You are on page 1of 69

CAE and Design Optimization – Basics Contents

Contents
Introduction ......................................................................................................2
About This Series ...........................................................................................2
About This Book .............................................................................................2
Supporting Material ........................................................................................3
Engineering Design Practice ...............................................................................4
Characteristics of Different Sectors ..................................................................4
CAE And The Design Cycle ..............................................................................5
The Impact of Optimization on CAE .................................................................6
Summary: How Engineers Should Design.........................................................8
Optimization Theory ........................................................................................ 10
What is an Optimum Design? ........................................................................ 10
Analysis and Design...................................................................................... 11
Finding An Optimum..................................................................................... 13
The Optimization Model ................................................................................ 16
Workable Implementations ........................................................................... 19
Summary ..................................................................................................... 19
FEA Essentials ................................................................................................. 21
Why use Numerical Methods at all? ............................................................... 21
What is Finite Element Analysis? ................................................................... 22
Choosing a Numerical Model ......................................................................... 24
The Role of Physical Testing ......................................................................... 25
Quick Summary of Analysis Terminology........................................................ 26
What are Elements? ..................................................................................... 30
Steps in FE Modeling .................................................................................... 31
Guidelines on Element Choice ....................................................................... 34
OptiStruct ....................................................................................................... 35
Before We Start ........................................................................................... 35
Techniques to Design Optimum Products ....................................................... 39
Putting it all together.................................................................................... 47
Summary ..................................................................................................... 49
Laminates ....................................................................................................... 51
The Miracle Material – Plastics....................................................................... 51
Reinforced Plastics: One Step Ahead ............................................................. 51
Data Required for Stress Analysis .................................................................. 53
Finite Element Approaches............................................................................ 57
Design Optimization Issues ........................................................................... 59
Glossary And References.................................................................................. 63

1
Introduction CAE and Design Optimization – Basics

Introduction
About This Series
To make the most of this series you should be an engineering student, in
your third or final year of Mechanical Engineering. You should have access
to licenses of HyperWorks, to the Altair website, and to an instructor who
can guide you through your chosen projects or assignments.

Each book in this series is completely self-contained. References to other


volumes are only for your interest and further reading. You need not be
familiar with the Finite Element Method, with 3D Modeling or with Finite
Element Modeling. Depending on the volumes you choose to read, however,
you do need to be familiar with one or more of the relevant engineering
subjects: Design of Machine Elements, Strength of Materials, Kinematics of
Machinery, Dynamics of Machinery, Probability and Statistics, Manufacturing
Technology and Introduction to Programming. A course on Operations
Research or Linear Programming is useful but not essential.

About This Book


Finite Element Analysis has traditionally been used as a design-verification
method. Recent developments in Mathematics, Software and Mechanics
have led to a dramatic change: Computer Aided Engineering (CAE) is now
widely deployed at the concept-design stage itself.

This volume presents these changes in the engineering industry and


introduces you to the theory necessary to effectively apply optimization
techniques using OptiStruct.

You should read Chapters 2 and 3 in their entirety. Chapter 4 can be skipped
if you are familiar with the Finite Element Method, but is essential reading if
you’re not. Chapter 5 is best read once, then referred to again when you are
working on your assignment. Chapter 6 is essential if you want to work on a
project addressing laminated composites, but can be safely omitted if you’re
working with other materials. The various references cited in the book will
probably be most useful after you have worked through your project and are
interpreting the results.

2
CAE and Design Optimization - Basics Introduction

Supporting Material
Your instructor will have the Student Projects and Student Projects
Summaries that accompany these volumes – they should certainly be made
use of. Further reading and references are indicated both in this book and in
the Projects themselves.

If you find the material interesting, you should also look up the HyperWorks
On-line Help System. The Altair website, www.altair.com, is also likely to be
of interest to you, both for an insight into the evolving technology and to
help you present your project better.

"Mach 2 travel feels no different." a passenger commented on an early


Concorde flight. "Yes," Sir George replied. "That was the difficult bit."

Sir George Edwards

3
Engineering Design Practice CAE and Design Optimization – Basics

Engineering Design Practice


It’s rare to meet a mechanical engineer who hasn’t, at one time or another,
been fascinated by automobiles and aircraft. With attractive looks, complex
designs, exciting performance, and a long history of colorful personalities,
they embody everything you ever wanted to create. Who’d ever want to
design anything else?

Lots of people, it turns out.

While the automotive and aerospace sectors continue to define technology


and trends in mechanical engineering even today, these are not the only
sectors that make widespread use of design software. Neither are they the
only segments where serious sums of money hinge on the efforts of the
Design Engineer.

The section below outlines the bases for evaluating designs in various
engineering sectors.

Characteristics of Different Sectors


First, let’s specify what we mean by the word “Design”. As engineers, we
usually ignore aesthetics, leaving that to Industrial Designers. One of the
tasks of an engineering designer is to come up with a design that is
functionally “satisfying”. That’s very often very hard to do, since there are so
many ways to define “satisfaction”. To understand how optimization fits into
engineering design practice, it’s instructive to look at the different ways in
which “satisfaction” is measured by some industry segments,

Cars, motorcycles, trucks, buses and other road-


transport vehicles are often grouped in one sector.
Appearance, ride quality, safety and fuel economy
are the most important factors in design – apart
from cost, of course. Plastics and Steel are the
most commonly used materials. Most often,
aesthetics and cost dictate the visible areas
(interiors and body), with the other parts being
designed for efficiency in function. Performance
and cost, in most cases, determine satisfaction
levels.

4
CAE and Design Optimization - Basics Engineering Design Practice

Aircraft and space vehicles are often clubbed in the “aerospace”


sector. Cost is rarely an issue, but performance is of paramount
importance. Safety regulations in passenger aircraft are among
the most stringent in the engineering world. Combat aircraft are
subject to harsh environments. Spacecraft can gain useful life
with every gram of weight shaved off. Advanced materials –
ceramics, composites, honeycombs, and exotic alloys – tend to
be widely used, along with advanced manufacturing techniques.

Ranging from the cardboard boxes that toothpaste


tubes are sold in to the containers of cars that are loaded on to cargo
ships, “packaging” is a multi-billion dollar industry. Not very surprising,
in fact, when you stop to consider that products like toothpastes and
soft-drinks are literally sold in billions. Packaging materials are often
“outside” standard engineering technology – Styrofoam and paper are
rarely treated as load-bearing materials in engineering courses!

Products such as cell phones, stereos, watches,


washing machines, etc. rarely cause loss of life if
they fail. Sometimes called “consumer goods”, these
products are usually designed for elegance and cost.
Product life is sometimes measured in weeks, translating
into extreme time pressures on designers. Plastics are
used very widely.

One thing that’s common to all sectors is the fact that


designers are always under pressure to create better products in less time
and at a lower price.

And this, of course, is why optimization plays such an important role in


product design.

CAE And The Design Cycle


It’s long been recognized that the designer is almost always under pressure
to meet time targets, performance targets and cost targets. The figure
below shows the “typical” design cycle:

5
Engineering Design Practice CAE and Design Optimization – Basics

The design cycle almost always originates with a drawing – a sketch to


illustrate a concept - and almost always ends with a drawing – the
manufacturing drawing. This is the biggest problem – how to translate the
sketch into an acceptable, manufacturable design. A typical design cycle
involves numerous trade-offs: appearance vs. function, cost vs. ease of
manufacture, etc. Every trade-off changes the design, and changes are
inevitable. One of the rules written by Kelly Johnson, legendary head of
Lockheed’s Skunk Works, demands that “A very simple drawing and drawing
release system with great flexibility for making changes must be provided”.

The widespread use of 3D CAD software has made it easier for engineers to
re-create manufacturing drawings when the design changes. But CAE1 is
often viewed as a visit to the dentist: put off as long as possible, and usually
painful.

The reason for this is easy to find. Since CAE has traditionally been used to
verify the preliminary design, analysts usually bear bad news: that the
design has failed the verification and must therefore be changed. If the
analyst carries good news, it’s often ignored since it’s too late in the design
cycle to implement the changes!

Wouldn’t it be great if the concept-designer had a tool that could help


suggest designs that are least likely to get rejected by subsequent CAE?

The Impact of Optimization on CAE


Relatively recent advances in mechanics and software have provided just
these capabilities: software can suggest the design that is best suited to the
conditions you specify. In other words, an “optimum” design.

1
Short for Computer Aided Engineering. Usually taken to mean simulation of
performance under operating conditions.
6
CAE and Design Optimization - Basics Engineering Design Practice

Design vs. Analysis


As opposed to analysis software, optimization tools are for designers. Put
forward your definition of a “satisfactory” design, and the optimization
software will suggest to you shapes and sizes that are most likely to pass
the subsequent analyst’s verification.

This means a tremendous change in the way we can now view CAE: the visit
to the dentist will not be as unpleasant! Since our design has been
“certified” by the optimization tool, we can approach the traditional
“verification” stage with a much higher level of confidence.

In the rest of this book you will learn how OptiStruct brings these
capabilities to the designer.

Other Designer Issues


This section summarizes some design approaches that are not addressed by
linear optimization – which is the focus of this book. If you think the below
are relevant to your problem or are of interest to you, you should read the
other volumes of this series.

Design-Of-Experiments
Statistics teaches us that in a “Normal Distribution”, a
large part of the population lies within 6 standard-
deviations of the mean. Engineering industries refer to
this as “six-sigma” quality – less than 3.4 failures per
million parts produced.

To achieve these levels of quality, designers setup


numerical experiments to account for the statistical
variations in various design parameters – the elasticity modulus of bars of
steel, the rigidity of supports, the variations in loads, and so on. Designers
also use non-linear equations to represent material behaviors.

As a result, optimization tools are often classified as linear-optimization


software, such as OptiStruct, and design-of-experiment tools such as
HyperStudy that interface with non-linear solvers like Radioss.

Multi-Disciplinary-Optimization
As mechanical engineers, we study the difference between mechanisms and
structures, and develop different sets of equations to design these. We
study equations that govern the flow of heat, equations that govern the flow

7
Engineering Design Practice CAE and Design Optimization – Basics

of fluids, equations that govern the stress-distribution, …each as a different


subject, with little if any interaction between the equations.

How would you optimize a product that has some parts that move rigidly
(i.e. act as mechanisms) and some that flex (i.e. behave as structures), and
that has to withstand stresses and also be aerodynamic?

MotionSolve, used in conjunction with OptiStruct, allows designers to find


optimal solutions to some of these problems, while HyperStudy addresses
others.

Process Optimization
A product that’s been designed and verified still stands the risk of rejection:
by the process engineers, if it’s too expensive to manufacture. How can you
design the manufacturing process to be most “satisfactory” – in other words,
how can you optimize the manufacturing process?

HyperForm is used to simulate the sheet-metal forming process, and,


together with HyperStudy, can be used to arrive at optimal process designs.

Summary: How Engineers Should Design


All these tools, then, change the way you should address the design cycle.
Rather than use CAE to verify (and in all probability reject!) suggested
designs, the preferred approach is to use software to suggest a design that
is much more likely to work. In many cases, the “redesign” cycle is all but
eliminated.

Computer software continues to redefine the way products are designed.


But that does not eliminate the need for engineering judgment. In fact, it
increases the burden on the engineer, who now has to act both as the
investigator and as an impartial and knowledgeable judge.
8
CAE and Design Optimization - Basics Engineering Design Practice

First, you phrase the design requirements in as realistic a way as possible.


You may even want to try different statements of the different
requirements.

Then you ask the software to come up with its suggestions.

Finally you sit in judgment: which of the statements was the most realistic,
and which of the suggestions do you want to adopt?

Engineering problems are under-defined, there are many solutions, good, bad
and indifferent. The art is to arrive at a good solution. This is a creative
activity, involving imagination, intuition and deliberate choice.
Sir Ove Arup

9
Optimization Theory CAE and Design Optimization – Basics

Optimization Theory
What is an Optimum Design?
It’s evident from the previous chapter that as a designer, you should search
for an optimum design. What is not so clear is how exactly we can recognize
the “optimum” design. The dictionary definition is a good place to start. An
“optimum”, says the dictionary, is “the greatest degree or best result
obtained or obtainable under specific conditions”.

It’s the phrase “specific conditions” that gives you your design freedom. As a
designer, you define the conditions that allow you to evaluate your design
alternatives. In engineering terms, this means you draw up mathematical
equations that quantify the performance of a design. The statement “good
ride quality” would translate, for instance, into a specification of the
maximum values of the components of acceleration that the passenger’s
seat can experience. The quantitative parameter that you use to evaluate a
design is called the objective.

