Spring 2003

Semester Project Report

Towards Finite Element-Based Thermal
Topological Design of Unit Cells for
Linear Cellular Alloys

















ANSYS
MATLAB

prepared for
ME 6124
The Finite Element Method
Dr. Suresh Sitaraman

By

Hae-Jin Choi
Marco Gero Fernndez

ME 6124 HJC & MGF
Tableof Contents

1. Overview 2
2. Objectives 2
3. Background Information 3
3.1 Linear Cellular Alloys 3
3.2 Topology Optimization 7
4. Purpose of this Investigation 8
5. Case Study Convectively Cooled Heat Sink for a Computer Chip 9
6. Modeling Details 11
6.1 Simplifying Assumptions 11
6.2 Finite Element Model for Heat Transfer Analysis 12
6.3 Modeling the LCAs in MATLAB

18
6.4 Modeling the LCAs in ANSYS

20
7. Results, Discussion, and Validation/Verification 21
7.1 The 2 x 2 LCA Design (MATLAB vs. ANSYS) 21
7.2 The 8 x 8 LCA Design (MATLAB vs. Finite Difference) 25
7.3 The 1 x 1 LCA Design (MATLAB vs. Hand Calculations) 27
8. Closure 27
9. Problems Encounteredand Learning 28
10. References 29
11. Appendices 30
11.1 Appendix A Derivation of Finite Element used for MATLAB

Heat
Transfer Analysis
11.2 Appendix B Hand Calculation for a 1 x 1 Cell
11.2 Appendix C Final Presentation Slides










1
ME 6124 HJC & MGF
1. Overview

The objective of this project was to facilitate the design of Linear Cellular Alloys
(LCAs), as developed by the Lightweight Structures Group at Georgia Tech, through
robust topology design techniques (i.e., a method combing topology optimization and
robust design that is currently being developed in the Systems Realization Laboratory,
also at Georgia Tech). Since topology design is very computationally expensive and
highly dependent on the ability to explore a given design space effectively, analysis
routines, used to judge the performance of the structures changing morphology are
critical. The primary obstacle to effective and efficient design space exploration is the
speed with which the algorithms used to conduct the analysis can be invoked. This is turn
is highly dependent on the ability to interface different software tools successfully and
update changing inputs/outputs automatically. Previous implementations have relied on
the coupled use of iSIGHT from Engineous Software for optimization and ANSYS

from
ANSYS Incorporated for analysis. Unfortunately, passing any information among
heterogeneous and potentially distributed software tools, though novel, introduces a
significant lag and greatly increases the amount of time required per iteration.
Considering that optimization can necessitate on the order of hundreds of runs, such
inherent lags can be detrimental.
In order to address this need for quick and efficient design space exploration, a 99
line MATLAB

code developed by Ole Sigmund from the Department of Solid

Mechanics at the Technical University of Denmark [1] has been modified for topology
design and analysis based on structural considerations. Since a key strength of LCAs is
their performance with regard to multifunctional criteria (i.e., structural and thermal), this
is not sufficient. The intent of this report is to provide an overview of the implementation
of thermal Finite Element Analysis in MATLAB

, as required for the expedition of

multi-objective topology design. The example considered here is an air-cooled heat sink
for a computer chip, as introduced in Section 5.

2. Objectives

The objectives addressed in this project are as follows:
1) Develop thermal analysis for LCAs that interfaces with existing structural
analysis routines in MATLAB

where topologies are modeled through the use of

two node frame elements with four degrees of freedom per node (i.e., 2
displacements, 1 rotational, and 1 temperature).
2) Develop a Finite Element for modeling convection that is compatible with the
frame elements currently employed for structural optimization.
3) Consider laminar flow for modeling convectively cooled heat sinks and
potentially extend to turbulent flow for modeling combustor liners.
4) Compare and contrast results obtained using the developed MATLAB

algorithms with results obtained using ANSYS

and via hand calculations

Deliverables include:
MATLAB

Thermal Analysis Code

Analysis of sample heat sink designs using the developed code
Validation of MATLAB

analysis code using ANSYS

2
ME 6124 HJC & MGF
3. Background Information

3.1 Topology Design

Topology design involves simultaneously adjusting both the external shape and the
number and shape of internal boundaries for a given 2D or 3D domain and associated
boundary conditions and design objectives [2,3]. Using topological design techniques,
vastly different topologies can be obtained from an arbitrary, initial domain. By
adjusting the topology of a structure, important properties like compliance, stiffness,
strength, eigenfrequencies, convective coefficients, and other properties sensitive to
material arrangement can be tailored. It is possible to distribute material strategically,
resulting in lightweight structures with desirable properties. Increasingly, manufacturing
processes, like additive fabrication and processing of cellular materials, are emerging that
facilitate the fabrication of structures with nearly arbitrary topologies. Thus, topological
design is a timely topic.
Topology Design, as implemented here, is defined as the sum total of methods used to
address the question: How can material be distributed efficiently in a given design region
to tailor properties that are sensitive to material distribution (e.g., compliance, stiffness,
strength, convection, etc.)? It is important to note that nothing is known about structure
or shape a priori in topology optimization. In fact, it is during the course of topology
optimization that the shape and number of discontinuities (i.e., voids) are determined. A
typical topology design approach, as proposed by Carolyn Conner Seepersad in her PhD
proposal involves the following steps:

Step1 - Establish design requirements, objectives, and domain. This step was completed
in the previous discussion.
Step 2 - Divide domain into finite elements. In Figure 1, the design domain (in this case
a simply supported cantilever beam with a load applied at one end) is divided into 2-D
planar finite elements.
Step 3 - Assign density variable to each finite element (
i
).
Step 4 - Modify density variables according to solution (optimization) algorithm. Small
density values for an element imply that the element is empty (i.e., part of a hole). Large
density values imply solid material.
Step 5 - Calculate effective properties of structure.
A. Select penalization power, p>3. The penalization power penalizes
intermediate densities and encourages convergence to regions of solid (full
density) and void (minimum density).
B. Calculate effective properties in each element. For example, a stiffness
matrix (K) for an element becomes: K
i
=
i
p
K
solid

C. Calculate effective properties for the structure.
Step 6 - Return to Step 4 until convergence is achieved.
3
ME 6124 HJC & MGF
The computational model for topology design used in this investigation stems from a
99 line MATLAB

code for compliance minimization of statically loaded structures,

developed by Ole Sigmund from the Department of Solid Mechanics at the Technical
University of Denmark [1]. The code was intended for engineering education and
contains both a mesh independency filter and a finite element code. It is based on a
number of simplifying assumptions aimed at reducing code complexity. For example, the
design domain is modeled as a rectangle and is discretized using square finite elements,
as indicated in Figure 1. Element and node numbering proceeds on a column-by-column
basis, starting in the upper left corner. The aspect ratio of the structure to be optimized is
determined by the number of horizontal (nelx) and vertical (nely) elements as specified
by the user.
The chosen implementation of topology optimization within this algorithm is based
on the power law approach or SIMP approach (Solid Isotropic Material with
Penalization), where properties are assumed constant within each element and design
variables are the element relative densities. The objective here is to minimize
compliance. The basic structure of a topology optimization problem formulated as such
is the following:
Finite
Element
Mesh
Finite
Element
Mesh

