control and machine learning, among others. These techniques
base their operation on the emulation of the mechanisms of natural evolution, identified by
Charles Darwin in 1859 in his famous work The Origin of Species by Means of Natural
Selection: natural selection, reproduction and genetic diversity of individuals.

In nature, during the evolutionary process, living beings try to solve problems related to
survival to ensure the perpetuation of the species. Evolutionary optimization techniques
emulate the biological process of adaptation of living organisms to their environment and
environmental conditions, applying it to problem solving in various areas.

Structural optimization theory consists of a set of optimization algorithms and numerical

methods focused on finding the best candidate among a collection of alternatives, without
having to explicitly evaluate all of those alternatives. We will then define optimization as:

“Optimization is a process through which one seeks to obtain or determine the best
possible solution, from among a group of results, which in turn satisfies the restrictions
presented by the system to which it is applied” [9].

Mathematically, optimization means:

Page | 3
 Chapter 1

“The search for the maximum or minimum of an objective function” [1].

In the general optimization problem, expressed as a mathematical programming problem,

both the objective function and the restrictions are formulated as a function of structural
behavior variables, such as displacements or tractions, which are a function of the geometric
or mechanical properties of the structure and the loads, which in turn are a function of the
design variables. Since these properties and loads can be known as a function of the design
variables, through a certain method of structural analysis the value of the behaviour
variables can be calculated and, therefore, the value of the objective function and the
constraints can be obtained as a function of the design variables. However, structural
behavior variables are, in general, highly non-linear functions of the design variables, and
their expression is usually unknown, although they can be calculated point by point.
Whenever structural calculations take a long time, and especially in structural systems with
a greater number of design variables, the operating cost of the successive analyses that must
be carried out in the design process may be unacceptable.

In mathematical terms, the optimization problem can be formulated as: find the vector of
design variables x that;

minimize f(x)

subject to hj(x) = 0 j = 1,...,mi

gk (x) ≥ 0 k = 1,...,md

x¡ ≤ x ≤
xi i = 1,...,n

Where

x n-dimensional vector of design variables

f (x) objective function, represents the optimal criterion
hj(x) equality design constraint j
gk (x) k-inequality design constraint
my number of equality constraints
md number of inequality constraints
xi lower limit of design variables Yo
xi upper limit of design variables Yo
xi design variable i
n number of variables

Page | 4
 Chapter 1

The n-dimensional space defined by the vector of variables x is called the design space.
Constraints define hypersurfaces that bound an environment in the design space. A set of
design variables defines a point in the design space. If a point is such that all constraints are
met, that point is a valid or feasible design; if, on the contrary, it violates any constraint, the
point in the design space corresponds to an invalid or infeasible design [18].

In order to obtain better structural designs, various optimization and approximation

methods of structural behavior have been proposed. In this work, some of them are
mentioned and the foundations necessary for the development of this thesis are detailed.

1.1.1 Design variables.

The different types of structural optimization depend on the design variables used. In the
most general case, four types of variables may intervene:

• Properties of the section of the piece: area, moments of inertia, and torsion, etc.
• Geometry of the structure: height, length, thickness, etc.
• Topology of the structure: material density, total perimeter, etc.
• Constitutive properties: modulus of elasticity, Poisson's ratio, etc.

Of the four types of variables, the material of the structure is usually considered as a
parameter, being initially set by the designer. Currently there are no optimization techniques
that can efficiently consider the four types of design variables. In this way, the most
common design variables that are handled are: the properties of the section, the geometry,
and the topology of the structure.

Variables can be classified as continuous and discrete. For the shape optimization of
continuous structures, geometry variables are usually associated with the position of control
points of curves or surfaces, which define the contours of the structure. In discrete
structures, the geometry variables are usually the coordinates of the nodes of the structure
[10].

1.1.2 Objective function.

The objective function is a part of great interest since it is through which designs can be
evaluated, so that we can identify which is the best design. The usual condition required of
this function is that, for the optimal solution, it takes a minimum (or maximum) value.

The objective function that has been used most frequently is the mass of the structure,
because the first structural optimizations were developed in the field of aeronautics, in

Page | 5
 Chapter 1

which the value of the weight is an essential condition. The mass is directly related to the
total cost of the same. However, in the evaluation of the total cost, aspects such as
manufacturing costs, labor, minimum production time, maximum production rate, minimum
energy used, maximum load, natural frequencies, maximum traction, displacements at
critical points, structural rigidity, etc. come into play. These aspects can also be used as an
objective function.

1.1.3 Restrictions.

Practical cases should be formulated as problems with constraints. This is because most
complex problems, such as the design of a structure, cannot be treated freely taking into
account all possible choices, but must be broken down into separate subproblems. Where
each of these subproblems has restrictions that have been imposed to limit the field of study.

Restrictions can be classified into two groups, in the first of which restrictions are divided
into explicit or implicit. The former act directly by imposing conditions on a variable, while
the latter impose conditions on magnitudes that in turn depend on the variables.

The second classification of restrictions groups them into equality or inequality

restrictions. Equality constraints are usually associated with the relationships that determine
the behavior of the structure, equilibrium conditions, compatibility, material behavior law,
symmetry conditions, tangency, curvature, etc. Inequality constraints are associated with
limitations imposed on the response of the structural element, such as stresses, maximum
displacements, vibration frequencies, etc. [25].

1.2 Structural optimization methods.

The most common design variables that are used to perform optimization cases have
already been mentioned above, and from these, the three types of structural optimization that
exist originate.

• Dimensional or property optimization. In property optimization cases, the design

variables are associated with geometric properties of the cross section of the
elements that make up the structure (areas of bars, moments of inertia, etc.). In the
case of continuous structures, the design variables are usually the thicknesses of
the elements.

• Shape optimization, also called geometry optimization. In cases of shape

optimization, the design variables control the geometry of the design and often
require an analysis model to be re-adapted during the optimization process. This

Page | 6
 Chapter 1

process consists of varying some contours of the model to be designed in order to

improve its mechanical behavior.

• Topology optimization. For cases of topological optimization, not only the

dimensions or position of characteristic points are modified, but also the existence
or not of structural elements, that is, the definition of the number of elements (or
cavities) that make up the structure. In many problems it is convenient to introduce
interior cavities that do not exist in the initial topology design, through which it is
possible, for example, to reduce the weight of the structure without necessarily
violating any of the imposed restrictions.

There are several optimization methods that are used depending on the design parameters.
These methods are classified into two categories: Gradient-based methods and Heuristic
methods.

Page | 7
 Chapter 1

Gradient-based methods: are those that make use of differential calculus to solve an
objective function and provide a minimum or maximum value for a specific function. In this
method it is assumed that the solution exists, that is, that there must be continuity in the
problem and the objective function.

• Unrestricted optimization.
• Constrained optimization.
• Lagrange Multiplier Method.
• Kuhn-Tucker conditions.
• Linear Programming (LP).
• Integral Linear Programming (ILP).
• Sequential Linear Programming (SLP).
• Nonlinear Sequential Programming (NLP).
• Homogenization Methods.
• Optimal distribution theory.
• Shape Optimization.

Heuristic optimization methods: are search methods across the field of design, developed
from any intuitive idea within the problem, or from convincing arguments of optimization
methodologies based on the observation of nature, that is, these methods do not involve
derivatives of the objective functions. Genetic algorithms require a greater theoretical basis,
which makes them more suitable for implementation in diverse cases. It must be taken into
account that the formulation of a genetic algorithm is independent of the problem, which
makes it a robust algorithm, as it is useful for any problem, but at the same time weak, as it
is not specialized.

• Annealing Simulation (SA).

• Genetic Algorithms (GA).
• Computer Aided Shape Optimization (CAO).

1.2.1 Evolutionary Structural Optimization (ESO)

Evolutionary structural optimization is based on the fact that, through a progressive and
slow elimination of the least efficient material from the structure, its shape (and/or
topology) evolves towards the optimum.

Xie and Steven (1993) introduced the ESO method, which, using a rejection criterion,
removes a small amount of material at each iteration. In its original form, the ESO method
determines the efficiency of the material from the stress in the elements. That is, if an
element presents a minimum stress (generally Von Mises), the Page | 8
 Chapter 1

element is inefficient. From an engineering point of view, the method presents the following
advantages: ease of programming, and rapid coupling with finite element programs.

The ESO method is between the heuristic method and the gradient-based method. This
means that ESO can search for solutions at the structural level, both locally and globally.
Due to its evolutionary characteristics, when a minimum has been located, the process
continues to evolve the structure in search of the best solution.

The Evolutionary Structural Optimization (ESO) method follows a concept that is very
simple and robust [18,27]. The principles and procedures that define the ESO method are as
follows:

1) Establish a mesh that completely covers the maximum domain that the structure is
allowed to occupy.

2) Apply all kinematic boundary conditions, loads, materials, element properties, etc.

3) Specify the criteria used to optimize the structure, for example, Von Mises.

4) Specify the driving ESO parameter, for example, the maximum or average Von Mises
stress.

5) Perform a finite element analysis of the structure.

6) Using the ESO inequality, ( )

maxuvm e—¿ -n
DD ^
uvm,e

RR = a0 + asS + a2 ss2

where
Von Mises stress for element e.

Von Mises's maximum effort in the elements.

V772.0

Rejection rate.

C0,d1,d2 Coefficients determined from the

experience with ESO.

H.H Number of local optima.

Page | 9
 Chapter 1
Determine whether there are elements in the structure that satisfy it. If an element satisfies
this equation, that element will be removed from the structural domain.

