153
Generative Design Optimization of Thermal Management Systems for High Output Power Electronics by Andrew Michalak A thesis submitted in conformity with the requirements for the degree of Master of Applied Science Graduate Department of Mechanical and Industrial Engineering University of Toronto © Copyright by Andrew Michalak 2019

Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

Generative Design Optimization of Thermal Management Systems for High Output Power Electronics

by

Andrew Michalak

A thesis submitted in conformity with the requirements for the degree of Master of Applied Science

Graduate Department of Mechanical and Industrial Engineering University of Toronto

© Copyright by Andrew Michalak 2019

Page 2: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

ii

Generative Design Optimization of Thermal Management

Systems for High Output Power Electronics

Andrew Michalak

Master of Applied Science

Graduate Department of Mechanical and Industrial Engineering

University of Toronto

2019

Abstract

Power electronic converters are becoming a critical part of the continuing electrification of

transportation technology. With the increasing popularity of electric vehicles, high demands are

being placed on the performance and reliability of these on-board modules. To meet these

challenges, novel architectures and advanced design techniques are being utilized to address the

growing issue of proper thermal management for compact power electronic devices. This thesis

proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid

topologies of compact heat sinks for power electronic systems. By incorporating precise

electrical design data into detailed thermal models, the optimization process accurately captures

the heat spreading within these complex systems. The intelligence nature of this iterative

program identifies ideal design characteristics to improve heat sink performance and generate

optimized cooling structures, specifically tailored to target converter systems.

Page 3: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

iii

Acknowledgments

The work outlined in this thesis would not have been possible without the support of several

crucial individuals that have played significant roles throughout this process.

First, I would like to express my sincere gratitude to my supervisors, Dr. James K. Mills and Dr.

Olivier Trescases. Their assistance and continual guidance through this process has helped me

overcome the difficult obstacles this project so graciously put in my way, time and time again. I

came to the University of Toronto hoping to learn more about the intricacies of electric vehicles,

their underlying technology and the design process behind it all. If nothing else, I know I have

been successful in this regard, thanks to the shared knowledge and experience of both these

professors.

Thanks to Dr. Wai Tung Ng and Dr. Kamran Behdinan for serving on my defense committee, I

appreciated hearing your thoughts, comments and advice on the presented work. I would also

like to thank to my awesome lab mates, Simarjot Sidhu and Ihab Abu Ajamieh, for always being

available to work through problems with me, brainstorm ideas or simply grab the occasion free

coffee. My good friends Omri Tayyara and Marinus Lurz were also an essential part of my

academic experience, always offering welcomed advice or needed distractions.

Special thanks to Dr. Steven Kinio for guiding me through the initial construction of my genetic

program. His advice and mentorship was an invaluable component of this project. I would also

like to acknowledge NSERC for providing the funding to make my project and my degree

possible.

Special thanks to my roommates, Bethany Litner, Jessica Germano and Maxime Larcheveque,

who always kept life light and fun, especially when school was not. Thanks also goes out to both

the rats in our first apartment, Toronto has definitely been a wild ride right from the get-go and I

wouldn’t have changed a thing.

Lastly, I would like to express my deepest gratitude to my parents, Timothy and Loreen, along

with my sister Sharon, for providing me with all the encouragement, patience, love and support I

needed as I pursued this degree. The University of Toronto has provided me with one of the most

memorable experiences of my life, my greatest thanks to everyone involved.

Page 4: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

iv

Table of Contents

Acknowledgments.......................................................................................................................... iii

Table of Contents ........................................................................................................................... iv

List of Tables ................................................................................................................................ vii

List of Figures .............................................................................................................................. viii

Nomenclature ............................................................................................................................... xiv

Chapter 1 ..........................................................................................................................................1

Introduction .................................................................................................................................1

1.1 Problem Statement ...............................................................................................................1

1.2 Proposed Solution ................................................................................................................3

1.3 Thesis Layout .......................................................................................................................4

Chapter 2 ..........................................................................................................................................6

Background .................................................................................................................................6

2.1 Basic Technologies and Practices ........................................................................................7

2.2 Air Cooling ..........................................................................................................................9

2.2.1 Natural Convection ..................................................................................................9

2.2.2 Forced Convection .................................................................................................11

2.3 Indirect Liquid Cooling......................................................................................................13

2.3.1 Liquid Cold Plates..................................................................................................13

2.4 Direct Liquid Cooling ........................................................................................................15

2.4.1 Packaging ...............................................................................................................16

2.4.2 Micro Channel Heat Sinks .....................................................................................20

2.4.3 Jet Impingement Heat Sinks ..................................................................................24

2.4.4 Integrated Coolers ..................................................................................................28

Page 5: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

v

2.4.5 Double-Sided and Stacked Modularized Cooling ..................................................30

2.5 Design Optimization ..........................................................................................................32

2.5.1 Practical Techniques ..............................................................................................33

2.5.2 Topology Optimization ..........................................................................................34

2.5.3 Genetic Algorithms ................................................................................................36

Chapter 3 ........................................................................................................................................40

Methodology and Construction of Optimization Program for Liquid Heat Sink Topologies ..40

3.1 Stage I: Genetic Optimization Logic .................................................................................41

3.1.1 Constructing Model and Defining Workspace ......................................................43

3.1.2 Grid Indexing .........................................................................................................44

3.1.3 Mutate Seed Design ...............................................................................................46

3.1.4 Validation ...............................................................................................................48

3.1.5 Evaluation ..............................................................................................................49

3.1.6 Selection, Crossover and Mutation ........................................................................52

3.1.7 Convergence and End Process ...............................................................................53

3.1.8 Preliminary Results ................................................................................................54

3.2 Stage II: Integrating Three-Dimensional Liquid Heat Sink Topologies ............................59

3.2.1 ANSYS Icepak .......................................................................................................59

3.2.2 Changes to Modeling Procedure and Genetic Functions .......................................63

3.3 Overview of Genetic Optimization Process for Three-Dimensional Liquid Heat Sinks ...71

Chapter 4 ........................................................................................................................................73

Simulation Results & Discussion ..............................................................................................73

4.1 Introducing the Test Model ................................................................................................73

4.2 Case Study .........................................................................................................................77

4.2.1 Preparing the Seed Model ......................................................................................78

4.2.2 Defining the Simulation Environment ...................................................................81

Page 6: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

vi

4.2.3 Initializing Genetic Logic ......................................................................................83

4.2.4 Design Optimization ..............................................................................................84

4.3 Optimization Testing .........................................................................................................89

4.3.1 Weighting Fitness Function ...................................................................................90

4.3.2 Inlet Temperature Variation ...................................................................................93

4.3.3 Flow Rate Variation ...............................................................................................96

Chapter 5 ........................................................................................................................................99

Conclusion ................................................................................................................................99

5.1 Contributions......................................................................................................................99

5.2 Prototyping Tool ..............................................................................................................100

5.3 Closing Remarks and Future Works ................................................................................103

Bibliography ................................................................................................................................105

Appendix: Details of Heat Sink Design Optimization .................................................................116

Page 7: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

vii

List of Tables

Table 1: Heat Dissipations of Various Cooling Methods for T of 353.15K. - [19] ....................... 6

Table 2: Manufactured Power Module Packing Technologies. - [53] .......................................... 19

Table 3: GaN Transistors Manufacturer Provided Thermal Characteristics. ............................... 76

Table 4: Icepak Material Assignments. ........................................................................................ 82

Table 5: Genetic Variables for Case Study Trial. ......................................................................... 84

Page 8: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

viii

List of Figures

Figure 1: Classification of Power Semiconductor Modules. - [15] ................................................ 3

Figure 2: Thermal Stack up for Conventional IPM. - [21] ............................................................ 7

Figure 3: Thermal Resistances of Common Heat Sink Materials. - [26] ........................................ 8

Figure 4: CFD Analysis of Passive Heat Sink Designs. - [29] ..................................................... 10

Figure 5: Study on Pin Design for Heat Sink Performance. a) Conventional Pin Fin Array.

b) Array with Expanded Pin Fin Diameter. .................................................................................. 11

Figure 6: Structural Examples of Commonly Manufactured Cold Plates. a) Deep Drilled Capped

Cold Plate. b) Pocketed Folded-Fin Cold Plate. - [46] ................................................................. 13

Figure 7: Common Embedded Tube Heat Exchanger. – [53] ...................................................... 14

Figure 8: Comparison of Conventional and Advanced Heat Sink Designs. a) Indirect Cold Plate

Arrangement. b) Direct Cooling Arrangement. - [26] .................................................................. 16