1

2

mxn

1

2

mxn
Figure 1 Dividing the Cantilever Beam Design Domain into
Finite Elements (Courtesy of CCS)
Minimize:
1
( ) ( ) U KU u k u
N
T p
e e o
e
c x x
=
= =

T
e
Subject to:
( )
o
V x
f
V
=
KU = F

min
0 1 x x <

U and F are the global displacement and force vectors, respectively, and K is the
global stiffness matrix; u
e
and k
e
correspond to the element; x
e
is the vector of design
variables bounded with respect to relative density by x
min
(non-zero to avoid singularity)
and x
max
; N = m x n is the number of elements within the discretized design space; p is
the penalization power; V(x) is the material volume; V
o
is the design domain volume; f is
the prescribed volume fraction.
4
ME 6124 HJC & MGF
Mesh-Independency Filtering
The purpose in including a mesh-independency filter within the topology
optimization algorithm is to avoid disjointed solid regions (i.e., chess/checker board
patterns). From a practical consideration, applying such a filter improves the likelihood
of manufacturability and can be used to control the minimum feature size. The parameter
used to controll the degree of filtering is r
min
. In filtering the mesh for
disjointed/unconnected elements an area equal to a square with side lengths of two times
r
min
around each element is searched. Setting
min
r 1 < essentially deactivates the filter.

Finite Element Code
The finite element code implemented within this MATLAB

routine takes advantage

of the sparse matrix capability within MATLAB

and as such is quite efficient, avoiding

what would otherwise be a significant amount of computational overhead associated with
required matrix inversion calls. Elements are modeled as four node quadrilateral
elements having a total of two degrees of freedom (horizontal and vertical displacement)
per node (i.e., 8 dof per element). Supports are modeled through the elimination of fixed
degrees of freedom from the linear equations. Elemental stiffness matrices are calculated
based on an analytic 8 x 8 matrix for a square bi-linear 4-node element. The user has the
ability to adjust both the Youngs Modulus E and the Poissons Ratio .
Note: For more information please refer to [1].

In order to make this code more useful for multi-objective topology optimization a
number of modifications were made, the most important among them being (1) the
implementation of frame elements with four degrees of freedom per node (i.e., 2
displacements, 1 rotational, and 1 temperature) in lieu of the more limited square cell
finite elements originally included and (2) the incorporation of a more powerful and
consistent optimization scheme. While the first was motivated by the need for increased
flexibility in modeling topology, the second requires a more detailed explanation.
Topology Optimization for the design of Linear Cellular Alloys (LCAs) is a highly
non-linear optimization problem. It involves either multiple nonlinear objectives or
multiple non-linear constraints, and very large numbers of continuous design variables.
The role of optimization algorithms in topology design is to aid in converging on a
particular solution, or rather, driving the process of topology optimization towards
achieving a structure that closely represents the posed requirements without violating any
constraints. As stated, topology design is characterized by high degrees of non-linearity
that involve either multiple nonlinear objectives or multiple non-linear constraints, and
very large numbers of continuous design variables. A first step towards effective
structural synthesis is the determination of stable optimization routines, capable of
effectively handling large numbers of variables and yielding consistent results. Previous
research was focused on comparing a number of different gradient-based algorithms with
regard to performance and stability, among them the Method of Moving Asymptotes
(MMA) [4,5], Sequential Quadratic Programming (SQP) [6,7], and Sequential Linear
Programming (SLP) [8,9]. The clear indication of the study was that MMA had the most
consistent performance among the three and the most favorable convergence behavior.
Developed by Krister Svanberg, this non-linear programming method focuses on an
iterative process in which strictly convex approximations in the form of sub problems are
5
ME 6124 HJC & MGF
generated and solved [10,11]. The generation of these approximations in the form of sub
problems is accomplished through the adjustment of asymptotes in a manner that can
both stabilize and speed up the convergence of the more general process. MMA was
developed explicitly for use in structural optimization and is based on a special type of
convex approximation. Consequently, MMA is capable of handling (1) general
requirements of non-linear programming problems (e.g., the ability to handle all types of
constraints, provided only that derivatives of constraint functions can be calculated either
analytically or numerically, etc.), as well as (2) requirements that are particular to
structural optimization (e.g., costly function evaluations and large numbers of variables).
Finally, MMA is designed to remain stable, capable of generating a sequence of
improved feasible, or quasi-feasible solutions, and be implementable with considerable
ease. The general form of the non-linear programming problem addressed in an MMA
problem is

Minimize:
2
1
( ) ( 0.5 )
n
o o i i i
i
i
f x a z c y d y
=
+ + +

Subject to: ( ) 0
i i i
f x a z y , i m 1,..., =

min max
j j j
x x x , 1,..., j n =
, y , i m 0 z 0
i
1,..., =

Thus, since MMA was developed explicitly for the type of problem under
consideration it is ideally suited for the dynamic and highly computationally intensive
nature of topology optimization. Gradient information is required. Although numerical
approximations are often easier to obtain than analytical expressions they are quite costly
and can significantly reduce algorithmic performance, especially considering the large
number of variables that lies at the core of topology optimization. In this particular
application analytical expressions for required gradients were implemented in order to
minimize the adverse impact of numerical approximations on computation time. Another
benefit of MMA is the inherent ability to influence convergence. Moving the asymptotes
away from the current iteration point, for example, speeds up convergence for monotone
or slow progression. Moving the asymptotes closer to the current iteration point, on the
other hand, has the effect of stabilizing the process in the case of unnecessary
oscillations. This is a key advantage in a process with the objective of distinguishing as
clearly as possible between solid and void regions (i.e., treating a variable that is modeled
continuously, despite being Boolean by nature) and stands in marked contrast to the
tendency of SLP, for example, to push the values of parameters to either extremum (i.e.,
due to linearization of objective function and constraints.