Of course, you may well have multiple objectives. For instance, it’s very
likely a car designer would simultaneously want excellent safety and low
cost. Unfortunately, in many cases, the objectives are contradictory, making
it increasingly difficult for the designer to reach the best compromise2. A
working design almost always involves a compromise of some sort or the
other.

To make things harder for you, few designers have the luxury of infinite
resources in the pursuit of their objectives. Whether the resources are the
money you can afford to spend on materials, the amount of fuel the
spacecraft can carry or the maximum drag coefficient permitted for a sports
car, there are usually limits you have to work between. These limits, or
constraints give rise to the subject named constrained optimization. A
solution that satisfies the constraints is called a feasible solution, while one
that does not is called an infeasible solution.

It’s important to realize that not all design is done from scratch. In several
cases, we have to start from existing designs and improve them to the best
extent possible. This could be for various reasons, ranging from the

2
MOO, or multi-objective optimization, is covered in CAE And Design Optimization -
Advanced.
10
CAE and Design Optimization – Basics Optimization Theory

necessity to liquidate existing inventory to the


modification of a manufactured design that has failed a
test.

If you’re starting from scratch, you can list the objectives


and constraints and search for the best solution. If you’re
working on modifying an existing design things are
usually a little harder since you have less flexibility to
change things.

Mechanical Engineers face one further requirement. Most


components you design have to assemble with other components. They
need to fit together. This means you have to work with a package space
within which your component needs to fit3, and assembly points that cannot
be varied since they’re decided by other components. In mathematics, the
package space is referred to as the design space or the optimization domain.

Finally, you may not be allowed to change every possible parameter. For
example, the material you can work with may be restricted by factors
beyond your control: working with sheet steel limits you to commercially
available thickness. The parameters that you have the freedom to vary are
called design variables.

The dependence of the objective on the design variables is expressed as an


equation, called the objective function.

The statement of the design optimization problem then, consists of the


package space, the design variables, the constraints and the objectives. If
you have any of these wrong, it’s pretty likely your design proposals will be
useless!

Analysis and Design


Coming up with a concept involves a synthesis of ideas to suggest different
alternatives or proposals. Evaluating the performance of each proposed
design involves analysis of function of that particular proposal. As a
designer, which should you focus on?

The nice part of using Design Optimization as a part of CAE is that you can
simultaneously do both, instead of doing them one after the other. As we

3
CAD users will be familiar with the use of “envelopes” in 3D modeling
11
Optimization Theory CAE and Design Optimization – Basics

saw in the previous Chapter, the separation of conceptual-design and


design-verification into distinct steps was one of the main
reasons analysis is frowned upon even though it’s essential for
good product design.

In the conventional design process, the designer would have to


rely on experience or insight to come up with proposals. The
analysis tool is then used to evaluate each proposal, with the
designer using these analysis results or responses to choose
the “best”.

Optimization changes this. The designer outlines the constraints, and leaves
it to the optimization tool to come up with proposals. The optimizer uses the
analysis tool to decide how to change the initial design to arrive at a better
one.

In qualitative terms, an analysis problem has only one correct answer4.

Design, of course, has no single “correct” answer.

There are always a variety of options that can satisfy the same
requirements, which is why it is extremely important to search for an
optimum design. This is the reason a good analyst often does not make a
good designer!

For the designer, then, analysis and optimization are very much
complementary functions. They are equally important parts of design
optimization: a design optimization model consists of an analysis model and
an optimization model. These are related and dependent but distinct areas,
so we will take some care to understand which parts of the design problem
will be defined in the analysis model and which in the optimization model.

This chapter outlines the background of optimization, while the next outlines
the basics of one of today’s most popular analysis methods, Finite Element
Analysis. Other analysis methods can also be used, of course, as in multi-
disciplinary optimization or non-linear optimization. These are covered in the
other volumes of this series.

4
At least in linear analyses, where uniqueness-of-solution is an important
mathematical proof.
12
CAE and Design Optimization – Basics Optimization Theory

The world of optimization is a hard one to live in. It’s a little like
being asked to search for a black cat in a dark room. You know
it’s in there somewhere, but have to feel your way forwards,
backtracking and changing direction frequently since the cat
changes its position every time you move5. In the world of linear
equations, at least we’re assured that there’s a cat in the room,
and that there’s only one cat to look for. In many real world
problems, we cannot always count on this, as we’ll see.

Since a person who analyses is called an analyst, perhaps a person who


seeks to optimize should be called an optimist!

Our objective, then, is to find a better design than the one we are starting
with. In some cases it will be the best while in other cases it may not.

Finding An Optimum
Since we will be happy to find a better solution even if it’s not the best, we
are looking for an optimum solution, not necessarily the optimum solution.
Why are we emphasizing this statement?

In optimization theory, by convention,


we search for the minimum of the
objective function. This is not a
limitation since maximization of an
objective is equivalent to minimizing its
reciprocal6. A function that has only
one minimum within the optimization
domain is called a convex function.

It’s useful to recall the basics of


differential calculus. In calculus, a minimum (as well as any other “turning
point”) of a curve is characterized by a zero slope (or first derivative). If the
objective function is a quadratic function of the design variables, we are
then guaranteed a global minimum. This is because a second order curve
has only one turning point – and therefore only one minimum in the design
space.

5
In mathematical terms, this behaviour is a characteristic of implicit equations. The
“knowns” and “unknowns” cannot be neatly separated into the right-hand-side and
left-hand-side.
6
Sometimes maximization of x is addressed as minimization of the negative value of
x, i.e. -x
13
Optimization Theory CAE and Design Optimization – Basics

A higher order curve may7 have multiple turning points within the design
space. If it does, then we may have multiple minima8. The turning point at
which the objective function has the least value is the global minimum, while
the other minima are called local minima.

A real life problem may well have


hundreds, if not thousands of design
variables. And the objective function
may well be a non-convex function,
with multiple local minima within the
design space.

How does optimization software


arrive at a better solution within a
reasonable time? How does it
interface with or use analysis
software?

The Mathematics of Optimization


Note that this book is restricted to “linear” problems only –where there is a
linear correlation between inputs and responses. For non-linear problems, an
entirely different approach is used, as described in CAE And Design
Optimization - Advanced.

To recap, to define a problem in design optimization you must specify the


design space, the design variables, the constraints, and the objectives.

The corresponding mathematical statement is:

Minimize f(x) = f(x1, x2, x3, …. xn)

Subject to gj(x) ≤ 0, j = 1, … m

xiL ≤ xi ≤ xiU

7
A higher order curve has more than one turning point, but some may lie outside
the design space.
8
Recall your calculus: a turning point can be a maximum, an inflection point or a
minimum.
14
CAE and Design Optimization – Basics Optimization Theory

where f(x) is the objective function, g(x) are the constraint


functions, and x is a vector of design variables.

An Example
We may be asked to design a light-weight bracket that has to fit in
a 300 mm x 300 mm x 600 mm volume. We want the bracket to
be made of steel, to carry a load of 100 Kg. The maximum
permissible deflection of the bracket is 0.1mm, and the maximum
permissible stress is 20 Kg/mm2. We are allowed to use sheet-steel
that can be 1 mm, 2mm or 4 mm thick.

In this case, our design space would be the 300 mm x 300 mm x 600 mm
volume. The objective would be to minimize the mass. The constraints
would be the permissible stress and deflection. The design variables would
be the thickness of the steel, and the layout of the steel (i.e. how the sheet
should “flow”– where material should be located - within the design space).

To solve a problem like this, the optimizer would start with an initial
configuration or proposal. It would ask the analysis software to evaluate the
mass, stress and deformations of this configuration – the values calculated
by the analysis package and tracked by the optimizer are called responses.
The optimizer would evaluate the sensitivity of the responses to the various
design variables, and decide which to change and by how much.

When the design variables change, the responses change too. If the steel
thickness changes, the mass of the bracket changes. The displacement
would probably change too, as would the stress. So the optimizer would
again need to ask the analysis package to evaluate the responses.

This iterative procedure would continue until the optimizer concludes it has
found the best possible design for the given constraints and variables.

Evaluating Sensitivity
Evaluating the sensitivity of the responses to changes in design variables is,
obviously, a very key part of the optimization process.

If we define the response as a function of the deformation u by the equation


g= T
u
then the sensitivity of the responses to the design variables is given by the
equation

15
Optimization Theory CAE and Design Optimization – Basics

∂g ∂ T ∂u
= u+ T

∂x ∂x ∂x
Some design problems have more constraints than design variables, while
others have more design variables than constraints. Different algorithms are
used by OptiStruct for each case, in order to efficiently arrive at the
optimum solution.

The Optimization Model


Asking the analysis package to evaluate the responses each time a variable
is changed can be very expensive in terms of computer time. OptiStruct
takes a different approach: the optimizer builds an approximate model, and
does most of its work within this approximate model, turning back to the
analysis software only when essential. This makes the optimization much
faster.

It also has another implication. The analysis model itself is an approximation


of the physical behavior of the product. Since the optimization model is an
approximation too, the responses evaluated by the optimizer are unlikely to
be very precise. They are twice removed from the physical product. This
means that as a designer you must subject the final proposal of every
optimized design to a verification-analysis.

There are various techniques the optimization model uses to reduce


computer-time and still get an accurate solution. Most of these are
programmed into OptiStruct. As a designer you can control these methods,
but doing that requires a good understanding of the mathematics. That’s not
the intent of this book. Here, we want to develop a good understanding of
better product design.

The following sections summarize some of the more important techniques.


Remember that the intention is not to be mathematically rigorous. Rather,
the intention is to provide you with an overview of the workings of the
optimization model. This should help you phrase your design problems
correctly.

Managing Local and Global Minima


Even if the analysis-model is linear, the optimization problem is frequently
not. Take, for instance, the deflection of a cantilever beam that has a
rectangular cross-section. The deflection equation is

16
CAE and Design Optimization – Basics Optimization Theory

wL3
δ=
3EI

The analysis model is linear since the equilibrium equation is a linear


function of the state variable δ. If the Elasticity Modulus (E) were a function
of the deflection, as in a plastic analysis, the analysis model would be non-
linear.

Suppose we want to choose an optimum depth for the cross-section. The


Moment of Inertia is given by

bd 3
I=
12

which is not a linear function of the design variable – d.

Depending on the objective function chosen, the optimizer might have to


search for the minimum of a non-convex function.

One of the parameters that determine whether the optimizer finds a global
or a local minimum is the starting point of the search – the initial
configuration or proposal. Another is the move size, which is the step that
the optimizer takes in the direction dictated by the Gradient Search
algorithm. If the step is too large, the optimizer may overshoot the
optimum, which means it will have to reverse its direction in the next
iteration. If it’s too small, the optimizer may take too long to locate the
optimum.

What does this mean to you as a designer? First, you can vary the move size
if the optimizer doesn’t converge. Second, an intelligent choice of the initial
configuration and design variables can significantly affect the design
suggested by the optimizer.

Convergence and Iteration Control


As the optimizer searches through the design space, it needs to check
whether proposals it comes up with are indeed optimal or not.

Mathematical formulations of the optimization problem usually include


Lagrange Multipliers, which can be interpreted as a method to find extrema

17
Optimization Theory CAE and Design Optimization – Basics

of a bounded surface. The objective function, of course, can be viewed as a


surface, with the constraints as boundaries. An interesting interpretation of
the Lagrange Multipliers is provided by S.Jensen9:

the constraint function g(P) can be thought of as "competing" with


the desired function f(P) to "pull" the point P to its minimum or
maximum. The Lagrange multiplier λ can be thought of as a
measure of how hard g(P) has to pull in order to make those
"forces" balance out on the constraint surface

The Karush-Kuhn-Tucker conditions, also called the Kuhn-Tucker conditions,


are a necessary condition for the solution of an optimization problem to be
optimal. The KKT conditions are often used not to find the solution but to
obtain information about the solution. This is useful to us, since we are
looking for a design solution, not a mathematically precise solution.

From our design perspective, it is important to understand that the search is


an iterative procedure. First, we can instruct the optimizer how long to
search, by telling it the maximum number of iterations. Further, we can tell
it how fine the search should be. If the difference between two successive
proposals is less than a convergence tolerance, the optimizer can be asked
to conclude that this is acceptable to us from a design perspective.

Gradient Search Methods


Most engineers are familiar with Newton’s method to find the roots of a
polynomial. As shown in the figure, this method uses the slope of the curve
to guess at which direction the initial guess should be adjusted in – to
increase or decrease it. In practice, the gradient is often computed using a
finite difference method.

