Vous êtes sur la page 1sur 89

A Thesis

entitled

Finite Element Modeling of Thermal Expansion


in Polymer/ZrW2 O8 Composites

by
Gregory J. Tilton

Submitted to the Graduate Faculty as partial fulfillment of the requirements


for the Master of Science Degree in Mechanical Engineering

Dr. Lesley M. Berhan, Committee Chair

Dr. Maria R. Coleman, Committee Member

Dr. Yong X. Gan, Committee Member

Dr. Patricia R. Komuniecki, Dean


College of Graduate Studies

The University of Toledo


December 2011
Copyright 2011, Gregory J. Tilton

This document is copyrighted material. Under copyright law, no parts of this


document may be reproduced without the expressed permission of the author.
An Abstract of
Finite Element Modeling of Thermal Expansion
in Polymer/ZrW2 O8 Composites
by
Gregory J. Tilton

Submitted to the Graduate Faculty as partial fulfillment of the requirements


for the Master of Science Degree in Mechanical Engineering
The University of Toledo
December 2011

Composite materials are being more frequently used in a wide variety of indus-

tries. Their high strength to weight ratio makes them a desirable material in many

applications. In some specific cases, polymer based composites can be subjected to

large changes in temperature causing undesirable amounts of expansion. To reduce

the composite’s thermal expansion, materials that have negative coefficients of ther-

mal expansion are used as a filler material. Zirconium tungstate (ZrW2 O8 ) is a metal

oxide which exhibits thermal behaviors not seen in most other materials. When sub-

jected to a positive temperature change, ZrW2 O8 will decrease in volume as opposed

to most other materials which show an increase in volume. This makes ZrW2 O8 an

ideal candidate to be used as filler material in these polymer composites to reduce

their overall thermal expansion.

While experimental research on ZrW2 O8 composites has previously been com-

pleted, this research looked at the finite element modeling of these composite materials

and tried to gain a better understanding of their possibilities. Initial two-dimensional

models were created using COMSOL Multiphysics with basic geometries for both the

matrix and filler. The results from these tests showed that the filler geometry had

little effect on the expansion results and volume fraction was the most important fac-

tor. To further test this, more complex models were created using three-dimensional

iii
geometries with the same volume fractions. These results confirmed the findings of

the two-dimensional tests by showing similar expansion. These results were then com-

pared to published experimental data where it was found that all the models showed

less expansion than the physical experiments of the same volume fraction.

The difference between the finite element analysis (FEA) and experimental results

was attributed to the interaction between the filler and matrix materials. In the mod-

els, the bond between the two was considered perfect, with no voids or separation,

leading to the filler material having more effect on the overall properties of the com-

posite. In real-world testing, this perfect bond would be nearly impossible to achieve.

To build on this idea and gain a better understanding of how the experimental test-

ing compared to the FEA, models with no bond between the filler and matrix were

created. Using the results from these models, as well as the models with a perfect

bond, an upper and lower bound of expansion were able to be created. All published

experimental data looked at was contained within these FEA-created bounds. This

showed that while some bond was likely made between the filler and matrix materials,

there was room for improvement if less expansion was desired.

iv
Table of Contents

Abstract iii

Table of Contents v

List of Tables viii

List of Figures ix

List of Abbreviations xii

List of Symbols xiii

1 Introduction 1

1.1 General Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Overview of Principles . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3.1 Composite Materials . . . . . . . . . . . . . . . . . . . . . . . 3

1.3.2 Thermal Expansion and Zirconium Tungstate . . . . . . . . . 5

1.3.3 Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . 6

1.4 Organization of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Literature Review 9

2.1 Negative Thermal Expansion Materials . . . . . . . . . . . . . . . . . 9

v
2.2 Experimental Research of ZrW2 O8 Composites . . . . . . . . . . . . . 11

2.3 Mathematical Models for Predicting the CTE of Composites . . . . . 12

2.4 Finite Element Modeling of Particulate Composites . . . . . . . . . . 15

3 Modeling Approach 17

3.1 Materials Used and their Properties . . . . . . . . . . . . . . . . . . . 17

3.2 Setup for COMSOL Multiphysics . . . . . . . . . . . . . . . . . . . . 18

3.3 Modeling to Compare Geometry . . . . . . . . . . . . . . . . . . . . . 23

3.3.1 Two-Dimensional Models . . . . . . . . . . . . . . . . . . . . . 23

3.3.2 Three-Dimensional Models . . . . . . . . . . . . . . . . . . . . 27

3.4 Modeling to Examine Matrix/Filler Bond . . . . . . . . . . . . . . . . 32

3.5 Analysis of Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

4 Results and Discussion 36

4.1 Two-Dimensional Models . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.2 Three-Dimensional Models . . . . . . . . . . . . . . . . . . . . . . . . 43

4.3 Comparisons to Experimental Data and Development of Upper and

Lower Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.4 Comparison to Mathematical Models . . . . . . . . . . . . . . . . . . 53

5 Conclusions and Future Work 55

5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

References 59

A MATLAB Codes 67

A.1 MATLAB Code for Creating Random Coordinates for Spherical Inclu-

sions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

vi
A.2 MATLAB Codes for Creating a Mapped Mesh and Randomly Assign-

ing Material Properties . . . . . . . . . . . . . . . . . . . . . . . . . . 70

A.2.1 Code for Developing Random Filler Elements in Mapped Mesh 70

A.2.2 Code for Assigning CTE Properties . . . . . . . . . . . . . . . 73

A.2.3 Code for Assigning Elastic Modulus Properties . . . . . . . . . 74

A.2.4 Code for Assigning Poisson’s Ratio . . . . . . . . . . . . . . . 75

A.2.5 Code for Assigning Density Properties . . . . . . . . . . . . . 76

vii
List of Tables

1.1 A list of materials and their coefficient of thermal expansion, α, at 20◦ C. 5

3.1 Materials used in FEA models and their required properties . . . . . . . 18

4.1 Numerical results from the two-dimensional modeling. . . . . . . . . . . 41

4.2 Numerical results from the three-dimensional modeling. . . . . . . . . . 48

viii
List of Figures

1-1 Comparison between positive and negative CTE materials . . . . . . . . 2

1-2 Example of different types of composites . . . . . . . . . . . . . . . . . . 4

1-3 Structure of ZrW2 O8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2-1 Diagram showing thermal contraction in oxides . . . . . . . . . . . . . . 10

2-2 Chart showing the effect of ZrW2 O8 on a polymer matrix . . . . . . . . 12

2-3 Figure displaying the opposing forces in a ZrW2 O8 composite . . . . . . 13

2-4 Chart comparing different mathematical models for predicting the CTE

of composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3-1 Starting COMSOL window . . . . . . . . . . . . . . . . . . . . . . . . . 19

3-2 Model builder tree in COMSOL . . . . . . . . . . . . . . . . . . . . . . . 20

3-3 An example of a two-dimensional mesh in COMSOL . . . . . . . . . . . 21

3-4 A screen-shot of the MBT showing changes made to the default solver . 22

3-5 Two-dimensional models with seven percent filler volume. . . . . . . . . 25

3-6 Two-dimensional models with nineteen percent filler volume. . . . . . . . 25

3-7 Two-dimensional models with thirty-seven percent filler volume. . . . . . 26

3-8 Three-dimensional model with seven percent filler volume made of one

cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3-9 Three-dimensional model with seven percent filler volume made of nine

cylinders. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

ix
3-10 Three-dimensional model with seven percent filler volume made of twenty-

five cylinders. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3-11 Three-dimensional model with seven percent filler volume made of ten

spheres. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3-12 Three-dimensional model with seven percent filler volume made of fifty

spheres. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3-13 Three-dimensional model with a 10 by 10 by 50 element mapped mesh. . 32

3-14 COMSOL results settings for determining the change in volume of the

models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4-1 Total displacement plot for a two-dimensional, pure polymer model. . . . 37

4-2 Total displacement plots for two-dimensional models with seven percent

filler volume. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4-3 Total displacement plots for two-dimensional models with nineteen per-

cent filler volume. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4-4 Total displacement plots for two-dimensional models with thirty-seven

percent filler volume. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4-5 Bar graph comparing all of the two-dimensional models studied. . . . . . 42

4-6 A graph showing volume fraction versus relative expansion from the two-

dimensional data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4-7 Total displacement plot for 3-D model with one cylindrical filler particle. 43

4-8 Total displacement plot for 3-D model with nine cylindrical filler particles. 44

4-9 Total displacement plot for 3-D model with twenty-five cylindrical filler

particles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4-10 Total displacement plot for 3-D model with ten spherical filler particles. 46

4-11 Total displacement plot for 3-D model with fifty spherical filler particles. 46

4-12 Total displacement plot for 3-D model made with a mapped mesh of cube-

shaped particles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

x
4-13 Bar graph comparing the three-dimensional models studied. . . . . . . . 48

4-14 Bar graph comparing the FEA models to experimental data. . . . . . . . 50

4-15 Graph showing the upper and lower bounds of expansion created by FEA

models compared to Tani’s data. . . . . . . . . . . . . . . . . . . . . . . 51

4-16 Graph showing the upper and lower bounds of expansion created by FEA

models compared to Sharma’s data. . . . . . . . . . . . . . . . . . . . . 52

4-17 Graph showing the upper and lower bounds of expansion created by FEA

models compared to Tani’s data. . . . . . . . . . . . . . . . . . . . . . . 54

4-18 Graph showing the upper and lower bounds of expansion created by FEA

models compared to Sharma’s data. . . . . . . . . . . . . . . . . . . . . 54

5-1 Geometry showing a possible way to model the interface between the

matrix and filler. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

xi
List of Abbreviations

ASME . . . . . . . . . . . . . . . . . American Society of Mechanical Engineers

CTE . . . . . . . . . . . . . . . . . . Coefficient of Thermal Expansion (α)

E . . . . . . . . . . . . . . . . . . . . . . Modulus of Elasticity

FEA . . . . . . . . . . . . . . . . . . Finite Element Analysis


FEM . . . . . . . . . . . . . . . . . . Finite Element Modeling

MBT . . . . . . . . . . . . . . . . . . Model Builder Tree

NTE . . . . . . . . . . . . . . . . . . Negative Thermal Expansion

PDE . . . . . . . . . . . . . . . . . . Partial Differential Equation


PI . . . . . . . . . . . . . . . . . . . . . Polyimide
PR . . . . . . . . . . . . . . . . . . . . Phenolic Resin

RAM . . . . . . . . . . . . . . . . . . Random-access Memory


ROM . . . . . . . . . . . . . . . . . . Rule of Mixtures

ZrW2 O8 . . . . . . . . . . . . . . . Zirconium Tungstate

xii
List of Symbols

Ac ....... total area of composite


Af ...... area of filler material
Vc ....... total volume of composite
Vf ...... volume of filler material

α ........ coefficient of thermal expansion


ν ........ Poisson’s ratio
ρ ........ density
E ....... Young’s modulus
K ....... bulk modulus
G ....... shear modulus
vf . . . . . . . volume fraction of filler material

T . . . . . . . . temperature
∆T . . . . . . change in temperature
K . . . . . . . Kelvin, unit of temperature

xiii
Chapter 1

Introduction

1.1 General Overview

Thermal expansion is the tendency of a material to change in volume when it

undergoes a change in temperature. It is referred to as thermal expansion because

most types of matter increase in volume when subjected to a positive change in

temperature. This property can often create problems for engineers who must account

for this change in volume in their designs. While a majority of solids do not undergo

significant changes for their given operating temperatures, certain cases exist where

a large increase in volume is observed (e.g. [1–4]). This often leads to less than ideal

situations where problems with fit and contact need to be addressed.

One way to combat the high thermal expansion in polymers is through the use

of composite materials. As stated earlier, most materials undergo thermal expansion

and expand given a temperature increase. Certain materials undergo thermal con-

traction, or, negative thermal expansion, and shrink when subjected to an increase

in temperature (Fig. 1-1). While none of these materials contain the properties to

completely replace polymers, they can be used to form a composite material that will

exhibit lower thermal expansion when compared to the original material.