Figure 9: Comparison of Material Stacks for Varying Heat Sink Design Principles. a) Indirect

Cooling Structure and Associated Materials. b) Direct Cooling Structure and Associated

Materials. - [58] ............................................................................................................................ 17

Figure 10: Coefficient of Thermal Expansion Values for Common IPM Materials. - [63] ......... 18

Figure 11: Micro Pin Fin Heat Sink Arrangement. - [70]............................................................. 21

Figure 12: Straight Channel, Staggered Pin and MDT In-Line Pin Heat Sink Formations. - [76]22

Figure 13: Demonstration of Fin Deformation Heat Sinks a) Micro Deformation Tools and

Process. b) Micro Deformed Heat Exchanger Array. - [76] ......................................................... 23

Figure 14: Common Jet Impingement Arrangements. a) Free Surface Jet. b) Submerged Jet.

c) Confined Submerged Jet. - [42] ................................................................................................ 24

Figure 15: Impingement Based Heat Exchanger Designs. - [86] ................................................ 27

Page 9: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

ix

Figure 16: Formation of Flow Vortices by Variations in Impingement Angles. a) Perpendicular

90° Impingement angle. b) 70° Impingement Angle. c) 45° Impingement Angle. - [86] ............ 27

Figure 17: Danfoss Shower Power Cooling Design. - [97] .......................................................... 28

Figure 18: Direct Integrated Heat Sink Arrangement. - [26] ....................................................... 29

Figure 19: Construction of MMC Heat Sink Prototype. - [102] ................................................... 30

Figure 20: Double Sided Heat Sink Approach. a) Conventional Liquid Cooling Arrangement.

b) Double Sided Cooling Arrangement. - [104] ........................................................................... 30

Figure 21: Demonstration of Double-Sided-Stacked Cooling Structure. - [106] ......................... 32

Figure 22: Heat Sink with Increased Fin Density Topology. - [121] ........................................... 35

Figure 23:Optimized Heat Sink Designs with Constant Gr Value and Mesh Size of 329 x 640 x

320 Elements. - [122] .................................................................................................................... 35

Figure 24: Basic Genetic Optimization Workflow. - [125] .......................................................... 36

Figure 25: First Stage Conventional Genetic Optimization. ......................................................... 37

Figure 26: Second Stage Perturbation Genetic Optimization. - [125] .......................................... 38

Figure 27: Two Stage Genetically Optimized Air-Cooled Heat Sink. - [132] ............................. 38

Figure 28: Content Breakdown for Chapter 3. .............................................................................. 41

Figure 29: Structure of Genetic Optimization Process. ................................................................ 42

Figure 30: Visualization of Structural Bit Array. ......................................................................... 43

Figure 31: Defining the Design Optimization Workspace. .......................................................... 44

Figure 32: Created Global Array. ................................................................................................. 45

Figure 33: Assigning Element Labels to Partitioned Workspace. ................................................ 45

Page 10: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

x

Figure 34: Linking Element Labels to Global Array. ................................................................... 46

Figure 35: Seed Mutation.............................................................................................................. 47

Figure 36: Utilizing Solution Vector and Global Array to Form Structural Bit Matrix for New

Individual. ..................................................................................................................................... 47

Figure 37: Validating New Design Candidate with Blob Analysis and Image Morphology. ...... 48

Figure 38: 2D Example Population of Initial Design Candidates. ................................................ 49

Figure 39: File Structure of Evaluation Stage............................................................................... 50

Figure 40: Evaluation Stages. a) Candidate Solution Vector. b) Representative Bit Array.

c) ANSYS CFD Model. ................................................................................................................ 50

Figure 41: Evaluation Process. a) Generated Bit Arrays. b) Converted to ANSYS CAD

Structures and Simulated in FLUENT. c) Ranked and Assigned Breeding Probability. ............. 51

Figure 42: Generating Child Design. ............................................................................................ 52

Figure 43: Convergence Window with Objective Tracking. ........................................................ 53

Figure 44: Genetic Optimization Process on a 2mm Grid Mini-Channel Design. a) Convergence

Plot. b) Intermediate Designs. c) Final Thermal-Flow Contours. ................................................. 55

Figure 45:Genetic Optimization Process on 1mm Grid Central Channel Design. a) Convergence

Plot. b) Intermediate Designs. c) Final Thermal-Flow Contours. ................................................. 57

Figure 46: Additive Topologically Optimized Heat Sink. a) Smoothed 3D View. b) 2D Profile. -

[140] .............................................................................................................................................. 58

Figure 47: Typical Icepak Workflow. ........................................................................................... 60

Figure 48: Example Icepak Project on Half-Bridge DBC Converter Module. ............................. 61

Figure 49: Icepak Semiconductor Package Design. ..................................................................... 62

Page 11: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xi

Figure 50: Icepak ECAD Import Structure. .................................................................................. 62

Figure 51: Workflow of Model Construction. .............................................................................. 63

Figure 52: Partitioning of Optimization Workspace with Three-Dimensional Voxels. ............... 64

Figure 53: Indexing Workspace Elements. ................................................................................... 64

Figure 54: Formation of Heat Sink Topology. a) Allocation of Structural Elements via Design

Modeler Boolean Functions. b) Forming Fluid Domain. c) Forming Solid Domain. d) Initial Seed

Model. ........................................................................................................................................... 65

Figure 55: Setting Simulation Conditions in Icepak GUI ............................................................. 65

Figure 56: Defining Output Parameters. a) Icepak GUI. b) Variables Exported from Workbench

Project as CSV File. ...................................................................................................................... 66

Figure 57: Formation of Three-Dimensional Global Array. ......................................................... 67

Figure 58: Mutating the Three-Dimensional Workspace to Achieve New Fluid Domain. .......... 67

Figure 59: Operations for Generating New ANSYS Models. a) Starting with Previous Design. b)

Clearing Liquid and Solid Boolean Functions. c) Selecting New Fluid Elements. d) Reapplies

Boolean Functions to Generate New Design Topology. .............................................................. 69

Figure 60: Stage II Convergence Window with Objective Tracking. .......................................... 71

Figure 61: Design Progression of Genetic Optimization Process. ................................................ 72

Figure 62: Base Model for Simulation Testing............................................................................. 74

Figure 63: Electrical DBC Design. ............................................................................................... 74

Figure 64: Compact HB Heat Sink Design. a) Integrated Cooler Approach. b) Inlet/Outlet

Manifold Design. .......................................................................................................................... 75

Figure 65: GaN Power Transistors. a) GaN Systems GS66508B Schematic. b) Corresponding

CAD Model. .................................................................................................................................. 76

Page 12: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xii

Figure 66: Footprint of Heat Sources on PCB. ............................................................................. 77

Figure 67: Electrical Simplification of HB Converter System. .................................................... 78

Figure 68: Simplification of Fluid Domain. a) Elimination of Manifold Sections. b) Comparison

of Old vs New Inlet/Outlet Connections. ...................................................................................... 79

Figure 69: Generating Optimization Workspace for HB Convert Model. a) Defining Active and

Passive Regions. b) Sizing Structural Voxel Elements. c) Partitioning Active Region into Array

of Workspace Elements. ............................................................................................................... 80

Figure 70: Forming Base Optimization Model into Starting Seed Design. .................................. 80

Figure 71: Icepak Modeling Environment & Example Surface Mesh. ........................................ 81

Figure 72: Objective Tracking Window for HB Converter Case Study. ...................................... 85

Figure 73: Design Progression of Case Study Optimization Process. .......................................... 87

Figure 74: Comparing PCB Temperatures Contours. a) Starting Seed Design. b) Final Optimized

Design. .......................................................................................................................................... 88

Figure 75: Comparing Fluid Domain Pressure Contours. a) Starting Seed Design. b) Final

Optimized Design. ........................................................................................................................ 88

Figure 76: Comparing Cross-Sectional Temperature Contours of Heat Sink Cooling Structure.

a) Starting Seed Design. b) Final Optimized Design. ................................................................... 89

Figure 77: Convergence and Optimized Fluid Topologies for Fitness Testing. a) Temperature

Dependent Fitness Scoring (a=1, b=0). b) Temperature Biased Fitness Scoring (a=0.7, b=0.3). 91

Figure 78: Performance Comparisons of Fitness Testing Models. ............................................... 92

Figure 79: Convergence and Optimized Fluid Topologies for Inlet Temperature Testing. a) 0°C

Inlet Fluid. b) 15°C Inlet Fluid. c) 50°C Inlet Fluid. .................................................................... 94