9
See http://www.slimy.com/~Esteuard/professional.html for an excellent
introduction.
18
CAE and Design Optimization – Basics Optimization Theory

The gradient search method, also called the method of steepest descent, is
one of the many methods used by the optimizer to move from the initial
configuration to the final solution. Non-linear optimization normally uses
other methods, as described in CAE And Design Optimization - Advanced.

Workable Implementations
Very often, an exact answer is worthless if it comes too late. An approximate
answer that is available in time is much more useful.

In order to speed up the optimization process, the optimization model uses


Constraint Screening, Constraint Linking and Constraint Deletion.

The first, constraint screening, is a technique used to identify which of the


constraints are critical to the current iteration. In an effort to reduce the
number of variables, the optimizer uses one or more criteria to choose a
subset of all variables for each iteration. This subset is likely to change from
one iteration to another as the optimizer moves through the design space.

Constraint linking is when you can use factors such as symmetry to reduce
the number of constraints that need to be considered. Suppose you want all
beams in a structure to use the same cross-section because it makes the
purchase process easier. In this case, it makes sense to link all of them
together, thereby reducing the load on the optimizer.

As the optimizer searches through the design space, the current


configuration may violate only 2 of 3 constraints. In this case, the third
constraint is not important for the iteration. It can be marked inactive and
ignored in other words, the constraint can be deleted for this iteration.

Summary
Part of the challenge of optimizing a product is that Design Variables
Things that can vary – thickness,
designers are not always able to clearly define their density, etc.
design problem or state their definition of “optimum”.
Don’t let this deter you too much. Even if you don’t Responses
arrive at the “best” design, any improvement over Things calculated by the analysis
your current proposal is better. model, and of interest to the
Optimizer. Mass, deflection,
stress, etc.
Optimization technology is fairly robust today. Most of
the methods outlined above are implemented Constraints
intelligently by the software. You can, however, make Limits on responses or design
things easier for the optimizer and reduce your variables.

Objective 19
Value that measures quality of
your design. Mass, frequency,
center of gravity, etc.
Optimization Theory CAE and Design Optimization – Basics

design time by intelligent choices in both phases of design optimization: the


Optimization Model and the Analysis Model.

You can also, of course, set the optimizer an impossible task if the
statement of your problem is itself wrong.

Therefore O students study mathematics and do not build without


foundations.
Leonardo Da Vinci

20
CAE and Design Optimization – Basics FEA Essentials

FEA Essentials
As we’ve seen, design optimization relies on CAE to calculate the response
of the product.

Computer Aided Engineering, unfortunately, is a catch-all phrase that’s not


very well defined. It can mean just about anything. And often does. In more
general usage, CAE, sometimes also called “Simulation”, is the use of
numerical analysis to study various behaviors of products. How they react to
forces, what drag or lift they experience in a fluid, how they react to
different thermal conditions, the forces generated as they experience
accelerations, and so on. The other volumes of this set10 address some of
these aspects.

In the context of our study, however, we’ll focus mainly on the Finite
Element Method. Without going into the mathematics of the method, we’ll
look at those aspects that let us understand how it fits into our goal: using
optimization to enhance product design.

Building models for analysis involves making approximations to the initial


geometry to omit irrelevant details, specifying conditions on the boundary or
at initial time, specifying solver options, choosing output options, and so on.
It can help, therefore, if you are familiar not just with the terms used but
also with some of the background. This can help you make intelligent
decisions when you prepare your models for optimization.

This chapter is not intended to be rigorous – numerous textbooks are


available that do that job admirably well. Rather, this is a quick summary of
some of the salient aspects of Finite Element usage and theory.

Why use Numerical Methods at all?


Most engineering problems can be solved using one of two methods:
analytical or numerical11. An “analytical solution” is a mathematical
expression that gives the values of the desired unknown quantity or
quantities at any location in the body, and at any instant of time. But
analytical solutions can usually be obtained only for relatively simple

10
See CAE And Design Optimization – Advanced and CAE For Multi Body Dynamics.
11
A third method, often neglected by engineering beginners, is Physical Testing, or
“Test”.
21
FEA Essentials CAE and Design Optimization – Basics

problems: as the geometries or mechanics or both become complicated, the


effort of finding an analytical solution is often so high that the solution
cannot be found at an acceptable cost or in an acceptable time.

For “complicated” problems, numerical methods provide approximate


solutions that are usually of adequate accuracy. One way of setting up these
numerical solutions is to discretize the original body. This means we break
the original geometry into several smaller geometries. We first solve the
equations governing the mechanics over these smaller bodies, then piece
the results back together to get the complete solution.

The two most widely used methods are Finite Element Analysis and Finite
Difference Methods. The latter are used mainly for problems in
Computational Fluid Dynamics (CFD), while the former is used in a wide
range of applications.

What is Finite Element Analysis?


Finite Element Analysis (FEA) simulates a physical part or assembly’s
behavior by dividing the geometry of the part into a number of elements of
standard shapes, applying loads and constraints, then calculating variables
of interest – deflections, stresses, temperatures, pressures, etc. The
behavior of an individual element is usually described by a relatively simple
set of equations. Just as the set of elements would be joined together to
build the whole structure, the equations describing the behaviors of the
individual elements are joined into a set of equations that describe the
behavior of the whole structure.

One way of looking at it is to recall the approach you studied in Engineering


Mechanics12. There, you drew free body diagrams of each member in the
structure, wrote equations that related the unknown forces in each member,
then wrote equations that had to be satisfied for the forces between
members if equilibrium is to be satisfied. Solving these equations gave you
the forces in each member.

Elements themselves are defined by specifying the nodes, which are the
vertices of the element. Just as 4 corners define a rectangle, the nodes
define the shape of an element.

12
A more complete discussion is presented in A Designer’s Guide To Finite Element
Analysis.
22
CAE and Design Optimization – Basics FEA Essentials

When you choose an element to represent a part of the product, you are
also specifying the parameters that define the behavior across the element.
For instance, in a stress analysis, if you know the 6 components of
deformation13 at any point, you can calculate the strain from this by taking
the first spatial derivative. And once you know the strain, you can use the
material properties to calculate the stress. For the Finite Element Method,
every node has these parameters associated with it, just as in a truss-
structure every member has forces associated with its end-points. From the
values at the nodes, you can interpolate for the values between the nodes.

Suppose you were asked to digitize a surface, using a Coordinate Measuring


Machine. Unless your surface were absolutely flat, you would not space the
measurement points evenly. Since you have to interpolate between
measured values, you would naturally choose to have more measurement
points at areas where the surface curves sharply. In maths, we say these
areas have a high derivative, or rate-of-change.

In a similar fashion, for an FE analysis you would create smaller elements


(which means more nodes) at areas where you expect the stress to be
high14. The choice of the sizes of elements depends on many things - the
anticipated stress levels of a certain area, the detail wanted in the results,
the stability of a solution algorithm, the available computational power, and
so on.

A Finite Element program takes the elements you have defined, lists the
equations for each unknown value, puts them together as a matrix equation,
then solves all these for the values of the unknown parameters.

The equilibrium equation is of the form

[K ]{u} = { f }
Since it’s analogous to the equations of spring-deflection, K is often called
the Stiffness Matrix, u is called the deformation vector, and f is called the
load vector. K is a square matrix, with one row (and column) for each

13
The 6 components are the translations along the 3 axes, and rotations about the 3
axes
14
A high stress means a high strain, from Hooke’s Law. Strain is the first derivative
of deformation. Hence a high stress area is one where the deformation has a high
derivative. And this, of course, means the rate-of-change of deformation is high in
areas of high stress.
23
FEA Essentials CAE and Design Optimization – Basics

unknown variable in the problem-definition. If, for instance, you have used
100 nodes in your model, and each node has 6 unknowns15, your stiffness
matrix would be 600 x 600. u and f are each column-matrices. In our
example, each has 1 column and 600 rows.

A computer is required because of the large number of calculations needed


to analyze a part or assembly. It is not uncommon for a model to have more
than 1,00,000 unknowns (called degrees of freedom). The power and low
cost of modern computers has made Finite Element Analysis available to
many disciplines and companies.

Finally, remember that most Finite Element Analysis models are applicable
only to “structures” – they cannot be applied to “mechanisms”. Components
such as the shackles that hold up the leaf-springs of a truck chassis require
different treatment. These are not treated in this volume – refer to CAE For
Multi Body Dynamics for that.

What are Finite Difference Methods?


In some areas, mainly in fluid flow, analysts often prefer to use a different
mathematical approach than FEA. In this approach, there are no elements –
the discrete points are referred to as grid points or grids. Some analysis
programs call for “structured” grids – the numbering of and positioning of
grid points must follow specific patterns. Other analysis programs are less
stringent in their requirements – unstructured grids or blocks are supported.

Choosing a Numerical Model


As a designer, you need to anticipate the behavior of the product you’re
designing. You will need to guess at the conditions it is likely to be exposed
to, and then to predict how it will respond to these conditions.

In some situations, the behavior is independent of time – these are called


steady state problems. In others, the solution varies with time – these are
called transient problems.

In some situations, the response of the body to stimuli is linear. That is,
there is a linear correlation between input and output. Such a model is,
obviously, called a linear problem. Other situations are non-linear because
there’s no linear dependence between stimulus and response.

15
The 6 components of deformation are the translations along 3 axes and the
rotations about the 3 axes
24
CAE and Design Optimization – Basics FEA Essentials

It’s important to remember that the product you’re analyzing does not know
whether it is “linear” or not. You, as the analyst, can choose to model it as
linear or as non-linear, depending on which is more likely to give you useful
results. Since we are designers, not mathematicians, we are not interested
in results that are “exactly correct”. We are willing to settle for
“approximately correct” provided we get the results in time and at a cost we
can afford.

As you know from your courses on Linear Algebra and Differential Equations,
linear equations are far easier to solve than non-linear equations. Therefore,
we very often choose to model behaviors as linear even if a non-linear
model is more precise. We tend to choose non-linear models only if there’s
no linear model that’s even reasonably accurate.

Non-linear models are of several types – the materials used, the geometry
involved, or conditions on the boundary can cause the “non-linear” nature.
Examples of material non-linearity are plastic deformation, melting and
solidification – the stiffness of the body changes as the material properties
change. In other problems, the stiffness changes as the body deforms even
if the material’s properties do not change – take for example the reduced
rigidity of a plastic bottle as it is crushed. Examples of boundary non-
linearities are contact and thermal radiation. In the former, the stiffness of
the part or assembly changes as sections come into contact with each other.
In the latter, the heat lost is proportional to the 4th power of temperature.

Some models, such as those required to simulate the behavior of a car when
it crashes, can involve several of these types of “non-linearities”.

The Role of Physical Testing


Remember that the the terms in the preceding paragraphs describe the
mathematical model of the physical behaviour. It’s your responsibility to
exercise your engineering judgement to ensure that the model you’ve
chosen does a reasonably good job of capturing the physical behavior. In
order to make it easier for the design engineer to verify this, in actual
engineering design, results of physical tests are used to verify that the
numerical model is capable of reproducing conditions obtained under test
conditions.

Then why not just test a physical model? Why simulate it at all using a
numerical or analytical model?

25
FEA Essentials CAE and Design Optimization – Basics

For two reasons. First, constructing test models is expensive and time
consuming. In many cases, there’s no way to reduce the time for the test.
This is starkly different from computer-methods. In the computer-world, a
faster computer produces results faster. This same time-compression effect
cannot be obtained in most tests! Second, tests themselves are not very
easily controllable.

As a result, most engineering practice requries that the analysis model be


validated against a test result. That is, the model is used to simulate
performance of the product under conditions similar to an existing test. If
the model is capable of doing this, then we assume it is capable of
reproducing behaviour under different conditions too. As a result, we can
dispense with the physical test for further studies.

Quick Summary of Analysis Terminology


Linear, Static
This model is used when the response of the body is linear, and there’s no
variation with time. In stress analysis, this model is appropriate when
operating within the elastic region (i.e. the stress-strain curve is linear) and
when the deformations are small16 and when the loads do not vary with
time.

This model is used widely since it’s quick to solve and relatively easy to
interpret the results. Very often, even if a non-linear model is more realistic,
a linear model is used to investigate likely behavior. Once the options have
been narrowed, a full non-linear analysis can be used.

The equilibrium equation is

[K ]{u} = { f }
where K, u and f are functions of x, y and z only – they are independent of
t.

16
Remember that large deformations need not mean inelastic behaviour.
26
CAE and Design Optimization – Basics FEA Essentials

Linear, Transient
In stress analysis, this model is appropriate when operating within the
elastic region (i.e. the stress-strain curve is linear) and when the
deformations are small but when the external conditions do vary with time.