7) If in an iteration there is no element in the design domain that satisfies the equation in
step (6), a stable and local optimum state has been reached. For the optimization process to
continue, the steady state (SS) number must be increased by 1, and step (6) repeated.

8) Repeat steps 5 through 7 until the stopping criterion is met.

The stopping criterion used in the explanation of this method is the Von Mises stress;
however, any criterion can be used to drive the optimization of a structure using the ESO
method. The important aspect is to understand that no matter what criterion is applied, the
ESO method does not change, only some parameters.

The flow chart representing this ESO method is composed as follows.

Page | 10
 Chapter 1

Fig. 1.1 Flowchart representing the ESO method

Page | 11
 Chapter 1
1.2.2 Additive Evolutionary Structural Optimization (AESO).

In additive evolutionary structural optimization (AESO), as its name indicates, elements

are introduced into the structure in the areas where they are needed (Querin, Steven, Xie,
2000), similarly to the elimination of elements in the ESO method.

The steps that define the AESO method are the following:

1) Specify the maximum volume that the structure can occupy.

2) Divide the domain using a finite element mesh.

3) Apply loads and support conditions.

4) Set the material properties.

5) Provide a minimum number of elements capable of connecting the structure supports

with the applied loads.

6) Define the criteria to be optimized. Generally, the Von Mises tension.

7) Analyze the response of the structure.

8) Add the elements that satisfy the AESO inequality.

( )
e - ID ^max -vm — 1 n IR = a0 — aSS — a Ss2
uv?n,e

Where:
Von Mises stress for element e.
c(e) vm
max Von Mises's maximum effort in the
Ovm,e elements.

GO Rejection Rate.

do , d , d2 Coefficients determined from experience

with AESO.

H.H Number of local optima.

9) If no new elements are added in an iteration, the value of the SS number is increased
by one unit.

10) Repeat steps 7 through 9 until the stopping criterion is met.

Page | 12
 Chapter 1

Fig. 1.2 Flowchart representing the AESO method

Page | 13
 Chapter 1
1.2.3 Bidirectional Evolutionary Structural Optimization (BESO)

In the two-way evolutionary structural optimization method, structural elements can be added or
removed. The addition and elimination processes are carried out using the ESO and AESO methods.
In order for the BESO method to work correctly, the equations leading to optimization must be
modified.

Optimization using BESO follows the following steps:

1) Specify the maximum volume that the structure can occupy.

2) Divide the domain using a finite element mesh.

3) Apply loads and support conditions.

4) Set the material properties.

5) Define the criteria to be optimized. Generally, the Von Mises tension.

6) Perform finite element analysis.

7) Remove the elements that satisfy the ESO inequality, and add the elements that satisfy the
AESO inequality. ( )
e ¿ PD ^r-Tnaxuvm
-nn (uvm,e
e)
HIP max
uvm — 1 - uvm,e

Where:
Von Mises stress for element e.

Von Mises's maximum effort in the elements.

RR Rejection rate

RR = ro + r * SS + aRR * ON0 (1.1)

< RR < 1

It is the number of stable states, a counter indicates all

the stable states in the evolution process, as well as the
H. local minima.
H

Page | 14
 Chapter 1
It is the oscillation number, a counter increments it by
one each time the optimization reaches an oscillatory state. It is defined as an
oscillatory state when an element is attached and the same element is removed in
the subsequent iteration and continues this pattern to infinity. The increase in this
variable each time an oscillatory state is reached allows
ON
the structure to continue its evolution.
=0

Constant determined with the experience of

KISS, usually equals 0.001
r0

aRR It is an oscillation number constant

r1 determined by BESO experience, typically 0.01

GO Addition rate

IR = io — i * Ss — aIR * ON (1.2)

O < IR < 1

10 =1

Constant determined with the experience of

11 KISS, usually 0.01

aIR Oscillation number constant, determined

from the experience of BESO

7) If a state is reached where there are no elements in the design domain that satisfy equations
(1.1) and (1.2), a stable state has been reached. To continue with the optimization process,
increase the steady state number by one (SS), and repeat step 7.

8) If an oscillatory state is reached before reaching the chosen optimization criterion, increase
the oscillatory state (ON) value by one unit and repeat step 7.

10) Repeat steps 6 through 9 until the optimization criteria(s) are met.

Page | 15
 Chapter 1

Select the optimization criteria

Fig. 1.3 Flowchart representing the BESO method

Page | 16
 Chapter 1

1.3 Finite Element Method (FEM).

The finite element method (FEM) is a numerical method for solving differential equations.
The method is based on dividing the body, structure or domain (continuous medium) on
which certain integral equations that characterize its physical behavior are defined, into a
series of non-intersecting subdomains called finite elements. This set of finite elements forms
a partition of the domain also called discretization. Within each element a series of
representative points called nodes are distinguished. The set of nodes considering their
adjacency relationships (connectivity) is known as a mesh. The calculations are performed on
a mesh or discretization created from the domain with mesh generating programs, in a stage
prior to the calculations called preprocessing. According to the connectivity relationships, the
value of a set of unknown variables defined in each node and called degrees of freedom is
related. The set of relationships between the value of a certain variable between the nodes
can be written in the form of a system of linear (or linearized) equations, the matrix of such a
system of equations is called the system stiffness matrix (K matrix). The number of equations
of such a system is proportional to the number of nodes [16].

Knowledge of what is happening inside this approximate body model is obtained by

interpolation of the known values at the nodes. It is therefore an approximation of the values
of a function based on the knowledge of a determined and finite number of said nodes.

Typically, the FEM is programmed computationally to calculate the displacement field and
subsequently, through kinematic and constitutive relations, the deformations and tractions
respectively, when it is a problem of mechanics of deformable solids or more generally a
problem of mechanics of the continuous medium. The finite element method is widely used
due to its generality and the ease of introducing complex calculation domains. Table 1.1 lists
some of the applications of this method [25].

Table 1.1 A Applications of the finite element method in engineering,

Fields of Study Balance Eigenvalues Spread
Structural engineering. Analysis of beams, Stability of structures. Propagation of stress
plates, sheets. waves.
Mechanical Engineering. Natural frequencies and
Two-dimensional vibration modes of Dynamic response of
Aerospace engineering. and three- structures. structures.
dimensional stress
analysis. Viscoelastic problems.

Page | 17
 Chapter 1

Soil mechanics. Two-dimensional Natural frequencies and Propagation of stress waves

and three- vibration modes of soil- in deformable porous
Foundation engineering. dimensional stress structure combinations. media.
analysis.
Dynamic soil-structure
Excavation problems interaction.
soil-structure
interaction.

Finally, an important property of the method is convergence: if successively finer finite

element partitions are considered, the calculated numerical solution converges towards the
exact solution of the system of equations.

1.3.1 Overview and steps for Finite Element analysis

The first difference between classical methods and finite elements is the way of viewing the
structure and the subsequent solution procedure. The finite element method considers the
structure as the assembly of a finite number of elements.

Typically, finite element analysis involves seven steps [4]. Steps 1, 2, 4, 5 and 7 require
information to be provided and decisions to be made, while the rest of the steps are
performed automatically by the algorithm.

1. Discretization of the structure: As part of the preprocessing, the structure is

divided into a finite number of elements. These elements generate a mesh that will
allow the mathematical management of the structure in question. This step is one of
the most crucial to obtain an exact solution to the problem; thus, the designer must be
very aware of determining the size or quantity of elements in a certain area or volume
of the structure for analysis and not generate unnecessary calculations or erroneous
solutions.

2. Define element properties: In this step the designer must define the
properties of the element.

3. Assembling element stiffness matrices: The stiffness matrix of an element

consists of coefficients which can be derived from equilibrium, weighted residuals, or
energy methods. The element stiffness matrix refers to the nodal displacements when
forces are applied at the nodes (K*F = U). The assembly of the stiffness matrices
implies the application of equilibrium to the entire structure.

Page | 18
 Chapter 1

4. Application of loads: Concentrated external forces or uniform forces and

moments are specified in this step.

5. Define boundary conditions: Support conditions must be given, for example,

if the displacement of certain nodes is known. Using the boundary elements, the
reactions in them can be determined.

6. Solving the system of linear algebraic equations: The sequential application

of the steps described leads to a system of simultaneous algebraic equations, where
the nodal displacements are unknown.

7. Calculate stresses: The designer can then calculate stresses, reactions,

deformations or other relevant information. The post-processor helps to visualize the
output in graphical form.

1.3.2 Types of Finite Elements.

A large number of types of finite elements can be used in the solution of mechanical
structures and applications. An example of the above is that in commercial finite element
programs a large number of elements can be found in their libraries. However, most
applications in the areas of structures, naval engineering and mechanics can be solved with
the basic elements indicated below.
Table 1.2 Types of finite elements. [29].

• Truss.
• Beam (viga). One-dimensional elements. (Line)
• Boundary

• Plane stress.
• Plane strain.
• Axisymmetric (Elements Two-dimensional elements. (Area)
Axisymmetric).
• Membrane
• Plate
• Shell (shell type)
• Solid or brick (solids)
• Tetrahedron
Three-dimensional elements. (Volume)
• Hexahedron

Page | 19
 Chapter 1

As shown in the Table. 1.2 Depending on the dimension, the basic elements can be divided
into three categories: line element, area and volume. A classification of the most commonly
used elements is shown in Table 1.3.