Figure 80: Performance Comparisons of Inlet Temperature Testing Models. ............................. 95

Page 13: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xiii

Figure 81: Convergence and Optimized Fluid Topologies for Flowrate Testing a) Inlet Flow 0.25

LPM. b) Inlet Flow 0.5 LPM. c) Inlet Flow 1.0 LPM. ................................................................. 97

Figure 82: Performance Comparisons for Flowrate Testing Models............................................ 98

Figure 83: Workflow of Genetic Optimization is Prototyping Process. a) Starting Design

Temperature Profile. b) Optimized Design Temperature Profile. c) Thermal Deformation of

Optimized Design. ...................................................................................................................... 102

Page 14: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xiv

Nomenclature

Acronyms

SiC Silicon Carbide

GaN Gallium Nitride

Si Silicon

DGT Dispersed Generation Technology

PV Photovoltaic

IPM Integrated Power Modules

EV Electric Vehicle

HEV Hybrid Electric Vehicle

CFD Computational Fluid Dynamics

FEA Finite Element Analysis

IGBT Insulated Gate Bipolar Transistor

MOSFET Metal-Oxide-Semiconductor Field-Effect Transistor

GTO Gate Turn-Off Thyristor

SCR Silicon Controlled Rectifier

HDD Heat Dissipating Device

PCB Printed Circuit Board

DBC Direct Bonded Copper

Page 15: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xv

DBA Direct Bonded Aluminum

AlN Aluminum Nitride Ceramic

Al2O3 Alumina Ceramic

Al Aluminum Alloy

Cu Copper Alloy

TIM Thermal Interface Material

CTE Coefficient of Thermal Expansion

MMC Metal Matrix Composite Materials

MDT Micro Deformation Technology

GA Genetic Algorithm

DM Design Modeler

HB Half-Bridge

Heat Transfer Variables

QFluid, Coolant Flow Rate

TFluid,In Coolant Inlet Temperature

RTH Thermal Resistance of Heat Sink

TD Device Temperature

ΔPInlet-Outlet Pressure Drop

TD,Base Device Temperature of Base Seed Design

Page 16: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xvi

ΔPBase Pressure Drop of Base Seed Design

a Temperature Weighting in Fitness Function

b Pressure Weighting in Fitness Function

ρ Material Density

kth Thermal Conductivity

cp Specific Heat Capacity

Genetic Optimization Variables

MB Design Candidate Bit Matrix

MP Partitioned Optimization Workspace Matrix

ΔxPixel x Dimension of 2D Structural Pixel Element

ΔyPixel y Dimension of 2D Structural Pixel Element

XWS x Dimension of Optimization Workspace

YWS y Dimension of Optimization Workspace

ZWS z Dimension of Optimization Workspace

MG Global Array of Indexed Workspace Values

xS Workspace Size Vector

xFE Design Candidate Solution Vector Identifying Fluid Elements

Δxvoxel x Dimension of 3D Structural Voxel Element

Δyvoxel y Dimension of 3D Structural Voxel Element

Page 17: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

xvii

Δzvoxel z Dimension of 3D Structural Voxel Element

Nx,voxels Number of Voxel Elements in x Direction of Workspace

Ny,voxels Number of Voxel Elements in y Direction of Workspace

Nz,voxels Number of Voxel Elements in z Direction of Workspace

Page 18: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

1

Chapter 1

Introduction

1.1 Problem Statement

With the increasing emphasis towards an electrified world, energy is quickly becoming a critical

requirement for almost all human ventures. One of the most vital aspects of this growing field is

the control and conversion of the energy itself, which is achieved through the utilization of

specialized electronic circuits and systems known as Power Electronics [1]. These units are

essential in almost all power conversation applications, controlling the flow of electrical energy

at much higher levels than conventional devices could handle. It is anticipated that all electrical

power will flow through a power semi-conductor in the very near future [2]. The more recent

developments in semiconductor materials, such as Silicon Carbide (SiC) and Gallium Nitride

(GaN) devices allows for higher breakdown voltages and forward current densities leading to

greater efficiency and better thermal stability as compared to conventional silicon (Si) devices

[3], [4].

Much of the rise in demand is attributed to the increasing popularity of dispersion energy

systems or Dispersed Generations Technology (DGT) [5]. These systems, both renewable and

non-renewable, include energy sources such as photovoltaic (PV) generators, micro-hydro

systems, wind turbines and fuel cells, which can operate at highly fluctuating levels of

intermediacy. Power electronics are the ideal interface technology to match the output

characteristics of these systems to the conventional grid connection requirements.

Due to the versatile nature in which they can operate, power electronics are also becoming a

fundamental component of industrial, commercial, residential, aerospace and military sectors.

Moreover, the energy range at which they can function make them ideal for use in mobile

transportation systems [6]. Specifically, in terms of the automotive industry, electronic Integrated

Power Modules (IPM) have an important role in the rising popularity of Hybrid Electric and

Electric Vehicles (HEV/EV). Their reduced size and cost along with high efficiency, reliability

and power capacity have made them an integral part of advanced modularized inverter systems

Page 19: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

2

[7]. In modern day designs, these can be found in electric drivetrains, battery charging units and

a variety of vehicular power accessories [8]–[11].

As mobile applications put higher objectives on power density a growing area of concern has

become proper thermal management of these semiconductor devices. While controlling the

energy transfer of electronic systems, power electronic devices experience losses in electrical

efficiency, leading to the generation of waste heat [12]. It is this waste heat that can lead to major

issues such as material degradation, internal thermal stresses, decreased efficiency and overall

system degradation [13]. The task of thermal-mechanical designers to achieve significant levels

of heat removal through the study and utilization of fluid cooled electronic heat sink structures.

One of the most important metrics in this area of thermal design is the semiconductor junction

temperatures1. To operate at peak performance, these semiconductor junctions must be

maintained at acceptable temperatures, depending on their material composition [14]. Different

applications can call for various levels, ranges and durations of power, depending on the function

[15]. Heat load can be categorized by the nature of the corresponding electrical design or the

structure of the associated power semiconductor devices, as shown in Figure 1. The problem then

falls on thermal designers to identify the required level of heat removal based on the expect

device efficiency and chose a suitable cooling method. With the move to electric mobility

applications, shape, weight and volume can become major cost variables while heightened

constraints on things like ambient conditions, fluid structures and performance reliability can

greatly limit design flexibility [16].

1 Junction Temperature: The highest operating temperature of a semiconductor within an electronic device or

package

Page 20: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

3

Figure 1: Classification of Power Semiconductor Modules. - [15]

By utilizing the fundamentals of heat transfer, the increasing power of Computational Fluid

Dynamics (CFD), Finite Element Analysis (FEA) and innovative methods of design and

optimization, thermal engineers can achieve advanced cooling structures to handle this rising

issue of waste heat removal. Developing novel solutions for these high energy applications is the

key to unlocking higher levels of power density and assisting in the continued electronification

of all the technology surrounding transportation and mobility.

1.2 Proposed Solution

The cases of electric cars present a unique and challenging environment for design optimization.

Being a mobile platform, it is always advantageous to reduce the size and weight of any internal

technology. Traditionally, electronic heat sinks and liquid cooling systems are comprised of

heavy metal components, requiring a significant amount of valuable volumetric space. Thus, this

thesis will seek to investigate the development of an optimization process to achieve compact

heat sink topologies for EV specific power electronics.

A custom designed program constructed in MATLAB 2018a utilizes the proven power of binary

genetic optimization and treats the layout of liquid cooling channels with heat exchangers as a

topological optimization problem to produce optimal cooling designs, unique to any electrical

Page 21: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

4

layout [17]. Using a specialized toolbox and Python scripting, this program is able to iteratively

communicate with the ANSYS 19.2 Workbench to model and simulate potential designs

generated by the genetic optimization algorithm coding. By pairing the principles of unique

optimization process with the modelling capabilities of ANSYS Icepak, CFD simulations can

accurately capture the influence of all components and materials within the electronic designs to

feed back into the genetic optimization learning process loop. This genetic optimization

approach, coupled with ANSYS and Icepack, results in three-dimensional heat exchanger

designs with optimized geometry to remove the maximum amount of heat generated form

integrated circuit power transistors.

Initial results show a robust functionality of the custom optimization program and a significant

ability to achieve novel designs and distinct improvements. A series of trials are run to determine

how the established program adapts to changes in the environmental conditions applied to the

simulation space. This, for example, includes optimization at various heat exchanger inlet

temperatures, and coolant fluid pressure. Key genetic variables are also investigated in an

attempt to identify key operating points in the code structure.