Transient problems themselves are often subdivided into further classes,


depending on whether the load varies with time in a periodic fashion or not.
If the load is periodic, as for instance if the excitation source is a rotating
unbalance, it is called “Harmonic” excitation. This is usually easier to solve
than when the load is non-periodic.

It’s important to note that Finite Element mathematics is applicable only to


spatial discretization. A Finite Difference method is usually used to step the
solution forward in time, from the initial time to the final time. To setup the
problem for analysis, then, the values at the boundary are specified at the
initial time (often referred to as “t = 0”). Time-variant solutions can also be
calculated by representing the solution17 as a weighted sum of the “mode
shapes”.

In the equilibrium equation

[M ] 2  + [C ] ∂u  + [K ]{u} = { f }
 ∂ 2u 
 ∂t   ∂t 

K is a function of x, y and z only – it is independent of t. F and u, however,


vary with t. M represents the mass, and C the damping.

Normal Modes
Sometimes our design problem is not just to calculate stresses or
deformations. We may be interested in identifying the resonance frequencies
of the system. In vehicle design, avoidance of resonance enhances ride
comfort by cutting out unwanted rattles. When designing a loudspeaker or a
megaphone, on the other hand, you may want resonance to occur.

In cases like these, we need to solve the “eigenvalue18” problem and


evaluate the natural frequencies of the body.

17
Look up the use of Rayleigh’s method or Dunkerley’s method. Also look up the
Rayleigh-Ritz method for a rough idea as to how this works.
18
Refer to a course or a text on Dynamics of Machinery for more details
27
FEA Essentials CAE and Design Optimization – Basics

The equilibrium equation is

 ∂ 2u 
[M ] 2  + [K ]{u} = {0}
 ∂t 

where K and u are functions of x, y and z only – they are independent of t.


The solutions to this equation are pairs of natural frequencies and the
corresponding “mode shapes”.

Random Response
In some situations, we cannot specify the exact value of the loads as a
function of time, but can specify the total energy in these loads. An example
would be the forces experienced by a plane when its engines are firing. We
know the total energy being transferred from the jet engines to the frame,
but cannot claim that we know the loads precisely as functions of time.

In cases like these, the loads are characterized by Probability Density


Functions, and the behavior is called stochastic. The designer’s goal then is
to predict a probability of safety.

The several ways to evaluate these responses is beyond the scope of this
book.

Inertia Relief
Setting up a Finite Element model for static analysis requires that the
structure be supported adequately. Some structures, like aircraft in steady
flight, are not supported explicitly but are still best represented by static-
analysis models. Inertia Relief is an approach used to model such problems.

Frequency Response
In many designs where vibration is important, and correlation with test-
results is essential, designers have to characterize the response of the
structure as a function of frequency-of-excitation instead of as a function of
time. In these cases a Fourier Transform converts the equilibrium equation
from the “time domain” to the “frequency domain”.

28
CAE and Design Optimization – Basics FEA Essentials

In the equilibrium equation, the variables are expressed as functions of ω


rather than time. This is called the frequency domain. Of course, the Inverse
Fourier Transform can convert the solution back to the time domain.

Linear Buckling
Designers sometimes have to take into account the fact that even if stresses
are less than permissible values, the structure may fail if it buckles – like a
tall column in compression. The equilibrium equation is similar to that of
Normal Modes, but the results are interpreted as a “buckling load factor”.

Buckling load factors are often important in the design of aerospace


structures, where the quest for a minimal weight and the use of advanced
materials leads to the frequent use of thin-walled designs.

Non-linear – Gap / Contact


The terms “gap” and “contact” are often used to mean the same thing – an
opening in the body that may close or widen under the influence of external
factors.

Clearly, if a gap closes or opens, the stiffness of the body changes. Since the
gap opens or closes depending on the deformation, this means the stiffness
depends on the deformation. In other words, in the equilibrium equation

[K ]{u} = { f }
K is a function of u, making the equation non-linear. The force, f, too can
be a function of u.

Component Mode Synthesis


When working with large models, resource constraints sometimes force the
analyst to break the problem into smaller parts. In static analysis, this
approach is called sub-structuring. When used in dynamic analyses, it is
called Component Mode Synthesis.

It’s sometimes impossible to treat the product purely as a structure or purely


as a mechanism. Consider, for example, the feed mechanism for a high-
speed packing machine. The rates of acceleration that the mechanism
experiences may be quite high. High enough that the deflection of the levers
is large enough, perhaps, for the feed mechanism to jam because of
misalignment.
29
FEA Essentials CAE and Design Optimization – Basics

Designing such a product requires that the equations of rigid-body-


mechanics be coupled with the equations of structural deformation.

Component Mode Synthesis also provides a way to do this.

What are Elements?


An element is a shape for which the Finite Element program can write out
the equations relating the unknown and known quantities. An element is
defined by its nodes – the unknowns at each node are called the degrees of
freedom.

Shapes that are accepted in most finite element programmes are triangles,
quadrilaterals, lines, tetrahedra, pentahedra and hexahedra.

The sizes of and the number of elements usually have a bearing on the
accuracy of the solution. As problems become more complex (advancing in
complexity from linear-statics to nonlinear-dynamics), the requirements on
shapes and sizes of elements become increasingly stringent. These
requirements are often referred to as mesh-specifications, and these are
usually strongly analysis-program dependent.

In most analyses, the more the number of elements, the better the results.
However, the computer time and disk-space required to solve the equations
also goes up. Most analysts have to settle for a quality of results that they
can afford, given the available computer resources.

Fortunately for us, optimization methods are less dependent on mesh


quality. As we saw in the previous chapter, the optimization algorithm makes
some simplifications and assumptions, so it’s not important that a mesh be
“perfect”. Only that it be “adequate”. In your assignments, you will learn
how to judge whether your mesh is sufficiently fine.

Element Types
Choosing the element type is an important part of any Finite Element
analysis. Elements are categorized based on their shape or topology, the
number of nodes needed to define them, and the mechanics or behavior
they represent.

Element types are usually solver dependent – they vary based on the solver
used. The elements listed below are specific to OptiStruct, but are available
in almost every commercially available analysis package.
30
CAE and Design Optimization – Basics FEA Essentials
Categorization based on Mechanics
Beams and Bars (or rods or trusses) are represented by one-dimensional
elements – lines or curves – but can lie in 3D space. Plain Strain, Plane
Stress and Axi-symmetric elements are two-dimensional shapes that can be
used only if the entire model lies in one plane only. Plates and Shells
represent surfaces that are two-dimensional in the sense that they have no
volume, but lie in 3D space. Solid Elements represent volumes.

Categorization based on Topology


Standard 2D Elements (plane strain, plane stress, axi-symmetric, plate and
shell) are either triangular or quadrilateral.

Standard 3D elements are either tetrahedral, pentahedral, or hexahedral. A


pyramid with a rectangular base is a pentahedron, as is a wedge. However
the two are different element types: the pyramid has 5 nodes while the
wedge has 6. Not all solvers support pentahedral elements, and some
support only one of the two pentahedra.

In most stress-analysis problems, quadrilateral and hexahedral elements are


preferred over triangular and tetrahedral elements. For reasons that you can
find in the references listed at the end of this volume, they give much better
results: more accurate and less CPU intensive.

1D elements are all (topologically) curves – either straight lines or arcs,


depending on the number of nodes. Typical applications are as beams, bars,
rods, pipes, springs, cold- or hot-runners, and axi-symmetric shells.

Categorization based on Order


The variation of the unknown quantity between nodes is assumed (by the
analysis code) to be linear, or quadratic, or cubic, etc. Linear and parabolic
elements are the most common. Linear elements have two nodes along each
edge, while parabolic elements have three nodes along each edge.

Further refinements do exist – for instance, parabolic quadrilateral elements


can have either 8 or 9 nodes.

Steps in FE Modeling
Geometry Preparation
While it is possible to build a model directly using elements and nodes, this
is not often done today. The geometry that defines the area to be analyzed

31
FEA Essentials CAE and Design Optimization – Basics

(also called the “domain”) is usually created first using a CAD program, and
elements are created to encompass that boundary or represent the volume.

CAD designers create models for manufacture. As many details are included
as possible. For a numerical analysis, we often choose to ignore aspects that
we think will not significantly affect the solution. For instance, a single hole
of 1 mm radius in a plate that is 2 meters wide can probably be ignored
safely when calculating the deformation of the plate.

Therefore the first task that most analysts are faced with is that of preparing
the geometry for analysis. This involves tasks like removal of features,
extraction of mid-surfaces, extrapolation of surfaces, etc.

Further, the CAD world has an abundance of data exchange formats, since
most CAD applications use proprietary data storage formats. A transfer of
data from the CAD package to the FE preprocessor sometimes results in a
loss of accuracy – gaps are introduced during the import process, for
example. Also, CAD assembly models are sometimes made up of parts that
were created in different CAD applications.

Therefore a cleaning-up of the geometry is often required. This involves


filling gaps, eliminating small edges or surfaces that will mislead the
automatic-mesh-generation routines, eliminating dangling faces, and so on.

Mesh Creation
Once the geometry is more or less ready for discretization, you then start to
subdivide the geometry into elements or grid points. The collection of
elements is usually referred to as a mesh. Meshes that consist of triangular
or quadrilateral elements can often be generated automatically, while
tetrahedral or hexahedral meshes usually require considerable manual
intervention.

Mesh Editing
Once a mesh has been created, the analyst checks if it meets the
specifications – several measures of quality are checked, depending on the
analysis requirements. Usually, some editing of the mesh is required.
Depending on the complexity of the mesh, this can be done either semi-
automatically or manually.

32
CAE and Design Optimization – Basics FEA Essentials

Preparing for Analysis


Once the mesh is ready, additional data is specified – the properties of the
materials used, the thickness or cross-sectional properties of shell or beam
elements, the conditions on the boundaries (restraints, loads or excitations),
initial conditions, data for the specific solution algorithm to be employed,
kind of output required for text and graphics records, and so on.

Once this is done, the data is turned over to the solution program for the
next phase – solving. Data is often written out in the form of a text file,
which is referred to as a deck. Each line of text in the deck is commonly
referred to as a card. A card image is the format followed by the analysis
program to interpret the text on the line.

The procedure of building the Finite Element Model is sometimes referred to


as FEM – short for Finite Element Modeling. Some books, however, use FEM
to refer to the Finite Element Method.

Solving
The model created in the earlier steps is now taken up for solution – the
computer program reads the data, calculates matrix entries, solves the
matrix equations and writes data out for interpretation.

This task is CPU-intensive, and is often called processing19. Most of the time,
very little interaction from the user is required. In some cases, the analyst
periodically monitors results to check that they are indeed on the right track.
If the solution seems to be evolving in an unexpected direction, the analyst
can stop the solver and modify the model, thereby saving valuable time.

Post-Processing
After the program has evaluated the results, the analyst examines and
interprets the data – looking for errors or improvements in design.

As with pre-processing, this calls for substantial interaction from the analyst.

19
Hence the term pre-processing for the preceding steps, and post-processing for
the subsequent steps.
33
FEA Essentials CAE and Design Optimization – Basics

Guidelines on Element Choice


Learning which element to choose is a little like learning driving. Guidelines
exist, but can’t be applied blindly. You need to adapt them to specific
situations. Remember this warning!

If your product has a region that is long and thin, you can probably model it
using beam elements. If this region is connected to the rest of the structure
by pin-joints, then you should use truss elements. Regions that are like
plates are best modeled using shell elements. Any areas that don’t fall in the
earlier categories should be modeled using solid elements.

If you have different element types in your model, there are rules that
govern the assemblage. For several models, we choose to use just one
element type to avoid these complications.

Engineering is the art of modeling materials we do not wholly understand,


into shapes we cannot precisely analyse so as to withstand forces we cannot
properly assess, in such a way that the public has no reason to suspect the
extent of our ignorance.
Dr.A.R.Dykes

34
CAE and Design Optimization – Basics OptiStruct

OptiStruct
Before We Start
The previous chapters outlined the reasons we want optimization to be a
part of the product design cycle, and introduced both the Optimization
Model and the Analysis Model.

The procedure, to summarize, is as follows:

1. As the designer, you decide the design variables,


the constraints and the objectives. You also
choose the design space, the loads and the
restraints – usually dictated by other components
in the assembly. It’s a good idea to list these as a
design-specification document.