1
(a) Positive CTE (b) Negative CTE

Figure 1-1: Comparison between the effects of a positive and a negative CTE
when subjected to an increase in temperature.

Extensive experimental research on these composites has been conducted and

results have shown that it is an effective way to reduce thermal expansion (e.g. [5–

8]). This research focuses on the finite element modeling of them in the hopes of

being able to correctly model the properties of these composites as well as gain a

better understanding of how the matrix and filler interact. With this understanding,

new experimental procedures could be developed to obtain better results and further

decrease the amount of thermal expansion in these composites.

1.2 Problem Statement

The objective of this work was to determine an accurate way to model composite

materials containing ZrW2 O8 as a filler material. These types of composites pose

problems in finite element modeling (FEM) because they contain a filler material

that exhibits negative thermal expansion while the matrix material exhibits positive

thermal expansion. In physical experimentation, this creates opposing forces within

the composite leading to imperfect bonds and separation between the two materials.

It is also likely that voids are created between the two materials during the production

2
of these composites. In a typical finite element model, the bond between the two

materials is modeled as a perfect connection with no separation and the voids are not

taken into consideration. This leads to inaccurate results that over-predict the effects

of the shrinking filler material.

To conduct this study, the FEM software package COMSOL Multiphysics was

used. Different approaches to modeling the composite’s internal geometry were looked

at and a better understanding of the interface region between the two materials was

gained. Comparisons were made to published experimental results that looked at

different volume fractions subjected to a large temperature change. The objective

was not only to develop a way to closely match the experimental results, but also to

do so using the simplest model. This was to ensure that the results of this study were

not only beneficial to those running computers with multiple processors and large

quantities of RAM.

1.3 Overview of Principles

This thesis looks at the finite element analysis (FEA) of thermal expansion in

composite materials containing zirconium tungstate as the filler material. In order

for the reader to better understand this work, a basic overview of the principals

used to conduct this research will be given. Topics covered will include composite

materials, thermal expansion, and FEA.

1.3.1 Composite Materials

A composite is a material which consists of two or more separate materials that

are combined. Their use has become more prevalent over the past few decades due to

the ability to obtain desirable properties that would not be achievable through the use

of a single material. Composite’s constituents are categorized into two components;

3
b b
b b b
b b b
b b b b b
b b
b b b b
b b b b
b
b b b
b
b b b b
b b b
b b
b b
b b b b b
b b b
b b b
b b
b b
b b
b b b b b
b b
b b
b
b b
b b b
b b b b b
b
b b b b b b
b b b b
bb b b
b
b b b b b
b b
b b
b b b
b b b b
b b b b
b b b b
b b b

(a) Sandwich Composite (b) Particle Composite (c) Fiber Composite

Figure 1-2: Three cross-sections showing different ways filler material can
be included in a composite. Each has its own advantages and
disadvantages.

matrix and filler. The matrix is the material that holds the composite together. It

can be made up of a polymer, metal, or ceramic material and is usually chosen for

its ductility, toughness, or electrical insulation property (e.g. [9–11]). The filler is the

material that is mixed into the matrix to form the composite. It is held in place by

the matrix and used to improve certain properties of the matrix material in order to

meet design criteria.

One important characteristic of a composite is its volume fraction (vf ). A com-

posite’s volume fraction is the ratio of one constituent’s volume divided by the entire

composite’s volume. For a two-part composite, the volume of the filler material is

usually referenced. For instance, a composite consisting of two materials and a volume

fraction of 25% would contain one-quarter filler material and three-quarters matrix

material.

Filler material in composite materials can be included in various forms. (Fig. 1-2)

These can range from tiny, nanoscopic particles to long, fibrous strands. In many

cases, the way the filler is included can affect the final properties of the composite.

When fibrous filler is used and the fibers are specifically oriented, the composite will

exhibit anisotropic behavior. This means the composite will behave differently de-

4
pending on the direction of the load relative to the orientation of the fibers. However,

those same fibers could be randomly oriented within the matrix and the composite

would exhibit isotropic behavior. This means the composite would exhibit the same

properties regardless of the loading direction. This paper looks at randomly oriented,

particulate filler in a polymer matrix, displaying isotropic behavior.

1.3.2 Thermal Expansion and Zirconium Tungstate

Thermal expansion is the tendency for a material’s volume to change in response

to a change in temperature [12]. Most materials undergo an increase in volume when

subjected to a positive change in temperature, hence the name thermal expansion.

However, some materials will exhibit thermal contraction, or negative thermal ex-

pansion (NTE), and decrease in volume when subjected to a positive temperature

change [6]. The materials that exhibit this behavior typically only do so over a small

temperature range, rendering them difficult to use in real-world situations. However,

zirconium tungstate is a material that is being heavily researched due to its negative

thermal expansion over a large temperature range (e.g. [5, 7, 8, 13–15]).

Table 1.1: A list of materials and their coefficient of thermal expansion, α,


at 20◦ C.

Material α × 10−6 (K −1 ) Ref.


Aluminum 23.9 [16]
Stainless Steel 17.3 [16]
Copper 17.6 [17]
Concrete 9.9 [17]
Polymide 64.2 [7]
Phenolic Resin 45.6 [5]
ZrW2 O8 -9.0 [6]

Zirconium tungstate is a metal oxide that has a coefficient of thermal expansion

(CTE), α, in its cubic phase of −9 × 10−6 K −1 [6]. Table 1.3.2 lists various materials

5
Figure 1-3: Structure of ZrW2 O8 [6].

along with their CTE for comparison. The NTE behavior of ZrW2 O8 is seen from 2 to

350K and can be attributed to its internal structure [6]. When ZrW2 O8 is subjected to

a positive temperature change, its rigid internal particles rotate, bringing them closer

together, decreasing the overall volume. Figure 1-3 shows the internal structure of

ZrW2 O8 that allows this. The NTE behavior can also be seen in other materials with

similar particle structure, however none have been found to occur over as wide of a

temperature range (e.g. [6, 18–20]).

1.3.3 Finite Element Analysis

Finite element analysis (FEA) is a numerical method to solving partial differential

equations (PDE). In engineering, it is often necessary to develop a mathematical

model to describe the behavior of a system. More often than not, this model will

contain differential equations that can be difficult to solve, making it challenging to

find an exact solution to the system. In FEA, this system is broken down into small

subdomains, or elements, that are connected to each other by nodes. An algebraic

expression can be developed at each of these nodes, allowing for an approximate

6
solution to the system to be obtained.

Today, FEA is a very common technique used among engineers. It allows for com-

putational models to be developed to real world engineering problems. The solutions

of these models are used to verify experimental results, which leads to fewer exper-

iments needing to be run, which leads to saving time and money. In this research,

COMSOL Multiphysics v4.0, a commercial FEA package, was used to numerically

solve thermal expansion problems. The results from the FEA were then compared to

published experimental data.

1.4 Organization of Thesis

The remaining chapters of this thesis will provide a detailed overview of the sim-

ulation approach taken to model ZrW2 O8 composites as well as how the simulations

compare to published experimental data. Chapter 2 consists of a literature review

of previous research in this area. It covers experimental research that has been done

on NTE composites, specifically those that deal with ZrW2 O8 /polymer composites.

It also reviews previous work in the mathematical and finite element modeling of

composite materials.

Chapter 3 covers the approach taken to model ZrW2 O8 composites in this research.

A detailed explanation of the 2-D and 3-D models are covered as well as the different

ways the filler material’s geometry was looked at. Details of the set-up and input

parameters used in COMSOL Multiphysics are also described.

Chapter 4 contains the numerical results to the models as well as discussion about

the results. In this chapter, results from the FEA are compared to similar experi-

mental results as well as analytical equations.

Finally, Chapter 5 includes the conclusions drawn from this research. It sum-

marizes the results from the modeling and published experimental data as well as

7
provides reasoning to the results. It also outlines possible future work, providing

goals and theoretical background to move forward with the research. The appendices

provided at the end of the document contain the computer codes used to develop

some of the geometries in FEA.

8
Chapter 2

Literature Review

Research in the areas of negative thermal expansion (NTE) and composite ma-

terials has been on the rise over the last two decades. Researchers are interested in

creating stronger, lighter, thermally stable materials to be used in a wide range of

applications [5, 7, 8, 21, 22]. This chapter will review previous research involving NTE

materials, specifically zirconium tungstate, as well as the mathematical and finite

element modeling of composite materials.

2.1 Negative Thermal Expansion Materials

As stated earlier, thermal expansion is the tendency for a material’s volume to

change in response to a change in temperature. It is a material property quantified

by a coefficient of thermal expansion (CTE), α, where

αV = V −1 (∂V /∂T ) (2.1)

for volumetric calculations. In Equation 2.1, V represents volume, T represents tem-

perature, and it is assumed that pressure is held constant. For most materials, the

value of α is positive, representing expansion when temperature is increased. How-

9
bc
rs bc rs
rs

rs
bc
bc bc bc bc bc bc

bc
rs

rs
rs bc rs

bc
Figure 2-1: Diagram showing thermal contraction in oxides. Circles repre-
sent oxygen and squares represent cations. As temperature is
increased, the movement of the oxygen atoms causes the cations
to become closer together. Therefore, the overall volume de-
creases [24].

ever, in some materials, the value of the CTE is negative in at least one direction and

in a few of those materials, the CTE is isotropically negative [6].

The study of NTE materials has been ongoing for roughly twenty years. Early

researchers found that there were several oxide systems that exhibited the behavior

[23–30]. In this research, however, it was found that the materials only exhibited

NTE behavior over a small, high temperature range. In addition, the contraction

of the material was usually small and anisotropic. From there, researchers sought

to determine the source of the NTE behavior in these materials. It was found that

the excited oxygen atoms linked to other elements were the cause [24]. The linkages

start linear, but then thermally bend when temperature is increased (Fig. 2-1). This

thermal bending causes a decrease in length between atoms which leads to a decrease

in volume in at least one direction.

One particular material that gained the attention of researchers of negative ther-

mal expansion was zirconium tungstate (ZrW2 O8 ). This material is a ternary oxide

first discovered by Graham et al. in 1959 [31]. In 1967, the phase relations of the oxide

were reported by Chang et al. [32]. This aided in the discovery of its NTE behavior

10
in 1996 by Sleight et al. [14]. This research shows that the NTE behavior exists from

0.3 K to its decomposition temperature of 1050 K. It also states that, because of

its cubic symmetry, the NTE properties are isotropic over its entire stability range.

Similar NTE behavior was also found by Sleight et al. for HfW2 O8 , but the cost to

produce it is much higher [21]. These findings have led to ZrW2 O8 become a highly

researched material for its potential uses in composite materials.

Production of ZrW2 O8 is somewhat challenging and has been carried out in differ-

ent ways [8, 15, 33–36]. For use in composites, small particles of ZrW2 O8 of uniform

size are desirable. This keeps particles from settling during the formation of the com-

posite, leading to a more homogeneous mixture [13]. Nanoparticles of ZrW2 O8 have

also been created to further improve homogeneity as well as mechanical strength [7].

2.2 Experimental Research of ZrW2O8 Composites

The knowledge of zirconium tungstate’s NTE behavior led to its use as filler in

composite materials. The use of polymers as the matrix material allows for a wide

range of applications with low cost and ease of production [37]. The addition of

filler materials within the polymer allows for fine tuning of the material properties

by combining the favorable properties of the polymer with favorable properties from

a filler material [38–42]. Incorporating ZrW2 O8 into matrix materials has allowed

researchers to reduce the overall CTE of the composite [5,8,13,21,22]. This is desirable

due to polymers’ usually high CTE, reducing their usefulness in certain applications

[43–45]. Figure 2-2 shows an example of this research conducted by Tani et al. where

ZrW2 O8 was added to a phenolic resin matrix. As the volume fraction of ZrW2 O8 is

increased, the overall expansion of the composite is reduced.