Table 1.3 Most commonly used finite elements

Type of Main uses Condition for use Illustration
element

TRUSS Roof structure (truss). Length much greater than

High voltage towers. his broad and
Bridge Structure. Etc. depth (≥ 8 to 10 times).

Does not allow moment

transfers.
u, A
z A , u2 "u
p/ v' x
GdL: Six degrees of freedom per node.
Translations: u1, u2, u3.
Rotations: u4, u5, u6.

BREAM Reinforced plates. Length much greater than

Props. his broad and 3
p 1 ug A éD J ua
Mark. depth (≥ 10 z
N I. • —-
Etc. times). ,2 U5 K 2
4
Allows moment and shear p And
transfers. X

GdL: Six degrees of freedom per node.

Translations: u1, u2, u3.
Rotations: u4, u5, u6.

PLATE Container. (h / L) ≤(1 / 10)

Ship hull.
Aircraft fuselage. h = Plate thickness.
Etc. c.
L = Dimension of the ua A
shorter side of the plate. u, gus'
and
z

GdL: Five degrees of freedom per node.

Translations: u1, u2, u3.
Rotations: u4, u5.

In the direction normal to the plate a very

small rotational stiffness must be specified.

Page | 20
 Chapter 1

~u,
BRICK Thick plates. (h / L) >(1 / 10)
Engine fundamentals. u2_____
Flanges. h = Thickness of the
Gears. piece. ugh ■
Etc. Z II-IE
L = Dimension of the
smaller side of the piece. and- - - AND
Ax

GdL: Three degrees of freedom per node.

Translations: u1, u2, u3.
Rotations: It does not present rotational
degrees of freedom.

1.3.3 Two-dimensional elastic elements

There are three types of two-dimensional elements:

1. Plane Stress Elements.
2. Plane Strain Elements.
3. Axisymmetric Elements.

The plane stress state is described below, whose characteristics are important for the
solution of the finite element used in the algorithm of this thesis.

1.4 Description of the State of Plane Stresses.

The state of stress in two dimensions, that is, biaxial, is also known as plane stress. Plane
stress requires that one principal stress be considered equal to zero. This is an important state
of stress because it represents the physical situation that occurs at a stress-free point on the
surface of a body.

To understand this, a body (continuous medium) subjected to a system of forces in static

equilibrium (with zero speed) is taken into account; in figure 1.4 it can be observed that it is
subjected to three equilibrium conditions: external equilibrium, internal equilibrium and
equilibrium between internal and external forces.

Page | 21
 Chapter 1

F1,F2,F3,F4 external forces FI/D,FD/I, Internal force resultant

Balance of external forces Balance of internal forces ∑Fi =
∑Fe = F1+F2+F3+F4 = 0 FI/D+FD/I=0

Fig.1.4 Balance between external forces and internal forces ∑Fe+∑Fi=0

[30].
Forces, whether body or contact, represent stresses on the continuous medium; their effects
will obviously depend on their magnitude and direction, but also on the geometric conditions
of the body. For this reason and to facilitate the analysis of their effect, it is necessary to
describe the area on which they occur, through which the concept of effort is defined. The
resultant force FI/D in Figure 1.4 is initially represented at the centroid of the area described on
the cutting plane; however, it is evident that the load is distributed over surface A.

Fig.1.5 ∆P resulting from the internal force applied to the area ∆A [30].

On the other hand, it is evident that ∆P can be broken down into its normal component on the
surface and its tangential component, in such a way that the so-called normal stresses and the
tangential or shear stresses can then be defined. Likewise, the reference coordinate system (x,
y, z) is also taken according to said normal.

The intensity of the forces per unit area, that is, the average stresses, are given by:
Oxx — , Txy — , Txz —
r xx AA xy xz

Page | 22
 Chapter 1

SiAA-0, the “instantaneous” stresses are given by:

1. APx dPx 1. APy dPy 1. APz dPz
Oy = limA A—0-=-----, Tyv = limAA—0 — = —TX7 = limAA—0-----=-----
X AA-O AA dA ' xy AA-O AA dA ' xz AA-O ^A dA

The intensity of the normal force to the surface is the normal stress σxx and can be
compression (when it goes towards the surface) or tension or traction (when it leaves the
surface). At the same time the efforts
Txy ^xz are parallel to the surface in question, so they are
known as tangent stresses or shear stresses.

Taking an internal differential element in the effort matter has the components of
shown in the figure.

Fig.1.6 Differential element taken from the Fig.1.7 Stress components

continuum
The stress components are grouped in the stress tensor as follows

[
^- ^x
xy z
^yz °yy Tyz
Xzy °zz
When in one direction, the stresses are zero, it is said that there is a plane state of stress, if the
direction of the principal stress which is zero is chosen arbitrarily on the z axis, the following
is obtained:

Oz Txz ^yz ^zx ^zy 0

[° ^-
The stress tensor is written in the form, a = xTy xy
z ° assumed to pass through the
The resultants of stresses acting in the plane stress state are
centroid of the body so that there are no torques acting directly on it. Since equilibrium of
forces is assumed in said body ∑ Mz , it is

Page | 23
 Chapter 1

say, (TyXdx(lz) dy + (TXydydz )

dx = 0 and since dV — dxdy dz + 0 = ^xy — ^yx,
This means that the stress tensor is symmetrical.

Fig.1.8 Plane stress state

The constitutive equation, that is to the one that relates stresses and deformations, for
plane stresses is: say

/1p cy V
F
V
F
0 0 eit
^y cy
1 V
V
0 0 0 ^x\
F ^y F cy F
^y Oh
V V 1 •
F 0 0 0 0
&z
F cy ^an F cy 1
ly

Yxy d
0 0 0 1 0
\ Yyz ) G
0
1
(2)
\Yzx^ 0 0 0 0 0
G

(0 0 0 0 0 G)

Page | 24
 Chapter 1

(5
Which can be simplified as:

)
'1
p
Dy
(
V
^■xy)
\YxyJ
He
y

Although the normal stress on the local Z axis is zero, the induced deformation in the z
axis is not. The above equation can be used to calculate the strain value for a linear elastic

&z=g ( V ( ))
material [23].

( )
Oz Ox +
Relocating the consideration for plane stress we obtain:

V Ox +

(§) ( 8, ) ( 8, ) )(
the calculation of efforts:

%)
-1

-,E,, (
He 0
V y 0
p p1 1 — v1 2
-and - 2(1 + v)
\Yxy/ \^xy/ \^xy/ \
(0
The properties matrix is assigned the letter C.
C Hey 0 }
2
1 0

22(1 +v) )

Page | 25
 Chapter 3

Chapter 2
Algorithm Development
“Most of the fundamental ideas of science are
essentially simple, and may, as a rule, be expressed
in a language comprehensible to everyone”
[Albert Einstein Koch]

The introduction of optimization techniques in the design of structures is necessary, as it

provides advantages in relation to cost reduction and decreased time to determine the best
solution compared to traditional design. Allowing to consider specifications, increasingly
numerous, that are required in current structures.

In the traditional design process it is formulated as can be seen in figure 2.1, but this
depends directly on the experience of the designer who, by not being able to check all the
possible solutions, due to the time costs that this involves, obtains good results but they are
not optimal.

Fig.2.1 Trial and error design process [11].

Optimization techniques reform the design process with the aim of automatically solving a
greater number of solutions, so that the result is closer to the optimum.

Page | 26
 Chapter 3

Fig.2.2 Optimized design process [11].

Three fundamental stages will be detailed below to better understand the evolutionary
structural optimization algorithm addressed in this thesis: geometric modeling, finite
element analysis and optimization criteria.

2.1 Mesh Generator.

Geometric modeling is the tool that allows us to transfer a certain physical object
(mechanical structures) to a symbolic representation, which is used to manipulate geometric
characteristics of the objects, allowing them to be edited, visualized, and to perform
calculations and simulations.

In essence, the geometric model is simply a set of data relating to the geometry, structure
and properties of the object. The data it contains will depend on the nature of the object to
be designed. That is, the structure of a geometric model to represent an electrical circuit will
be different from that used to represent the fuselage of an aircraft. Furthermore, at the model
design stage it must be taken into account that the structure will be used to perform certain
operations mentioned above: editing, visualization, calculation of properties, etc.

“Geometric modeling can be defined as: a mathematical representation of a real geometry,

in a two- or three-dimensional domain, through known curves, such as lines, parabolas,
hyperbolas, etc., using a computational algorithm.” [21].

A mesh is the geometric discretization of a region, formed by nodes and elements

connected to each other. The connectivity of a mesh is the description of the nodes that
make up each of the elements that form it. At the same time, the interaction between

Page | 27
 Chapter 3
elements is given by the nodes. The solution to these nodes are the primary variables in the
finite element procedure. For each analysis an appropriate set of element equations is
required and the proposed element solution is written in terms of selected variables at the
nodes. The variables at the nodes are generally referred to as Degrees of Freedom DoF.

Finite element analysis depends on the number of elements and their arrangement to
define the model. In the same way, depending on the type of element and the precision
required, a more or less fine mesh of elements must be defined.

There are numerous mesh generation methods for finite elements [7]. These methods are
classified, according to their degree of automation, into manual, semi-automatic and
automatic [17]. In manual methods, all information is provided manually, including the
shape of the elements, the coordinates of the nodes, physical attributes, etc. Semi-automatic
methods take advantage of the fact that the domain to be discretized has simple geometric
shapes to use a mixed model, in which information is given on some aspects of the mesh,
and a routine provides the rest. Finally, automatic methods generate the mesh without
requiring any additional information; they only need to know the domain outline and the
physical characteristics that must be taken into account.