6
ME 6124 HJC & MGF
3.2 Linear Cellular Alloys

Linear Cellular Alloys (see Figure 2) are metallic cellular materials with a constant
cross section, fabricated through a process developed by the Lightweight Structures
Group at Georgia Tech. The process combines extrusion of ceramic slurry, composed of
metal oxides and water through a die, allowing for the achievement of quasi-arbitrary
two-dimensional cellular topologies. Extrusion of the ceramic is followed by exposure to
thermal and chemical treatments that cure the composites. The inherent advantage in
producing materials using this process is the ability to tailor properties of the resulting
structure such as the effective moduli of elasticity and conductivity by altering the
topologies of the cells. Structures may be composed of either periodically repeating unit
cells or functionally graded, non-uniform cells of various topologies. LCAs are
particularly suitable for multifunctional applications that require not only structural
performance but also lightweight thermal or energy absorption capabilities
Accordingly, LCAs have potential for use in applications such as actively cooled
supersonic aircraft skins or engine combustor liners [12].
Figure 2 -- Square-Cell Linear Cellular Alloy [13]











7
ME 6124 HJC & MGF
4. Purpose of this Investigation

Typically, topology design and optimization involve the general steps outlined in
Figure 3. As indicated, every change in geometry requires renewed analysis to evaluate
system performance with regard to desired objectives. Considering that such changes in
geometry also require the recalculation of temperature dependent (i.e., inlet, outlet, and
bulk) properties such as fluid viscosity , convective coefficient h, Prandtl Number Pr,
Reynolds Number Re, Hydraulic Diameter D
h
, etc. and the reevaluation of potentially
huge stiffness matrices computational expense is considerable. This is especially true
when a number of different software applications are involved.
The overarching objective in this project is to investigate the potential of conducting
optimization alongside analysis exclusively within MATLAB

, in order to avoid the

computational overhead associated with relying on distributed computing. Relying on
iSIGHT, for example, results in significant delays due to reading and writing numbers
from/to input/output files, that are amplified further by the large number of design
variables required for topology optimization). Porting FEA analysis required for the
topology optimization routine from ANSYS, as used in conjunction with iSIGHT, to
MATLAB

native code eliminates the required use of iSIGHT to wrap the

optimization procedure. Having self-standing code has the additional advantage of
reducing the time required for completing optimization runs by an order of magnitude
or more. This task is accomplished through an adaptation of the 99 line topology
optimization algorithm, developed by Ole Sigmund and extended by Carolyn Conner
Seepersad, as described in Section 3.1.

Discretize Domain
and ID Appropriate
Variables,
Objectives, and
B.C.s.
Assign Element
Material Properties
Build and Solve
Finite Element
Model for Specific
Boundary
Conditions
Calculate Values
of Objectives
Based on FE
Results
Update Element
Material Properties
to Improve
Objective Function
Figure 3 General Steps for Topology Design and Optimization (Courtesy of CCS)
This investigation is concerned with developing, validating, and smoothly integrating
thermal analysis capabilities into the modified MATLAB

code in order to make quick

exploration of multi-objective design spaces possible. With this in mind, a motivating
example, necessitating the consideration of multiple objectives is introduced in the next
section.

8
ME 6124 HJC & MGF
5. Case Study Convectively Cooled Heat Sink for a Computer Chip

The example LCA application considered in this report is that of a convectively
cooled heat sink for a computer chip. A sample schematic of the structure is given in
Figure 4. The general requirements for a CPU heat sink are that it 1) remove enough heat
from the chip so as to ensure steady state operation and 2) withstand the relatively high
compressive forces exerted by clamps used to attached the heat sink to the chip as tightly
as possible, see Figure 5. With this in mind, it is important to note that constant
temperature at the chip interface is assumed in this investigation. Although, it may seem
more intuitive to model constant heat flux instead, the idea is to design a heat sink that is
capable of removing enough heat to keep the chip below 1) its maximum operating
temperature or 2) its melting temperature (in the case of potential over-clocking)

Figure 4 Compact, Forced Convection Heat
Exchanger with Graded Rectangular
LCAs [12]
W
H
D
w
1
w
2
w
3
w
Nh
. . .
h
2
h
1
h
Nv
t
h
t
v
Heat
Source
T
source
Air Flow, T
in
.
.
.
x
y
z
Figure 5 Steps Involved in CPU/Heat
Sink Assembly
The basic structure of the problem being addressed in this report (see Figure 4) is the
following :

Given:
1) Overall dimensions of the structure (width (W) x depth (D) x height (H)).
2) A potential LCA heat sink design, as defined by the number of cells in the
horizontal and vertical directions (i.e., N
h
and N
v
), respective Cell Heights (C
h
)
and Cell Widths (C
w
), Wall Thicknesses in the x-direction (W
t-x
), and Wall
Thicknesses in the y-direction (W
t-y
).
3) A heat source (i.e., the CPU), maintained constant temperature T
s
, tightly clamped
to the top face of the heat sink.
4) Insulated boundaries (i.e., zero heat flux) at the right, bottom, and right side faces.
5) Air flow rate M

and initial bulk temperature (T

bulk
), T
in
.
6) Conditions are representative of fully developed laminar flow.

Determine:
1) The steady state temperature distribution in the structure
2) The overall rate of steady state heat transfer Q
total

9
ME 6124 HJC & MGF
The values of design variables, used in this investigation are as follows:


W = 25 mm
D = 75 mm
H = 25 mm
N
h
= varies
N
v
= varies
C
h
= varies
C
v
= varies
W
t-x
= varies
W
t-y
= varies


350
s
T K =
300
in
T K =
kg
M varies
s
=

W
363
m-K
s
k =










The boundary conditions and coordinate system used for the thermal finite element
analysis are provideD in Figure 6.


Z
X
Y
Figure 6 - FEA Boundary Conditions


10
ME 6124 HJC & MGF
6. Modeling Details

6.1. Simplifying Assumptions
In implementing thermal analysis for multi-objective topology design, a number of
simplifying assumptions were made. Some were a direct consequence of the requirement
to interface with the existing MATLAB

code. Others resulted from modeling

considerations. Each is explained in detail as follows:

Type of Finite Element Used
In order to add modeling/analysis capability of the two-dimensional cross section
(determined as a result of topology design with regard to structural considerations) in the
third dimension, a fundamental requirement was the reliance on previously determined
information. Since structure in topology design is determined through the successive
determination of solid and void areas, the most important piece of information is the
status (i.e., activity/inactivity) of nodes.
The nodes in question comprise the endpoints of frame elements that determine the
constant cross-section of the LCAs. Each node has four degrees of freedom (i.e., 2
displacements, 1 rotational, and 1 temperature). The information derived from the
structural considerations is a two dimensional grid of active nodes, as determined by the
active frame elements. The challenge now becomes effectively adding a third dimension,
used to analyze the convective properties of the structure. Since the cross section is
modeled in terms of one-dimensional elements, this translates to modeling thermal
characteristics in the third dimension using two-dimensional elements. Although a three
dimensional element might be better suited to model the thermal behavior of a three-
dimensional structure, this would require the specification of two-dimensional elements
for modeling the cross section. This is indicated in Figure 11. In essence, the element
developed in Section 6.2 is a plate element, capturing two-dimensional geometry and
variation of DOFs and having constant behavior in the third.