Lastly, the resulting geometries produced by the code are analyzed with the goal of

manufacturability. Several options are presented of how to tie the optimization capabilities of the

program to some advanced manufacturing techniques in order to produce these unique models

for real world applications. Specific variables and functions within the coding structure are

identified as key areas to improve the efficiency, functionality and effectiveness moving forward.

The programs flexibility for optimizing any style of heat sink as well as applications to any other

topology problems, capable of being simulated in ANSYS, is discussed as a final note on the

usefulness and future potential.

1.3 Thesis Layout

This thesis is organized as follows: Chapter 2 reviews the wide spectrum of existing methods for

electronic cooling as well as techniques for heat sink design. Various examples of conventional

thermal management systems are presented along with more innovative approaches to high

density cooling, all with the focus of power electronic applications for EV/HEVs. Chapter 3

details the construction of the proposed design optimization program for liquid heat sinks. A

genetic optimization base logic is developed in MATLAB and linked to ANSYS Workbench

Page 22: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

5

using an AAS toolbox in order to simulate and evaluate various topologies for these liquid heat

sink geometries. Chapter 4 demonstrates the operation of the Generative Design Process applied

to a compact, power dense converter module. The optimization process is run on this model at

varying operating parameters in an attempt to characterize the programs behavior and investigate

how changes in system parameters influence the optimal geometries generated. Chapter 5

discusses the conclusions drawn from the tested results, the contributions of the work presented

in this thesis and considers a variety of future opportunities to further improve the efficiency and

effectiveness of this process for generative topology optimization.

Page 23: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

6

Chapter 2

Background

In this section, the basics of power semiconductor thermal management is reviewed. This chapter

provides a review the methodologies for the thermal management of power electronics specific

to hybrid and electric vehicles. Basic design approaches are presented as well as a review of the

more recent advancements being made to address the growing issue of electronic cooling and

waste heat dissipation. Power semiconductor devices can be classified by their application and

the corresponding current and voltage levels [18]. Some current industry examples of common

power electronic devices include: Insulated Gate Bipolar Transistors (IGBT), Metal-Oxide-

Semiconductor Field-Effect Transistors (MOSFET), Gate Turn-Off Thyristors (GTO) and

Silicon Controlled Rectifiers (SCR). A variety of cooling options are available depending on the

extend of the waste heat generated by these power electronic devices. These different approaches

to electronic thermal management have be categorized by Scott [19] in Table 1 by the

corresponding level of heat removal that can be achieved.

Table 1: Heat Dissipations of Various Cooling Methods for T of 353.15K. - [19]

Although there exists a large variety of cooling options for high power electronics, several major

constraints must be considered when identifying suitable options for a given system design.

Particularly, in the case of on-board mobility applications, spatial restrictions and weight

requirements may lead to significant constraints on thermal designs. EV’s can also be a

challenging environment for electronics devices. Standards on ambient conditions, coolant

temperatures and performance reliability create a harsh environment for thermal designers to

work within [20].

Page 24: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

7

2.1 Basic Technologies and Practices

As a result of the growing performance requirements, most IPM’s are designed with integrated

cooling systems to extract the waste heat and maintain suitable overall temperature levels. A

convectional IPM stack is presented in Figure 2. The main heat dissipating device (HDD) usually

takes the form a silicon chip or package. However, as mentioned earlier, advances in device

packaging and the utilization of bare die SiC or GaN components help reduce thermal resistances

in close proximity to the junction heat source [3], [4]. These devices are usually solder bonded to

a printed circuit board (PCB), which provides electrical isolation as well as housing any other

passive/active components required by the power system layout [21]. More recent designs seek

to replace conventional PCBs with Direct Bonded Copper (DBC) or Direct Bonded Aluminum

(DBA) modules. The ceramic materials within DBC stacks, such as Aluminum Nitride (AlN) or

Alumina (AL2O3) allow for much higher levels of heat transfer through the isolation material

when compared with the FR-4 epoxy commonly used for PCB isolation [22].

Figure 2: Thermal Stack up for Conventional IPM. - [21]

While the electrical aspect of IPM designs seek to reduce thermal resistances cause by material

layers, the mechanical components attempt to utilize heat spreading and convective heat transfer

to dissipate the waste energy away from the HDD [23]. The system represented in Figure 2

depicts a conventional finned heat sink design, usually machined from a highly conductive metal

alloy such as Aluminum (Al) or Copper (Cu). These heat sinks are in turn mounted to the

corresponding electrical design via a Thermal Interface Material (TIM) which usually takes the

form of thermal grease or solder [21].

The final component required by any thermal management system is a working fluid or coolant.

The role of this moving fluid is to act as the mechanism for convective heat transfer from the

Page 25: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

8

heat sink, carrying the excess heat away from the IPM stack, and rejecting it from the system

altogether [24]. With regards to HEV/EV systems, there are historically two main fluids

available for electronic cooling, which are air and automotive antifreeze (water/ethylene glycol

mix) [25]. Each carries its own set of benefits and draw backs, which will be discussed in the

coming sections. The inclusion of a coolant system introduces several variables, such as: flow

rate (QFluid), fluid temperature (TFluid,In), required plumbing, ducting and reservoir storage, which

can all have major influence on the performance and architecture of the overall thermal system

design [24].

The data presented in Figure 3 compares the associated Thermal Resistance (RTH) values

expected from conventional IPM materials as well as the convective resistance between the heat

sink and working fluid [26]. Such information is important for thermal designers looking to

improve on the current approaches to electronic cooling.

Figure 3: Thermal Resistances of Common Heat Sink Materials. - [26]

In order to achieve innovative designs and novel solutions, all materials and layers within the

IPM stack must be considered individual design components and analyzed accordingly. By

integrating these different components into the design process, advanced thermal management

systems can meet the rising needs of high-level power electronics.

Page 26: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

9

2.2 Air Cooling

One of the most well established and widely utilized methods of electronic cooling is that of

convective heat transfer with air as the working fluid. Air is still the most flexible and least

expensive options in terms of a heat transfer medium. In addition, air cooled systems require

very little complexity, resulting in significantly lower material and equipment costs when

compared with other methods [27]. Regarding thermal management in automotive settings, high

velocity air can be readily available, further reducing the system requirements and parasitic

loads. It is also important to note that for moving vehicles all heat, either directly or indirectly,

must be rejected to the surrounding ambient air. Thus the use of air cooling can greatly reduce

the overall size and of a coolant loop [21].

2.2.1 Natural Convection

Many designers in the past have utilized the heat transfer phenomenon known as Natural

Convection to achieve very simplistic, robust designs for handling low power electronic systems.

Bouknadel et al. carried out extensive CFD analyses on several heat sink configurations,

investigating various fin arrangements as well as conductive metals. Results indicated that heat

sinks composed of Graphite-metal provided lower thermal resistances when compared to

conventional Aluminum and Copper designs. Elliptical style fins were also found to achieve

higher levels of heat dissipation, as shown in Figure 4, outperforming parallel plate fins as well

as staggered circular and square fin arrangements [28].

Page 27: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

10

Figure 4: CFD Analysis of Passive Heat Sink Designs. - [29]

A similar study carried out by Arefin found that expanding the outer diameter of conventional

pin fin arrays had the potential to increase the heat transfer capabilities of passive aluminum heat

sinks. A 1° expansion of the pin diameter along the length of the fins [Figure 5] was found to

results in a 5°C temperature drop for a 50-Watt heat load [30].

Page 28: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

11

Figure 5: Study on Pin Design for Heat Sink Performance. a) Conventional Pin Fin Array.

b) Array with Expanded Pin Fin Diameter.

Christen et al. sought to compare the efficiency of forced convection to natural convection for air

cooled heat sinks. Considering the power consumption required by active cooling fans and

blowers, it was found that thermal losses and volumetric requirements can be reduced by parallel

mounting the associated semiconductor devices as well as increasing the number of switching

devices within the system. These methods were found to make passive cooling a more feasible

option for specific systems, reducing the complexity of the associated thermal system while

increasing the power density [31]. With a focus on the formation of laminar versus turbulent air

plumes, Kitamura et al. characterized the geometric variables of a vertical cylinder array in

relation to heat transfer and natural convection [32].