2. Next, you prepare the FE model. To do this, you

i. inherit the product definition as a CAD model. If


necessary, you must modify it to omit unnecessary
details. Note that this step is optional. It makes sense if
you are trying to improve an existing design, or if it is
easier to build the design-space in a CAD modeler. You
may choose to define the design space within the FE
pre-processor itself if you’re working on a new concept
with a geometrically simple design space.

ii. mesh the product-geometry or the design space,


depending on which you are starting with. The design
space can, but need not, span the entire product. For
instance your design may not allow you to change
mounting points. In this case, the restraint-areas will not
be a part of the design space, although they will be a
part of the analysis model.

iii. specify material data for the elements – Modulus of


Elasticity, etc.

iv. specify element properties – the thickness of shell


elements or the cross-section for beam elements
35
OptiStruct CAE and Design Optimization – Basics

v. define the forces acting on the body

vi. specify the restraints on the body – where and how it’s
supported

vii. choose the type of analysis you want to perform –


linear-static, modal, etc.

3. Before running the optimizer, you should check that the Analysis
Model is adequate. A good way to do this is to run the analysis for
meshes of different element-sizes. If the reported results
(deformation, stress, frequency, etc., depending on your interest)
do not vary with the mesh, it’s reasonable to conclude that it’s
adequate.

4. Once the FE model is ready, you prepare the Optimization Model.


This means you specify

i. the design variables. Remember that different parts of


the design space can have different variables. You may
have the freedom to place cutouts in one region, but
only to vary the thickness in others.

ii. the responses that the optimization model needs from


the analysis model. The optimizer will use these to
evaluate sensitivities.

iii. the design constraints.

iv. the objective function.

5. Now you are ready to perform the optimization.

6. After the optimization is done, you review the results to check that
the optimization has proceeded in line with your design
requirement. You may have to revise or restate the optimization
model to better reflect the statement of the design requirements.

7. When you are satisfied with the design configuration proposed by


the optimizer, you take this geometry back to your CAD modeler
for further CAD-related work such as drawing generation, etc.
36
CAE and Design Optimization – Basics OptiStruct

In the subsequent sections we’ll review the specific methods OptiStruct


uses20.

Terminology
OptiStruct includes both an FE solver21 and an Optimizer. In other words, it
can be used to solve the Analysis Model and the Optimization Model.

The models themselves are created using HyperMesh, which is the pre-
processor. HyperMesh is used to define both the Analysis Model and the
Optimization Model. The table below lists the key terms used by HyperMesh
and correlates them with the Analysis and Optimization Models.

Analysis Model
Collector A way to group related items together. For instance
all elements that have the same thickness would be
in the same collector.
Load External forces acting on the boundary. Includes
concentrated forces, moments, pressures, gravity,
etc.
SPC Short for Single Point Constraint. Refers to restraints
applied to the analysis model at locations where the
body is supported22.
Subcase Combination of SPCs and Loads. Since they
represent values on the boundary, these are often
clubbed together as Boundary Conditions. A subcase
is sometimes called a Load Case.
Card Some data in the analysis model, such as the
material properties, cannot be displayed graphically.
Such data is entered as a card image by typing in
text or numerical values.

Optimization Model
Response Any quantity calculated by the Analysis Model, and of
interest to the Optimization Model. This could include
20
Some of these features are unique to OptiStruct.
21
See A Designer’s Guide To CAE
22
Do not confuse these with design constraints, which are applicable to the
optimization model.
37
OptiStruct CAE and Design Optimization – Basics

interest to the Optimization Model. This could include


mass of the model, volume, deformation, stress,
frequencies, etc.The Analysis Model calculates a lot
of things, not all of which are relevant to the design
problem. Any quantity or variable that you want to
use as a design-constraint or as an objective must be
identified as responses.
Design Space One or more component collectors that contain
entities that can be altered as part of the design
effort. All other component collectors are non-design
areas.
Desvar Short for Design Variable.
Discrete DVs Design variables that can vary only in predefined
steps rather than continuously.
Dconstraint Design Constraint. Limits on the values selected
responses can take. For instance the maximum
permissible stress, or a range of frequencies to
avoid.
Objective The goal of the optimization. This must be a quantity
that has been tagged as a response. An objective
can either be minimized or maximized. Multi-
Objective Optimization is not covered in this book.
We restrict our attention to a single objective
function23.
Minmax This means you want to minimize the maximum
value of the objective function.
Maxmin Used when you want to maximize the minimum
value of the objective function.
Opti Control Optimization Control. Parameters to control the
optimization algorithms

23
While MOO is covered in CAE And Design Optimization - Advanced, note that you
can define an equation that clubs a series of responses together to form a single
response, which can then be used in single-objective optimizations. This is similar to
the use of an average-mark in exams.
38
CAE and Design Optimization – Basics OptiStruct

Techniques to Design Optimum Products


Using optimization at the design stage itself is a good way to reduce the
chances that the design will fail. Thus far, this has been our main motivation
for modifying the CAE cycle to include optimization.

But your design could be rejected for other reasons too.

Having a design criticized on the grounds that it is too hard or too expensive
to manufacture is no fun for the designer. It is even worse if your proposal
is impossible to manufacture.

Mechanical engineering designs can be manufactured in a wide variety of


ways: they can be cast, machined, molded, drawn, blanked, forged, welded,
and so on. As we saw in the earlier chapter, materials used can range from
steel to paper. You may be limited to working with predefined sizes and
shapes, as for instance with standard-section beams or metal sheets of
standard thickness.

With all these complications, how can you work towards designs that are not
only likely to pass verification, but also likely to be approved as
manufacture-able? In this section, we’ll first run over the techniques
OptiStruct offers, and then see how to put these together in a meaningful
way for product design.

Types of Analyses
OptiStruct’s Finite Element solver can address several different types of
linear analyses: static, inertia relief, normal modes, linear buckling,
transient, and frequency response24. Component mode synthesis and multi-
body dynamics are also supported. Non-linear analysis is restricted to
simulation of gaps.

From an optimization perspective, this means that we can obtain responses


for any one of, or any combination of these analyses. All we need to do is
tell the Optimization Model which responses of the Analysis Model it should
track.

24
Transient analysis and frequency response analysis can be performed using either
direct-integration or modal-superposition.
39
OptiStruct CAE and Design Optimization – Basics

Types of Responses
Responses are quantities that are calculated by the Finite Element solver
and are important for the optimizer.

They can be any of, or any combination of, mass, volume25, center of
gravity, moments of inertia, compliance, natural frequency, deformation, von
Mises stress, strain, and buckling load factor. For transient and frequency
response analyses, velocity and acceleration can be identified as responses.
For laminates26, the failure index can be a response.

Types of Objectives
The objective can be any of the responses, but some care needs to be
exercised. Choosing an objective that’s insensitive to the design variables
makes it harder for the optimizer to reach a minimum. Some guidelines on
appropriate choices for the objective function are included in the
assignments that accompany this book. A more detailed description for each
optimization method is available in the on-line documentation.

Types of Constraints
Constraints are of two types; first, design constraints, which are any
responses that can be obtained from the Finite Element solver.

Second, manufacturing constraints. These are guidelines we give the


optimizer to reflect our preferred manufacturing method. It would certainly
be nice if the optimizer could tell us which of the available materials is best
for the design. Engineering designers, unfortunately, know that the factors
that govern material choice are often impossible to quantify. And we have
been clear that the objective function should be clearly quantifiable.
Therefore the choice of material is the responsibility of the designer, as is
the choice of the manufacturing method. Having made this choice, we use
manufacturing constraints to restrict the set of available designs to those
that can be manufactured.

The various manufacturing constraints available, and their relevance, are


summarized below, with examples to illustrate the difference between
design-optimization with and without these options.

25
For a model that has only a single material, mass and volume are equivalent. But
for products with multiple materials, there’s a significant difference.
26
See the next chapter for details
40
CAE and Design Optimization – Basics OptiStruct

Member Size: Extremely


thin or extremely large
members can pose
problems in metal-casting
or injection molding. Size
control avoids designs like
the one shown on the
left, producing “cleaner”
designs, as shown on the
right.

Draw Direction: Cast or


molded components must
be ejected from the
cavity, which means
undercuts must be
avoided, and a positive
draft must be provided.
Split molds require drafts
on either direction of the
parting surface. Draw
direction constraints yield
designs such as the one
on the right, which is far
better than the one on
the left.

Extrusion: Materials like


Aluminum are often
extruded. The design on
the right, obtained using
extrusion constraints, can
be manufactured in an
extrusion die. The one on
the left cannot.

Pattern Repetition:
Components like airplane
wings have different
sections (ribs) that are
linked together by spars,
covered with a skin and
41
OptiStruct CAE and Design Optimization – Basics

then subjected to a load.


The different ribs must
match topologically so
that the spar can connect
all of them, as shown on
the right. Pattern
repetition constraints
enforce this behavior. The
wing sections on the left
are worse from a
manufacturing
perspective.

Pattern Grouping:
Symmetry affects the
aesthetics of a product.
Repetitive patterns often
look nicer. OptiStruct’s
pattern grouping
constraints enforce
several types of
symmetry: planar, linear,
cyclical, radial and
circular. The figure on the
right uses radial grouping,
while the one on the left
doesn’t. It’s easy to
decide which the better
looking design is!

Laminated Composites require different manufacturing methods. Some of


these are covered in the next Chapter.

Types of Optimization
There are six distinct approaches that OptiStruct can take to arrive at the
optimum design. These can be used either singly, or in sequence, or
simultaneously. In order to decide which to use when, you obviously need to
understand the methods themselves. As with our review of the Finite
Element method, we will focus on a qualitative effort to grasp the spirit of
the methods.

42
CAE and Design Optimization – Basics OptiStruct
Topology
Topology is a well-established branch of mathematics, dealing with
continuous transformations27. From one point of view, it is the study of the
relationships between the edges, faces, vertices and volumes of 3D objects,
independent of the dimensions. A circle and an ellipse, for example, are
topologically identical, though their dimensional characteristics are obviously
different. Several theorems in mathematics require that the topology of the
domain remain the same. From a stress analysis perspective, vertices are
particularly troublesome. Stress is unbounded or singular at a vertex, so
introducing new vertices can change the scenario drastically.

Obviously, restricting a designer to a single topology is unacceptable: even a


simple cutout changes the topology of a design, but cutouts are an essential
part of design!

OptiStruct’s Topology Optimization approach provides an excellent way to


allow the design configuration to change, neatly sidestepping the
mathematical problems caused by changes in topology. There are two
approaches used by OptiStruct. These are called the density method and the
homogenization method, but for the purpose of our qualitative discussion we
will not distinguish between them.

Think of the difference between a steel sieve and a steel plate, both of the
same size. The plate is solid metal, while the sieve has a mesh of holes. As
the holes in the sieve get larger and larger, the sieve not only gets lighter, it
also gets weaker. Making the holes smaller results in a heavier, stronger and
stiffer sieve. From a mathematical perspective, we can treat the solid plate
as a sieve that has holes of zero diameters. We can also calculate an
“equivalent density” for the sieve. When the diameter of the holes is zero,
the equivalent density is the density of steel. As the holes get larger and
larger, the “equivalent density” approaches zero.

In the Topology Optimization approach, OptiStruct treats each element as a


“sieve”. It treats the “equivalent density” of each element as a design
variable. The equivalent density is normalized so that 1 is equivalent to no
holes or 100% material, while 0 is equivalent to no material in the element.
Now given the constraints and the objective, OptiStruct can calculate what
equivalent density to assign to each element in order to yield the best
design.

27
Interested students should look up the litho-cuts of M.C.Escher for some
remarkable topology-related art.
43
OptiStruct CAE and Design Optimization – Basics

In most cases, you cannot take the same “sieve” approach in your design.
Manufacturing constraints will limit the sizes of holes as well as the thickness
of material between holes28. Therefore, you take this “equivalent density”
plot and exercise your design discretion to decide where to omit material
completely (i.e. create holes) and where to retain material.

The density plots closely follow the flow of forces in the domain. As a result,
this approach provides a very intuitive understanding of the package space:

where material is wasted, and where it’s most effective. The original-design
and the topology-optimized design are shown above. The improvement is
clearly visible in the performance-evaluation shown below.

The technique is applicable to shell elements, to one-dimensional elements


like bars, and to solid elements. In the assignments that accompany this
book, we will restrict our attention to shell and solid elements only.

Topography
Every mechanical engineer knows of the fact that the further a fiber is from
the neutral axis, the better it is at resisting bending forces. A plate with ribs
can be much lighter than, but just as strong as, a thicker plate.