Researchers are now beginning to look into the interface between the filler and

the matrix [7, 46–49]. This is especially important in composites with ZrW2 O8 as a

11
Figure 2-2: Chart showing the effect of ZrW2 O8 on a phenolic resin ma-
trix during a temperature increase. As the volume fraction of
ZrW2 O8 increases, the overall expansion of the composite de-
creases [5].

filler because of the opposing forces created during a positive temperature change.

Because of this, the interface between the two materials is subjected to high stress.

When the stress becomes too high and the two materials separate, the effect of the

filler material is reduced. Researchers are currently trying different surface treatments

to the ZrW2 O8 to improve the bond between the filler and matrix [7]. Improvement

to this interface region allows the ZrW2 O8 to further reduce the composite’s CTE.

2.3 Mathematical Models for Predicting the CTE

of Composites

Several mathematical models have been developed to predict the overall CTE

in composite materials [50–53]. These range from simple, linear relationships to

12
Figure 2-3: Figure displaying the opposing forces in a ZrW2 O8 composite.
The gray area represents the ZrW2 O8 filler and the white area
represents the matrix material. The dotted line is the interface
between the two materials where high stresses would be seen
when subjected to a temperature increase.

complex equations that account for material stiffness and matrix/filler interaction.

This interaction can greatly affect the resulting CTE of the composite which allows

these models to give more accurate results [5]. A chart comparing results from each

model can be seen in Figure 2-4.

The simplest of the mathematical models is called the Rule of Mixtures (ROM)

[53]. The ROM is a linear average of the matrix’s and filler’s CTE based on volume

fraction. The first order equation is expressed as

αc = vf αf + (1 − vf )αm (2.2)

where αc , αf , and αm are the coefficients of thermal expansion of the composite, filler,

and matrix, respectfully and vf is the volume fraction of the filler. The ROM model

does not account for any relationship between the filler and matrix and also does not

account for the stiffness of either material. It assumes a uniform stress distribution

throughout the composite [7].

Another mathematical model for predicting the CTE of a composite is called the

13
Turner model [50]. This model builds upon the ROM and incorporates each materials

bulk modulus, which accounts for stiffness. The model is expressed as

vf Kf αf + (1 − vf )Km αm
αc = (2.3)
vf Kf + (1 − vf )Km

where Kf and Km represent the filler and matrix bulk moduli, respectively. αc , αf ,

αm , and vf are the same as from the ROM. One will notice that if Kf is equal to Km ,

then the equation simplifies to the ROM.

The last model to be covered is called the Schapery model [51]. This model

consists of two equations that develop an upper and lower bound for the effective CTE

of composites. Schapery used energy principles to develop this model that works for

isotropic and anisotropic composites and also is independent of filler geometry. The

Schapery model also uses a model for elastic modulus developed by Hashin-Shtrikman

(H-S) [54] within the CTE equations. The upper bound is expressed as

Kf (Km − Kcl )(αf − αm )


αcu = αm + (2.4)
Kcl Km − Kf

and the lower bound is expressed as

Kf (Km − Kcu )(αf − αm )


αcl = αm + (2.5)
Kcu Km − Kf

where the superscript u and l refer to the upper and lower bounds and Kcu andKcl

are calculated from the H-S model using the equations below.

1 − vf
Kcu = Kf + 1 3vf (2.6)
Km −Kf
+ 3Kf +4Gf

vf
Kcl = Km + 3(1−vf )
(2.7)
1
Kf −Km
+ 3Km +4Gm

In the Schapery and H-S equations, all variables are the same as in the previous

14
Figure 2-4: Chart comparing different mathematical models for predicting
the CTE of composites. All calculations were done for a com-
posite with a polymer matrix and ZrW2 O8 filler.

two models and Gm and Gf represent the shear moduli of the matrix and filler,

respectively. A chart comparing the results of the three models can be seen in Figure

2-4.

2.4 Finite Element Modeling of Particulate Com-

posites

Finite element analysis (FEA) is a numerical procedure to solve differential equa-

tions that are difficult or impossible to solve analytically [55]. The finite element

15
method takes a given geometry and breaks it down into smaller parts (elements)

and uses a system of algebraic equations to describe the relationship at the points

(nodes) connecting the elements [6]. This allows the original differential equations to

be simplified into a system of algebraic equations that can be quickly solved using a

computer to come up with an approximate solution to the problem.

FEA allows researchers to test ideas and theories without having to purchase

materials and produce samples. It can also be used to verify experimental results. In

the study of thermal expansion in composites, FEA is a tool that can give an idea of

the effect certain filler materials will have on the overall composite and what volume

fractions researchers should test experimentally.

Research in the area of FEA of particulate composites has been mostly limited

to composites with positive CTEs for both the filler and matrix [56–61]. In these

studies, the effect of particle size, shape, and distribution within the FEA model have

been looked at and compared to mathematical models and experimental results. To

date, little research on the FEA of composites with negative CTE fillers has been

completed. One study in the area looked at the FEA of closely packed tetrahedra

ZrW2 O8 at 60% volume loading in a copper matrix [62]. This study used FEA to

calculate the thermal mismatch stresses within the composite and found that they

were high enough to trigger a pressure-induced phase transformation. Another pub-

lication looked at how thermal and mechanical properties, rates of cooling/heating,

geometry, and packing fraction influenced the overall expansion and thermal stress in

composites with ZrW2 O8 filler and either copper or ziconium oxide matrix [6]. Here

researchers concluded that the thermal stresses are larger when the amount of filler

is larger and that it is advantageous to use matrix and filler materials with similar

magnitudes of CTE. The research of this thesis built upon these works and studied

the FEA of polymer/ZrW2 O8 composites.

16
Chapter 3

Modeling Approach

This research focused on the finite element modeling of thermal expansion in

polymer based composites with zirconium tungstate as a filler material. The FEA

was completed with the software package COMSOL Multiphysics version 4.0 using the

structural mechanics module. This software allowed for two and three-dimensional

models to be created with different numbers of filler particles and different filler

geometries. This chapter will provide a detailed explanation of the approach to

modeling including the materials used and their properties, the various geometries

of the 2-D and 3-D models, the setup for COMSOL, and how the solutions to the

simulations were analyzed.

3.1 Materials Used and their Properties

The materials studied in this research were polymer based composites with the

negative thermal expansion material ZrW2 O8 as a filler. Two different polymer matrix

materials were looked at in order to compare the results from the FEA to experimental

results published by Tani et al. [5] and Sharma et al. [7] The experiments conducted

by Tani and Sharma were on ZrW2 O8 composites using phenolic resin (PR) and poly-

imide (PI) as the matrix materials, respectively. To run FEA of thermal expansion

17
on these composites, four material properties are required for each material. These

are Young’s modulus (E), Poisson’s Ratio (ν), density (ρ), and coefficient of thermal

expansion (α). Then, in order to calculate the CTE using the mathematical models

discussed in section 2.3, the materials’ bulk moduli (K) and shear moduli (G) must

also be known. All of the materials used are listed in table 3.1 with each of their

required properties.

Table 3.1: Materials used in FEA models and their required properties

kg −6
Material E(GP a) ν ρ( m3) α( 10K ) K(GP a) G(GP a)
ZrW2 O8 88.3 0.33 5090 -9.0 74.5 33.9
PI 2.2 0.35 1430 64.2 2.4 0.8
PR 3.5 0.37 1370 45.6 4.5 1.3

3.2 Setup for COMSOL Multiphysics

This research was conducted using the finite element software COMSOL Mul-

tiphysics version 4.0. This software was used to create two and three-dimensional

models, simulate a change in temperature applied to the models, and analyze the ef-

fects of the temperature change on the models’ dimensions. This section will overview

the settings and initial setup of COMSOL for this research.

When first opening COMSOL multiphysics, a window similar to the one in figure

3-1 is seen. When starting a new model, this is where options such as space dimension,

physics, and type of study are selected. For this research, both 2-D and 3-D models

were used, so one of those two options was chosen first. Next, COMSOL prompts

users to choose which physics modules they would like to use. COMSOL is capable of

running multiple physics modules on one model, however, for this research, only the

solid mechanics module under the structural mechanics section was needed. Lastly,

18
Figure 3-1: Opening window seen when COMSOL is first launched. This
research used both 2-D and 3-D modeling, the solid mechan-
ics module under structural mechanics physics, and a stationary
study.

COMSOL needs to know what type of study will be run, which in this case was

stationary.

After inputing the initial settings, the model is ready to be created. At this

point, the geometries of the matrix and filler are drawn. Details about the different

geometries studied and how they were created can be seen in sections 3.3 and 3.4.

Once the geometry of the model is made, the matrix and filler material properties

need to be assigned. To do this, two materials were created for each model under the

material branch of the model builder tree (MBT). (Fig. 3-2) Then, the properties

listed in Table 3.1 were input to the appropriate matrix or filler material.

The next step was to apply boundary conditions to the model. This is done under

the solid mechanics branch of the MBT. For all models, roller supports were used as

the only boundary conditions. The rollers were added to the x = 0, y = 0, and z = 0

19
Figure 3-2: An example of what the model builder tree in COMSOL looked
like for the finite element models created for this research.

surfaces (or just the x and y if the model was 2-D) to allow for thermal expansion

in the positive directions similar to what is seen in Figure 1-1(a). After applying

the boundary conditions, the loads were applied to the models. While there were no

external forces applied in any of the simulations, the loading was applied through a

change in temperature. To do this, the thermal expansion option was added to the

model under the linear elastic material model branch of the MBT. Once added, this

option allowed for two temperatures to be defined. The first was the strain reference

temperature, Tref , which was the initial temperature of the model. Then, the user

defined temperature, T , was the final temperature. Subtracting Tref from T allowed

for the change in temperature, ∆T , to be calculated. For all models, ∆T was positive

and Tref was kept at 293.15 K.

After the boundary conditions and loads were applied, the model was ready to be

meshed. Meshing in FEA is where the geometry of the model is broken down into

20
Figure 3-3: An example of a two-dimensional, free triangular mesh in COM-
SOL set at the ‘finer’ size setting.

smaller pieces, called elements, which are connected by nodes. Each element is then

solved individually creating a simpler way to obtain an approximate solution. When

meshing models, it is generally understood that a finer mesh (more elements) will give

a more accurate solution. However, a finer mesh will also require more time for the

computer to solve the study. Therefore, it is necessary to determine the appropriate

balance between computation time and an acceptable level of accuracy. A mesh was

created in COMSOL by right-clicking on the mesh branch of the MBT and selecting

free triangular for 2-D models and free tetrahedral for 3-D models. Then, the size

of the mesh was set to the ‘finer’ setting. This gave an initial starting point for the

mesh which could then be adjusted either finer or courser based on the time it took

to solve the problem. An example of a two-dimensional mesh set to the ‘finer’ setting

can be seen in Figure 3-3.

The last step in setting up COMSOL for this research was to adjust the default

21
Figure 3-4: A screen-shot of the MBT showing the changes made to the de-
fault solver. Changes were made under the ‘Advanced’, ‘Iterative
1’, and ‘Multigrid 1’ branches to achieve faster solution times
with no effect to the accuracy of the results.

solver settings. This allowed for the solutions to be found in a slightly faster time

while having no negative effects on the accuracy of the solution. To do this, the study

branch of the MBT was right-clicked on and ‘show default solver’ was selected. Once

all the drop-down menus are expanded, various options are available to change on

the solver. For the models run in this research, the first change was made under the

‘Advanced’ branch which was found under ‘Stationary Solver 1’. Here the matrix

symmetry was changed from automatic to symmetric. The next change was made

on the ‘Iterative 1’ branch. Here the solver was changed from GMRES to conjugate

gradients. The last change was made under the ‘Multigrid 1’ branch where the solver

setting was changed from a geometric to an algebraic multigrid. The branches these

changes were made under can be seen in Figure 3-4.