2.2.2 Forced Convection

In general, air offers low thermal conductivity and density, which can result in low rates of

convective heat transfer across heat exchangers. Thus, much of the work in this area has focused

on maximizing the available heat transfer area for the fluid and improving the exchange of

energy. This is why, following the work of Tuckerman and Pease on the enhancement of liquid

cooling via micro-channels, many sought to apply these same principles to active air cooling

Page 29: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

12

concepts [33]. As was the basis for the design work by Hilbert et al. which investigated the use

of laminar air flow through micro channels to offer low thermal resistivity level at <1.7K/W with

very low internal pressure drops [34]. Following this Knight, Goodling and Gross found that

higher rates of heat transfer could be experimentally achieved if channel heat exchangers were

optimized to induce turbulent flow [35]. A later investigation carried out by Azar, McLeod and

Caron defined a new style of Narrow channel heat exchangers capable of cooling high powered

components at dissipation levels of 20W/cm2 [36].

Gromoll was able to alter heat exchanger stacking techniques and integrate micro-heat-pipes,

direct air cooling and thermosyphons to his air cooled heat exchangers and reach dissipation

levels otherwise only possible via liquid cooling [37]. Through the use of tubes for directing air

flow, Kleiner et al. was able to theoretical and experimentally attain much higher levels of

cooling than open air systems [27]. Many also found the use of staggered pin arrays could

increase turbulent flow and thus the power density levels heat exchangers could handle. As was

the case with the work of Marques and Kelly who investigated the use of micromachining to

achieve compact, high performing air cooled models [38].

One of the more recent areas of this field that has seen significant improvements has been the use

of jets for the distribution of air across electronic units. Due to its low viscosity, air can be

delivered at high velocities through very small diameter jet arrays at reasonable pressure levels,

dissipating waste heat fluxes to upwards of 4kW/m2 [39]–[41].

Most designs involving air flow will incorporate large bulky heat sinks composed of thermally

conductive metals, making them a large, heavy accessory for vehicular applications where space

and weight are key issues. Moreover, it is becoming a growing consensus that air cooling

techniques are approaching their peak dissipation levels of ~800 W/cm2 via direct jet arrays [42].

The low conductivity and high convective resistances offered by the fluid will keep it from

achieving higher levels of performance. Thus, as power densities and size reductions become key

design factors, the industry moves away from air cooling methods and towards higher

performing methods of convective heat transfer [43].

Page 30: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

13

2.3 Indirect Liquid Cooling

2.3.1 Liquid Cold Plates

As power levels rise to meet the needs of more commercial and industrial scale energy

applications, so too does the associated heat losses. Hence more stable and reliable cooling

schemes are required. Liquid cold plate technology is seen as a viable solution for greater heat

removal capability. Cold plates commonly utilizes fluids with high heat capacity pumped

through machined passages in metal bodies compose of thermally conductive metals [44]. The

ideal heat transfer fluid is usually plain water as it offers good thermal conductivity, high density

at a relatively low viscosity, indicating reasonable internal pressures. However this is commonly

supplemented with an ethylene glycol solution to raise the boiling point and lower the freezing

point of the working fluid [45], [46].

Figure 6: Structural Examples of Commonly Manufactured Cold Plates. a) Deep Drilled Capped

Cold Plate. b) Pocketed Folded-Fin Cold Plate. - [46]

Cold plates can come in many different forms, usually dependent on the manufacturing

capabilities of the designer and the heat flux requirements of the specific system. Models such as

Deep Drilled Cold Plates, as shown in Figure 6a, are simple to manufacture and offer an

inexpensive, reliable solution to relatively high heat flux applications. While more complex

designs, such as the Pocketed Fin Cold Plate shown in Figure 6b, require a more intricate

manufacturing process, but thus yield much high rates of heat dissipation [46]. Due to their

widespread use, the performance and optimization of cold plate model is important practice

across various industries. Much of the work with cold plate models has been focused around

improving the design characteristics through a variety of optimization methods. One of the most

commonly used forms for the optimization of heat exchangers is Topology Optimization. Much

work has gone into analyzing the physical parameters of these thermal systems and using

Page 31: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

14

established equations for flow and heat transfer to achieve high performing designs [47]–[49].

An investigation by Sparrow et al. analyzed the effect of baffles on flow characteristics and heat

transfer via numerical CFD methodology [50]. Work Nam et al. achieved an algorithm capable

of designing different serpentine channel geometries for fuel cell technology, although further

optimization was recommended to supplement this process [51]. A process carried out by

Fesanghary, Damangir and Soleimani combined Global Sensitivity Analysis and Harmony

Search Algorithms to optimization the design of shell-tube heat exchangers with respect to

multiple variables simultaneously [52].

The most common and widely used styles of heat exchanger is the Formed Tube Cold Plate. This

design, presented in Figure 7, places embedded copper tubes into the machined body of made

from thermally conductive material, usually aluminum alloy. One of the more attractive features

of this design is that the fluid flow within the tube remains completely isolated from the externals

of the system, requiring no rubber seals or hydraulic interfaces that could possible leak due to

wear [46].

Figure 7: Common Embedded Tube Heat Exchanger. – [53]

Due to the safe, reliable and simple design nature of cold plates they have implemented in a

variety of heat removal systems, including various automotive applications. Thermal

management of battery packages has emerged as an ideal area for implementing cold plate

technology. Through CFD analysis, Ghosh showed without stable heat removal from specific

thermal ‘Hot Spots’ the life of HEV/EV batteries can be severely reduced [54]. Pesaran showed

that although the level of heat flux removal by HEV battery packs at a suitable level for air

cooling, EV and more complex HEVs benefit from the ability of cold plate to both heat and cool

effectively [20]. The work presented by Jarrett and Kim showed how, with design modeling

Page 32: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

15

optimization, cold plates were ideal at management the non-uniform nature at which of battery

stacks generate waste heat [55].

Although they provide a very stable cooling solution, cold plates still require the purchase and

machining of metals, adding additional space and weight to the electrotonic units they are

assisting. The thermal resistance of these materials has also raised concerns on the technologies

ability to scale with rising power densities [43], [53], [56]. Therefore, much of the research at the

forefront of the automotive power electronics industry is seeking new ways to effectively

eliminate these materials from the system, reducing size, weight, thermal resistance and bring the

fluid closer to the heat sources. The next section of this thesis will focus on the more innovative

solutions being investigated for high level heat flux applications where conventional cooling

systems fall short.

2.4 Direct Liquid Cooling

Conventional indirect liquid cooled plates, such as the one represented by Figure 8a, are a

practiced technology that has been implemented and optimized over a wide range of

applications. One area of focus working towards increasing performance looks at the sequential

thermal resistance network that exists between the heat source device and the coolant fluid. As

presented by the summarized data in Figure 3, the biggest contributors to thermal impedance are

the base of the heat sink body and the TIM responsible for providing a thermal pathway between

the electrical board and the heat exchanger body [26]. A new style of design, known as Direct

Backside Cooling or Impingement Cooling and seen in Figure 8b, addresses this issue by

eliminating these layers of material and bringing the base plate of the electrical module into

direct contact with the working coolant. Usually some geometry is formed into the base of the

electrical module of IPM to induce turbulence or provide greater interface area for the coolant,

increasing the effect of heat spreading as well as convective transfer [26], [57].

Page 33: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

16

Figure 8: Comparison of Conventional and Advanced Heat Sink Designs. a) Indirect Cold Plate

Arrangement. b) Direct Cooling Arrangement. - [26]

Direct backside cooling can offer significate reductions in size and weight while simultaneously

reducing in total thermal resistance of electronic models or IPMs. For HEV/EV technology this

can mean more compact systems, operating at lower junction temperatures allowing for more

efficient energy conversion [58], [59]. Yet several drawbacks can result from this method of

design. Without the use of a separated cold plate, the interface between the fluid and IPM must

be sealed via O-Ring or rubber gasket. If not designed properly this seal could fail under thermal

cycling, causing coolant to leak in close proximity to the electrical components. In the likely case

that the working fluid is not a dielectric any leakage could have severe impacts on the system

integrity and safety [46]. Effects of corrosion and material degradation on the IPM base have to

be addressed depending on the materials selected for manufacturing [60].

Direct liquid cooling is a very versatile technology and can be integrated with various

conventional and advanced heat sink geometries to enhance performance. However, if this

method is to be pursued, they are important systematic considerations that must be made with

regards to materials and processing techniques that will be discussed in this section.

2.4.1 Packaging

With the fundamental principle of Direct Liquid Cooling being to eliminate intermediate layers

of materials and provide more compact stacking structures, the matching of material properties

becomes a critical design issue. Comparing the material structure of direct cooling to that of

indirect cooling, presented in Figure 9, several key differences should be noted. With the

elimination of the base plate and thermal grease, due to its low conductivities, the electronic

module is directly bonded to the heat sink geometry. This is done using convectional solder,

brazing or pressure sintering [61].