This effect has long been used by designers to strengthen thin plates by
providing them with beads or swages. The presence of beads stiffens the
plate by moving the fibers away from the neutral surface of the plate. In
effect, this alters the topography of the plate.

Applicable only to shell elements, OptiStruct’s Topography Optimization


makes use of the change in stiffness with depressions in the plate. With an

28
Sometimes referred to as a ligament.
44
CAE and Design Optimization – Basics OptiStruct

intelligent use of manufacturing constraints, this can yield structures that are
very efficient, and outside the realm of experience-based design.

Asked to stiffen a water tank, most designers would suggest “regular”


stiffening, as shown on the right. The water tank shown on the left is
unlikely to be suggested by any “experience”, but is the optimal shape and
is eminently manufacturable!

Remember that beads can also be interpreted as ribs, which means


topography optimization can also be used to decide where to add ribs!

Size
Suppose you are assigned the task of increasing the strength of a pressure
vessel that has already been manufactured because subsequent testing has
identified weaknesses. Ribs are ruled out because of assembly requirements.
You obviously cannot form beads on the vessel. The only alternative you
have is to weld reinforcing plates over selected areas, thereby increasing the
effective thickness, and consequently the strength. Using reinforcement
plates indiscriminately is not a sensible answer. Not only is it expensive, it
may alter the dynamic response of the vessel. How can you address this?

Size Optimization is perfectly suited for such requirements.

Scenarios where the design parameters are limited to the thickness of


plates, or the cross-section of beams are best addressed by this method.
Changes in topology or topography are not relevant here.

Free-size Optimization also treats the thickness of shell elements as design


variables, but with a slight difference. Unlike size optimization, which treats
elements in groups, it allows the thickness to vary for each element
individually. To achieve the same effect using size optimization, you would
have to define one collector for each element.

45
OptiStruct CAE and Design Optimization – Basics

Ideally suited for machined components or laminates, it provides an effect


somewhat similar to topology optimization. In the latter the “equivalent
density” is allowed to vary continuously, while in Free-Size, the element
thickness can vary continuously. If you specify a minimum thickness of 0,
Free-Size Optimization can even provide for cutouts.

If you have a thick structure that must be modeled with 3D elements,


topology optimization will do the job. If you model a plate with layers of
hexahedral elements, the equivalent-density approach allows you to remove
elements, thereby achieving a variation in thickness.

Free-size is ideal if you have a thin structure that would be prohibitively


expensive, from a CPU-time point of view, to model using 3D elements.
Free-size also provides a powerful approach for the optimization of
laminates, as addressed in the next Chapter.

Shape
Stress is a very local phenomenon. It often dies away very rapidly.
Unfortunately, it also peaks surprisingly rapidly. Any designer who has been
faced with a cracked product would rue the fact that adequate fillets were
not used. Unlike other types of optimization, here you are dealing with the
very definition of the external boundary of the design space. In some sense,
this is a little like CAD modeling, treating the dimensions of selected feature
as design variables.

While Topography, Topology and Size Optimization tell you about overall
behavior, Shape Optimization helps you pay attention to detail. The external
boundaries of the analysis model are modified by OptiStruct to improve
performance. This can dramatically increase your power as a designer,
allowing you to step beyond the limits defined by the Finite Element method,
giving you capabilities that have traditionally been seen as “pure-CAD”.

The figure on the left shows the


initial design, with the shape-
optimized design on the right. As
you can see, the optimizer has
changed the fillet radius
significantly.

Shape Optimization requires you to specify the perturbation of the boundary.


This is not an easy task if the product has more potentially-modifiable
boundaries than the designer can specify with reasonable effort.
46
CAE and Design Optimization – Basics OptiStruct

Free-shape Optimization gets around this. OptiStruct automatically


determines the allowable movements and includes these in the Optimization
Model.

Putting it all together


Remember that the optimizer cannot choose the materials, the loads or the
restraints. These come from you. The optimizer can work with the variables
you provide to it – these are either element properties, “density” as in
topology optimization, or bead-parameters in topography optimization.

An intelligent choice of the design variables and constraints helps. It’s not
essential, but can help the optimizer find the solution faster if you restrain
yourself from taking the brute force approach and leaving all decisions to
the optimizer. However a wrong choice of design variables and constraints
can completely destroy your design effort, since it may result in no feasible
solution, or, even worse, a wrong solution. As you will see in the assignment
problems, we frequently track the objective function even as the solution is
in progress, to check that it’s proceeding in the right direction!

Remember to check that the Finite Element model is adequate. You can help
speed up things by organizing elements intelligently.

Which of the optimization methods you use depends to some extent on your
problem, and on the time you have available. In an ideal scenario, you
would first use topology optimization to get an idea as to the best path for
load to flow in your design space. Remember that “best” is defined by your
objective: weight, frequency, etc. Then you would look at the manufacturing
process you can choose from.

If you can cast the component, you would set appropriate manufacturing
constraints (symmetry, draw direction, grouping, etc.) and run topology
optimization again. When you’re satisfied with this, shape optimization can
help choose details like fillet radii.

If restricted to using sheet-metal, you would lay out a configuration that can
be guided by the results of the earlier topology optimization. You would then
use topology optimization again on this to identify areas where you can have
cutouts, topography optimization to stiffen the product and size optimization
to choose the sheet thickness.

47
OptiStruct CAE and Design Optimization – Basics

An excellent example of usage of OptiStruct can be found at www.altair-


india.com/edu/chassisdesign. This discusses how design optimization was
used to arrive at a better design for the chassis of an SUV.

Controlling the Optimizer


There are several parameters that control the performance of the optimizer.
Some, such as design variable linking and manufacturing constraints, relate
to the algorithms used by the software. Others relate to the computer
resources available to you, such as the RAM on your computer.

OptiStruct makes an effort to choose these intelligently. While you can set
these in HyperMesh, choosing appropriate values requires a deeper
understanding of the mathematics and algorithms than has been presented
in this book.

List of Optimization Results


An OptiStruct analysis-and-optimization creates several files, all created in
the same directory where your model resides.

In general, the best way to view and interpret results is using the post-
processor, HyperView. If things go wrong, though, it sometimes helps to
look at other ways to review the information logged by the optimizer. The
files all have the same name as the model, but have different suffixes, as
summarized below:

modelname.hm The model you create using HyperMesh. This is


a binary file containing the geometry, analysis
model and optimization model.
modelname.fem This is an intermediate file. It contains the
analysis and optimization models only, without
any geometry. It is created by HyperMesh and
read by OptiStruct. It’s a text file and can be
interpreted using the format-definitions listed
in the OptiStruct On-line Help.
modelname.out This is a text file created by OptiStruct. The
contents depend on the instructions you
specify in HyperMesh when creating the model.
modelname.spcf This file is created only if you explicitly tell
OptiStruct to record all reactions. It is a useful

48
CAE and Design Optimization – Basics OptiStruct

check to ensure that you have applied loads


correctly.
modelname.stat This is a text file created by OptiStruct,
containing statistics on CPU usage.
modelname.h3d People who don’t have access to HyperWorks
licenses but want to view results of analyses
use HyperView Player, freely downloadable
from www.altair.com. The Player reads this
binary file that is created by OptiStruct.
modelname.html This is a quick summary of analysis and
optimization results. Viewable using any web-
browser.
modelname.mvw This is a text file, intended for use by
HyperView. You will use this file to view
stresses, displacements, density, convergence
history, constraint violation, etc.
modelname.res This is a binary file containing the results of the
analysis and optimization. It’s readable only by
HyperMesh and HyperView.
modelname_frames.html If you have Macromedia’s Flash installed on
your computer, you can use this file to view
the results using HyperView Player.
modelname_oss OSSmooth is a module that helps you take
your optimized design back to a CAD model in
either STL or IGES format. This file controls the
performance of OSSmooth.

There are several other files29, most of which you can ignore in the normal
course of events.

Summary
You’re now well equipped to use Design Optimization as an integral part of
the design process30. Remember that while this changes the way CAE has

29
Detailed in the on-line help documentation
30
If you want to learn about design using laminated composites, you should read the
next Chapter before going on to the assignments.
49
OptiStruct CAE and Design Optimization – Basics

traditionally been viewed, it puts some burden on you to think of your


problem as one of design rather than as a problem in analysis.

Don’t be discouraged if your optimization does not yield results at first.


Learning from failure is an integral part of design. Use your understanding of
the Optimization Model to revisit the design variables, constraints and
objective. Verify that you have indeed used your knowledge of
manufacturing technology and machine design properly, and you will be
rewarded with safe, elegant, and effective designs.

Some Examples of Usage


Worldwide, Conferences include papers presented by designers highlighting
applications where they have successfully used the very same technologies
that are available to you. www.altair-india.com.edu is periodically updated to
reflect the latest publications, so browsing through these is well worth your
time.

Aeroplanes are not designed by science, but by art in spite of some pretence
and humbug to the contrary. I do not mean to suggest that engineering can
do without science, on the contrary, it stands on scientific foundations, but
there is a big gap between scientific research and the engineering product
which has to be bridged by the art of the engineer.
British Engineer to the Royal Aeronautical Society, 1922.
Quoted by Walter G Vincenti in 'What Engineers Know and How They Know It'.

50
CAE and Design Optimization – Basics Laminates

Laminates
The Miracle Material – Plastics
First used extensively in packaging products like PVC pipes and polythene
bags, plastics are today the “visible” part of almost any product, from
expensive automobiles to cheap MP3 players. The reason for this is easy to
find: plastics are an industrial designer’s delight because of the ease with
which they can be shaped into almost any form you can think of. With
metals, manufacturing planners are likely to reject shapes that it’s hard to
hammer the metal into. Plastics, on the other hand can be formed,
extruded, molded, sintered and even machined with a fraction of the effort it
takes to work metals.

Most plastics are also much lighter than metals.

Unfortunately, from a stress-analysis perspective, while few engineering


students have trouble finding design equations that govern the behavior of
steel, plastic is another story altogether. While the material is present
everywhere you look, calculating the stresses in plastics is a bit of an art.

Another disadvantage of plastics, from a designer’s point of view, is just as


easy to find. As the name shows, they are not “elastic”. They cannot carry
loads as well as most metals can, and certainly not as well as steels.

As with many other milestones in engineering history, Nature showed the


way out of this. Wood has long been used as a structural material. Its fibers
give it strength. As carpenters know, it is easier to saw a piece of wood
“along” the grain than “against”.

Reinforced Plastics: One Step Ahead


The use of fibers to reinforce plastics, then, led to the development of Fiber
Reinforced Plastics, also called Composites. While preserving their weight
advantage over steel, they can match or even exceed the strength of even
special-alloys. If we look at the strength-to-weight ratio, it’s almost no
contest. Kevlar, for example, is used to build bullet-proof vests that are light
enough to be worn, yet strong enough to stop speeding bullets.

51
Laminates CAE and Design Optimization – Basics

The development of composites is a science all by itself. Composite


materials come in many forms, including MMCs – metal matrix composites.
Our focus, however, is restricted to FRPs – fiber reinforced plastics.

Laminated Composites
Laminates are one form of composite materials, in which the reinforcing
fibers are laid out in thin mats or “laminae”. The lamina or mat is also often
referred to as a ply. Remember that fibers are like cables. They are good at
resisting tension, but poorer at handling bending and compression. To take
advantage of this behavior, the mats are laid out so that the fibers lie along
the direction in which you expect tensile forces to act. If the mats are
oriented along one direction only, the laminate is called a uniaxial laminate.
It has good strength in one direction, but much poorer resistance to forces
applied in any other directions. Most laminates, however, have multiple
layers of mats. Also, a lamina may consist of fibers woven together to form
a mat. The designer’s job is to determine how many layers to use in which
part of the product, and how to orient each layer.

Now something’s needed to fill the gaps between the fibers, and to hold the
layers together. This is achieved using a binder. The binder, which is usually
an epoxy, is sometimes referred to as the matrix.

Laminates are widely used in applications that demand excellent strength


and low weight. They are also sometimes used in non-load-bearing
situations, such as the cowls of motorcycles, but these stylistic applications
are not in the scope of our study. Aircraft, spacecraft, and sports equipment
are excellent examples of sectors where laminates are very widely used for
their excellent strength-to-weight ratio. Most high-performance bicycles, for
instance the ones ridden in races such as the Tour de France or the
Olympics, have frames constructed from laminated composites. Aircraft like
the Stealth Bomber also make extensive use of these advanced materials.

You will sometimes hear them referred to as CFRPs and GFRPs. These stand
for Carbon Fiber Reinforced Plastics and Glass Fiber Reinforced Plastics.
Numerous fibers can be used for reinforcement. As with plastics such as
Kevlar, trade names are widely used to refer to the materials.