22
3.3 Modeling to Compare Geometry

The first portion of this research looked to compare the FEA results of polymer/ZrW2 O8

composites with the same volume fraction but different geometries. The goal was to

determine if varying the geometry of the model had a significant effect on the results.

Two-dimensional and three-dimensional models were compared as well as different

filler geometries and numbers. Details of the geometries modeled and compared can

be seen in the following two sections.

3.3.1 Two-Dimensional Models

The initial modeling completed for this research was done using two-dimensional

models. These models were created using the same materials and volume fractions

and then subjected to the same temperature change. The goal was to compare these

models to see the effect different filler geometries and number of inclusions has on the

relative expansion of the composite after being subjected to a positive temperature

change. The first model created was a completely isotropic polymer to use as a

control. The geometry was a one-meter by one-meter square created by right-clicking

on the ‘Geometry’ branch of the MBT and selecting the ‘Square’ option. The material

properties of phenolic resin were applied to the square geometry and the model was

subjected to a 70K temperature increase. The geometry was meshed and the study

was run using the settings described in Section 3.2. After running the control model,

filler material was added to the geometry.

To model the composites in two dimensions, smaller geometries, or inclusions,

were added to the polymer square. This portion of the research looked to compare

the effects of changing the shape and number of these inclusions. To do this, three

sets of models differing only by volume fraction were created. Each set consisted of

eight models with different filler geometry that equated to the same amount of matrix

23
and filler materials. The three volume fractions modeled were 7%, 19%, and 37%.

For each volume fraction, eight different models were simulated. These consisted of

four different filler arrangements with two different filler geometries. The two filler

geometries used were square and circular and the four different filler arrangements

consisted of models with one-fourth of an inclusion (using symmetry), one inclusion,

nine inclusions, and twenty-five inclusions. All the variations of the filler created a

matrix of twenty-four models to be simulated for this section. Figures 3-5, 3-6, and

3-7 show all of the models simulated for the two-dimensional portion of the study.

To create the models in COMSOL, the geometry of the filler was added to the

control model of pure polymer. This was done through the use of the ‘Square’ or

‘Circle’ option under the ‘Geometry’ branch of the MBT. The size of the square or

circular inclusions depended on the volume fraction and number of filler particles. To

quickly determine these sizes, equations 3.1 and 3.2 were used for square and circular

inclusions, respectively.

nx2sq
r
n(Af ) n[(lf )(wf )] n[(xsq )(xsq )] vf
vf = = = = → xsq = (3.1)
Ac (lc )(wc ) (1)(1) 1 n

πd2 2 2
r
n(Af ) n( 4cir ) n(πrcir ) nπrcir vf
vf = = = = → rcir = (3.2)
Ac (lc )(wc ) (1)(1) 1 nπ

In Equation 3.1, vf is the volume fraction and n is the number of square inclusions

within the composite. Af is the area of one inclusion, which is equal to the length

times the width of the square (lc and wc ) which are both equal and called xsq . Ac

is the area of the entire composite, which is equal to the length times the width of

the model (lc and wc ). Both of these values are equal to one meter making the area

one square meter. Equation 3.2 is for circular inclusions and all the variables are the

same as in equation 3.1 except for rcir which is the radius of one inclusion.

Once the size of the filler particles was determined, the ‘Array’ tool under ‘Trans-

24
Figure 3-5: Geometry of two-dimensional models with 7% percent filler vol-
ume. Each grid division represents 0.5 m and all models have an
initial total volume of 1 m2 .

Figure 3-6: Geometry of two-dimensional models with 19% percent filler vol-
ume. Each grid division represents 0.5 m and all models have an
initial total volume of 1 m2 .

25
Figure 3-7: Geometry of two-dimensional models with 37% percent filler vol-
ume. Each grid division represents 0.5 m and all models have an
initial total volume of 1 m2 .

forms’ was used on models with more than one inclusion to create a symmetric pattern

centered within the polymer square. After the full geometry of the filler was created,

the ‘Difference’ tool under ‘Boolean Operations’ was used to subtract the filler ge-

ometry from the matrix. When doing this, the ‘Keep input objects’ box was checked

so that the filler’s geometry was not deleted after the operation. However, because

that option was selected, the original matrix square needed to be removed using the

‘Delete Entities’ tool. This left two domains within the model. One being the matrix

with voids, and the other being the filler inclusions. Material properties were then

added to both domains with the matrix and filler consisting of phenolic resin and

ZrW2 O8 , respectively. All models were subjected to a 70 K temperature increase

with boundary conditions, mesh settings, and solver settings as described in section

3.2. Results from these models can be seen in Chapter 4.

26
3.3.2 Three-Dimensional Models

Three-dimensional models were also tested and compared to the results of the

two-dimensional models. A similar matrix of tests was created that varied the shape

of the filler as well as the number of inclusions. The only volume fraction tested

was 7% due to the computation time required to run a simulation. While only one

volume fraction was tested, results were still conclusive and able to be compared to

the two-dimensional models.

The exterior dimensions of the three-dimensional models were based off of the

samples tested by Tani et al. [5] with a length of 20 mm, a width of 4 mm, and a

height of 4 mm. Filler was added to the block in the shape of either cylinders, spheres,

or cubes. Cylindrical filler models consisted of either one, nine, or twenty-five, 20 mm

tall cylinders with radii depending on the number of inclusions. Equation 3.3 was

used to determine these radii and is similar to the equations used in two-dimensional

modeling. The geometry of these models can be seen in Figures 3-8, 3-9, and 3-10.

πd2 hcyl 2 2
r
n(Vf ) n( cyl4 ) n(πrcyl (20)) nπrcyl 16vf
vf = = = = → rcyl = (3.3)
Vc (lc )(wc )(hc ) (4)(4)(20) 16 nπ

For three-dimensional models containing spherical filler particles, Equation 3.4

was used to determine the radius of the spheres based on the volume fraction and

number of particles. Models were made using ten and fifty spherical inclusions located

randomly within the matrix. To generate the random coordinates, a MATLAB code

was used. This code ensured that the correct number of inclusions were used and

that the spheres did not overlap or break the surface of the matrix. This code can

be seen in Appendix A. Because the location of the spherical inclusions was random

within the matrix, the results of the simulation differed slightly with each run. To

account for this, five different simulations were run and the results were averaged to

27
Figure 3-8: Three-dimensional model with seven percent filler volume made
of one cylinder. The initial volume of the model is 320 mm2 .

Figure 3-9: Three-dimensional model with seven percent filler volume made
of nine cylinders. The initial volume of the model is 320 mm2 .

28
Figure 3-10: Three-dimensional model with seven percent filler volume made
of twenty-five cylinders. The initial volume of the model is
320 mm2 .

obtain the data point for each specific case. The geometry of these models can be

seen in Figures 3-11 and 3-12.

n( 34 πrsph n( 43 πrsph
3 3 3
r
n(Vf ) ) ) nπrsph 3 240vf
vf = = = = → rsph = (3.4)
Vc (lc )(wc )(hc ) (4)(4)(20) 240 nπ

The last style of three-dimensional model studied was quite different than the

previous two. This model used MATLAB and COMSOL together to create a model

with a specific mesh and then randomly assigned either matrix or filler material

properties to individual elements until the desired volume fraction was achieved. To

do this, five MATLAB files were created, all of which can be seen in Appendix A.

The first of these codes defined the size of the mapped mesh (a grid pattern mesh

consisting of a certain number of blocks) as well as the volume fraction and then

randomly assigned each block as either matrix or filler material. This code output

29
Figure 3-11: Three-dimensional model with seven percent filler volume made
of ten spheres. The initial volume of the model is 320 mm2 .

Figure 3-12: Three-dimensional model with seven percent filler volume made
of fifty spheres. The initial volume of the model is 320 mm2 .

30
the coordinates of the elements to become filler material which was then used by the

other four codes to assign material properties. Each required material property (E,

nu, rho, and alpha) has a MATLAB code that looked at the coordinates generated by

the first code and then assigned the correct values for that specific material property

to each particular element. Once the codes for each material property were written,

the m-files needed to be placed in the COMSOL root folder so they could be found

and used when the models were run.

To use these MATLAB files within COMSOL, a few settings needed to be changed

from previous models. Firstly, the codes needed to be defined within COMSOL so that

they could be found when needed. To do this, a MATLAB function was added under

the ‘Global Definitions’ branch of the MBT. Here, each MATLAB file that defined a

material property was added to the list and the variables it used were defined. Next,

the geometry of the models was created. This was, simply, a solid, 20 mm by 4 mm

by 4 mm bar with roller supports on three sides. The only load applied to this bar

was, again, a temperature increase of 90 K. To define the material properties of

this bar, no values were input under the ‘Material’ branch of the MBT. Instead, one

material was added and the names of MATLAB files defined above were input into

the ‘value’ fields, telling COMSOL to look to the results of the MATLAB code to

find the material property. The last step to create these models was to create the

mesh. For these tests, a mapped, grid-pattern, mesh was used. It was important that

the mesh defined within COMSOL matched the grid that was input in the MATLAB

codes. For this research, a mapped mesh of ten blocks, by ten blocks, by fifty blocks

was used. To create this within COMSOL, the first step was to add a mapped mesh

under the ‘Mesh 1’ branch of the MBT. Then, the x = 0, y = 0, and z = 0 faces

were added to the selection box. The next step was to define the size of the mesh

by adding the ‘Size’ option under the ‘Mapped 1’ branch of the MBT. Here, the

‘custom’ option was selected and the maximum element size was set to 0.0004 m to

31
produce the desired 10 by 10 by 50 element mesh. The last step in the creation of

the mesh was to add the ‘Swept’ option under ‘Mesh 1’ to convert the face mesh to a

three-dimensional mesh of the whole block. Once all the proper options were applied

to the mesh, the build button was pressed and COMSOL output that 5, 000 elements

were created. Figure 3-13 shows what this mesh looked like within COMSOL.

Figure 3-13: Three-dimensional model with a 10 by 10 by 50 element mapped


mesh. The initial volume of the model is 320 mm2 .

After creating the mesh the models were ready to be run. The solver was set-

up the same way as all previous models which is shown in Figure 3-4. As with the

spherical three-dimensional models, five separate models were run and the results

were averaged. These results can be seen in Chapter 4.

3.4 Modeling to Examine Matrix/Filler Bond

The models created in the previous section were based off the experiments of

Tani et at. [5] where expansion data was collected for phenolic resin/ZrW2 O8 com-

posites. After running the simulations it was found that the FEA models tended to

32
over-predict the negative CTE effect of the ZrW2 O8 filler material. More detailed

results can be seen in the next chapter, but this section will describe the modeling

techniques that were used to gain a better understanding of the interface between the

two materials.

It was assumed that the difference between the experimental data and the FEA

models could be attributed to the bond between the matrix and filler materials. In

the FEA models, this was created as a perfect bond with no voids or separation. In

experimental procedures, this perfect bond is nearly impossible to achieve. Therefore,

the goal of this modeling was to establish the upper and lower bounds of the data by

creating a model with a perfect bond between the filler and matrix as well as a model

with no bond between the filler and matrix.

The model used for the perfect bond case was already created for the previous

section and the results could be carried over. It was decided to use the geometry of

the model that showed the most expansion of all the two or three-dimensional models.

This way, when the bond between the matrix and filler was removed, it would show

the most expansion and better represent the worst-case scenario. However, choosing

a different geometry would make little difference as all the models showed similar

results.

The next step was to create a model with no bond between the filler and matrix.

To do this, the filler particles of the model were completely removed, leaving empty

voids within the matrix. By doing this, the effect of the filler was removed from the

model, which gave the same effect as if there was no bond between the two materials.

Therefore, these models would show much more expansion than the models with a

perfect bond and represent the worst-case scenario in experimentation.