Modeling Considerations
Modeling three-dimensional geometry in terms of a combination of one-dimensional
and two-dimensional elements has several consequences, the most significant of which
can be explained by elaborating on the nature of plate elements. As plate elements are
two-dimensional analogues to beams they do not capture all the effects associated with
the three dimensional geometries they are meant to model. In our case, conduction
throughout the thickness of the plate element was not modeled explicitly. In order to
differentiate between elements of different thickness the thermal conduction coefficient
on an area basis was multiplied by the thickness of the structure.
In order to model the three-dimensional structures effectively, several other
assumptions were made.
Negligible pressure drop at the inlet
Negligible heat conduction along the direction of extrusion
Constant surface temperature, equal to average duct wall temperature (because
temperature difference of adjacent duct walls is very small)
Fluid temperature difference between inlet and outlet is very small

11
ME 6124 HJC & MGF
6.2 Finite Element Model for Heat Transfer Analysis

In this section, the linear (four node) rectangular element, employed to model the
thermal characteristics of LCAs along the direction of extrusion (in MATLAB), is
presented in light of the simplifying assumptions discussed in Section 7.1. An overview
of the elements derivation is provided alongside governing equations, stiffness matrices,
convective coefficients, etc. The reader is referred to Appendix A for mathematical
details.
Thermal analysis as implemented in MATLAB consists of two parts. The first is
aimed at modeling the thermal behavior of the structural material, the second at modeling
the changing properties of the fluid flowing through structural voids. The finite element
approach employed in this investigation (and mapped out in Figure 7) is somewhat
unique in so far that sections of the structure, rather than the structure as a whole, are
analyzed one at a time. This is akin to the way in which finite difference codes work.
The inherent advantage is that there is no need to calculate immense stiffness matrices,
reducing computational complexity. Additionally, this procedure facilitates the
determination of changing properties of the convective fluid and the incorporation of
these to the thermal analysis of the structure. In essence, this is what a coupled
ANSYS/FLOTRAN analysis accomplishes.

Fluid
Rectangular
Duct
Fluid
Rectangular
Duct
W
H
D
w
1
w
2
w
3
w
Nh
. . .
h
2
h
1
h
Nv
t
h
t
v
Heat
Source
T
source
Air Flow, T
in
.
.
.
x
y
z
Plane element
In FEA
Fluid Cell
Element
Plane element
In FEA
Fluid Cell
Element
Figure 7 - Modeling Approach for Thermal Analysis of LCA Performance
With this in mind, a brief overview of the linear (four node) rectangular and fluid
elements, developed for the purposes of this investigation, follows.




12
ME 6124 HJC & MGF

1 2
3 4
a
b
t t
Linear (Four Node) Rectangular Finite
Element
Linear (Four Node) Rectangular Finite
Element
The finite element used for LCA thermal
analysis is shown in Figure 8. As indicated,
geometrically, the element is defined by length
(a) and width (b). Each of the four nodes (1
The finite element used for LCA thermal
analysis is shown in Figure 8. As indicated,
geometrically, the element is defined by length
(a) and width (b). Each of the four nodes (1, 2,
3, and 4) has one degree of freedom
temperature. The element also takes into
consideration internal heat generation as
indicated by the black arrows, as well as
convection effects on either face. This is
indicated by the light blue arrows pointing away
from either face in Figure 8. Figure 8 - Linear (Four Node)
Rectangular Thermal Elemen
The governing equation for steady state heat
transfer in plane systems is given by

( ) ,
x
T
f x y k k
y
T
x x y y
| | |
=
|

\ .
\ .
|
|
[1]

where T is the temperature (in
o
K), k
x
and k
y
are the thermal conductivities of the
material (in W m
-1

o
K
-1
) along the x and y directions respectively, and f is the internal
heat generation per unit volume (in W m
-3
). For convective boundary conditions, the
natural boundary conditions are a balance of energy transfer across the boundary due to
conduction and/or convection (i.e., Newtons Law of Cooling) [14]:


( )

x x y y
T T
k n k n T T
x x



+ +

n
q = [2]

where b is the convective conductance (or the convective heat transfer coefficient) (in
W m
-2

o
K
-1
), T

is the ambient temperature of the (ambient) surrounding fluid, and q is

the specified heat flow. The first two terms account for heat transfer by conduction, the
third for heat transfer by convection; the term on the right hand side accounts for the
specified heat flux, if any [14].

n

The system illustrated in Figure 4 is thus described by the following relationship:

{ } { } { }
e e e e
K H T F P ( + = +

e
[3]

where K
e
is the stiffness of the element, H
e
describes the influence the convection on
each node, T
e
is the temperature of the element, F
e
is the internal heat generation of the
element, and P
e
is used to define the convection on the top and bottom faces of the
element.

The interpolations functions for a linear (four node) rectangular element used here are:
13
ME 6124 HJC & MGF

1
1 1
x y
a b

| ||
=
|
\ .\
|
|
.
[4]
2
1
x y
a b

| || |
=
| |
\ .\ .
[5]
3
x y
a b

| || |
=
| |
\ .\ .
[6]
4
1
x y
a b

| ||
=
|
\ .\
|
|
.
y
[7]

The elemental stiffness matrix (for an isotropic material with
x
k k = ) is defined by:

0 0
b a
j j
i i
ij
K k dxdy
x x y y


(
= +

(
dy
dy

[8]

The convective coefficients on an elemental basis are determined through:
( )
0 0
b a
ij i j i T B
H T dxdy = +

[9]

Internal heat generation is accounted for by:
| |
0 0

b a
n
F wq dx =

[10]

Convection on the top (T) and bottom (B) surfaces is given by

| |
0 0 0 0
b a b a
T T B B
P w T dxdy w T dx

= +

[11]

14
ME 6124 HJC & MGF
Using the interpolation functions for a linear (four node) rectangular element, the
stiffness matrix (for an isotropic material with
x
k k
y
= ) may be determined to be:

2 2
2
2 2
2
[ ]
6 2 2
2
2 2
2
a b a b a b a b
b a b a b a b a
a b a b a b a b
b a b a b a b a k
K
a b a b a b a b
b a b a b a b a
a b a b a b a b
b a b a b a b a
| | | | | | |
+ +
| | |
\ . \ . \ . \
| | | | | | |
+ +
| | |
\ . \ . \ . \
=
| | | | | | |
+ +
| | |
\ . \ . \ . \
| | | | | | |
+ +
| | |
\ . \ . \ . \
(
(
(
(
(

(
(
(
(
|
(
.
|
|
.
|
|
.
|
|
.
|
(
(
[12]