52
CAE and Design Optimization – Basics Laminates

As shown in the picture31 of the Airbus, developments in composites


technology are driven by the fact that the fraction of composites used in
aircraft is increasing steadily!

Data Required for Stress Analysis


Stress analysis entails two distinct steps. The first is to estimate the stiffness
of the structure, from which we can calculate the stresses or resonant
frequencies. Once this is done, we check if the stress is within permissible
limits using an appropriate failure theory.

There are two main differences you should keep in mind when moving from
the design of steel components to the design of laminated composites.

First, steels are isotropic. That is, their properties are the same in any
direction. Plies are different. They resist tension well in the direction of the
fiber, but not in other directions. Such behavior is called orthotropic.
Orthotropic materials can be characterized by specifying the properties along
orthogonal axes. If there are a number of plies stacked up, each oriented at
different angles, the material may be better characterized as anisotropic –
the properties vary in all directions, not just along orthogonal axes.

31
The image is by Phillipe Cognard, www.specialchem.com
53
Laminates CAE and Design Optimization – Basics

Second, Steels are nicer for the designer since the Elasticity Modulus and
Poisson’s Ratio are largely independent of heat treatment, rolling, etc. The
permissible stress is the property that varies. In fact, the Modulus and
Poisson’s Ratio of steel do not even vary much with alloying elements. This
means the first step in stress-analysis, estimating the stiffness, can be done
using standard properties.

This is not the case for laminates.

Unfortunately for the designer, as with all plastics, the properties of the
laminates are strongly dependent on processing conditions or manufacturing
conditions. Accordingly, it’s essential that properties be obtained from the
manufacturer. It is a foolhardy designer who relies on “standard” properties
for a plastic or a laminate!

The next question that the designer faces is whether the laws of elastic
stress analysis are applicable at all. Plastics are known to creep at lower
temperatures than metals, they behave differently under fatigue loads, and
most important of all they’re not elastic. In other words, the behavior of
plastics, and therefore of laminates, is best described using non-linear
relations between stress, strain and deformation.

Remember what we discussed earlier – linear models of behavior are often


chosen even if a non-linear model is more accurate. Designers of plastic
components often use the same approach. Even if the material behavior is
best modeled as non-linear, linear models are used at the preliminary design
stage to narrow down the choices. If the component is critical enough,
however, you may need to use non-linear analyses right from the start32.

To sum up, there are two classes of data we require to design laminates.
First, the material constants that characterize the stiffness. Second, the
failure criteria we can use to estimate if calculated stress is within
permissible limits.

Material Constants
Strain is the first derivative of deformation – that’s a definition that’s
independent of the material properties. The stress-strain relationship,
however, is material dependent. The 3D equivalent of Hooke’s Law relates

32
This is covered in the volume on CAE And Design Optimization – Advanced.
54
CAE and Design Optimization – Basics Laminates

the six components of stress with the six components of strain. In the
equation,

σ 11   c11 c12 c13 c14 c15 c16  ε 11 


σ 12   c 21 c 22 c 23 c 24 c 25 c 26  ε 12 
  
σ 13   c 31 c32 c33 c 34 c35 c 36  ε 13 
 =  
σ 22   c 41 c 42 c 43 c 44 c 45 c 46  ε 22 
σ 23   c 51 c52 c53 c 54 c55 c 56  ε 23 
    
σ 33   c 61 c 62 c 63 c 64 c 65 c 66  ε 33 

the constants Cij represent material behavior, and the matrix C is called the
Constitutive Matrix. Since the stress and strain components are symmetric
(that is, σij = σji), C is symmetric too. This means that the general
relationship between stress and strain in 3D requires 21 constants to fully
characterize the behavior of the material.

An anisotropic material, then, requires 21 constants to be specified for


stress-analysis.

For an isotropic material, it can be shown that two constants are enough. All
others can be derived from these two. Most often, we specify the Modulus of
Elasticity (E) and the Poisson’s Ratio (ν).

For an orthotropic material in 3D, 9 constants are required, while for an


orthotropic material in 2D (as in a ply) 4 constants are required. Usually, we
specify the Elasticity Modulus along the two principal directions, and the
Shear Modulus and Poisson’s ratio in the “12” direction: E1, E2, G12 and ν12.
In this notation, the subscripts refer to the ply’s coordinate system. The “1”
direction is along the direction of the fiber, the “2” direction is in the plane
of the fiber but perpendicular to the fiber. The “3” direction can be obtained
by taking the cross-product of the “1” and “2” vectors.

55
Laminates CAE and Design Optimization – Basics

Y
2
1 - Fiber Direction

Failure Theories
From a design perspective, there are a number of different ways in which a
component can “fail” – it may buckle, it may deform too much and interfere
with another part, it may deform permanently, it may rupture, and so on. In
this section, we will restrict our attention to stress as a measure of failure.
You are familiar with the use of the von Mises condition, the Tresca
condition, etc. to characterize the failure of steel under the permissible-
stress mode of design. We will look for a similar method to quantify
laminates: how can we determine whether the stresses within the laminated
composite are within safe limits or not?

Failure criteria, historically, have been derived from experimental


observations, not from fundamental laws. One extreme view is that “most
Failure Criteria are meaningless curves passed through unrelated data
points”33! Experimental Mechanics is the field in which engineers subject
materials to various conditions, then attempt to define boundaries on
acceptable environments. These are called Failure Theories.

For ductile materials some theories extend till yield (or permanent
deformation, which is the elastic limit) while others extend till rupture or
fracture.

Composites are no different. There are several different failure theories that
have been propounded to allow the designer to decide whether or not the
stresses are within safe limits. Since composites are unlike ductile materials,
they have different failure modes. A composite, for example, can fail by
“delamination” – when the binder fails and a layer peels off, like the bark on

33
J. Hart-Smith, cited in Mechanics of Composite Materials, R. Jones
56
CAE and Design Optimization – Basics Laminates

a tree. Since the failure can occur either within the fiber or within the binder,
characterizing failure is far more difficult. Just as there are no “standard”
properties that you can use for a composite, there are no “standard” failure
theories you can apply. Most failure theories are valid only for a select set of
materials or within a select range of conditions. Extensive testing is often
resorted to, with numerical analysis being used as a guideline.

Some commonly used theories are the Hill theory, the Hoffman theory, the
NASA Larc02 theory, the Tsai-Hill theory, and the Tsai-Wu theory, all of
which apply to failure of plies. The Strain Invariant Failure Theory is often
used to estimate failure of the matrix.

Finite Element Approaches


The sections above outline the theory. How can we use this theory to frame
and solve actual problems? From an analysis perspective the question, of
course, is how we can determine the stress components from the given
loads.

Numerical analysis has proved to be very useful in dealing with the


complexity posed by laminated composites. The Finite Element Method is
more or less the standard numerical analysis method used in engineering
design. As we’ve seen earlier, FEA consists of defining the Finite Element
model, running the analysis to create the matrix equations and solve them,
and then interpreting the results.

Choosing Elements
Most laminated composites are thin, which means they are best modeled
using shell elements. If the component is not thin, as for example in the hub
of the propeller, shear stress plays a more prominent part than bending
stress. In this case, solid elements are better used than shells.

From this perspective, the guidelines for laminated composites are no


different than the general guidelines we saw earlier. The main difference
from FE models of isotropic materials lies in the way the material data is
specified.

The ABD Matrix


The mechanics of composite materials is not very well understood. Several
approaches have been propounded, but as the science is still evolving, most
commercially available software takes an approximate approach.

57
Laminates CAE and Design Optimization – Basics

Rather than investigate the behavior of each ply and the matrix separately,
an “average’ approach is taken. In this approach, we assume the stiffness of
the laminate can be separated into bending, extensional, and coupled
bending-extensional stiffness. Each of these is represented by a 3 x 3
matrix. To calculate the strains in a laminate, given the forces and moments,
the equation used is

 Nx   A11 A12 A13 B11 B12 B13  εox 


 Ny   A21 A22 A23 B 21 B 22 B 23  ε 0 y 
  
 Nxy   A31 A32 A33 B 31 B 32 B 33  γ 0 xy 
 =  
Mx   B11 B 21 B 31 D11 D12 D13   Kx 
My   B 21 B 22 B 32 D 21 D 22 D 23  Ky 
    
Mxy   B 31 B 22 B 33 D 31 D 32 D 33   Kxy 

where N and M are the Force and Moment components, A represents the
extensional-stiffness, D represents the bending stiffness, and B represents
the coupled bending-extension stiffness. The 6 x 6 matrix is often referred
to as the ABD matrix.

In this formulation, the entries in the ABD matrix depend on the thickness
of each ply, the orientation of each ply, the distance of each ply from the
neutral surface of the laminate, and the material properties of the plies and
the binder.

If the ABD matrix is available, the stresses and strains can be calculated
from the forces. In other words, the problem can now be solved.

Material Data Specifications


For the Finite Element solver to do its work, then, the data required is now
clear.

First we need to supply the material properties (Modulus of Elasticity, Shear


Modulus, etc.) separately for the binder and for each ply.

Next, we need to specify the “stackup sequence”. That is, we need to


describe the order in which the laminae are stacked, the thickness of each

58
CAE and Design Optimization – Basics Laminates

lamina, and the orientation of each lamina with respect to some reference
coordinate system.

Finally, for each element, we need to specify the coordinate system with
reference to which laminae orientations are specified.

Given this data, the Finite Element Solver calculates the ABD matrix for
each element and proceeds with the solution. Remember that the analyst
does not directly specify the ABD matrix itself: it is derived from the
data of the constituents of the composite. This is very convenient. Why?

Remember that we make frequent use of experimentally measured data. It


would be impossible for an experimental scientist to generate data for every
single possible combination of plies! This is because a fabricator buys the
lamina and binder, then lays up the laminae as required. There are literally
infinite ways in which the same laminae can be stacked up and oriented.
With the approach described above, the analyst only needs to ask the
material supplier for the properties of each constituent. Then, when building
the model, the analyst specifies the specific stackup-sequence used.

Interpreting Results
Having solved the problem, how do you use the results for your design?
Unlike a ductile material where a single equivalent stress can be compared
to the permissible stress of the material, the composite designer is often
faced with the necessity to examine the laminate layer by layer. You’ll see
how to do this in the project that addresses optimization of composites.
Fortunately, the HyperWorks approach makes it fairly easy to present the
data in the form that’s most convenient.

Design Optimization Issues


What we’ve seen so far is the theory of laminate behavior, and the analysis
approach. Our interest however is not in analysis. We are interested in
design optimization. Given the understanding we have of optimization
theory, and of how OptiStruct works, we’ll spend the rest of this Chapter
understanding the recommended method to use OptiStruct for the design of
laminates.

What we do not Optimize


It’s important to realize that there are some parts of a laminate that are
omitted from our design-approach. Take, for example, an I-section beam

59
Laminates CAE and Design Optimization – Basics

constructed using laminates. The figure shows the junction in the detailed
view. The approach we will follow ignores these areas.

Further, we will restrict our attention to shell elements.

The Poverty of Abundance


Let’s restate the design problem before we proceed further.

You know the loads you want the component to carry. You have the design
space that the component can occupy. You have available a set of plies of
predefined thickness, and a binder. You also know your design objective –
stress, deformation, frequency, buckling load factor, etc. The problem is to
determine how many layers to use, in what sequence to stack them, and
what orientation each ply should have. An engineer familiar with the rate at
which complexity rises when faced with permutations and combinations can
be forgiven for giving up. Given all this complexity, is it even worth it trying
to come up with an optimal design? Why not just settle for the design-then-
verify approach, even if the accepted design is sub-optimal?

It’s like a starving man who is offered a choice of hundreds of dishes: faced
with the impossibility of making a “best” choice, he settles for any available
food. There are just too many choices!

The main motivation to cope with this abundance, of course, lies in the
reason we use composites in the first place. Their strength-to-weight ratio
60
CAE and Design Optimization – Basics Laminates

makes a compelling case in the design of weight sensitive equipment. A


reduction of even a few percentage points can make the difference between
a product that’s “good enough” and a product that’s “outstanding”.

With OptiStruct, it’s surprisingly easy to specify an initial configuration and


ask the software to come up with a suggestion that’s better.

Discrete Optimization
When optimizing laminates, one of the challenges lies in the fact that the
laminae can have finite thickness only. Further, the designer should have the
freedom to cut a layer off at any point. In other words, some of the design
variables can vary in steps, rather than varying continuously. As you will
recall from your calculus courses, gradients of a discontinuous function are
not defined at the discontinuity. Gradient Search Methods, therefore, need
to be modified to work effectively.