Using the modeling techniques just described, FEA was completed to compare to

the experiments of Tani et al. and Sharma et al. To compare to the Tani experiments,

models contained the properties of phenolic resin and zirconium tungstate at volume

33
fractions of 7%, 19%, and 37%. To compare to the Sharma experiments, models

contained the properties of polyimide and zirconium tungstate at volume fractions of

5%, 10%, and 15%. The results of these models as well as how they compare to the

published experimental results can be seen in Section 4.3.

3.5 Analysis of Models

The final step to the modeling conducted in this research was to evaluate the

models. This consisted of simply determining the final total volume after the 70 K

temperature change. To do this, post-processing was conducted within COMSOL

after the models had been run.

After a model finished its study, COMSOL displays a surface plot of total dis-

placement and puts the user in the ‘Results’ section of the MBT. The first step was

to adjust the plot to show the deformation of the model. To do this, under ‘2D

Plot Group 1’, ‘Surface 1’ is right-clicked on and ‘Deformation’ is selected. The plot

should now show a scaled, deformation plot that extents past the black lines of the

original shape. The next step was to have COMSOL calculate the change in volume

of the model. To do this, the ‘Derived Values’ section was right-clicked on and either

‘Surface Integration’ or ‘Volume Integration’ was selected depending on if the model

tested was two-dimensional or three-dimensional, respectively. In the settings of the

integration, all domains were selected and the expression was set to ‘solid.evol’ to

integrate over the volumetric strain. With those settings chosen, the equals button

was selected and COMSOL then output a value for volumetric strain in the results

table. This value represented the change in volume of the model. Therefore, know-

ing the initial volume of the model (1 m2 for 2-D, and 320 mm2 for 3-D), the final

volume of the model was calculated by adding the value of volumetric strain to the

initial volume. From there, the relative expansion was calculated using Equation 3.5

34
to allow for direct comparison between the 2-D and 3-D models. Comparisons were

also made between these results and the mathematical models discussed earlier. The

results and discussions about them can be seen in Chapter 4.

Vf inal
RelativeExpansion = −1 (3.5)
Vinitial

Figure 3-14: COMSOL results settings for determining the change in volume
of the models.

35
Chapter 4

Results and Discussion

The models described in Chapter 3 were run and evaluated as described in Section

3.5. The results of these models will be given and discussed in this chapter. First,

the two-dimensional models will be looked at, followed by the three-dimensional ones.

Then, the models will be compared to published experimental results and modifica-

tions will be made to create an upper and lower bound for expansion. Finally, the

results will be compared to the mathematical models described in Section 2.3 to de-

termine which model the data most closely followed and how the upper and lower

bounds compare.

4.1 Two-Dimensional Models

The two-dimensional models detailed in Section 3.3.1 were run and evaluated as

described in Section 3.5. The results of these tests will be discussed in this section.

The first model run was a one-meter by one-meter square made up of only phenolic

resin; the matrix material. This model was created to determine the baseline of

expansion for the pure polymer and to better understand the effects of adding the

filler material. The total displacement plot for this model can be seen in Figure 4-1

and the relative expansion can be seen in Table 4.1.

36
Figure 4-1: Total displacement plot for a two-dimensional, pure polymer
model. This model represents the baseline of all future models
and has an even, linear growth in the positive x and y directions.
It was subjected to a temperature change of 70 K and the scale
factor of the displacement is 200.

When looking at the displacement plot, take note of its linear shape and even

displacement in both the positive x and y directions. It should be noted that the plot

has a scale factor on the displacement of 200. This allows the changes in shape to be

more easily seen and will be used on all 2-D plots. After using COMSOL to determine

the change in area, Equation 3.5 was used to calculate the relative expansion using

the initial and final areas in place of the volumes. This was found to be 0.292%.

Next, the two-dimensional models containing 7% filler material were run. These

models were all created in the same COMSOL file and run at the same time, but

analysed individually. The total displacement plot for these models can be seen

in Figure 4-2 and the expansion data can been seen graphically in Figure 4-5 and

numerically in Table 4.1. When looking at the total displacement plot, the effects of

the ZrW2 O8 can be seen, but only slightly, and the displacement of all eight models

37
appears to be similar. This was confirmed by calculating the relative expansion for

each model and plotting the results. The graph in Figure 4-5 shows that there was

little deviation in the expansion of the eight models suggesting that filler shape and

arrangement have little effect on the overall expansion. The average value of relative

expansion for the 7% models was 0.252% with a standard deviation of 0.0010%.

This equates to a 0.070% reduction in relative expansion when compared to the pure

polymer model.

Figure 4-2: Total displacement plots for two-dimensional models with seven
percent filler volume. The average relative expansion of the mod-
els was 0.252% meaning an average reduction in expansion of
0.070% was seen when compared to the pure polymer model.

Next, the models with 19% filler were run. Like the 7% models, these were all

created in the same COMSOL file and run and the same time, but analysed separately.

The total displacement plot for these models can be seen in Figure 4-3. In these plots,

the filler material’s effect is more noticeable, especially in the models with only one

particle where the expansion is not as evenly distributed as the models with multiple

inclusions. Despite these differences, the relative expansion of the models were all

38
similar with an average value of 0.187% and a standard deviation of 0.0024%. This

gives an average reduction of expansion of 0.135% when compared to the pure polymer

model. This data can be seen graphically in figure 4-5 and numerically in Table 4.1.

Figure 4-3: Total displacement plots for two-dimensional models with nine-
teen percent filler volume. The effects of the ZrW2 O8 filler can
be seen in the shape of the displaced plots and an average re-
duction in expansion of 0.135% was calculated when compared
to the pure polymer model.

The last set of two-dimensional models run were those containing 37% filler. These

models were run the same way as those with 7% and 19% and the total displacement

plot can be seen in Figure 4-4. The negative effect of the filler material is very

noticeable in these plots with the expanded shape appearing only slightly larger than

the original geometry, even with the scale factor of 200. These models also gave the

largest variation in relative expansion with an average value of 0.096% and a standard

deviation of 0.0026%. This can be seen graphically in Figure 4-5 and numerically in

Table 4.1. The 37% filler reduced the expansion of the polymer by 0.226%.

To compare the results of the 2-D models, the graph seen in Figure 4-5 was

39
Figure 4-4: Total displacement plots for 2-D models with 37% percent filler
volume. These models had the largest variance in expansion with
an average value of 0.096% and a standard deviation of 0.0026%.
The 37% filler reduced the expansion of the polymer by 0.226%.

created. This graph shows that the addition of the ZrW2 O8 reduced the expansion of

all models when compared to the pure polymer model, represented by the horizontal,

orange line. It also compares the different arrangements of filler particles tested for

each volume fraction. It was found that at each volume fraction, the arrangement

or shape of the filler particles was not an important factor in the overall expansion.

This can be seen by each volume fraction producing very similar expansion data for

all models tested. Knowing this, the expansion data was averaged and plotted versus

volume fraction, as shown in Figure 4-6. A linear best-fit line was added to these

data points and fit very well with an R2 value of 0.98. This means that as more

filler material is added to a composite, the reduction in expansion should be linear.

If this prediction is projected past the data of the models, a composite exhibiting

no expansion over the tested temperature range would require a volume fraction of

approximately 52%.

40
Table 4.1: Numerical results from the two-dimensional modeling. This data
shows that the filler geometry of 2-D models has little effect on
the expansion results with low standard deviations calculated for
each volume fraction. This is represented graphically in Figure
4-5.
Volume Filler Num. of Final A Relative Avg Std
Fraction Shape Inclusions (m2 ) Exp. Exp. Dev.
0% NA 0 1.00322 0.322% 0.322% NA
7% Sq 0.25 1.00252 0.252%
7% Sq 1 1.00251 0.251%
7% Sq 9 1.00251 0.251%
7% Sq 25 1.00251 0.251% 0.252% 0.0010%
7% Cir 0.25 1.00253 0.253%
7% Cir 1 1.00253 0.253%
7% Cir 9 1.00253 0.253%
7% Cir 25 1.00253 0.253%
19% Sq 0.25 1.00186 0.212%
19% Sq 1 1.00183 0.210%
19% Sq 9 1.00184 0.210%
19% Sq 25 1.00184 0.210% 0.187% 0.0024%
19% Cir 0.25 1.00189 0.214%
19% Cir 1 1.00189 0.214%
19% Cir 9 1.00188 0.214%
19% Cir 25 1.00188 0.214%
37% Sq 0.25 1.00095 0.095%
37% Sq 1 1.00094 0.094%
37% Sq 9 1.00094 0.094%
37% Sq 25 1.00094 0.094% 0.096% 0.0026%
37% Cir 0.25 1.00100 0.100%
37% Cir 1 1.00099 0.099%
37% Cir 9 1.00098 0.098%
37% Cir 25 1.00097 0.097%

41
Figure 4-5: Bar graph comparing all of the two-dimensional models studied.
The orange line at the top represents the expansion of the pure
polymer model. It can be seen that all models reduced the expan-
sion of the polymer and that each volume fraction gave similar
results for all filler arrangements.

Figure 4-6: This is a plot of volume fraction versus relative expansion cre-
ated from the averaged data of the two-dimensional models. A
best-fit line was added to show the linearity of the data and also
projects that a volume fraction of approximately 52% will pro-
duce a composite with no expansion.

42
4.2 Three-Dimensional Models

After running and evaluating the two-dimensional models, the three-dimensional

models were looked at. The goal was to see if modeling in three-dimensions affected

the outcome of the results in a significant manner. To do this, various models were

created at a volume fraction of 7% and compared to each other, as well as the results

from the two-dimensional testing. The models were based off the experiments of Tani

et al. [5] where composite bars of 4 mm × 4 mm × 20 mm were subjected to a 70 K

temperature increase.

The first model run had a filler geometry of one cylinder running through the

center of the bar. This model’s total displacement plot can be seen in Figure 4-

7. When looking at the plot, it is apparent that there was an even displacement

throughout the bar, mostly in the y and z directions. The long, cylindrical filler

prevented the bar from expanding in the x direction. The relative expansion of this

model was calculated to be 0.255%.

Figure 4-7: Total displacement plot for 3-D model with one cylindrical filler
particle. Expansion is seen mainly in the y and z directions with
little growth along the x axis.

43
For the next model, an array of nine cylinders was used for the filler geometry.

Like the model created with one cylinder, this one also showed even growth in the y

and z directions with the cylinders preventing much expansion in the x direction. This

can be seen in Figure 4-8. The relative expansion was calculated to be 0.256%. The

model shown in Figure 4-9 with twenty-five cylinders, again, showed similar behavior.

Its relative expansion was found to be 0.262%.

Figure 4-8: Total displacement plot for 3-D model with nine cylindrical filler
particles. Expansion is similar to the model with one cylindrical
particle.

For the next 3-D models, the filler was made up of spherical particles. One model

was made with ten inclusions and another with fifty. The total displacement plots

of these models can be seen in Figures 4-10 and 4-11, respectively. When looking

at these plots, the first thing noticed is that they are drastically different than the

ones of the models with cylindrical filler geometry (the displacement scale is kept the

same for all models). The plots for the spherical models show large displacement in

the x direction while the plots for the cylindrical models showed very little change

in that direction. This would lead one to believe that the expansion of the spherical

44
Figure 4-9: Total displacement plot for 3-D model with twenty-five cylindri-
cal filler particles. Little difference was seen between the three
models with cylindrical-shaped filler.

models is significantly greater than the expansion of the cylindrical models, however,

this is not the case. The relative expansion of the ten sphere model was found to be

0.265% and the fifty sphere model was 0.258% (both values obtained by averaging

five models’ results) showing that they exhibited a similar change in volume as the

cylindrical models. This meant that the spherical particles allowed the models to

expand primarily in the x direction while the cylindrical models prevented this growth,

which then was then made up for in the y and z directions.