Similarly, the matrix of convective coefficients computed to be:

( )
1 1 1
1
2 4 2
1 1
1
2 2
[ ]
1 1 1 9
1
4 2 2
1 1 1
1
2 4 2
T B
ab
H

(
(
(
(
+ (
=
(
(
(
(
(
(

1
4
[13]


The following matrix gives heat generation on a nodal basis:

1
1
[ ]
1 4
1
n
abq
F
(
(
(
=
(
(

[14]


Finally, the effects of convective boundary conditions are accounted for through:

| | ( )
1
1
1 4
1
T T B B
ab
P T T

(
(
| |
(
= +
|
(
\ .
(

[15]

15
ME 6124 HJC & MGF
Fluid Element
The fluid element, detailed in Figure 9, is used to calculate the conductivity of the
fluid, viscosity, Reynolds Number, Prandtl Number, Nusselt Number, convective
coefficients, and hydraulic diameter of the fluid passed through the structural voids, using
the inlet temperature and mass flow rate. The convective coefficient h is stored in the
fluid cell slot and used as both
T
and
B
in the P matrix. The following equations are
used to calculate the convective coefficients.

(1.329 ) / 2.889
6.163/(1 )
ratio
Nu e

= + (Laminar Flow) [16]

where ratio is the aspect ratio of the duct.

0.5 2/ 3
( / 8) Re Pr
1.07 12.7( / 8) (Pr 1)
fr
Nu
fr

=
+
(Turbulent Flow) [17]
where fr is the friction factor.

h
Nu k
h
D

= [18]

Nodal temperatures for each of the four elements making up a rectangular duct are
then calculated using the FEA formulation of the Linear (Four Node) Rectangular Finite
Element. An average of the four nodal temperatures pertaining to each element is taken
and assigned as the surface temperature. This information, in turn, is then used to
calculate the exit temperature T
out
and the total heat transfer rate by the fluid within each
cell Q, respectively by

/
( )
p
hA mC
out surface surface in
T T T T e

=

[19]

(
p out in
Q mC T T =

) [20]








16
ME 6124 HJC & MGF
Finally, the calculated T
out
is assigned as T
in
of the next layer and the process
repeated.

Calculate
Viscosity,h, k,
Pr, Re, h, Dh
Given
Tin, mass flow rate
Get
Tout? Q?
Calculate
Viscosity,h, k,
Pr, Re, h, Dh
Given
Tin, mass flow rate
Get
Tout? Q?
CalculatingFilm Coefficient and
Bulk Temperature
1. Given
Tin, mass flow rate
2. Calculate
k, viscosity, Re, Pr, Nu, h, Dh
(Hydraulic Diameter)
3. Solve for node temperatures using
FEA formulation (previous slide)
Plug h and Tin into FEA matrix
4. Assign average node temperature as
surface temperature and calculate
Tout and Heat transfer rate by fluid
within each cell.
5. Assign Tout as Tin of next layer
Figure 9 - Overview of Fluid Element Calculations
17
ME 6124 HJC & MGF
6.3 Modeling the LCAs in MATLAB

The routine used to conduct thermal analysis (i.e., model
and solve) for LCA structures in MATLAB

is shown in
Figure 10. The thermal module in MATLAB

starts with
specifying input parameters, such as node numbers in the x, y,
and z directions, inlet temperature and mass flow rate of the
fluid, and thermal conductivity of the structural material. The
program then forms nodes and elements for the LCA structure
based on these input parameters. These nodes and elements
are first formed in the x-y plane and extruded along the z-axis
(i.e., the direction of extrusion). This is done in order to
ensure consistency of structural and thermal analyses. It is
thus possible to model any type of cross sectional geometry as
long as it is composed of rectangular (shown in Figure 11) or
triangular (not shown) cells.



















The node numbering scheme, implemented within the
MATLAB code, is illustrated at the hand of a simple 2 by 2
LCA design in Figure 11. The thermal model is fully
compatible with the structural model. In fact, it is formed via
extrusion along the z-axis, as indicated in the previous
paragraph. Although, this example shows only a single slice
of the structure along the z-direction, an arbitrary number may be used, as required for a
particular analysis. As indicated, the thermal element is a rectangular plane while the
structural element is a one-dimensional line. Nevertheless, thermal elements are
numbered in the same pattern as nodes corresponding to the structural elements, so as not
to duplicate global node and element numbers in the z direction. More specifically, node
Specify input
parameters
Formulate Nodes
and Elements
Figure 11 An example of node numbering scheme in structural model
and equivalent thermal model
Structural Model
18
9 3
2
1
6
4
5
7
8
17
16
15
14
13
Thermal Model
12
11
10
z
x
y
9
8
7
6
5
4
3
2
1
Calculate fluid exit
temperature and Q
in each fluid cell
No
Yes
Display Results
Sum the total Q
Last
Layer?
Solve for nodal
Temperatures
Formulate Global
Stiffness, H, P, and
F matrices for each
layer
Calculate Fluid
Properties for each
layer
Define Boundary
Condition
Figure 10 Flow chart
of Matlab FEA Routine
18
ME 6124 HJC & MGF
numbering for structural elements follows the pattern indicated in the left half of Figure
11. The equivalent thermal model then builds on the numbering scheme of the structural
model as shown in the right half of the same figure. Consequently, material properties
such as conductivity, thickness, etc., assigned to the structural model, can be directly
inherited by the thermal model. This is especially important since the thickness of any
given element is likely to change based on the results of the preceding optimization loop
and is thus consistently applied to both structural and thermal models.
After formulating nodes and elements, boundary conditions are defined by fixing the
attributes of the corresponding nodes. The temperature attribute, for example has two
properties one being the temperature and the other being its status (fixed or not).
Temperatures that are imposed as a boundary conditions are thus marked as fixed. Much
the same is true for heat flux, the other possible boundary condition for thermal analysis
as implemented here.
Once the boundary conditions are imposed the program begins the process of solving
the structure. As mentioned previously, this is done on a layer-by-layer basis (along the
direction of extrusion). For each layer, the property calculation module first calculates
fluid (air) properties (e.g., film coefficient, Re, Pr, k, kinematic viscosity, Nu (for laminar
or turbulent flow), etc.), given inputs of air inlet temperature and mass flow rate. Global
stiffness, H, P, and F matrices are then formulated for each layer. Since each of these
matrices is just for the layer in question, sizes are considerably smaller than global
stiffness matrices corresponding to the entire structure. Since the solver makes use of the
sparse matrix capability of MATLAB (which is significantly faster than ordinary matrix
operations involving matrices with large numbers of zeros), computational efficiency is
aided further. The output of the solver is the numerical value of temperatures for all
nodes corresponding to the given boundary condition.
Calculated nodal temperatures are then used to calculate other fluid properties,
specifically the outlet temperature and heat transfer rate. The outlet temperature of the
fluid is stored alongside other fluid cell properties and used to assign the outlet
temperature of the current cell as the inlet temperature of the interfacing cell within the
next layer. Heat transfer rates for each fluid cell are also stored as fluid cell properties, to
be summed over all slices at the conclusion of the analysis routine to determine the total
heat transfer rate by the fluid (corresponding to the total heat drawn from the CPU in the
example considered here). At the conclusion of the thermal analysis, the results (i.e.,
node temperatures, graphically represented using color swatches, a depiction of cross
sectional LCA geometry, maximum and minimum temperature readings, and total heat
transfer rate by the fluid) are displayed in a window.