Meshes can be structured or unstructured. In the former, connectivity can be described by

some indexing scheme (identification by means of i, j, k indices); while in the latter, this
relationship does not exist and a special data structure is necessary to represent the
information of said connectivity [22].

The type of element used in the mesh that discretizes the structure to be analyzed is of
vital importance. Figure 2.3 shows the configuration of three two-dimensional elements
classified as 'good', 'poor' and 'illegal'. Good elements are characterized by their proportion
(length/width ratio) being close to unity; thus, for example, in elements formed by four
nodes the angles should be as close to 90 degrees as possible and the internal angle for
elements of different configuration depends on the number of nodes that the element
contains. Poor elements should be avoided, as these can lead to inaccurate results. Illegal
elements are unacceptable, and should not be used for finite element models. When illegal
elements are inadvertently drawn, they generate invalid finite element models [29].

Page | 28
 Chapter 3

Fig.2.3 Elements (a) Good; (b) Poor; (c) Unacceptable.

Asymmetry is defined as the variation of the vertex angle of the element, from 60° for
triangles, and from 90° for quadrilaterals. Two of the “poor” elements in Figure 2.3(c) can
be characterized as asymmetric.

For triangular elements, it is recommended to avoid acute angles less than 30°, and for
quadrilateral elements, obtuse angles greater than 120°, see Figure 2.3 (b). As a general rule,
using asymmetry is acceptable when you are primarily interested in displacements.
However, if the stresses are to be calculated accurately, then the model must have less
asymmetry.

2.2 Finite Element Analysis for Plane Stresses (General Plane Finite Element)

Plane stress and strain problems are special cases of elastic solid problems, in which the
variables of interest correspond to displacements, strains and stresses. Displacements can be
considered as the main variables from which, once known, the other two are derived. In flat
cases, with Cartesian reference, the variables or degrees of freedom to be determined are the

Page | 29
 Chapter 1

displacements in the X and Y directions, i.e. the values of the displacement distributions u[x, y]
and v[x, y] at the nodes of a finite element mesh (2 DoF per node).
Common to any finite element analysis, the displacement distributions are calculated from
interpolation functions which we will handle with the letter (N) and nodal values of these. If
we assume any FEM with n number of nodes, we can say that:

( u1 )
v1
u
[x, y] N1 0 N2 0 ... NT
.d
v
[x, y] 0 N1 0 N2 u2
v2
The deformation can be calculated as:

/ON
/ du \ dx (-- either dN2 or dx .\ /"TO
dx
£and )
&x \ dv 0 dy dN. dN2
0 —É dy V1 BT
dy — u2 = ■d
and, / 12,
r xy/ you dv , ON. dN± dN2 dN2 ( )
\-----1---)
\dy dx/ \ dy dx dy dx "7

and the stresses are calculated from the constitutive equations [3].

The finite element equation for plane situations is (without considering loads)
punctual):
Kd = rq + rb

where the stiffness matrix is: Khf AB.C.BT dA

where h is the thickness of the finite element; the body force vector is:

rb h
/ / N. bx )d A

where bx and by are the components of the body force per unit volume applied at the center of
mass of the finite element.

Regarding the vector rq, which is the equivalent of distributed loads, it is necessary to consider
the following. Each finite element for plane stress and strain has a
Page | 31
 Chapter 3

geometry defined by n nodes. These n nodes, in the simplest case, generate n edges (a triangular
element has three nodes and three edges, for example). Therefore, on a plane finite element there
can exist up to n distributed loads qc={qxc,qyc}; where c=1,2,...,n refers to the edge in question.
For each distributed load, a vector rqc is calculated and the corresponding interpolation function
NC is used:

Fig.2.4 Finite element for plane stress, geometry defined by n nodes.

rqc

therefore, the vector rq is:

rq Nc 1.( qx1
Assigned Jury:..............................................................................................................................3
Acknowledgements.....................................................................................................................iv
1.1 Justification of the proposal...............................................................................................7
1.2 Objectives of the thesis....................................................................................................10
Chapter 3........................................................................................................................................37
Implementation of the Algorithm...................................................................................................37
Chapter 4........................................................................................................................................50
Application and Results..................................................................................................................50
Conclusions....................................................................................................................................74
References......................................................................................................................................76

- ncn qyn

where Nc -i will be the interpolation functions for each edge that we have [23].

Page | 32
 Chapter 3

The triangular finite element used in the algorithm of this thesis is explained in subtopic 3.2 of
chapter 3.

2.3 Evolutionary Rules Algorithm

In general, optimization results in a result that could have been achieved by an experienced
designer, provided he or she had the time and money to directly explore all the design
alternatives. The introduction of optimization techniques in structural design is necessary in view
of the advantages it offers, such as: cost reduction and time savings for designers; allowing them
to devote more time to the more creative phases of the design and the possibility of

Page | 33
 Chapter 1

Consider the increasingly numerous and difficult-to-meet design specifications

that is required of current structures.

The diagram below shows in general terms the optimization methodology of this work, under
the methodology of structural optimal design in combination with finite element analysis.

Fig.2.5 Flow chart, corresponding to the optimization methodology.

The classical concept of evolutionary structural optimization developed by Xie and Grant [27]
is based on the fundamental idea that an object evolves towards a fully stressed structure by
slowly removing material that does not present stress levels.
Page | 33
 Chapter 3

of high efforts and considers that the optimal structure is one in which each element is working at
the maximum admissible stress.

The Evolutionary Rules algorithm is as follows:

1) Define the initial geometry of the model using the dimensions of the rectangular
domain.

2) Divide the domain using a fixed mesh of finite elements.

3) Apply the loads as well as the applied load cases (CFN and CFE) and
non-design zones.

4) Set the material properties.

5) Define Von Mises stress as the criterion to be optimized.

6) Perform the calculation of the Von Mises stress in each of the elements by means
of finite element analysis and compare the Von Mises stress of the element or
(
6 ) against the maximum permissible stress of the structure
^vm,e .
Where
(

6 ) Von Mises traction of element e.

^vm,e Maximum Von Mises traction on the elements.

7) Geometry is modified based on a process of removing and/or adding material,

defined by the stress level of each of the elements of the structure. Removal is carried
out on elements that present a low level of stresses.
(6)

— RR ^vm,e . On the other hand, the addition of material is

applied to those elements whose stress level is high or whose Von Mises stress is greater
than the yield stress of the material.
( )
5
— IR ^vm,e .
Where
RR Rejection rate 0 ≤ RR ≤ 1
IR Addition rate 0 ≤ RR ≤ 1

Page | 34
 Chapter 3
RR = ro r * SS + aRR * ON IR=i- i * ss — aIR * ON

SS is the number of stable states, a counter indicates all the stable

states in the evolution process.

ON is the oscillation number, a counter increments it by one each

time the optimization reaches an oscillatory state. It is defined as an
oscillatory state when an element is attached and the same element is
removed in the subsequent iteration and continues this pattern to
infinity. The increase in this variable each time an oscillatory state is
reached allows the structure to continue its evolution.

ro = O

r Constant determined with BESO experience, normally equal to

0.001.

aRR is an oscillation number constant determined by BESO

experience, typically 0.01.

8) If a state is reached where there are no elements in the design domain that satisfy
the equations in step 7, a stable state will have been reached. To continue with the
optimization process, increase the steady state number by one (SS), and repeat step 7.

9) If an oscillatory state is reached before reaching the chosen optimization

criterion, increase the oscillatory state (ON) value by one unit and repeat step 7.

10) Repeat steps 6 through 10 until the optimization criteria(s) are met.

Page | 35
Chapter 3

Page | 36
 Chapter 3

Chapter 3
Implementation of the Algorithm

"There is greatness in this conception of life,... that as

this planet has been revolving according to the constant
law of gravitation, there have been and are being
developed, from so simple a beginning, an infinity of
forms more and more beautiful and wonderful."
[Charles Robert Darwin]

This chapter details the implementation of the algorithm described in Chapter 2. For this thesis,
three modules were developed (Meshing Module, Finite Element Analysis Module and
Optimization Module) through which the optimal solution for the desired structure is found.

3.1 Meshing Module

The geometric modeling of the domain in Finite Element Analysis is represented by a mesh of
elements and by the determination of stiffness matrices which is performed by numerical
integration. This stage is of vital importance since for an optimization process the design
geometry is not definitive, it is continually modified, which requires generating a new mesh of
finite elements to reanalyze the new structure, which is computationally expensive due to the
time it takes to generate it.

In this thesis the concept of fixed mesh is used, whose main objectives of this representation
are:

1. Establish an efficient link between the mesh structure and the finite element
analysis model.

2. Reduce the computation time to obtain the stiffness matrix of the structure.

The first phase consists of defining the domain of the finite elements. If we allow that
2fg be the smallest domain (in the form of a “rectangle”) that fits and encloses

Page | 37
 Chapter 1

completely to region 22 , being aligned with the principal axes, then ílfg is defined as the
domain of the fixed mesh (Fig. 3.1).

A node x € ílfg is considered interior if x € íl . A point x € ílfg is considered exterior if x

íl .

Fig. 3.1 Fixed mesh domain Ωfg , design region Ω , and structure outline Γ (Area where material
has been removed)

When the fixed mesh is superimposed on the domain Ωfg we obtain a discrete finite
element mesh. This mesh is divided into elements with a fixed geometry (triangular
elements in two dimensions). Depending on the position of the elements with respect to the
structure, they can be classified as: interior elements I (In) if all their nodes are contained in
Ω, exterior O (Out) if all their nodes are outside Ω, and edge B (Boundary) [18].