The last three-dimensional model created used a mapped mesh to create cube-

shaped elements. Matlab was then used to randomly select enough elements to give

the composite the desired 7% filler material and the rest were given matrix material

properties. The total displacement plot of this model can be seen in Figure 4-12

where it can be seen that it expanded similarly to the spherical models with large

displacement in the x direction. The relative expansion, found by averaging five runs,

was found to be 0.262%.

45
Figure 4-10: Total displacement plot for 3-D model with ten spherical filler
particles. The shape of the filler allows this model to expand
primarily in the x direction, which is different from the models
with cylindrical filler. However, the final volumes did not differ
significantly.

Figure 4-11: Total displacement plot for 3-D model with fifty spherical filler
particles. This model behaved similarly to the model with ten
spheres and produced a similar final volume.

46
Figure 4-12: Total displacement plot for 3-D model made with a mapped
mesh of cube-shaped particles. This model used Matlab to ran-
domly assign material properties to the mesh and contained the
smallest particles tested. The results were most similar to the
spherical models.

In Table 4.2, all of the numerical data from the three-dimensional models can be

seen. Looking at the relative expansion and standard deviation for all the models

shows that little changes were seen when the filler geometry was changed, which

agrees with the results from the two-dimensional modeling. This trend can be seen

graphically in Figure 4-13 where the average two-dimensional, 7% volume fraction

results were added as the last column. This shows that the three-dimensional models

showed similar expansion to the two-dimensional ones. All of the three-dimensional

models exhibited slightly more growth than the two-dimensional models, but not

enough to be deemed significant. The standard deviation of the three-dimensional

models was also higher than the two-dimensional ones, but no pattern to explain this

was seen.

47
Table 4.2: Numerical results from the three-dimensional modeling. This data
shows that the filler geometry of 3-D models has little effect on
the overall expansion of the model. The standard deviation of the
data is low and follows the trend seen in the 2-D modeling. This
data is represented graphically in Figure 4-13.

Filler Initial Final Relative Standard


Geometry V (mm3 ) V (mm3 ) Expansion Deviation
1 Cylinder 320.0 320.815 0.255%
9 Cylinders 320.0 320.818 0.256%
25 Cylinders 320.0 320.839 0.262%
10 Spheres 320.0 320.838 0.262% 0.0038%
50 Spheres 320.0 320.816 0.255%
Mapped Mesh 320.0 320.838 0.262%

Figure 4-13: Bar graph comparing the three-dimensional models studied.


The orange line at the top represents the expansion of the pure
polymer model. It can be seen that all models reduced the ex-
pansion of the polymer and that each filler configuration showed
similar expansion behavior. Results from the 2-D models are
also included and shown to be similar.

48
4.3 Comparisons to Experimental Data and De-

velopment of Upper and Lower Bounds

When looking at the results of the two and three-dimensional models evaluated

for the previous sections, it can be seen that all models of the same volume fraction

produced similar expansion results. While the consistency of the results is a plus, how

the models compare to real-world experimental data is also important. To do this,

the results of the experiments of Tani et al. [5] were looked at as the same materials

were used. Figure 4-14 compares the results of the 7% models with the experimental

data of Tani. The two models chosen for the graph were the 7% models that showed

the highest and lowest expansion. The model that showed the least expansion was

the two-dimensional model with nine square-shaped filler particles. The model that

showed the most expansion was the three-dimensional model with twenty-five cylin-

drical filler particles. Looking at the plot shows that even the model with the most

expansion is considerably less than the experimental data published by Tani. This

means that all the FEA models over-predicted the effect of the ZrW2 O8 filler material

and reduced the expansion more than what is seen in physical experiments.

While exactly matching experimental results was not a goal of this research, the

comparison in Figure 4-14 does give a better understanding of the FEA models. In the

models, the filler and matrix materials are considered to be perfectly bonded together

with no voids or separation. In real-world experiments, this is nearly impossible to

achieve. When a composite material is made, small air bubbles can get trapped be-

tween the two materials which leads to imperfect bonds. In composites with ZrW2 O8

filler material, this has larger consequences as the opposing forces created during a

temperature increase only accentuate these imperfections. This explains why the all

the FEA models showed less expansion than the experimental data. The perfect bond

between the filler and matrix gave the negative thermal expansion filler more effect,

49
Figure 4-14: Bar graph comparing the FEA models to experimental data.
The two models chosen represent the highest and lowest expan-
sion of all the 2-D and 3-D, 7% models. This shows that all
modeling produced results that showed less expansion that the
experiments of Tani et al.

decreasing the final volume.

Knowing that the bond between the filler and matrix is perfect in the FEA models,

it is considered to be the lower bound of possible expansion for the composite. This

means that in a physical experiment, a test specimen will not show less expansion than

this lower bound provided they are tested at the same volume fraction. The obvious

next step is to then determine the upper bound for the expansion. A specimen

would undergo the most expansion if the ZrW2 O8 filler was not connected to the

matrix material at all, or, in other words, the complete opposite of the previous FEA

models. To do this in FEA, models were created similarly to the previous models, but

the filler particles were deleted leaving voids within the matrix material. To reduce

the amount of modeling needed, the results from the previous two sections were used

to guide this section. Geometry for the models was chosen based on the previous

50
data where the model showing the most expansion was used. This ensured that the

upper bound represented the worst case scenario and the lower bound maintained

consistency. The model showing the most expansion was the three-dimensional model

with twenty-five cylindrical filler particles and the expansion data from it was used

for the lower bound. It was then modified to represent no bond between the matrix

and filler by deleting the twenty-five cylindrical filler particles, leaving twenty-five

cylindrical voids within the matrix. Volume fractions of 7%, 19%, and 37% were

tested and the results can be seen in Figure 4-15.

Figure 4-15: Graph showing the upper and lower bounds of expansion cre-
ated by FEA models compared to Tani’s data. The data ob-
tained by Tani et al. falls between the upper and lower bounds
in blue and green, respectively, for all volume fractions.

In Figure 4-15, it can be seen that the experimental data obtained by Tani et

al. falls between the upper and lower bounds created by the finite element models

for all volume fractions. The experimental data also tends to be closer to the upper

bound, meaning that there was likely many voids and separation between the filler

51
and matrix within the specimens. To further validate the FEA upper and lower

bounds, the experiments of Sharma et al. [7] were also compared to the models. In

these experiments, ZrW2 O8 filler was added to a polyimide (PI) matrix material and

the effects on the CTE were looked at. Surface treatments were also done to the

ZrW2 O8 to improve the bond between the filler and matrix. The FEA models were

adjusted to account for the different matrix material properties and different volume

fractions of 5%, 10%, and 15%. The results can be seen in Figure 4-16.

Figure 4-16: Graph showing the upper and lower bounds of expansion cre-
ated by FEA models compared to Sharma’s data. Again, the
data falls within the bounds and follows a similar trend as the
Tani data.

Looking at the graph in Figure 4-16, it can be seen that the upper and lower

bounds created by the FEA models encased the results of Sharma’s experiments.

In these experiments, Sharma et al. tested samples of PI/ZrW2 O8 composites with

and without surface treatments to the ZrW2 O8 in attempt to improve the bond

between the two materials. In the graph, the unmodified tests are represented by

52
the red bar and labelled Sharma Exp. and the samples with surface modifications are

represented by the orange bar and labelled Sharma Exp. Mod. While little difference

was seen between the two experiments, the modified tests did show a slight reduction

in expansion for the 5% and 10% samples but exhibited an increase in expansion

in the 15% samples. Despite these differences, the upper and lower bounds still

remained above and below all the results and the data followed a similar trend to

Tani’s, tending to be closer to the upper FEA bound. This means that there is room

to improve the bond between the filler and matrix if less overall expansion is desired.

For a better understand of Sharma’s data and the effect of the surface treatments on

the expansion of the composites, please see item [7] in the references section.

4.4 Comparison to Mathematical Models

The mathematical models discussed in Section 2.3 were developed to predict the

CTE of composite materials. These included the rule of mixtures, the Turner model,

and the Schapery upper and lower bounds and can be seen graphically in Figure 2-4.

Figures 4-17 and 4-18 show the same plot with the bounds created by the FEA for

Tani’s and Sharma’s data added. It can be seen that the line for the lower bound falls

between the Schapery limits. This makes sense as both of the experimental results

closely followed the Shapery upper limit and the FEA models showed that a lower

CTE was possible. The upper bound predicts the CTE to be higher than any of the

other models, climbing above the rule of mixtures and showing the greatest difference

between the lower bound at a 50% filler loading. This makes sense because with a high

filler volume fraction, there is more surface area that could or could not be bonded

to the matrix material leading to a better chance for variance in the results. After

50%, the difference starts to become smaller again as the filler becomes the dominant

material and less bond area is seen because of the reduction in matrix material.

53
Figure 4-17: Graph showing the upper and lower bounds of expansion cre-
ated by FEA models compared to Tani’s data.

Figure 4-18: Graph showing the upper and lower bounds of expansion cre-
ated by FEA models compared to Sharma’s data.

54
Chapter 5

Conclusions and Future Work

5.1 Conclusions

The main goal of this research was to determine the effect of filler geometry on

the finite element modeling of polymer/ZrW2 O8 composites. Through the research it

was determined that the filler geometry had little effect on the expansion results of

the FEA. All two and three-dimensional models of the same volume fraction showed

similar expansion over the 70 K temperature change despite drastic differences in

their geometry. It should be noted, however, that all of the three-dimensional models

showed slightly more expansion than all of the two-dimensional models. This means

that the two-dimensional models gave the filler material more effect, slightly reducing

the amount of expansion witnessed when compared to the three-dimensional models.

These results can be seen in Sections 4.1 and 4.2.

After determining the effect of filler geometry on the outcome of the models, the

results were compared to the experiments of Tani et al. where it was found that all the

models showed less expansion than the experimental data of the same volume fraction.

While the FEA results were close enough to rule out any errors in modeling, the

difference was significant and consistent enough that an explanation was required. It

55
was decided that the discrepancy was most likely caused by the interface between the

filler and matrix materials. In the FEA models, the bond between the two materials

was considered perfect, with no voids or separation. In real world experimentation,

a perfect bond is nearly impossible to achieve. During the creation of particulate

composites, voids can be created and separation between the two materials can occur,

all of which lead to the filler material having less effect on the overall properties. This

meant that the modeling completed represented the best possible case for reducing

the coefficient of thermal expansion in the composite. To gain a better understanding

of the experimental results, the opposite case needed to be modeled where there was

no bond between the filler and matrix material. Together, the results of the models

with a perfect bond and the results of the models with no bond created a lower and

upper bound for the possible expansion of the polymer/ZrW2 O8 composite. These

bounds were verified using the experimental results of Tani and Sharma et al. where

their results fell within the bounds at all volume fractions. These results can be seen

in Section 4.3.

5.2 Future Work

Research in the field of negative thermal expansion materials is headed in the

right direction. The experiments of Sharma et al. that look to improve the bond

between a polymer matrix and ZrW2 O8 filler are areas that need to be addressed

if researchers wish to create materials with lower coefficients of thermal expansion.

While, currently, the polymer/ZrW2 O8 composites created are able to reduce the

CTE of the polymer, this research has shown that further reduction could be seen

with a more efficient bond between the two materials.

The work conducted for this paper also leaves an opportunity for future research.

To start, the findings of this research could be verified through the recreation of the

56
models alongside real world experiments. This would ensure full confidence in the

research having gathered both the experimental data as well as completing the finite

element analysis. While completing the modeling, more volume fractions could be

modeled to add more data points to the upper and lower bounds created as in Section

4.3. The data used to create the bounds in Figure 4-17 were based on only three data

points and a best-fit line. These could be better represented if more volume fractions

were modeled.

Figure 5-1: Geometry showing a possible way to model the interface between
the matrix and filler. The effect of the ZrW2 O8 could be reduced
by adjusting the size and properties of the red layer surrounding
the filler.