19
ME 6124 HJC & MGF
6.4 Modeling the LCAs in ANSYS

The main purpose of analyzing the sample LCA designs discussed in Section 7 using
ANSYS was to verify and validate the performance of the MATLAB code, developed in
order to address the needs enumerated in Section 2. Unlike the MATLAB model, no
restrictions were placed on the geometry/nodal configurations of the elements used here.
In order to obtain the most accurate results possible, both a two-dimensional and three-
dimensional analysis were conducted. The elements, chosen to model the LCA
geometries in ANSYS were the following:

PLANE 55 2D 4-Node Thermal Solid
SOLID 70 3D 8-Node Thermal Solid

These elements were chosen for their capability of modeling both the thermal
properties we were concerned with in this investigation as well as structural properties
potentially considered for any further validation of our robust topology design efforts in
the future. Since the geometry of the LCAs is fairly simple and structural and thermal
gradients do not change drastically for the applications taken into consideration here, no
higher order elements (e.g., PLANE 77 2D 8-Node Thermal Solid, SOLID 90 3D 20-
Node Thermal Solid) were considered.
LCA geometries were initially modeled in three dimensions. After establishing that
the use of average temperature and film coefficient values produced good quality results
for thermal analysis, however, ANSYS models were simplified to two dimensions.
Convergence of results (i.e., temperature distributions) was then established through
mesh refinement for the two dimensional models. It is quite likely that the upper limit
(128,000 elements) of ANSYS computational capabilities would have been breeched, had
the use of average fluid properties not proven accurate. Although finer granularities with
respect to the two-dimensional cross section would not have been required, as indicated
by the results shown in Figure 16, this effect is likely to have been offset by the increase
in granularity required along the direction of extrusion, mandated by gradients and
elemental aspect ratio considerations. This line of reasoning is supported by the
MATLAB results given in Section 8.1, indicating that increasing the number of elements
used to model a given cross section is not quite as important as increasing the number of
slices used to analyze a structure.
In both the two and three-dimensional cases, ANSYS free mesh features consistently
produced large numbers of bad elements with poor aspect ratios. This is due to the
irregular geometry that results from the extremely thin walls, characteristic of LCAs.
Meshes were thus created manually by dividing the cross sectional geometry into myriad
rectangular elements. Having provided a brief overview of modeling considerations,
results are discussed in Section 8.
20
ME 6124 HJC & MGF
7. Results, Discussion, and Validation/Verification

7.1 The 2 X 12 LCA Design (MATLAB vs. ANSYS)















MATLAB FEA Results
Figure 12 Input Specifications, Boundary Conditions, and LCA Cross-Sectional Geometry for the 2 x 12
LCA Design
Geometry of 2 x12 LCA
Given Specification
Material: Copper (k = 363 W/m K)
Fluid: Air
Width : 0.025 m (even width for each cell)
Height: 0.025 m (0.01365, 0.00701 for each
cell)
Wall Thickness: 0.001 m
Depth: 0.075 m
Inlet Air Temperature = 273.15 K
Inlet Air Velocity = 2.97 m/s
Boundary Condition
Heat Source Temperature at Top = 373.15 K
Complete Insulation at side and bottom walls
MATLAB FEA results are shown in Figure 13. The numbers of Layers in the z-
direction were varied from 20 to 90 layers in increments of 10. As shown in the figure,
maximum temperature, minimum temperature, total heat transfer rate by the fluid, and
average film coefficient (since film coefficients are calculated for each fluid cell) are
provided as alongside visual color swatches representing the temperature distribution
throughout the structure. The left side of each figure is the inlet and the right side the
outlet. The temperature of the top face of the results shown was fixed at the given
temperature of 373.15 K.










Figure 13 MATLAB Results for the 2 x 12 LCA Design
90 80
70 60 50
40 30 20
21
ME 6124 HJC & MGF
ANSYS Results
ANSYS FEA results are shown in Figure 14. After establishing that the use of
average fluid temperature and film coefficient values gave acceptable results, the LCA
structures were modeled in two dimensions using the PLANE 55 (4-Node Thermal Solid)
element. Mesh refinement was conducted by increasing the number of elements used to
model geometrical features of the LCA cross section. The total number of elements
varied from 179 at the lowest resolution to 10425 at the highest resolution.
As shown in the figure, maximum and minimum temperatures are provided alongside
a visual presentation of the temperature distribution throughout the structure. The left,
bottom, and right sides of each cross section were insulated, while the top face was held
at a constant temperature of 373.15 K.



















Figure 14 - ANSYS Results for the 2 x 12 LCA
10425
6672
3753
1668 417
179















22
ME 6124 HJC & MGF
Comparison Minimum Temperature of MATLAB FEA with ANSYS Result

As shown in Figures 15 and 16, the results obtained for the minimum temperature
within the 2 x 12 LCA design in both in MATLAB and ANSYS are very close.
MATLAB FEA results converged to 360.18 K while ANSYS results converges to
361.95K. The difference between the two results is attributable to differences in the film
coefficient and bulk temperatures used for computations. Specifically, film coefficients
and air bulk temperatures were dynamically calculated and updated in MATLAB FEA
from slice to slice, capturing the variation of the fluid properties along the length of the
LCA. In the ANSYS analysis, o the other hand, film coefficients and air bulk
temperatures were fixed at their average value as determined through the use of our
MATLAB code. For this reason, it is reasonable to assume that MATLAB FEA results
are more accurate than ANSYS results.