Elements I, O and B differ only in the constitutive properties of the material. The I
elements have the properties of the structure material (real material), the O elements the
properties of a virtual medium, and the B elements present a combination between the
properties of the real material and the virtual material.

The characteristics of the evolutionary optimization method allow to further simplify the
application of the fixed mesh concept for this thesis, since in the search for the optimal
form, we start from the discretization of a rectangular area (nodes that belong to ílfg) to
which elements are removed or added according to the Xie criterion, obtaining a different
structure each time, but without the regeneration of the mesh being necessary. The new
shape found contains only nodes
Page | 38
 Chapter 3

interiors (nodes belonging to 22) because only entire elements are modified and not factions
of them. Consequently, there are no edge elements B that have a combination of real and
virtual material in their properties.

In mesh generation, quadrilateral elements are usually preferred over triangular ones for
reasons of geometric isotropy. However, the use of triangular elements is suggested when
there are irregularities in the geometry of the element to be modeled [4]. The optimal shapes
that can be found generally show these irregularities, from the above, for the generation of
the mesh in the development of the algorithm of this thesis, triangular elements are used
which alternate in the manner shown in Fig. 3.2.

For meshing, the domain of the fixed mesh must be defined, providing the lengths in the
two-dimensional Cartesian plane X and Y. It is then required to provide the level of
discretization on both the X and Y axis (Fig. 3.2), according to these dimensions a point
cloud is generated in the design domain.

Fig. 3.2 Rectangular design domain and mesh element characteristics.

The nodes (generated point clouds) are assigned a connectivity which describes the
triangular element; this assignment is carried out using a node ordering rule. The design
domain contains “i” number of nodes and the elements that define the mesh are generated by
the following rules:

Page | 39
 {i, i+1, npx+i} {i+1, npx+i+1, npx+i} {i+1, npx+i+1, npx+i} {i, i+1, npx+i}
Where:
i the i-th node.
, , ,1 length in the X direction of the design domain
npx number of nodes along the X axis, npx = length differential in X (AX)

Fig. 3.3 Connectivity of triangular elements that form the mesh.

When forming the mesh, the pair of triangular elements alternate as required, only the
information of the i-th node being necessary for their generation. The connectivity of the
elements is grouped into a vector which allows mathematical management of the mesh.

This thesis work considers two degrees of freedom (DOF) per node, which is applied to
the connectivity of the mesh and finally the connectivity information is obtained based on
the degrees of freedom.

Connectivity of the element by GdL{U;, V;, Ui+i, V+1, Unpx+i, Vnpx+i}

Fig. 3.4 Triangular element with three nodes and six degrees of freedom.

Page | 40
 Chapter 3

Figure 3.5 shows three meshes with different numbers of elements and dimensions that
this module can generate. For each of the meshes, the most significant data are presented,
such as the number of nodes, elements and execution time.

Mesh 2: 2,500 elements, 1,326 nodes.

Time: 0.282 seconds.

Mesh 3: 1,200 elements, 651 nodes.

Time: 0.094 seconds.
Fig. 3.5 Meshing with different number of elements.
3.1 Finite Element Analysis (FEM) Module

The numerical finite element method is the module by which the solution of the nodal
displacements of each of the discretized elements that form part of the structure under
analysis is obtained.

For the implementation of this module, essential information is required for the resolution
of the optimal design. Under the premise of plane stress analysis, the thickness (h) of the
plate must be indicated, as well as the essential boundary conditions (EBC), natural
boundary conditions (NBC) and different stresses such as: body forces, distributed loads,
tangential and normal loads.

Page | 41
 Chapter 3

Likewise, the set of relationships between the value of a certain variable between the
nodes of the generated mesh can be written in the form of a system of linear equations, the
matrix of said system called stiffness matrix and the finite element in general for triangular
elements, is explained below.

3.2.1 Triangular Finite Element

Fig. 3.6 Triangular finite element subject to different loads

In Figure 3.5 we can see a simple triangular finite element formed by three nodes of two
GdL each, in total 6 GdL u1, v1, u2, v2, u3, v3. To interpolate these values in the full
element domain, the following linear interpolation functions are defined, which represent
planes [3].

1
N1 (xb1 + yc1 + f1 );
1
N2 ■ (xb2+ yc2 + f2 );
1
N3 - (x b3+ and c3 + f3 );

where A is the area of the triangle; bi, ci, fi for i=1, 2, 3 are the following geometric
constants of the triangle

b1 y2 y3; b2 y3 y1; b3 y1 y2;

c1 x3 x2; c1 x1 x3; c3 x2 x1;
f1 x2y3 x3 y2; f2 x3 y1 x1 y3; f3 x1 y2 x2 y1
A1-
(f1 + f2 + f3 )

Page | 42
 Chapter 3

Substituting these interpolation functions in the deformation calculation, we obtain:

In turn, replacing the above with the calculation of the stiffness

matrix: BCBT BCBT
Khf dA hA ( )

h dA
h fj N1 bx dA

h fj N2 bx dA h fj AN2

h
h fjby
by dAfjAN1 N3 bx dA ha

bx
rb h ( bx by bx

fj AN3 by

dA
ff N. 1

The body force vector is:

n1 = ( nx1 ny1 );
n2 = ( nx2 ny2 );
n3 = ( nx3 ny3 );

by bx by )

The loads distributed along each edge are generally defined in normal and transverse
directions. To transform them to global coordinates x, y, the following calculation is
performed for each of them: where i=1, 2, 3; qni and qti are the normal and tangential

Each edge of the triangle has a normal direction;

distributed loads of the i-th edge, respectively.
qxi nxi nyi qni qti
;
qyi nyi nxi

Page | 43
 Chapter 3

Fig. 3.7 Triangular finite element subjected to normal and

tangential distributed loads.

For the edges of the triangle, the Nc-i interpolation functions are defined as:
L1 t 0 t 0 00)
NTc L1 L1
1 L1 t t
0 L1 0 00
L1 /
L2 t 0 t 0 "1
NTc
2 L2 L2
0 L2 t t ;
L2 0 L2 )
t 0 t
00 0 N
NTc L3
3 LL33
t 0 00 L3 t ;
0
L3 L3 )

Where Li corresponds to the edge length.

These functions are used to calculate distributed load vectors.

u0L1 LL11t qx1 dt]
( qx1 )
dt
J0L1 LL11t qy1
qy1
dt
rq1 Nc 0L1 Lt1 qx1 qx1
-*1. rq1 hL1
qy1 c—>1 qy1

qy1
0
0 k0)
< 0 )
Similarly 0L1 Lt1

Page | 44
 Chapter 3
(0N 0 ( qx3
) qy3
rq2 qx2 0
2
; rq3 h L3

qy2 0
qx2 qx3
. .
qy2 ) qy3 )
Using the equations obtained for a triangular element, together with the
interpolation functions given by N1, N2 and N3, the matrix B can then be calculated,
which is necessary to obtain the Stiffness matrix K for each element, in addition to
introducing the properties matrix C (see chapter 1 section 1.4) for analysis under
plane stress [23].

In assembling the equations of each of the generated elements, it is necessary to

know both the number of nodes as well as the number of elements generated during
meshing. Connectivity allows each node and element to be located in a position
within the global assembly matrix, as well as in the assembly vector called the
global load vector, these initialized to zero. This global ensemble matrix is a square
matrix of dimensions nxn, where:

n = 2 * Total number of Nodes, since each node has 2 GDL.

To introduce essential boundary conditions (EBC) at a node, a value of zero is

assigned to either the degree of freedom ui or vi depending on the direction in which
the constraint is desired. The global assembly matrix and the global load vector can
then be reduced.

To solve the system kd = rq+ rb, the linear solution method is used where the
displacement vector “d” is cleared. The reduction of the system implies that
inverting the global assembly matrix has a solution. In this way, the value of the
nodal displacements or nodal solution is obtained; to this vector, the values of the
boundary conditions that were removed when reducing the system must be added, in
the positions that correspond to them.

From the nodal solutions vector “d”, the calculation of quantities necessary for the
application of the optimization method is then carried out, such as: Von Mises stress,
deformations, etc.

The calculation of the stress per element is obtained by applying the dot product to

Page | 45
 Chapter 3
the obtained deformations and to the properties matrix for plane stress C. The von
Mises stress value for each element is calculated as:

Gym = 2 √ (0, - 02)2 + (02)2 + (—01)2 + 6(xy2)

From the von Mises stress values for each element of the mesh, a color scale is

Analysis 1: Analysis 2:
2500 elements, 1326 nodes. 3200 elements, 1681 nodes.
Analysis time: 19.281 seconds. Analysis time: 29.188 seconds
Efforts: Efforts:
Maximum 0.0581728 [N/mm2] Maximum 0.999393 [N/mm2]
Average 0.0037004 [N/mm2] Average 0.201017 [N/mm2]
Minimum 0.0000209 [N/mm2] Minimum 0.0000904 [N/mm2]
calculated which behaves as follows: from Blue to Cyan, from Cyan to Green, Green
to Yellow and Yellow to Red, in this scale the blue color corresponds to the
minimum stress value, successively the following colors will have higher values and
finally red corresponds to the maximum stress value. This information is displayed
in the graph together with the mesh to have a visual appreciation of the distribution
of stresses in the analyzed structure.

Fig. 3.8 Stress distribution in two different cases generated by the FEM module.