Page 34: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

17

Figure 9: Comparison of Material Stacks for Varying Heat Sink Design Principles. a) Indirect

Cooling Structure and Associated Materials. b) Direct Cooling Structure and Associated

Materials. - [58]

The only major remaining material components in a direct cooling stack are that of the ceramic

substrate within the DBC/DBA board, used for electrical isolation, and the heat sink body itself.

The selection of these materials greatly impacts the thermal conductivity of the material stack.

However a much more important property to now consider is that of thermal expansion,

specifically the associated Coefficient of Thermal Expansion (CTE) values [62]. If proper

material selection is not carried out, the components will expand at different rates during thermal

cycling. This can cause thermo-mechanical stresses resulting in fatigue, internal cracking and

overall system degradation. The various CTE values for some common IPM materials are

presented in Figure 10.

Page 35: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

18

Figure 10: Coefficient of Thermal Expansion Values for Common IPM Materials. - [63]

A report compiled by Aranzabal et al. summarizes various commercial IPM systems being

implemented in current automotive designs and the innovative aspects that correspond to each

[53]. These include such models as: Toyota Prius (2004), Nissan LEAF, Toyota Lexus and more.

A summary of this information is seen in Table 2. The report discusses various packaging

aspects of IPM systems, identifying die attachment techniques and interconnection wiring. Most

importantly, the corresponding IPM ceramic substrate materials were discussed, being one of the

critical design areas that can affect the overall performance and lifespan of the modules.

Page 36: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

19

Table 2: Manufactured Power Module Packing Technologies. - [53]

Page 37: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

20

The early work of Romero et al. set out to evaluate the use of Metal Matrix Composite Material

(MMC) as a base plate/heat sink material for power electronics applications in leu of

conventional aluminum or copper. They found that after 4000 thermal cycles the MMC heat

sinks models (AlSiC in particular) offered much higher reliability over conventional options.

This was due to the low CTE associated with the material that corresponds well with that of IPM

ceramic’s, inducing less thermos-mechanical stress at the bond interface [64]. More recent

innovations have been made with regards to material analysis, such as that of Ivanova et al.,

achieving a 40% drop in IPM junction temperatures by utilizing a process that integrated micro

heat pipes directly into the DBC layering [65]. Weyant et al. determined that temperature

gradients could be reduced by an additional 50% by combining embedded heat pipe technology

with MMC heat sink designs [66]. A recent evaluation of industry trends, carried out by

Stockmeier, identified several areas at the forefront of IPM material stacking procedures. The

replacement of solder contacts by high pressure sintering, bond wires with weld contacts and the

elimination of base plate materials are all major areas of improvement [67]. Uhlemann and

Herbrandt investigated the use of aluminum clad materials for heat sink designs with the notion

that the raw material costs of MMC are too high for mass production applications. Al-Cu clad

materials offer the same matching of CTE values to the IPM ceramics but are more easily

available and machinable. When compared to basic aluminum designs, Al-Cu clad heat

exchangers provided a 10% reduction in thermal resistance and a 30% reduction in overall

weight [68]. Some have even gone to adjust the material structure of the ceramics themselves,

like Xu et al. who simulated a ladder shaped DBC arrangement capable of reducing thermal

stresses and plastic strains within the system during thermal cycling [69].

2.4.2 Micro Channel Heat Sinks

Since the emergence of electronic overheating, one of the most versatile and reliably high

performing methods of cooling has been that of mini and micro channel flow structures. The

basic concept of microchannel flow is to reduce the cross-sectional area of the fluid as is it

passes across the heat sink, improving the local convective heat transfer by increasing the

velocity of the coolant, reducing the development of thermal boundary layers and possibly

inducing turbulent flow [33]. These passages are formed by micro-machining extrusions in the

base plate of the heat sink. Two very common patterns, shown in Figure 11 and Figure 12 are

finned channels and pin-fin arrays [70]. These arrangements are commonly setup up in areas of

Page 38: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

21

high temperature gradients such that an even flow of coolant occurs across the width of the

pattern. Although microchannel heat exchangers are a widely established technology, there are

still several associated drawbacks with this method. Machining on this small of a scale can be

difficult or expensive if the required level of equipment is not readily available and the reduction

in cross sectional area can result in high pressure drops across fluid inlet and outlet locations.

Most importantly Thermal Runaway2 is a well-known issue when utilizing liquid cooled

microchannels, as the coolant further downstream of the inlet experiences higher temperatures,

which result in higher junction temperatures of any electronic devices located downstream [71].

This temperature differential between devices can lead to increase thermo-mechanical strain,

decreased efficiency and accelerated material degradation.

Figure 11: Micro Pin Fin Heat Sink Arrangement. - [70]

As mentioned in previous sections, much of the efforts made in this area of electronic cooling is

built off the pioneering work of Tuckerman and Pease, who found that straight forward

integration of compact 50um channeled heat exchangers could greatly reduce thermal resistances

for power dense IC packages [33]. Lee and Vafai compared the performance of microchannel

coolers to that of jet coolers for high heat flux dissipation and found that the microchannel design

options were much more preferable to small dimensional (<5mm2) applications [72]. An

innovative structure, developed by Harris, Despa and Kelly, combined microchannel liquid

cooling with cross-flow forced air to decrease thermal diffusion lengths and provide a function

similar to automotive radiators [73]. An extremely compact integrated design with low thermal

resistance of 0.087K/W was achieved by Steiner and Sittig who also found water to offer much

2Thermal Runaway: A thermodynamic phenomenon in situations where an increase in temperature changes conditions, which in

turn leads to further temperature increases, possibility leading to a destructive result

Page 39: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

22

better thermal performance as a working fluid when compared to dielectric fluids [74].

Analyzing the high accuracy of numerical simulations for single phase microchannel flow

structures, Qu and Mudawar also found that a basic fin analysis model, accounting for thermal

entrance effects, could provide reasonably accurate predictions when compared to experimental

results [75].

Figure 12: Straight Channel, Staggered Pin and MDT In-Line Pin Heat Sink Formations. - [76]

Parametric studies carried out by Zhang et al. determined that performance of microchannel heat

exchangers are greatly influenced by the channel width. This variable defines a significant trade-

off between thermal resistance and pressure drop. It was also found that the effect of base plate

thickness is minimized when aluminum material is replaced with more conductive substances

such as copper or diamond [77]. Building on this Kim investigated the use of different methods

for optimizing microchannel coolers, focusing on analytical fin and porous medium models as

well as a numerical three-dimensional approach. The findings indicated that the main design

variables for optimizing performance are channel height, width and fin thickness [78]. An

analytical and experimental procedure carried out by Peles et al. indicated that pin fin

arrangements provide better design flexibility and thermos-hydraulic performance over channel-

based designs. Moreover, decreasing fin length and increasing array density is preferable when

dealing with high Reynolds number flows [70]. An interesting study performed by Moreno et al.

evaluated the influence of Micro Deformation Technology (MDT) on the performance of

conventional microchannel geometries, presented in Figure 13. MDT is an advanced technique

that seeks to add small scale deformations to fin structures in order to increase the heat transfer

area and induce turbulence. Examples of this technique are provided in Figure 13 .It was found

Page 40: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

23

that the MDT pin array provided significant thermal improvements over the conventional

structures [76].

Figure 13: Demonstration of Fin Deformation Heat Sinks a) Micro Deformation Tools and

Process. b) Micro Deformed Heat Exchanger Array. - [76]

Although numerical simulations and analytical models have been found to provide reasonably

accurate predictions for flow and heat transfer in microchannel arrangements, a significant

challenge in this area has been that of flow visualization. However Sajith et al. were able to

attain valid optical measurements of a mini channel system using Mach-Zehnder Interferometry,

eliminating the need for flow disturbing thermocouples [79]. More recent studies have sought to

build on the cooling capabilities of microchannel designs by integrating them with other

advanced flow techniques. As was the case with Husain et al. who investigated the capabilities

of various hybrid designs incorporating different combinations of microchannel, pin fin and jet

arrangements. It was found that the inclusion of microchannels decreases the convective

effectiveness of jet swirl. Thus the most effective design, achieving acceptable thermal

resistances, high convective heat transfer and low junction temperature was the combination of

jet and pin arrays [80].

Microchannel cooling schemes have proven to be a highly utilized means of thermal

management in the field of power electronics. With regards to automotive applications, they

provide compact designs that can be easily integrated into the structures of IPM and IGBT

modules. It has proven to also be a flexible technology, open to a variety of optimization and

hybrid design techniques. The main drawbacks that result from this method of heat sink design

are with respect to manufacturing. Fabrication methods tend to increase in cost as the scale of the

channel dimension decrease [81]. Moreover, as flow areas are reducing factors such as surface

finish and roughness can start to have influence on the already high levels of applied pressure.