M A T L A B M e s h C o n v e r g e n c e
3 5 9 . 9 5
3 6 0
3 6 0 . 0 5
3 6 0 . 1
3 6 0 . 1 5
3 6 0 . 2
2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0
N u m b e r o f L a y e r s ( L C A L e n g t h )
A N S Y S M e s h C o n v e r g e n c e
3 6 1 . 8
3 6 1 . 8 5
3 6 1 . 9
3 6 1 . 9 5
3 6 2
3 6 2 . 0 5
3 6 2 . 1
3 6 2 . 1 5
3 6 2 . 2
1 7 9 4 1 7 1 6 6 8 3 7 5 3 6 6 7 2 1 0 4 2 5
N u m b e r o f E l e m e n t s
Figure 16 - ANSYS 2D Mesh Convergence with Respect to the Minimum Temperature within the
Structure (2 x 12 LCA)
Figure 15 - MATLAB Mesh Convergence along Direction of Extrusion with Respect to the
Minimum Temperature within the Structure (2 x 12 LCA)

















23
ME 6124 HJC & MGF

Comparisons the total heat transfer rate of FEA in MATLAB and Finite Difference
Analysis

In order to validate the results of the total heat transfer rate computed by MATLAB FEA,
these were compared to Finite Difference results for the same scenario (i.e., geometry,
input specification, and boundary conditions) [12]. As shown in Figure 17, the
MATLAB FEA results converged to a value of 106.1 W. The total heat transfer rate
documented by Seepersad, et al., however, is 94.96 W. Although we are not certain, it is
likely that this difference is due to a simplification of cell wall widths as having constant
thicknesses of 1 mm for walls in both the in x and y direction. This stands in marked
contrast to the thicknesses shown on the right-hand side of Figure 17 that were used
within the Finite Difference code. Overall comparison with Finite Difference code
results for the 8 x 8 structure, described in Section 8.1, which featured constant wall
thicknesses showed, close agreement. In fact, taking into consideration that the
experimental setup is likely to indicate higher heat transfer rates due to the inability to
reproduce perfectly insulated boundaries, it seems that the MATLAB FEA is in closer
agreement with theoretical results.


MATLAB Mesh Convergence
105.7
105.8
105.9
106
106.1
106.2
20 30 40 50 60 70 80 90
Number of Layers (LCA Length)
T
o
t
a
l

H
e
a
t

T
r
a
n
s
f
e
r

(
W
)
Figure 17 MATLAB Mesh Convergence along Direction of Extrusion with Respect to the Total
Heat Transfer Rate (2 x 12 LCA)

























24
ME 6124 HJC & MGF
7.2 The 8 X 8 LCA Design (MATLAB vs. Finite Difference)

MATLAB Results
In order to validate the results of the MATLAB FEA code using experimental data,
the identical conditions (see Figure 18) cited for the experiments documented by
Seepersad, et al. [12] were simulated. Case study 1 consisted of 6 different setups with
top temperatures of 305, 312.5, 315, 319, 323, and 325 K and Reynolds number 1200
respectively. The second case study was conducted at Reynolds number 891 with top
temperatures of 305, 312.5, 315, 317.5, 319, 323, and 325 K respectively. A sample
result of MATLAB FEA analysis for a top temperature as 327 K and air velocity 8.8513
m/sec, corresponding to Re=891, is given in Figure 19.



Given Specification
Material: Copper (k = 363 W/m K)
Fluid: Air
Width: 0.0138 m (even width for each cell)
Height: 0.0138m (even height for each cell)
Depth: 0.0508 m
Wall Thickness: 0.0012 m
Inlet Air Temperature = 273.15 K
Inlet Air Velocity 1 = 11.92 m/s (Re = 1200)
Inlet Air Velocity 2 = 8.8513 m/s (Re = 891)
Boundary Condition
Heat Source Temperature at Top = 305~327K
Complete Insulation at side and bottom walls
































Figure 19 - MATLAB Results for the 8 x 8 LCA
Figure 18 Input Specifications, Boundary Conditions, and LCA Cross-Sectional Geometry for the 8 x 8
LCA Design
Geometry of 8 x 8 LCA
25
ME 6124 HJC & MGF
Comparisons the total heat transfer rate of FEA in MATLAB with Experimental Data

As can clearly be seen in Figures 20 and 21, the total heat transfer rate calculated for
each Heater Temperature (i.e., the top surface temperature) tested, using the MATLAB
FEA code is very close to experimental data corresponding to both Reynolds number
1200 and 891. In fact, the total heat transfers between 312.5 K and 315 K are almost
identical for both FEA and the experiments. The difference of the total heat transfer rate
at higher top temperatures can be explained by the fact that it is virtually impossible to
implement completely insulated boundary conditions assumed during MATLAB FEA
analysis (at the left, right and bottom walls of structure) in physical setups.
Consequently, slightly higher heat transfer is to be expected, as shown in both cases.









Total Heat Transfer vs. Heater Temperature
(Re = 1200)
0
5
10
15
20
25
300 305 310 315 320 325 330
Heater Temperature (K)
T
o
t
a
l

H
e
a
t

T
r
a
n
s
f
e
r

R
a
t
Matlab Analysis Experimental Data
00
Total Heat Transfer vs Heater (Re=891)
0
5
10
15
20
25
310 315 320 325 330
Heater Temperature (K)
T
o
t
a
l

H
e
a
t

T
r
a
n
s
f
e
r

R
a
t
e
Matlab Analysis Experiment
00
Figure 20 Comparison of MATLAB FEA and Experimental Results for Total Heat Transfer at
RE = 12
Figure 20 Comparison of MATLAB FEA and Experimental Results for Total Heat Transfer at
RE = 12
26
ME 6124 HJC & MGF
7.3 The 1 x 1 LCA Design (MATLAB vs. Hand Calculations)

Hand calculations for a simple 1 x 1
channel indicated that MATLAB code is in
agreement with theoretical results. This
should come as no surprise since the
MATLAB algorithm essentially comprises
a computational instantiations of the
element derived in Appendix A. Any slight
differences in the numerical results obtained
for larger structures can thus be attributed to
round off error. For detailed hand
calculations, please refer to Appendix B.

8. Closure

Finite Element Thermal analysis has
thus been developed and successfully
deployed in MATLAB. Results obtained
using these analysis modules have been
validated against (1) theoretical results (i.e.,
hand calculations for a simple rectangular
duct), (2) numerical results using both finite
difference and finite element code, and (3)
experimental results. Assumptions,
underlying this development were also
verified satisfactorily. Specifically, it was determined that (1) heat conduction along the
LCA length at layer interface is indeed negligible, (2) constant surface temperature
(Internal Pipe Flow) holds, and (3) approximating the Inlet Temperature as Bulk
Temperature is valid. Furthermore, the ability to calculate Fluid Heat Transfer Rate
(Internal Pipe Flow) was added to current analysis capabilities. This is a fundamental
step towards successful multi-objective topology design.
Figure 21 MATLAB FEA Results for a 1 x 1
Cell
Having developed thermal analysis capabilities in MATLAB, future work will consist
of extending the applicability of fluid cell analysis to different geometries. In essence,
this effort will focus on developing an algorithm to (1) determine the extent of void
regions and (2) calculate the corresponding hydraulic diameters, thus allowing us to
determine the nature of the flow in question (i.e., laminar or turbulent), when given mass
flow rate and inlet temperature. Thermal analysis will also have to be integrated with
preexisting structural analysis modules in order to run the multi-objective optimization
routines, required for robust topology design.