Page | 46
 Chapter 3
3.3 Optimization Module

The optimization module is responsible for performing iterations in which the

stresses of a structure are calculated and optimized using the evolutionary rule. By
integrating this module with the meshing and FEM modules, we have the elements
required for the application of the proposed algorithm for structure optimization.

This module is what makes the difference between the traditional design method
and design by optimization, since the algorithm directs a series of routines that allow
the interaction between the meshing and finite element analysis modules in a
controlled manner, without the need to make manual changes in each series of
calculations.

It is in this part of the algorithm where the fixed mesh fulfills its main benefit to
the method. The assembly of the stiffness matrix requires generating a series of
functions, connectivities and geometric characteristics, as well as the assembly of
said matrix. While the module can handle this task, this represents a waste of
resources both in terms of computational memory and time. The fixed mesh concept
allows us to overcome the disadvantages that occur when it comes to generating a
different stiffness matrix for each structure modified by iteration.

The initial stiffness matrix is then taken as a base because; if elements of the
structure and consequently nodes are removed, a relatively simple management can
be done where the connectivity of the removed element is found and through this its
information is disassembled from the place that corresponds to it within the matrix.
In the case where a node is removed from the mesh, the rows and columns
corresponding to the degrees of freedom contained in the node must be eliminated.

The handling of disassembly of elements and elimination of nodes can also be

done in the opposite direction, adding both elements and therefore connectivities or
adding degrees of freedom contained in a node. By following these steps a
considerable part of the finite element analysis process is generated, without the
need to recalculate values for each element. Obtaining the stresses is similar to that
described in section 3.2.

In the optimization method described (see chapter 2 section 2.3), it is a general rule
that in each iteration elements are removed or added to the structure. This
modification is directly related to the level of effort of these elements through:

Page | 47
 Chapter 3
RR Rejection rate 0 ≤ RR ≤ 1

RR= r0 + r1*SS + aRR* ON

A(e) < RR nmaxuvm

— -n uvm,e
IR Addition rate 0 ≤ RR ≤ 1

IR= i0 - i1 * SS - aIR* ON

A(e) > IR nmax

-vm — 1 — uv?n,e

Where initially the evolutionary rates will take the following values:

Rejection rate Addition rate

r0 = 0 i0 = 1

r1 = 0.001 i1 = 0.01

aRR = 0.01 aIR = 0.1

ON = 0 and SS = 1

% RR = 0.1 % IR= 99%

Iteratively, stress values are obtained from each of the elements that make up the
fixed mesh. These values are normalized using the maximum stress value obtained
for each iteration. As a next step, the rejection (RR) and addition (IR) rates are
applied.

It is very important to note that the behavior of the rates, both addition and
rejection, is governed by their respective equations, which are made up of constant
terms obtained through experience when calibrating the programming of the
algorithm, as well as two variable terms SS and ON on which finding the optimal
structure depends. Stable states throughout the optimization process are identified by
increasing SS, making it possible to identify local optima. The ON increment is
defined as an oscillatory state, which means that if the same element or group of
elements are added and the same element(s) are removed in the subsequent iteration
and the rates are not modified, this pattern repeats itself to infinity. The increase of
this variable each time it reaches an oscillatory state allows the structure to continue
its evolution.

Page | 48
 Chapter 3
After identifying the elements to be rejected and the elements that require addition,
during the elimination process it is sufficient to remove the indicated element,
however for the addition, it is established to add elements on the three edges of the
element that presents the greatest stress according to the addition rate. The algorithm
is able to identify the elements that are previously contained in the mesh, so that
there are no repeated elements. This process can be observed in Figure 3.9.

Page | 49
 Chapter 3

Fig.3.9 a) Element to be rejected (blue) and elements where addition is required (red).
b) Structures generated after applying the addition and rejection criteria.

The structure generated in each new iteration is analyzed using finite elements.
The previous procedure is carried out iteratively until no improvements can be
obtained, that is, the stopping criterion is reached.

3.3.1 Stopping Criterion

There are different stopping criteria which depend on the design variables and
characteristics considered. The stopping criterion of the algorithm can be a stress, a
volume, a displacement, etc. In this thesis work, volume is used as said criterion,
which will stop the optimization process in the iteration that meets the specified
volume.

In the following chapter, a brief discussion is presented on the results obtained

from the application of the described algorithm, as well as case studies of structures
that involve plane stress analysis, and the graphs obtained by said algorithm are
reviewed.

Page | 50
 Chapter 3
Chapter 4
Application and Results

“Intelligence consists not only of

knowledge, but also of the ability to
apply knowledge in practice.”
[ Aristotle of Stagira]

In the field of optimal structural design, the analysis depends on: the nature of the
problem (static, dynamic calculation, etc.), the type of structure, the mathematical
calculation model and the numerical model ultimately used. It is evident that it is
impossible to create a single calculation model for all structural analysis problems.
However, a large part of the problems that appear in professional practice can be
solved effectively by matrix methods, in the case of discrete structures, or by the
Finite Element Method in the case of continuous structures. In this way, taking the
appropriate considerations based on the modules developed in chapter 3, the
optimization algorithm for structures under plane stress generated for this thesis is
applied to three case studies, from which the optimal design solution is obtained
with volume as the optimization criterion.

4.1 Case Study I

Michell type structures

Michell's work [13] poses standard optimization problems, the solutions to these
problems are used to test the performance of different structural optimization
methods. However, it is evident that if the solutions obtained by the different
methods are compared, they are similar but not equal. This inequality arises from
differences in the formulation of optimization methods. The above prevents us from
speaking of a general optimal solution, despite this drawback, the solution of two
different methods applied to the same problem will have a great similarity between
solutions.

Figures 4.1 and 4.8 show two standard problems performed by Michell (common
examples in the optimization literature), where rectangular area structures are

Page | 51
 Chapter 3
subjected to a force in the central part and two supports at their ends [27].

4.1.1 Rectangular domain with fixed supports and vertical force centered at
its bottom

For the first structure, the design domain is given by the following dimensions:
the domain length is L= 10 m, the height H= 5 m and a plate thickness of 0.1 m. An
elasticity modulus E= 100 GPa and a Poisson coefficient v=0.3 are considered [27].

The discretization of said domain is carried out by means of 0.2 m divisions in

both the X and Y directions in accordance with the provisions of chapter 3, section
3.1 of this work. 1326 nodes and 2500 triangular elements are generated.

The restrictions for this case are located in the two lower corners, so the degrees of
freedom corresponding to these nodes are restricted in both directions. The vertically
applied point load is 1000N as shown in Figure 4.1.

Fig.4.1 Rectangular domain with fixed supports and vertical force centered at its bottom.

Page | 52
 Chapter 4
The result of the representative efforts of the finite element analysis is presented below with
the respective graph.

V on Mises stresses [N/mm2]

• Maximum 0.05817284788
• Average 0.00370048
• Minimum 0.00002090291

Fig. 4.2 Finite element analysis of Michell type structure with fixed supports.

For this optimization, the process starts with a rejection rate of 3.1% and an addition rate of
69%, the stopping criterion defined for this case study was based on volume (a structure with
30% of the initial volume is required).

Figure 4.3 shows the evolutionary history of the different stable states of the structure (local
optima). In each of the figures the corresponding rejection and addition rate is shown, the
optimal structure for the imposed conditions is shown in Figure 4.3 (i).
(to (b)
Iteration:4 %RR=3.1
) %IR=69 Iteration:36 %RR=4.6 %IR=54

Page | 53
 Chapter 4

(c) (d)
Iteration:128 %RR=4.6 %IR=54 Iteration:158 %RR=7.8 %IR=27

(and) (F)
Iteration:180 %RR=8.9 %IR=27 Iteration:205 %RR=10 %IR=27

(g) (h)
Iteration:226 %RR=10.9 %IR=27 Iteration:232 %RR=11.7 %IR=27

(Yo)
Iteration:270 %RR=13 %IR=27

Fig. 4.3 Evolutionary history of the Michell type structure with fixed
supports.

Page | 54
 Chapter 4

The evolutionary history of the maximum, average and minimum effort corresponding to
the iterations throughout the optimization is presented in the graph in Figure 4.4.

Fig.4.4 Graph of maximum, average and minimum effort vs rejection rate.

Figure 4.5 reviews the graph of average stresses, where the stable states are displayed
throughout the optimization. This graph also represents the tendency to homogenize the
stress values as an optimal structure is obtained.

Fig.4.5 Graph of average effort vs rejection rate.

Page | 55
 Chapter 4

The history of volume reduction is shown in Figure 4.6 according to the change of
rejection rate or steady state.

Fig.4.6 Graph showing the behavior of volume vs. rejection rate.

Table 4.1 compares the stress levels, both maximum and average, between the initial
domain of the structure corresponding to Figure 4.2 and the optimized structure shown in
Figure 4.3 (i).

Table 4.1Comparison between the initial design and the optimal design.
^max [N ⁄mm2] ^average [N ⁄mm2] 3 V/V0
Volume [m ]
Initial model 0.0581728 0.00370048 5.0 1
Optimal design 0.0581347 0.0121003 1.5 0.3

The structure shown in the comparison in Figure 4.7-a was designed using the ESO
methodology. The modifications presented by this structure are similar to those obtained in
this work using the BESO algorithm.

to) b)

Fig.4.7 Comparative results, a) result generated by Xie, b) solution generated by the author.