Page 41: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

24

Microchannels are a suitable option to any application that may require very high-power density

cooling at the tradeoff of cost and pumping power.

2.4.3 Jet Impingement Heat Sinks

A longstanding method of increasing the rate of convective heat transfer between fluids and

heats sinks has been the use of Forced Surface Impingement. The foundation of this method is

that higher turbulence can occur if flow is directed normal to the heated surface and the fluid can

impinge directly on the area of heat transfer. The area around the stagnation point then

experiences a higher level of heat transfer than if the flow was to pass parallel to the heat source

plane through a series of pins or channels [82], [83]. For many years this practice was almost

exclusively utilized for air cooling setups, but with the heat flux levels currently being reach by

power electronics it is becoming more commonly adapted to liquid cooling schemes. The

impingement flow is usually established through the use of small diameter jets or channels that

induce substantial pressure drops, forcing the liquid into a high velocity, turbulent trajectory

perpendicular to the heat source [84]. The style of impingement can be broken down into the

main categories of Free, Submerged and Confined impingement, as shown in Figure 14. These

are dependent on the arrangement of the physical structure as well as the allocation of liquid and

gas. Although confined impingement has be more prevalent in automotive applications, due to

the predicable and controllable nature of the associated flow, all methods offer high levels of

thermal dissipation with a variety of design variables that correspond differently to performance

[42].

Figure 14: Common Jet Impingement Arrangements. a) Free Surface Jet. b) Submerged Jet.

c) Confined Submerged Jet. - [42]

Liquid impingement systems have been seen to achieve some of the highest rates of convective

heat transfer in the industry. There are also several additional benefits associated with this

Page 42: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

25

technology. The use of liquid impingement means that the design complexity is usually limited

to the jet or channel structure, allowing the impingement surface to remain untouched. This can

eliminate the need for machined fins or pathways directly on base of the heat sink, or in some

cases, eliminate the need for a base plate altogether [85]. In addition, since heat transfer

completely takes place around the stagnation point of the fluid, the material used for the

impingement structure does not have to handle any of the heat flux. To clarify, the thermal

conductivity of the structure does not matter and could theoretically be made from any material

capable of supporting the flow of coolant [86]. However, all these benefits come at the trade-off

of internal pressure. Since the nature of this cooling method requires high pressure at the

impingement jets or channels, there are usually high pumping requirements associated with these

systems, which translates to larger parasitic loads.

An early report compiled by Jambunathan, Moss and Button analyzed the variables affecting the

local heat transfer occurred at the stagnation point of impinging flow. It was found that

geometric nozzle characteristics along with flow confinement, turbulent intensity and the

dissipation of jet temperatures are the significant factors when attempting to predicting local

Nusselt numbers [87]. The work of Gerimella and Nenaydykh, focused only on the influence of

nozzle geometry for submerged and confined liquid jet arrangements. It was found that the

highest heat transfer coefficient corresponded to smaller nozzle aspect ratios (length/diameter <

1), however the influence of these ratio values reduced as the displacement between the nozzle

and target surface increased [88]. Oliphant, Webb and McQuay carried out an experimental

comparison of common jet array and spray droplet impingement, finding that jet cooling,

dependent on flow velocity and array size, performed at equal levels to spray cooling, which was

found to mostly be dependent on mass flow alone. However, spray impingement was able to

achieve this level of heat transfer at significantly lower mass flow rates of coolant. It was

theorized that this was due to the formation of an evaporative film along the impingement

surface and the unsteady nature of thermal boundary layers experienced in spray impingement

[89]. The more recent work of Mertens et al. focused solely around the performance of spray

impingement, achieving an air-water cooling system capable of dissipating 825 W/cm2 of heat

from and IGBT module [90]. Bhunia, Chandrasekaran and Chen compared the cooling

capabilities of liquid micro-jet arrays to that of conventional air and cold plate technology. Not

only was a significantly low thermal resistance of 0.013 °C/W achieve for the liquid jet based

Page 43: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

26

heat exchanger, but the temperature variation between the various IGBT and Diodes being tested

was reduced to approximately 2°C [91]. As the practice has become more establish many

designs, such as the one implemented by Turek et al., have sought to combine impingement

arrangements with two-phase cooling approaches. The pressure atomized evaporative spray

cooling method they presented utilized high temperature fluid (100°C) as to reduce the

condenser size and requirements. The system was able to provide a 3.3 increase in the heat flux

output by the electrical converter under test while maintaining junction temperatures under

125°C [92].

Another example of hybrid cooling scheme was that pursued by Barrau et al. which combined

liquid jet impingement and microchannel design. It was found that the inclusion of the

microchannel structure provided higher rates of heat transfer and more even temperature

distribution across the heat sink at the cost of increasing overall pressure. The channel density

was also found to be the major scaling factor of the hybrid system with respect to Reynolds and

pressure values [93]. Hadziabdic and Hanjalic conducted an in-depth analysis on the influence of

liquid vortices cause by jet impingement on heat transfer. Through direct Eddie simulations the

different flow regimes were characterized and provided insights into the relationship between

stagnation, turbulence and resulting Nusselt values [94].

In recent years the many have noticed that liquid impingement can be an ideal cooling method to

keep up with the ever-increasing requirements of automotive power applications, offering

extremely high levels of thermal dissipation in the form of light, compact designs. Parida, Ekkad

and Ngo compared a variety of jet arrangements utilizing separation walls and different angles of

impingement, as shown in Figure 15 for the cooling of HEV/EV microelectronics.

Page 44: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

27

Figure 15: Impingement Based Heat Exchanger Designs. - [86]

It was found that the presence of center walls can increase conduction and assist in the formation

of liquid vortices, shown in Figure 16, caused be low impingement angles, which are found to

increase the convective heat transfer between the fluid and the heat exchanger [86].

Figure 16: Formation of Flow Vortices by Variations in Impingement Angles. a) Perpendicular

90° Impingement angle. b) 70° Impingement Angle. c) 45° Impingement Angle. - [86]

A design achieved by Morozumi et al. integrated the principles of Direct Cooling with

impingement technology to improve the reliability of an HEV power control unit while

simultaneously reducing overall size. Also, the use of an Sn-Sb solder was demonstrated to

account for the disproportional CTE values of ceramic substrates and aluminum heat sinks,

extending fatigue lifetime and improving the overall reliability of the module [95]. Gould et al.

found that a jet impingement cooling solution was ideal for the thermal management of power

Page 45: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

28

electronics in military hybrid vehicles, where harsh ambient conditions and high fluid

temperatures can apply. Using water-glycol coolant at inlet temperatures of 100°C at constant

flow rates, the jet impingement design was found to maintain junction temperatures at 169°C, a

substantial improvement other the commercial cold plate and microchannel models also tested

[96]. New innovations utilizing the compact nature impingement cooling are constantly being

developed, such as the “Shower Power” automotive cooling concept presented by Danfoss

Silicon Power, shown in Figure 17. This original design avoids the high pressure drops

commonly associated with impingement cooling by utilizes small-winding microchannel

passageway, attaining a uniform cooling across HEV/EV IPMs [97].

Figure 17: Danfoss Shower Power Cooling Design. - [97]

As automotive designs become more reliant on the performance of power electronics modules,

the trade-offs of pumping demands for high level heat flux dissipation and uniform temperature

distribution, offered by impingement cooling are becoming more appealing. Many see jet

impingement as a promising field for power dense, mobile applications.

2.4.4 Integrated Coolers

The continuous goal to reduced thermal resistance within power electronic structures has led to

one of the more complex but extremely high performing design concepts, referred to as Direct

Integrated Cooling. This cooling arrangement , depicted in Figure 18 , seeks to integrate the heat

sink structure directly into the DBC structure of a power electronic module [26]. This can

essentially eliminate all the conductive thermal resistances associated with any solder, TIM or

Page 46: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

29

heat sink materials that would overwise impede the flow of heat between the DBC and the

working fluid [26].

Figure 18: Direct Integrated Heat Sink Arrangement. - [26]

The implementation of this design structure requires the careful selection of materials and

assembly processes. A design by Colgan et al. achieved significant cooling levels of 300W/cm2

and over, using a microchannel design formed by integrated silicon fins. The final thermal

resistance associated with their prototype was 10.5 C-mm2/W [98]. A similar idea was proposed

by Tang et al. who sought to combine the ideal corrosion resistance properties of aluminum with

the superior heat spreading characteristics of copper. A hybrid substrate model was achieved

through a carbon nanofiber bonding process. Performance and reliability testing has yet to be

reported, but early thermal modeling suggests a high level of performance [99].