Another possible area of future research would be to attempt to closely match

experimental results using FEA. The unmodified results of Sharma and Tani both

showed similar differences to the bounds created by the finite element models. This

means that if one could determine how to properly model the interface between the

two materials when standard composite production techniques are used, accurate

57
results could be seen using FEA. One way to do this could be to add a third domain

to the FEA models surrounding the filler particle as shown in Figure 5-1. This

would allow researchers to reduce the effect of the ZrW2 O8 by adjusting the material

properties and thickness given to this layer. If the correct size and properties are

found, it would provide researchers with a simple and repeatable process to predict

the CTE in these composites. It would also aid in the understanding of the interface

between the matrix and filler which could lead to improving their bond and further

reducing the CTE.

58
References

[1] C. Zweben, “Advances in composite materials for thermal management in elec-

tronic packaging,” JOM Journal of the Minerals, Metals and Materials Society,

vol. 50, no. 6, pp. 47–51, 1998.

[2] S. Numata, S. Oohara, K. Fujisaki, J. Imaizumi, and N. Kinjo, “Thermal ex-

pansion behavior of various aromatic polyimides,” Journal of Applied Polymer

Science, vol. 31, no. 1, pp. 101–110, 1986.

[3] M. Gasch, D. Ellerby, E. Irby, S. Beckman, M. Gusman, and S. Johnson, “Pro-

cessing, properties and arc jet oxidation of hafnium diboride/silicon carbide ul-

tra high temperature ceramics,” Journal of Materials Science, vol. 39, no. 19,

pp. 5925–5937, 2004.

[4] F. C. Nix and D. MacNair, “The thermal expansion of pure metals: Copper,

gold, aluminum, nickel, and iron,” Phys. Rev., vol. 60, pp. 597–605, Oct 1941.

[5] J. Tani, H. Kimura, K. Hirota, and H. Kido, “Thermal expansion and mechanical

properties of phenolic resin/ZrW2O8 composites,” Journal of Applied Polymer

Science, vol. 106, no. 5, pp. 3343–3347, 2007.

[6] M. Jakubinek, C. Whitman, and M. White, “Negative thermal expansion mate-

rials,” Journal of Thermal Analysis and Calorimetry, vol. 99, no. 1, pp. 165–172,

2010.

59
[7] R. Sharma et al., Polyimide nanocomposites based on cubic zirconium tungstate.

PhD thesis, The University of Toledo, 2010.

[8] L. Sullivan and C. Lukehart, “Zirconium tungstate (ZrW2O8)/polyimide

nanocomposites exhibiting reduced coefficient of thermal expansion,” Chemistry

of Materials, vol. 17, no. 8, pp. 2136–2141, 2005.

[9] R. Gibson, Principles of composite material mechanics, second edition. CRC

Press Taylor & Francis Group, 2007.

[10] B. Agarwal, L. Broutman, and K. Chandrashekhara, Analysis and performance

of fiber composites. Wiley, 2006.

[11] K. Chawla, Composite materials: science and engineering. Springer Verlag, 1998.

[12] P. Tipler and G. Mosca, Physics for scientists and engineers, vol. 1. Macmillan,

2008.

[13] C. Lind, M. Coleman, L. Kozy, and G. Sharma, “Zirconium tungstate/poly-

mer nanocomposites: Challenges and opportunities,” Physica Status Solidi (B),

vol. 248, pp. 123–129, 2011.

[14] T. Mary, J. Evans, T. Vogt, and A. Sleight, “Negative thermal expansion from

0.3 to 1050 kelvin in ZrW2O8,” Science, vol. 272, no. 5258, p. 90, 1996.

[15] P. Lommens, C. De Meyer, E. Bruneel, K. De Buysser, I. Van Driessche, and

S. Hoste, “Synthesis and thermal expansion of ZrO2/ZrW2O8 composites,” Jour-

nal of the European Ceramic Society, vol. 25, no. 16, pp. 3605–3610, 2005.

[16] J. Shigley, C. Mischke, and R. Budynas, Mechanical engineering design, seventh

edition. McGraw-Hill, 2003.

[17] A. Boresi, R. Schmidt, and O. Sidebottom, Advanced mechanics of materials,

sixth edition. Wiley, 2003.

60
[18] P. Forster and A. Sleight, “Negative thermal expansion in Y2W3O12,” Interna-

tional Journal of Inorganic Materials, vol. 1, no. 2, pp. 123–127, 1999.

[19] J. Li, A. Yokochi, T. Amos, and A. Sleight, “Strong negative thermal expansion

along the O-Cu-O linkage in CuScO2,” Chemistry of Materials, vol. 14, no. 6,

pp. 2602–2606, 2002.

[20] J. Evans and T. Mary, “Structural phase transitions and negative thermal ex-

pansion in Sc2(MoO4)3,” International Journal of Inorganic Materials, vol. 2,

no. 1, pp. 143–151, 2000.

[21] J. Shi, Z. Pu, K. Wu, and G. Larkins, “Composite materials with adjustable

thermal expansion for electronic applications,” in Materials Research Society

Symposium Proceedings, vol. 445, pp. 229–234, Materials Research Society, 1997.

[22] W. Miller, C. Smith, P. Dooling, A. Burgess, and K. Evans, “Tailored thermal

expansivity in particulate composites for thermal stress management,” Physica

Status Solidi (B), vol. 245, no. 3, pp. 552–556, 2008.

[23] R. Roy, D. Agrawal, and H. McKinstry, “Very low thermal expansion coefficient

materials,” Annual Review of Materials Science, vol. 19, no. 1, pp. 59–81, 1989.

[24] A. Sleight, “Thermal contraction,” Endeavour, vol. 19, no. 2, pp. 64–68, 1995.

[25] N. Marzari, D. Vanderbilt, A. De Vita, and M. Payne, “Thermal contraction

and disordering of the Al (110) surface,” Physical Review Letters, vol. 82, no. 16,

pp. 3296–3299, 1999.

[26] V. Korthuis, N. Khosrovani, A. Sleight, N. Roberts, R. Dupree, and W. Warren,

“Negative thermal expansion and phase transitions in the ZrV2-xPxO7 series,”

Chemistry of materials, vol. 7, no. 2, pp. 412–417, 1995.

61
[27] D. Taylor, “Thermal expansion data,” British Ceramic Transactions and Jour-

nal, vol. 84, no. 1, p. 9, 1985.

[28] R. Baughman and D. Galvão, “Crystalline networks with unusual predicted me-

chanical and thermal properties,” Nature, vol. 365, no. 6448, pp. 735–737, 1993.

[29] A. Sleight, “Isotropic negative thermal expansion,” Annual Review of Materials

Science, vol. 28, no. 1, pp. 29–43, 1998.

[30] C. Perottoni and J. Jornada, “Pressure-induced amorphization and negative ther-

mal expansion in ZrW2O8,” Science, vol. 280, no. 5365, p. 886, 1998.

[31] J. Graham, A. Wadsley, J. Weymouth, and L. Williams, “A new ternary oxide,

ZrW2O8,” Journal of The American Ceramic Society, vol. 42, no. 11, pp. 570–

570, 1959.

[32] L. Chang, M. Scroger, and B. Phillips, “Condensed phase relations in the systems

ZrO2-WO2-WO3 and HfO2-WO2-WO3,” Journal of The American Ceramic So-

ciety, vol. 50, no. 4, pp. 211–215, 1967.

[33] C. Lind, A. Wilkinson, Z. Hu, S. Short, and J. Jorgensen, “Synthesis and prop-

erties of the negative thermal expansion material cubic ZrMo2O8,” Chemistry of

Materials, vol. 10, no. 9, pp. 2335–2337, 1998.

[34] C. Georgi and H. Kern, “Preparation of zirconium tungstate (ZrW2O8) by the

amorphous citrate process,” Ceramics International, vol. 35, no. 2, pp. 755–762,

2009.

[35] E. Liang, T. Wu, B. Yuan, M. Chao, and W. Zhang, “Synthesis, microstructure

and phase control of zirconium tungstate with a CO2 laser,” Journal of Physics

D: Applied Physics, vol. 40, p. 3219, 2007.

62
[36] J. Chen, G. Huang, C. Hu, and J. Weng, “Synthesis of negative-thermal-

expansion ZrW2O8 substrates,” Scripta Materialia, vol. 49, no. 3, pp. 261–266,

2003.

[37] T. Osswald and G. Menges, Materials science of polymers for engineers. Hanser

Gardner Publications, 2003.

[38] O. Breuer and U. Sundararaj, “Big returns from small fibers: a review of poly-

mer/carbon nanotube composites,” Polymer Composites, vol. 25, no. 6, pp. 630–

645, 2004.

[39] J. Kim and D. Reneker, “Mechanical properties of composites using ultrafine

electrospun fibers,” Polymer Composites, vol. 20, no. 1, pp. 124–131, 1999.

[40] A. Balazs, T. Emrick, and T. Russell, “Nanoparticle polymer composites: where

two small worlds meet,” Science, vol. 314, no. 5802, p. 1107, 2006.

[41] B. Jang, “Advanced polymer composites: principles and applications,” ASM

International, Materials Park, OH 44073-0002, USA, 1994. 305, 1994.

[42] S. Ahmed and F. Jones, “A review of particulate reinforcement theories for

polymer composites,” Journal of Materials Science, vol. 25, no. 12, pp. 4933–

4942, 1990.

[43] G. Schwarz, “Thermal expansion of polymers from 4.2 K to room temperature,”

Cryogenics, vol. 28, no. 4, pp. 248–254, 1988.

[44] R. Boyer and R. Spencer, “Thermal expansion and second-order transition effects

in high polymers: Part I. experimental results,” Journal of Applied Physics,

vol. 15, no. 4, pp. 398–405, 1944.

[45] D. Krevelen, “Properties of polymers,” Properties of Polymers, 1990.

63
[46] G. Papanicolaou and P. Theocaris, “Thermal properties and volume fraction of

the boundary interphase in metal-filled epoxies,” Colloid & Polymer Science,

vol. 257, no. 3, pp. 239–246, 1979.

[47] Z. Hashin, “Thermoelastic properties of particulate composites with imperfect

interface,” Journal of the Mechanics and Physics of Solids, vol. 39, no. 6, pp. 745–

762, 1991.

[48] N. Lombardo, “Effect of an inhomogeneous interphase on the thermal expan-

sion coefficient of a particulate composite,” Composites Science and Technology,

vol. 65, no. 14, pp. 2118–2128, 2005.

[49] P. Theocaris, G. Papanicolaou, and G. Spathis, “Physical model for the thermal

expansion behaviour of fibre-reinforced viscoelastic composites,” Fibre Science

and Technology, vol. 15, no. 3, pp. 187–197, 1981.

[50] P. Turner, “Thermal-expansion stresses in reinforced plastics,” J. Res. NBS,

vol. 37, p. 239, 1946.

[51] R. Schapery, “Thermal expansion coefficients of composite materials based on

energy principles,” Journal of Composite Materials, vol. 2, no. 3, p. 380, 1968.

[52] H. Vo, M. Todd, F. Shi, A. Shapiro, and M. Edwards, “Towards model-based

engineering of underfill materials: CTE modeling,” Microelectronics Journal,

vol. 32, no. 4, pp. 331–338, 2001.

[53] M. Orrhede, R. Tolani, and K. Salama, “Elastic constants and thermal expansion

of aluminum-SiC metal-matrix composites,” Research in Nondestructive Evalu-

ation, vol. 8, no. 1, pp. 23–37, 1996.

[54] Z. Hashin and S. Shtrikman, “A variational approach to the theory of the elastic

64
behaviour of multiphase materials,” Journal of the Mechanics and Physics of

Solids, vol. 11, no. 2, pp. 127–140, 1963.

[55] L. Segerlind, Applied finite element analysis. Wiley, 1976.