Page | 56
 Chapter 4

4.1.2 Rectangular domain with mobile support and vertical force centered at the
bottom

This rectangular domain shown in Figure 4.8 has the same dimensions as the Michell-type
structure presented in the previous section 4.1.1. Same material characteristics E= 100 GPa
and v=0.3 [27]. Likewise, it is discretized with 1326 nodes that generate 2500 triangular
elements.

In this case, the structure is restricted in the two lower corners; in its left corner, the
degrees of freedom corresponding to this node are restricted in both directions, and in the
right corner, only the degree of freedom corresponding to the “Y” axis is restricted. The
vertically applied point load is 1000 N as shown in Figure 4.8.

Fig. 4.8 Rectangular domain with mobile support and vertical force centered at its bottom.

Page | 57
 Chapter 4

Figure 4.9 shows the result of the finite element analysis and the most relevant stresses.

V on Mises stresses [N/mm2]

• Maximum 0.05850835808
• Average 0.00447698706
• Minimum 0.00003226284

Fig. 4.9 Finite element analysis of Michell-type structure with sliding supports.

The optimization process starts with a rejection rate of 3.1% and an addition rate of 69%, a
volume stop criterion of 32%.

Figure 4.10 shows the evolutionary history of the different stable states of the structure. In
each of the figures the corresponding rejection and addition rate is shown, the optimal
structure for the imposed conditions is shown in Figure 4.10 (i).
(a) Iteration: 3 %RR=3.1 %IR=69 (b)
Iteration:17 %RR=3.8 %IR=62

Page | 58
 Chapter 4

(c) (d)
Iteration:42 %RR=5.3 %IR=47 Iteration:68 %RR=5.9 %IR=41

(and) (F)
Iteration:90 %RR=7.4 %IR=27 Iteration:113 %RR=8.9 %IR=27

(g) (h)
Iteration:147 %RR=11.4 %IR=27 Iteration:201 %RR=13.9 %IR=27

(Yo)
Iteration:241 %RR=16.4 %IR=27

Fig. 4.10 Evolutionary history of the Michell type structure with sliding
support.

Page | 59
 Chapter 4

Figure 4.11 presents in a graph the evolutionary history of the maximum, average and
minimum stress against the corresponding rejection rate throughout the optimization. The
graph in Figure 4.12 shows only the average effort to better appreciate the trend.

Fig.4.11 Graph of maximum, average and minimum effort vs rejection rate.

Fig.4.12 Graph of average effort vs rejection rate.

Page | 60
 Chapter 4

The behavior of the volume in each stable state is presented in the following figure, where
convergence is observed when approaching the optimal state when the rejection rate has a
value of 16.4%.

Fig.4.13 Chart showing volume behavior vs rejection rate.

The following table shows the stress and volume data of the initial model against the results
of the optimal structure.

Table 4.2Comparison between the initial design and the optimal design.
^max [N ⁄mm2] ^average [N ⁄mm2] Volume [m3] V/V0
Initial model 0.0585084 0.00447699 5.0 1
Optimal design 0.0607159 0.0146098 1.608 0.32

Xie solved this same problem by obtaining the geometry shown in Figure 4.14- a. In this
comparison the differences between the ESO solutions and those obtained here can be
appreciated. This is not surprising since Xie only removes material, while in this work there
is an evolutionary rule of addition.

to) b)

Fig.4.14 Comparative results, a) developed by Xie, b) solution generated by the author.

Page | 61
 Chapter 4

4.2 Case study 2

Bridge-type structure with fixed supports at the corners

In this case, a domain with the dimensions specified in Figure 4.15 is optimized. In this
structure, a non-designable area is considered at the top, and there is also a distributed load
with a magnitude of 2500 [N/m]. It is restricted with fixed supports at the lower corners at 2
nodes.

Fig.4.15 Rectangular bridge-type domain with fixed supports at the corners.

The characteristics that are taken into account for the material are E= 210 GPa and
v=0.33. Figure 4.16 shows the result of the FEA, applied to the structure of case study 2 for
this thesis.

Von Mises stresses [N/mm2]

• Maximum 0.147648
• Average 0.00863493
• Minimum 0.00063136

Fig.4.16 Finite element analysis of the structure.

Page | 62
 Chapter 4

The optimization process starts with a rejection rate of 0.6% and an addition rate of 94%, a
structure with 30% of the initial volume is required. Figure 4.17 shows the stable states and
the optimum reached at iteration 315 (Figure 4.17 (i)).
(to) (b)
Iteration: 5 %RR=0.9 %IR=81 Iteration:77 %RR=3.1 %IR=59

(c) (d)
Iteration:106 %RR=3.5 %IR=55 Iteration:139 %RR=4.1 %IR=49

(and) (F)
Iteration:172 %RR=4.8 %IR=42 Iteration:219 %RR=5.9 %IR=31

(g) (h)
Iteration:250 %RR=7.2 %IR=20 Iteration:284 %RR=8.3 %IR=20

(Yo)
Iteration:315 %RR=15.9 %IR=27

Fig. 4.17 Evolutionary history of the typical

structure

The results of the maximum, average and minimum stress values of the optimization are
represented in the graph in Figure 4.18.

Page | 63
 Chapter 4

Fig.4.18 Graph of maximum, average and minimum effort vs rejection rate.

Fig.4.19 Graph of average effort vs rejection rate.

The volume reduction history for this case study is shown in Figure 4.20.

Page | 64
 Chapter 4

Fig.4.20 Chart showing volume behavior vs rejection rate.

Table 4.3 Comparison between the initial design and the optimal design.
^max [N ⁄mm2] ^average [N ⁄mm2] Volume [m3] V/V0
Initial model 0.147648 0.00863493 8 1
Optimal design 0.16419 0.0256596 2.536 0.31

The obtained structure shown in Figure 4.21-b will be compared with that reported by J.
H. Zhu [28] in Figure 4.21-a, from which topological similarity is observed.

to) b)

Fig.4.21 Comparative results, a) developed by Zhu, b) solution generated by the author.

4.3 Case study 3

Bridge-type structure with fixed lateral supports

For the latter case the length of the design domain is 12 [m], the height 4 [m], and the
thickness 0.1 [m]. The design domain has been discretized with 2400 elements, 60 divisions
in longitudinal direction and 20 in transverse direction. The first row of elements below the
load line is treated as a non-design zone, and represents the bridge deck (not subject to the
optimization process). The load applied to

Page | 65
 Chapter 4
the structure is a uniformly distributed force of 2500 [N]. This structure is restricted by
fixed supports on both sides of the top Figure 4.22.

Fig.4.22 Rectangular bridge-type domain with fixed lateral supports

An elasticity modulus E= 210 GPa and a Poisson coefficient v=0.33 are considered. In the
case of optimization, the process starts with a rejection rate of 1.1% and an addition rate of
89%, the stopping criterion defined for this case study was based on volume (a structure
with 40% of the initial volume is required). The stress distribution of the structure,
generated by the finite element analysis of this thesis, is shown in Figure 4.23.

V on Mises stresses [N/mm2]

• Maximum 0.107657
• Average 0.00842483
• Minimum 0.000162723

Fig.4.23 Finite element analysis of the rectangular bridge-type domain.

Figures 4.24 show the most representative optimal topologies that arise during the
optimization, obtaining the optimal design in iteration 298.

Page | 66
 Chapter 4

(to) (b)
Iteration:2 %RR=1.1 %IR=89 Iteration:26 %RR=2.7 %IR=73
R

(c) (d)
Iteration:52 %RR=3.8 %IR=62 Iteration:88 %RR=4.9 %IR=51

(and) (F)
Iteration:125 %RR=6.2 %IR=38 Iteration:158 %RR=7.4 %IR=27

(g) (h)
Iteration:215 %RR=10 %IR=27 Iteration:245 %RR=12.3 %IR=27

(Yo)
Iteration:298 %RR=14.2 %IR=27

Fig. 4.24 Evolutionary history of the typical

structure

Figure 4.25 shows the evolution of the maximum, mean, and minimum Von Mises
stresses against the rejection rate.

Page | 67
 Chapter 4

Fig.4.25 Graph of maximum, average and minimum stress vs rejection rate.

In Figure 4.27 it can be observed that the volume fraction decreases rapidly until it
reaches the target volume value.

Page | 68
 Chapter 4

Fig.4.27 Chart showing volume behavior vs rejection rate.

The comparison of maximum, average stress and volume between the initial rectangular
design and the optimal design in Figure 4.24 (i) is shown in Table 4.4.

Table 4.4 Comparison between the initial design and the optimal design.
oMax[N ⁄mm2] ^average [N/mm2] 3 V/V0
Volume [m ]
Initial model 0.107657 0.00842483 4.8 1
Optimal design 0.100295 0.0217882 1.96 0.408

Similarly to the previous case study, Figure 4.28 presents a comparison between the design
obtained by Zhu's algorithm [28] and the algorithm developed in this thesis.
to b)

Fig.4.28 Comparative results, a) developed by Zhu, b) solution generated by the author.

4.4 Discussion of Results.

The Bidirectional Evolutionary Structural Optimization (BESO) methodology was applied

to three rectangular initial domain case studies, using volume as the objective function.
The volume and, therefore, the weight of the structure were considerably reduced in all the
study cases, above the 59% reduction of the initial value, complying with the design criteria
imposed on each of the structures analyzed. Table 4.5 shows the percentage of reduced
volume for the different structures.