The mathematics used to solve such problems is called Discrete


Optimization.

Topology optimization, as we’ve seen, uses a density-based approach. Free-


size, on the other hand, uses the element’s thickness as the design variable.
Note that this is not the same as size optimization. Using free-size, we can
obtain results similar to those of a topology-optimization, but with a more
even spread of material through the design space. Problems where buckling-
stiffness is the main objective would be better off with topology
optimization, while problems where stress is a constraint would yield better
results with free-size optimization.

Free-size optimization changes the thickness of the laminae during the


optimization cycle. Remember that the A matrix represents the extensional
or in-plane behavior. Consequently, it’s largely independent of the sequence
in which the plies are laid up. (The D matrix is sequence sensitive, since
bending stiffness varies with distance from the neutral surface). For most
shell-like structures, the extensional stress dominates the behavior, so Free-
size works quite well.

This makes things much easier for the designer, since you only need to
specify which plies you want to choose from – you do not need to specify
the number of layers of each ply. OptiStruct will add or remove layers of
plies as required!

61
Laminates CAE and Design Optimization – Basics

It’s important to note that laminate optimization problems often have a


number of local minima, so the optimization-for-design precept holds good:
you are looking for a better design, not necessarily for the best design.

Putting it all together


The flow of steps in OptiStruct is as follows:

0. Ensure the FE model is complete. In addition to the usual definition of


the elements, loads and restraints, use HyperLaminate to define the
properties of the matrix and plies, and to specify the stacking sequence.
Each combination of plies+orientation+thickness+stacking-sequence is
saved as a PCOMP or a PCOMPG. Then use the “Composites” command
to specify the relevant coordinate system and PCOMP / PCOMPG for the
elements in the model.

1. Choose the Design Variables.

2. Specify the Responses that the optimizer needs from the Finite Element
solver.

3. Specify the constraints and the objective.

4. Run the optimizer.

This is pretty much the same as running any other OptiStruct optimization,
except for the fact that you are dealing with discrete optimization.

The inspirational value of the space program is probably of far greater


importance to education than any input of dollars... A whole generation is
growing up which has been attracted to the hard disciplines of science and
engineering by the romance of space.
Arthur C. Clarke

62
CAE and Design Optimization – Basics Glossary, Tips and References

Glossary And References


CAD Computer Aided Design. Usually means creation of 3D geometric
models of parts and assemblies.

CAE Computer Aided Engineering. Said to have been coined by Dr.Jason


Lemon in 1980, and meant to consist of CAD + FE Modeling + FE
Analysis + Design. Today includes Multi-Body Dynamics (MBD). Often
separated into MCAE (for Mechanical CAE, or the analysis of
structures) and FCAE (for Fluid CAE, or the analysis of flow of heat
and liquids / gases).

CFD Computational Fluid Dynamics. The use of computers to solve various


forms of Navier-Stokes equations to analyze the flow of fluids.

DOF See Degree of Freedom

Degree of Variable we want the FE analysis to solve for. In an FE model every


node / grid has at least 1 dof, often more. For a stress-analysis
Freedom
problem, each node can have upto 6 dofs – 3 rotations and 3
translations. Other variables such as temperature at the nodes,
pressure, etc. are also included in some models.

Bandwidth The stiffness matrix of a typical FE model has zeroes in most entries
except for a band about the diagonal. The bandwidth measures this
“spread” of non-zeroes in the matrix. A smaller bandwidth means
faster computation.

Core Memory High-speed memory, usually RAM. Lower-speed storage, such as disk,
is used to provide virtual memory to augment available core memory.

Design Space In a topology optimization, this space contains the elements whose
density can be varied by the optimizer.

Packaging The art of laying out components in 3D such that best use is made of
available space, keeping in mind required clearances between
components.

Response In the absence of a continuous function relating the objective to


design variables, numerical experiments can be used to generate a
Surface
table of objective-function values vs. design-variable values. A surface
fitted through this table of points, called the Response Surface, is then
used to find optimal locations.

Sensitivity Reflects the rate of change of the objective function with changes in
the design variables. A zero sensitivity is an indicator of a badly
63
Glossary, Tips and References CAE and Design Optimization – Basics
the design variables. A zero sensitivity is an indicator of a badly
phrased problem: If the objective is independent of the design
variable, the optimizer is lost!

Convex Function A function that has only one minimum in the domain. This minimum is
the “global minimum”.

Convergence An iterative process can be terminated when it has “converged” to a


solution – that is, when subsequent iterations do not change the
solution. OptiStruct offers a choice between soft and hard
convergence. The former is faster while the latter is more precise.

MDO Multi-disciplinary Optimization. Used, for example, when your product


needs to be designed for optimal performance as a mechanism and as
a structure.

Compliance Reflects the reciprocal of the stiffness. Maximizing the compliance is


the same as minimizing the stiffness.

Stiffness Usually referred to as the “stiffness matrix” in FE models, relates the


applied loads to the deformation of the structure. The matrix is
square, with “n” rows and columns. “n” is the number of unknowns
(dofs) in the FE model.

Stochastic Something that involves chance or probability, but with an overall and
measurable trend or direction – this makes it possible to predict the
behavior. Engineers frequently encounter stochastic processes and
stochastic variables.

Robust Design A design method to reduce sensitivity of the design to inherent


unpredictability of design parameters.

Isotropic Material whose properties are independent of direction. Applies to


most metals. 2 elasticity constants are required to fully specify the
material for stress analysis. The Modulus of Elasticity and the Poisson’s
Ratio are most frequently used. In OptiStruct, these materials are of
type MAT1.

Orthotropic Material whose properties vary along principal or orthogonal


directions. Applies to many fibrous materials, and to composites that
have 2 ply directions. Upto 9 elasticity constants are required to fully
specify the material for stress analysis. In OptiStruct, these materials
are of type MAT8 for shell elements.

Anisotropic Material whose properties vary with direction, but not necessarily
along orthogonal directions. Several fused or sintered materials are
anisotropic. 21 elasticity constants are required to fully specify the
material for stress analysis. In OptiStruct, these materials are of type
MAT2 for shell elements. MAT9 should be used for solid elements.

64
CAE and Design Optimization – Basics Glossary, Tips and References

Supported Responses
Responses can be used either as constraints or as an objective. Choices can
be quite complex, particularly when multiple optimization-techniques are
used simultaneously.

Exercise your discretion when assigning responses as objectives. For


instance, a topography optimization is unlikely to have much affect on mass,
since no material is being added or removed.

The responses that OptiStruct supports include:

compliance, frequency, compliance index, volume, mass, volume fraction, mass


fraction, moments of inertia, center of gravity, displacements, velocities,
accelerations, buckling factor, stresses, strains, composite failure, forces, synthetic
responses, external (user-defined) functions.

It’s a bad idea to work with a response that you do not understand from an
engineering and a mathematical point of view. It’s a good idea to search for
a tutorial problem (part of the on-line documentation) that uses the
responses you plan on using.

Errors
It’s frustrating, and often confusing, to have the solver reject your model
because it contains errors. As a general practice, use the “check” option
before running an analysis or optimization. It can save you considerable
time.

Reading the output file (modelname.out) is usually the best way to figure
out what went wrong.

Common errors include:

• putting elements in a wrong collector. For instance, shell elements in a


PSOLID collector, or mass elements in a PSHELL collector
• forgetting to create loads and / or restraints
• creating restraints and loads in the same load-collector
• neglecting to specify density when using gravity loads or calculating
dynamic response
• mixing up units

65
Glossary, Tips and References CAE and Design Optimization – Basics

References
Engineering Optimization: Theory and Practice, 3rd Edition,
Singiresu S. Rao,

Engineering Optimization: Methods and Applications, 2nd Edition,


A. Ravindran, K. M. Ragsdell, G. V. Reklaitis

Optimization Methods for Engineering Design, R.L.Fox, Addison


Wesley

Arora, J., Introduction to Optimum Design (McGraw-Hill, 1989).

Bendsoe, M.P., and Sigmund, O., Topology Optimization - Theory,


Methods and Applications (Springer, 2003).

Haftka, R.T., and Guerdal, Z., Elements of Structural Optimization


(Kluwer Academic Publishers, 1996).

Rozvany, G.I.N., Topology Optimization in Structural Mechanics


(Springer, 1997).

Other Resources
www.altair-india.com/edu, which is periodically updated, contains case
studies of actual usage. It also carries tips on software usage.

OptiStruct and the FE Model


The table below shows the names of some of the data types OptiStruct and
HyperMesh use, along with the relations between them. Use this as a
guideline to remember when you need to specify which data.

Normally, you would proceed down the table: first the mat collector, then
the component collector, then elements (grids or nodes are implicit), and
then the load collector. While load collectors can contain both loads and
restraints, it’s a good practice to keep them in separate load-collectors so
that you can organize them into sub-cases.

Data Entity Refers to Data 1 Data 2 Data 3 Data 4


Mat& Matid E Nu etc.
Component Mat CompName ElemType+ Mat Id Prop Data#
Property% PropId
Grid Coordsys Gridid X Y Z
Elem Grids, Elemid Propid Grid Id 1 Grid Id 2
66
CAE and Design Optimization – Basics Glossary, Tips and References
Props etc.
Load@ Grids/ Loadid Gridid Value
Elems
Restraints! Grids Spcid Gridid Dofs Values
SubCase$ Loads, SubCaseid
Spcs

Notes:
& A Mat needs a card image. Use Mat1 for linear isotropic, Mat8 for
orthotropic shells, Mat9 for linear anisotropic.

+ The component collector is either a PSOLID or a PSHELL. Composites use


a PCOMP or PCOMPG.

# Data depends on the element type. For a solid, there’s nothing. For a
shell, there’s the thickness. For a composite, thickness is derived from the
PCOMP or PCOMPG data.

% A Prop collector is needed only for 1D elements like beams or special


elements such as springs, connectors, etc.

@ Forces and Moments need no card image. Loads such as gravity, which
cannot be depicted graphically, require card images.

! Restraints normally will not require a card image. Remember that non-
zero displacements may be specified, in which case you will need to enter
values. Restraints on non-existent dofs are ignored (for instance,
specifying restraints on all the rotational dofs of a solid element).

$ Sub-Case definitions are followed by a set of cards, each with a keyword


(Load, Spc, etc.) followed by the relevant id. These can be viewed in the
“fem” file, not using the card editor.

Common Material Properties


Be careful in using these properties. Some properties vary widely with
alloying elements or processing parameters, so treat these as indicative. It’s
probably safe to use them in exploratory design efforts, but not in designs
that will be manufactured. For those, you should look for values from the
material supplier.

Also remember to check the units in your model – they must be consistent!

Material Modulus Poisson’s Density Sample


of Ratio Permissible
67
Glossary, Tips and References CAE and Design Optimization – Basics

of Ratio Permissible
Elasticity Stress
Units N/m2 Kg/m3 N/m2

Steel 200x109 0.29 7800 250x106


Aluminum 69 x109 0.33 2700 110 x106
Wood34 13 x109 0.029 480 50 x106
Cast Iron 190 x109 0.21 7150 170 x106
ABS Plastics 2.3 x109 40x106
Epoxy 1790

Sample properties for a Carbon Fiber Composite material:

E1 181 GPa
E2 10.3 GPa
G12 7.17 GPa
ν12 0.28
Density 1.60 gm/cm3

Consistent Units
Mixing up units is one of the most common errors. It’s also the least
forgivable if committed by an engineer who is allowed to use the SI
system of units. While FPS can be challenging, the SI system is very
straightforward. The table below lists some common properties of Steel in
consistent units.

Mass Length Time Force Stress Density Young’s Acceleration


Modulus due to
Gravity
kg m s N Pa 7.83e+03 2.07e+11 9.806
kg cm s 1.0e-02 N 7.83e-03 2.07e+09 9.806e+02
g cm s dyne dy/cm² 7.83e+00 2.07e+12 9.806e+02
g mm s 1.0e-06 N Pa 7.83e-03 2.07e+11 9.806e+03
ton mm s N MPa 7.83e-09 2.07e+05 9.806e+03

34
In compression
68
CAE and Design Optimization – Basics Glossary, Tips and References
lbfs2/in in s lbf psi 7.33e-04 3.00e+07 386
slug ft s lbf psf 1.52e+01 4.32e+09 32.17
kg mm s mN 1.0e+03 Pa 7.83e-06 2.07e+08 9.806e+02

69

You might also like