[56] B. Agarwal and L. Broutman, “Three-dimensional finite element analysis of

spherical particle composites,” Fibre Science and Technology, vol. 7, no. 1,

pp. 63–77, 1974.

[57] P. Davy and F. Guild, “The distribution of interparticle distance and its applica-

tion in finite-element modelling of composite materials,” Proceedings of the Royal

Society of London. A. Mathematical and Physical Sciences, vol. 418, no. 1854,

p. 95, 1988.

[58] N. Chawla, V. Ganesh, and B. Wunsch, “Three-dimensional (3D) microstruc-

ture visualization and finite element modeling of the mechanical behavior of

sic particle reinforced aluminum composites,” Scripta Materialia, vol. 51, no. 2,

pp. 161–165, 2004.

[59] C. Nan and D. Clarke, “The influence of particle size and particle fracture on

the elastic/plastic deformation of metal matrix composites,” Acta Materialia,

vol. 44, no. 9, pp. 3801–3811, 1996.

[60] H. Shen and C. Lissenden, “3D finite element analysis of particle-reinforced alu-

minum,” Materials Science and Engineering A, vol. 338, no. 1-2, pp. 271–281,

2002.

[61] S. Qin, C. Chen, G. Zhang, W. Wang, and Z. Wang, “The effect of particle shape

on ductility of SiCp reinforced 6061 Al matrix composites,” Materials Science

and Engineering A, vol. 272, no. 2, pp. 363–370, 1999.

65
[62] S. Yilmaz and D. Dunand, “Finite-element analysis of thermal expansion and

thermal mismatch stresses in a Cu-60vol% ZrW2O8 composite,” Composites Sci-

ence and Technology, vol. 64, no. 12, pp. 1895–1898, 2004.

66
Appendix A

MATLAB Codes

A.1 MATLAB Code for Creating Random Coor-

dinates for Spherical Inclusions

%Make random s p h e r e s w i t h i n a bar and p l o t

clc

clear a l l

%Volume f r a c t i o n ( p e r c e n t )

Volumefraction = 7;

%Volume f r a c t i o n ( d e c i m a l )

vf = Volumefraction /100;

%Dimensions o f bar

l = 20;

w = 4;

67
h = 4;

Vbar = l ∗w∗h ;

%Numer o f s p h e r e s

q = 50;

%Radius o f s p h e r e s

r = ( ( 3 ∗ v f ∗( Vbar ) ) /(4∗ q∗ pi ) ) ˆ ( 1 / 3 ) ;

%Create i n i t i a l random c o o r d i n a t e s

x1 = ( rand ∗(w−2∗ r ) )+r ;

y1 = ( rand ∗( h−2∗ r ) )+r ;

z1 = ( rand ∗( l −2∗ r ) )+r ;

%Create non−o v e r l a p p i n g c o o r d i n a t e s

m = 1;

t = 0;

while m < q

i f t >1e6

error ( ’ Could not compute enough non−o v e r l a p p i n g

points . Process aborted . ’ )

break

end

Xm = ( rand ∗(w−2∗ r ) )+r ;

Ym = ( rand ∗( h−2∗ r ) )+r ;

Zm = ( rand ∗( l −2∗ r ) )+r ;

C1 = ( x1−Xm) .ˆ2+( y1−Ym) .ˆ2+( z1−Zm) . ˆ 2 ;

68
i f any (C1<4∗( r ˆ 2) )

t = t +1;

continue

end

m = m+1;

x1 = [ x1 ;Xm] ;

y1 = [ y1 ;Ym] ;

z1 = [ z1 ;Zm ] ;

end

%Output s p h e r e c o o r d i n a t e s

[ x1 y1 z1 ]

%P l o t

[ xs , ys , z s ] = sphere ( 3 0 ) ;

for n = 1 :m

surf ( xs ∗ r+x1 ( n ) , ys ∗ r+y1 ( n ) , z s ∗ r+z1 ( n ) , ’ FaceColor ’ , ’ g ’ , ’

EdgeColor ’ , ’ r ’ )

hold on

camlight ( ’ l e f t ’ )

material dull

end

axis e q u a l

xlim ( [ 0 w ] )

ylim ( [ 0 h ] )

zlim ( [ 0 l ] )

hold o f f

69
A.2 MATLAB Codes for Creating a Mapped Mesh

and Randomly Assigning Material Properties

A.2.1 Code for Developing Random Filler Elements in Mapped

Mesh

%S e l e c t random Elements w i t h i n a bar and p l o t

clc

clear a l l

%S e t t h e s t a t e t o t h e c l o c k t o e n s u r e random

rand ( ’ s t a t e ’ , sum(100∗ clock ) ) ;

%Volume f r a c t i o n ( p e r c e n t ) ( a c t u a l w i l l be l o w e r )

Volumefraction = 7 . 2 3 ;

%Volume f r a c t i o n ( d e c i m a l )

vf = Volumefraction /100;

%Enter t h e d i m e n s i o n s o f t h e bar

xd = 4 ;

yd = 2 0 ;

zd = 4 ;

%Enter number o f e l e m e n t s i n mapped mesh i n x , y , and z

directions

70
xn = 1 0 ;

yn = 5 0 ;

zn = 1 0 ;

t o t a l e l e m e n t s = xn∗yn∗ zn ;

%C a l c u l a t e e l e m e n t s i z e

xe = xd/xn ;

ye = yd/yn ;

z e = zd / zn ;

%C a l c u l a t e number o f p o i n t s needed t o meet volume f r a c t i o n

N = c e i l ( v f ∗xn∗yn∗ zn ) ;

%Create i n i t i a l random c o o r d i n a t e s

x1 = c e i l ( xn . ∗ rand (N, 1 ) ) ;

y1 = c e i l ( yn . ∗ rand (N, 1 ) ) ;

z1 = c e i l ( zn . ∗ rand (N, 1 ) ) ;

%S h i f t p o i n t s t o m i d p o i n t o f e l e m e n t s

x1 = x1∗xe−xe / 2 ;

y1 = y1∗ye−ye / 2 ;

z1 = z1 ∗ ze−z e / 2 ;

%Check i f u n i q u e and o u t p u t a c t u a l volume f r a c t i o n

A = [ x1 , y1 , z1 ] ;

B = unique (A, ’ rows ’ ) ;

SizeB = s i z e (B, 1 ) ;

71
SizeA = s i z e (A, 1 ) ;

ActualVf = ( SizeB / t o t a l e l e m e n t s ) ∗100

%S e l e c t o n l y u n i q u e p o i n t s

x2 = B ( : , 1 ) ;

y2 = B ( : , 2 ) ;

z2 = B ( : , 3 ) ;

N2 = SizeB ;

%P l o t

s c a t t e r 3 ( x2 , y2 , z2 , 1 )

axis e q u a l

xlim ( [ 0 xd ] )

ylim ( [ 0 yd ] )

z l i m ( [ 0 zd ] )

%Save o u t p u t

save x2

save y2

save z2

save xe

save ye

save z e

save N

72
A.2.2 Code for Assigning CTE Properties

function Alpha = a l p h a d a t a e l e m e n t s ( x , y , z )

%I n p u t m a t r i x and f i l l e r p r o p e r t y v a l u e s

a l p h a m a t r i x = 1 5 . 2 4 e −6;

a l p h a f i l l e r = −9e −6;

%Load c o o r d i n a t e f i l e s

load ( ’N. mat ’ ) ;

load ( ’ x2 . mat ’ ) ;

load ( ’ y2 . mat ’ ) ;

load ( ’ z2 . mat ’ ) ;

load ( ’ xe . mat ’ ) ;

load ( ’ ye . mat ’ ) ;

load ( ’ z e . mat ’ ) ;

[m, p ] = s i z e ( x ) ;

%A ssign p r o p e r t i e s t o each e l e m e n t

a l p h a l o g i c = f a l s e (m, p ) ;

for j = 1 : length ( x2 )

a l p h a l o g i c = a l p h a l o g i c | ( abs ( x−x2 ( j ) )<=xe /2 & abs ( y−y2 (

j ) )<=ye /2 & abs ( z−z2 ( j ) )<=z e / 2) ;

end

Alpha = a l p h a l o g i c ∗ a l p h a f i l l e r + ˜ a l p h a l o g i c ∗ a l p h a m a t r i x ;

73
A.2.3 Code for Assigning Elastic Modulus Properties

function E = e d a t a e l e m e n t s ( x , y , z )

%I n p u t m a t r i x and f i l l e r p r o p e r t y v a l u e s

e m a t r i x = 3 . 5 1 e9 ;

e f i l l e r = 8 8 . 3 e9 ;

%Load c o o r d i n a t e f i l e s

load ( ’N. mat ’ ) ;

load ( ’ x2 . mat ’ ) ;

load ( ’ y2 . mat ’ ) ;

load ( ’ z2 . mat ’ ) ;

load ( ’ xe . mat ’ ) ;

load ( ’ ye . mat ’ ) ;

load ( ’ z e . mat ’ ) ;

[m, p ] = s i z e ( x ) ;

%A ssign p r o p e r t i e s t o each e l e m e n t

e l o g i c = f a l s e (m, p ) ;

for j = 1 : length ( x2 )

e l o g i c = e l o g i c | ( abs ( x−x2 ( j ) )<=xe /2 & abs ( y−y2 ( j ) )<=ye

/2 & abs ( z−z2 ( j ) )<=z e / 2) ;

end

E = e l o g i c ∗ e f i l l e r + ˜ e l o g i c ∗ ematrix ;

74
A.2.4 Code for Assigning Poisson’s Ratio

function Nu = n u d a t a e l e m e n t s ( x , y , z )

%I n p u t m a t r i x and f i l l e r p r o p e r t y v a l u e s

numatrix = 0 . 3 7 ;

nufiller = 0.33;

%Load c o o r d i n a t e f i l e s

load ( ’N. mat ’ ) ;

load ( ’ x2 . mat ’ ) ;

load ( ’ y2 . mat ’ ) ;

load ( ’ z2 . mat ’ ) ;

load ( ’ xe . mat ’ ) ;

load ( ’ ye . mat ’ ) ;

load ( ’ z e . mat ’ ) ;

[m, p ] = s i z e ( x ) ;

%A ssign p r o p e r t i e s t o each e l e m e n t

n u l o g i c = f a l s e (m, p ) ;

for j = 1 : length ( x2 )

n u l o g i c = n u l o g i c | ( abs ( x−x2 ( j ) )<=xe /2 & abs ( y−y2 ( j ) )<=

ye /2 & abs ( z−z2 ( j ) )<=z e / 2) ;

end

Nu = n u l o g i c ∗ n u f i l l e r + ˜ n u l o g i c ∗ numatrix ;

75
A.2.5 Code for Assigning Density Properties

function Rho = r h o d a t a e l e m e n t s ( x , y , z )

%I n p u t m a t r i x and f i l l e r p r o p e r t y v a l u e s

rhomatrix = 1370;

r h o f i l l e r = 5090;

%Load c o o r d i n a t e f i l e s

load ( ’N. mat ’ ) ;

load ( ’ x2 . mat ’ ) ;

load ( ’ y2 . mat ’ ) ;

load ( ’ z2 . mat ’ ) ;

load ( ’ xe . mat ’ ) ;

load ( ’ ye . mat ’ ) ;

load ( ’ z e . mat ’ ) ;

[m, p ] = s i z e ( x ) ;

%A ssign p r o p e r t i e s t o each e l e m e n t

r h o l o g i c = f a l s e (m, p ) ;

for j = 1 : length ( x2 )

r h o l o g i c = r h o l o g i c | ( abs ( x−x2 ( j ) )<=xe /2 & abs ( y−y2 ( j ) )

<=ye /2 & abs ( z−z2 ( j ) )<=z e / 2) ;

end

Rho = r h o l o g i c ∗ r h o f i l l e r + ˜ r h o l o g i c ∗ r h o m a t r i x ;

76

Vous aimerez peut-être aussi