Page | 69
 Chapter 4
Table 4.5 Volume reduction of optimal structures.
Case study 1 Case study 2 Case study 3
69.6% and 67.84% 68.3 % 59.16 %

In the three case studies presented in this work, it is observed through the volume graphs
against rejection rates that the stress distribution tends to increase. Which is due to the
elimination of areas with inefficient elements of the structure, taking the structure to a state
of high levels of stress under a system of particular loads and restrictions.

The good performance of the algorithm is due to the calibration of the values of the
addition and rejection rates. According to the analysis of the different cases, the following
table is presented with the values of rejection rates, addition and the computational analysis
time.

Table 4.6 Initial values of evolutionary rates.

Case study 1 Case study 2 Case study 3
2106 nodes 1281 nodes
1326 nodes
SS=5 SS=10
SS= 30
ON=0 ON=0
ON = 0
IR=94% IR=89%
IR =69 %
RR=0.6 % RR=1.1 %
RR =3.1 %
Optimization time 3.4 hrs Optimization time 1.6 hrs
Optimization time in the two
cases presented 1hr

The tests were performed on a computer with the following characteristics: Windows 7,
64-bit operating system, Intel Core Duo 3.16 GHz processor, 8 GB of RAM, and the
commercial package Mathematica 7.

The optimization algorithm is a tool for structural design, which allows obtaining the
optimal solution for a case study and whose performance does not depend on the experience
of the designer as in a traditional methodology; however, the interpretation of the results
does require the knowledge of the engineer to achieve the implementation of a final
structure.

Conclusions
In this study, a computational application based on Bidirectional Evolutionary Structural
Optimization (BESO) has been developed, a general methodology applicable to a wide
range of conceptual design optimization problems and in particular for this thesis in plane
stress. Evolutionary structural optimization can be used as a tool in design, replacing the
traditional method and providing advantages such as conceptual simplicity, automated
analysis and wide applicability.

Page | 70
 Chapter 4
From the work carried out in this thesis the following conclusions are drawn in
accordance with the objectives set:

• A two-dimensional triangular finite element meshing module was

generated to discretize rectangular domain areas.

• The FEM theory was applied to structures under plane stress conditions,
obtaining as a result a module in charge of performing the finite element analysis.

• An optimization module was obtained by applying the BESO methodology

and incorporating into the application the main advantage of the fixed mesh
concept, which influences the resolution speed of the finite element method for each
iteration during the evolutionary process.

• These modules were implemented in the development of software for the

optimal design of structures.

• The computational tool was implemented in three case studies, validating

the method based on volume reduction and average effort increase. Likewise, a
topological comparison was made of each of the structures obtained in this thesis
with the results presented by the most representative authors of the literature.

Besides:

The respective interdependencies existing between the various modules implemented in

this thesis show that the quality of the information provided by a module determines the
effectiveness of the remaining ones.

It is concluded that to develop efficient optimal structural design systems it is necessary to

have decision modules and, therefore, mathematical programming algorithms, especially
conditioned by the nature and possibilities of the calculation modules based on the Finite
Element Method, and in turn these methods

They must be supplemented so that they provide the additional information required by the
decision modules.

The methodology presented and structured in the form of modules gives rise to the
construction of an optimal design system, where different modules coexist and, although
they have different concepts, perform the same function. In this way, the efficiency, power
and versatility of the system are increased, since the modules that best adapt to the
peculiarities of each particular problem can be used, achieving greater use of resources,
since the same module can be used to solve problems of a very diverse nature.

Page | 71
 Chapter 4

Future work
As a continuation of the work developed in this thesis, the following lines of future work
are proposed:

The development of automatic mesh generation techniques specially adapted to this type
of problems, and the application of automatic refinement techniques, so that in the first
iterations simple meshes are generated and as the process progresses, the discretization size
is increased, obtaining more precise results while improving the design.

The development of new, more efficient mathematical programming algorithms,

specifically designed for this type of problems, considering other types of restrictions and
load conditions.

Extend the developed methodologies to the optimal design of three-dimensional

continuous structures.

Page | 72
 References

References
[1] Annicchiarico, William, “A general scheme for the Structural Optimization of Shapes
by means of Genetic Algorithms and Geometric Design Elements”, Thesis Summary, Central
University of Venezuela, October 2000, pp. 226-234.

[2] Bendsoe Mp, Kikuchi N. “Topology and generalized layout optimization of elastic
structures.” Topology Design of Structures. 1993.

[3] Bhatti, M. Asghar, “Fundamental Finite Element Analysis and Applications”, 2nd ed.,
Ed. John Wiley & Sons Inc., USA 2005, pp. 490-551.

[4] Cubillos Alfonso “Introduction to the Finite Element Method” University of Ibagué –
Mechanical Engineering Program. Course notes 2004.

[5] GP arsons Michael “Applications of Optimization in Early Stage Ship Design”

Cartagena (Colombia). 2009, pp. 7-29.

[6] Gonzalez Goose J. A., Basket weavers Morante B., Morcillo López M. A., Mayor
Moreno L. TO. “Analysis and Optimization of Aeronautical Components Made of CFRP
Laminated Composite Materials Using the Finite Element Method (FEM)”, Coimbra,
Portugal. University of Coimbra.2008.

[7] Ho-Le K. “Finite element mesh generation methods: a review and classification”
Computer-Aided Design, 20:27–38, 1988.

[8] Keane, A.J. and Brown SM “The design of a satellite boom with enhanced vibration
performance using genetic algorithm techniques” In, Proceedings of the Second
International Conference, IC University of Plymouth, 1996, pp.107-113.

[9] Lozano T., David, Velazquez V., Fernando, Zepeda S., Antonio, “Proposal for an
Adaptive Mesher for Profile Modeling in Incompressible Flows”, XIV Annual SOMIM
Congress, September 2008, pp. 1-8.

[10] Marti, Pascual. “Application of optimization techniques in the design of lightweight

structures for large spans” Second International Meeting on Lightweight Structures:
Murcia.1995.

[11] Martínez, Pedro, “Simultaneous Optimal Design of Topology and Geometry of

Articulated Structures using Evolutionary Techniques”, Doctoral Thesis, Polytechnic
University of Cartagena, 2003, pp. 5-37.

Page | 73
 References

[12] Matteck C, Burkhardt S. “A new method of structural shape optimization based on

biological growth.” Int. Journal of Fatigue 1990.

[13] Michell Anthony George Maldon. "The limit of economy of material in frame
structures". Philosophical Magazine .London .1904.

[14] Navarrina Martinez Fermin “A General Methodology for Structural Optimization in

Computer Aided Design” Doctoral Thesis. Barcelona, March 1987, PP. (I-1)-(VII-46).

[15] Oda, J. and Yamazaki, K . "On a technique to obtain an optimum strength shape by the
finite element method." Bull. 1977.

[16] Ortiz P. Armando. Ruiz C. Osvaldo. “Basic Concepts of the Finite Element Method”.
Materials Research and Technical Support Unit, Faculty of Engineering, UNAM. 2007.
[17] PL George. “Automatic mesh generation. Application to finite element methods” John
Wiley and Sons; Masson, 1991.

[18] Querin, Osvaldo M., “Evolutionary Structural Optimization”, Doctoral Thesis,

University of Sydney Australia, April 1997, pp. 9-40.

[19] Rodriguez J, Seireg A. “Optimizing the shapes of structures via a rule-based computer
program.” ASME-Computers in Mechanical Engineering”. 1985.

[20] Salas M., Del Río C., Hesse M., Bertram V. and Castro. B. “Introduction of the
Poseidon Program in the Design and Calculation of Naval Structures”. Institute of Naval
and Maritime Sciences, Austral University of Chile.

[21] Shackelford, James F., “Introduction to Materials Science for Engineers”, 5th ed., Ed.
Prentice Hall, Mexico, 1999, pp. 643 – 782.

[22] Shewchuk, J. “Lectures notes on delaunay mesh generation, Technical report”

Department of Electrical Engineering and Computer Science, University of California at
Berkeley. 1999.

[23] Velázquez Villegas Fernando., “Notes on finite element analysis”. Faculty of

Engineering, UNAM. 2009.

[24] Velázquez Villegas Fernando and SD Santillán-Gutiérrez “Optimal Evolutionary

Design” Design and Manufacturing Center, Faculty of Engineering, UNAM. 2005.

[25] Victoria Nicolas, Mariano. “Shape and topology optimization with fixed mesh and
genetic algorithms”. Polytechnic University of Cartagena Department of Structures and

Page | 74
 References

Construction. 2006. pp. 4-40.

[26] Woon SY, Tong L, Querin OM, Steven GP. “Optimizing Topologies through a
MultiGA System” The Fifth World Congress of Structural and Multidisciplinary
Optimization (WCSMO-5). 2003.

[27] Xie, Y. M., Steven, G. P., “Evolutionary Structural Optimization”, 1st ed., Ed.
Springer, Great Britain . 1993, pp. 1-40.

[28] Zhu, J. H., Zhang, W.H., Qui, K.P., “Bi-directional evolutionary topology optimization
using element replaceable method” Paper. 2006.

[29] Zúñiga Marín Edgardo Antonio “Determination of Stresses in the Stern Azimuthal
Zone of a Tugboat, by Means of a Linear Static FEM Analysis” Bachelor's thesis. Austral
University of Chile Faculty of Engineering Sciences School of Naval Engineering. 2008, pp.
25-26.

Mesography

[30] Faculty of Engineering, Autonomous University of the State of Mexico (UAEM).

Notes on Mechanics of Materials.
http://fi.uaemex.mx/adgc/MM/ESFUERZO.htm

1 -and 2

Page | 75