27
ME 6124 HJC & MGF
9. Challengesand Thoughts on Learning

Challenges
The main challenge in this investigation was to develop a thermal analysis module
that interfaced directly with existing structural analysis codes in MATLAB. Since
structural analysis is currently based on one-dimensional frame elements (chosen to
increase flexibility in modeling two-dimensional topologies) this limited our flexibility in
answering this need. Since elements are continuously removed in topology optimization,
a key requirement was to use existing node locations within a given two-dimensional
cross section and simply model the corresponding elements with regard to their thermal
characteristics along the direction of extrusion.
Other challenges included reducing the computational time required for conducting
thermal analysis as much as possible. This aim was accomplished by slicing the structure
along the direction of extrusion and solving for the thermal performance of the structure
on a layer-by-layer basis. Our investigation showed that even the lowest number of slices
gave results that were extremely good. Additionally, we found that improvements in the
quality of results to be obtained by increasing the number of elements used to model a
given LCA cross section beyond the minimum were negligible. Slicing the structure and
determining thermal characteristics in a sequential manner thus proved to greatly reduce
computational intensity, in part through global stiffness matrix reduction.
With respect to modeling the given LCA designs in ANSYS, the prime difficulty
turned out to be obtaining elements with good aspect ratios. Any of the automatic
meshing features within ANSYS proved incapable of dealing with the rather thin
geometries of the LCAs effectively. Consequently, each mesh had to be created
manually, a task that was complicated by the fact that edges of inner and outer geometric
features did not match up. Creating a suitable mesh thus turned out to involve quite a bit
of artistic finesse.
Although we initially attempted to model the fluid in ANSYS using FLOTRAN,
further investigation showed that applying average values of bulk temperatures and
convective coefficients, as obtained through our MATLAB code gave us results that were
accurate enough. Using such constant values, further made it possible to rely on two
dimensional ANSYS models. Although several geometries considered were also
modeled using symmetry considerations, this avenue of investigation was no pursued
once the models were reduced to two dimensions. Considering that LCA cross sections
are not necessarily symmetric and may be functionally graded in more than a one
direction and using two-dimensional analysis (relying on average fluid properties)
seemed more appropriate, at least for the case at hand.

Learning
As a whole, we feel that this project was an excellent vehicle for gaining a more
fundamental understanding of the Finite Element Method that complemented the
theoretical foundation we were given in class rather well. Developing thermal analysis
code in MATLAB required us to draw on underlying theory in deriving our thermal and
fluid elements as well as take into consideration computational aspects, fundamental to
the implementation of FEA. We developed a solid understanding of ANSYS in terms of
its execution as well as an appreciation of its rather complicated underpinnings.
28
ME 6124 HJC & MGF
10. References

1. Sigmund, O., 2001, A 99 Line Topology Optimization Code Written in Matlab,
Structural Multidisciplinary Optimization, Vol. 21, pp. 120-127.
2. Eschenauer, H. A. and N. Olhoff, 2001, Topology Optimization of Continuum
Structures: A Review, Applied Mechanics Reviews, Vol. 54, No. 4, pp. 331-389.
3. Rozvany, G. I. N., 2001, Aims, Scope, Methods, History, and Unified Terminology
of Computer-Aided Topology Optimization in Structural Mechanics, Structural and
Multidisciplinary Optimization, Vol. 21, pp. 90-108.
4. Sigmund, O., 2001, Design of Multiphysics Actuators Using Topology Optimization
-- Part I: One-Material Structures, Computer Methods in Applied Mechanics and
Engineering, Vol. 190, No. 49-50, pp. 6577-6604.
5. Neves, M. N., O. Sigmund and M. P. Bendsoe, 2002, Topology Optimization of
Periodic Microstructures with a Penalization of Highly Localized Buckling Modes,
International Journal for Numerical Methods in Engineering, Vol. 54, pp. 809-834.
6. Diaz, A., R. Lipton and C. A. Soto, 1995, A New Formulation of the Problem of
Optimum Reinforcement of Reissner-Mindlin Plates, Computer Methods in Applied
Mechanics and Engineering, Vol. 123, pp. 121-139.
7. Saxena, A. and G. K. Ananthasuresh, 1999, "Towards the Design of Compliant
Continuum Topologies with Geometric Nonlinearity," ASME Advances in Design
Automation, Las Vegas, NV, ASME DETC99/DAC-8578.
8. Frecker, M. I., G. K. Ananthasuresh, S. Nishiwaki, N. Kikuchi and S. Kota, 1997,
Topological Synthesis of Compliant Mechanisms Using Multi-Criteria
Optimization, ASME Journal of Mechanical Design, Vol. 119, No. 2, pp. 238-245.
9. Sigmund, O. and S. Torquato, 1997, Design of Materials with Extreme Thermal
Expansion Using a Three-Phase Topology Optimization Method, Journal of the
Mechanics and Physics of Solids, Vol. 45, No. 6, pp. 1037-1067.
10. Svanberg, K., 1987, The Method of Moving Asymptotes - A New Method for
Structural Optimization, International Journal for Numerical Methods in
Engineering, Vol. 24, pp. 359-373.
11. Svanberg, K., 1999, MMA and Some Modeling Aspects, Optimization and
Systems Theory, KTH.
12. Seepersad, C. C., B. M. Dempsey, J. K. Allen, F. Mistree and D. L. McDowell, 2002,
"Design of Multifunctional Honeycomb Materials," 9th AIAA/ISSMO Symposium on
Multidisciplinary Analysis and Optimization, Atlanta, GA, AIAA, Paper Number
AIAA-2002-5626.
13. Hayes, A. M., A. Wang, B. M. Dempsey and D. L. McDowell, 2001, "Mechanics of
Linear Cellular Alloys," Proceedings of IMECE, International Mechanical
Engineering Congress and Exposition, New York, NY.
14. Reddy, J. N., 1993, An Introduction to the Finite Element Method, McGraw-Hill,
Boston, MA.


29
ME 6124 HJC & MGF
11. Appendices

11.1 Appendix A Derivation of Finite Element used for MATLAB

Heat Transfer
Analysis

11.2 Appendix B Hand Calculation for a 1 x 1 Cell

11.2 Appendix C Final Presentation Slides
30
ME 6124 HJC & MGF
Appendix A
Derivation of Finite Element used for MATLAB

LCA Heat Transfer Analysis

31
ME 6124 HJC & MGF
Appendix B
Hand Calculation for a 1 x 1 Cell







































32
ME 6124 HJC & MGF
33
Appendix C
Semester Project Presentation Slides