More recently, Jung et al. achieved experimentally validated embedded silicon microchannel

cooling design with the potential to extract up to 850C/cm2 of waste heat from vehicular power

electronics. Using single-phase water as a working fluid and a unique 3-D manifold design, it

was observed that one of the biggest areas of uncertainty is the accurate prediction of pressure

drop. This is due to the difference between “target” micro channel dimensions and “actual”

channel dimensions as a result of microfabricating [100]. Chen et al. were able to apply the

principles of integrated cooling into the area of thermal management for lithium-ion battery

packs. Combined with a custom optimization process, temperature reductions of ~1.87°C and

deviations of 0.35°C were attained for easily manufacturable design [101].

An innovative design recently developed by Erp et al. utilized a combination of metalized heat

spreading via integrated silicon micro channels and low-pressure liquid distribution via PMMA

Page 47: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

30

manifolds. The proposed system, shown in Figure 19, seeks to minimize the pumping

requirements of conventional liquid cooling systems while keeping volumetric sizes tight by

bringing the cooling liquid in close proximity to the active devices[102] .

Figure 19: Construction of MMC Heat Sink Prototype. - [102]

Although their currently exists a high trade-off between assembly costs and complexity, with the

rising cooling requirements of power electronics, DBC Direct Integrated heat exchangers may be

applicable for extremely power dense applications.

2.4.5 Double-Sided and Stacked Modularized Cooling

Another more recent straight forward but effective approach to high power density electronics

double sided and modularized cooling. By integrated a second heat exchange into the IPM stack,

as shown in Figure 20, thermal designer can effectively double the rate of conductive heat

transfer away from power electronics and HDD [103].

Figure 20: Double Sided Heat Sink Approach. a) Conventional Liquid Cooling Arrangement.

b) Double Sided Cooling Arrangement. - [104]

Page 48: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

31

This principle is reinforced by the findings of Wang et. Al who computational and

experimentally compared a conventional one-sided IGBT stack to that of a double-sided cooling

stack. The addition of the second liquid heat sink was found to increase system efficiently by

47.9% [105]. With the potential to greatly increase power density in IPMs, this style of design

has become popular in the field of automotive power electronics and EVs [106].

The concept of double-sided cooling allows for simpler and easily manufactured heat sink

geometries, such as macro pin fin arrays or mini channels, to be utilized for high power dense

applications. However, it should be noted that the application of double-sided liquid cooling

arrangements put substantial constraints on the corresponding electrical design. The mounting of

an additional ceramic substrate on the topsides of the electronics devices means that

conventional wire bonds can no longer be utilizes as electrical interconnects. The addition of this

new criteria heavy constraints the design and assembly of the and electrical layout [26].

The issue of electrical interconnects was investigated by Gillot et al. who sought to replace wire

bonds with flip chip solder bumps, allowing the associated IGBT devices to be mounted on both

sides to DBC substrates for double sided cooling. Through thermal simulations and experimental

testing it was found that the ‘sandwiched’ cooling structure provided a 76% increase in heat

dissipation as compared to a corresponding single-sided design [107]. Charbonneau et al. were

able to utilizes Embedded Power Packaging structures, similar to Figure 20b , to achieve the

electrical requirements of a proposed double-sided cooling scheme. Through experimental

analysis a 60% improvement in thermal performance was achieved [104]. A high performing

design was achieved by Schneider-Ramelow, Fraunhofer and Hoene by combining the principles

of Direct Cooling with a double-sided structure. A 40% increase in thermal resistance resulted

from the additional heat sink. It should also be noted that very low thermal resistances of ~0.08-

0.09 K/W were achieved with very low pumping requirements (<2 lpm) [63]. A different

approach carried out by Chang et al. used copper-clip packaged for the electrical connection of

wafer thin IGBTs between two liquid cooled heat sinks. The major finding were a 200% increase

in the power handling capabilities of the IGBTs, a 40% reduction in thermal resistance and a

260% increase in the predicated number of cycles due to the wireless connections [108].

As the technology around, double sided cooling becomes further establish, more inventive

structures are being developed. One of the major innovations in this area is the formation of

Page 49: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

32

Stacked liquid cooling schemes, shown in Figure 21. Utilizing the double-sided cooling

approach, multiple IPMs and liquid heat sinks can be bundled to provide extremely compact,

multipurpose, high performance electrical systems.

Figure 21: Demonstration of Double-Sided-Stacked Cooling Structure. - [106]

The benefits of this approach have been especially apparent in the power electronics sector of the

transportation industry, where many companies such as: Toyota, Denso, Alstom are developing

compact innovative designs based on multi-level stacked cooling [26]. As the automotive

industry continues to electrify, the highly compact nature of double-sided and stacked liquid

cooling may lead this to be a standard practice in years to come.

2.5 Design Optimization

Despite all the new technologies and methods presented in this section thus far, a major aspect of

electronic cooling design that cannot be ignored is that of performance and system optimization.

In a mass -production industry, such as automotive manufacturing, small improvements in size,

weight and performance can translate to major reductions in processing costs [109]. Many

optimization methods have been practiced over the years and are tailored to a specific

optimization goal. Common focus include: the physical design variables of heat exchangers, the

layout of heat sink geometries and fin/channel ratios, or the influence of controlled system

variables, such as the mass flow of coolant and the associated incoming fluid temperatures [110]

Page 50: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

33

2.5.1 Practical Techniques

Early work in field of heat sink optimization relied on the governing equations of fluid dynamics

as well as the principles of conduction and convection heat transfer. Analytical modeling

provided functional relationships between design variables and performance variables. Like the

major work of Knight et al., who developed dimensionless, generalized equations relating the

geometrical aspects of microchannel heat sinks to their thermal resistance values. This

methodology was applied to several existing designs and found to have significant improvements

on cooling performance [111]. Ritzer and Lau utilized thermal modelling and experimental

variations to optimize the performance of a laboratory chiller responsible for managing transient

heat loads. The basis of this optimization process was set on economic feasibility of the heat sink

under review. It was found that the bulky nature of passive air cooled heat sinks made them an

unsuitable option while high fin density heat sinks were found to be price competitive with low

density models while out performing all competitors with regards to thermal performance [112].

In a similar fashion, Lee developed analytical simulation models base on parametric analysis,

capable of predicting and optimizing the heat transfer capability of bi-directional heat sinks

within a confined configuration [113].

The optimization of thermal systems saw a major shift with the rise CFD analysis. These high

performing, reasonably reliable software packages allowed for the complete modeling and

numerical simulation of Multiphysics systems. This tool was quickly adapted into optimization

procedures. An early example is that of Lee, who was able to reduce the operating temperatures

of the IGBT devices to below 100°C associated with a given power module by simulating

various pin fin arrangements within the liquid cooling design [114]. A more advanced technique

was implemented by Li and Peterson, utilizing numerical simulations to attain a simplified three

dimensional conjugate heat transfer model capable of optimizing the geometric structure of

microchannel heat sinks [115]. Husain and Kim relied on FEA simulations to solve 3-D Navier-

Stokes and conjugate heat transfer equations for rectangular micro-channel heat sinks. The size

of the designs was then optimized for a constant heat source by a variety of surrogate models

which included: Response Surface Approximation, Kriging and Radial Basis Neural Network

methods. Although these models were found to provide slightly different geometric designs they

all predicted similar objective function values [116]. Thermal FEA analysis was also used by

Wang, Hung and Chen to model the heat transfer of a finned air-cooled heat sink working in

Page 51: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

34

tandem with a thermos-electric generator. A two-stage optimization process was carried out on

the design, first utilizing an analytical approach to determine the optimal fin-to-fin spacing, then

applying a compromised programming method to determine the maximum generator

performance at the cost of the heat sink performance when the heat sink volume was fixed. Final

results showed an 88.7% increase in the generators power density at the cost of a 20.93%

decrease in the heat sink efficiency [117].

2.5.2 Topology Optimization

With the increasing capabilities of advanced manufacturing techniques, such as CNC machining

and 3D printing an area of growing popularity within the field of automotive design is Topology

Optimization. This method seeks to move beyond simple size variables but instead optimize the

structural layout of a product or model within a predefined design space [118]. A common

application for this method is the volume or mass minimization of material within an object

while satisfying load or force requirements. This is useful when dealing with automotive

structural problems, such as vehicle chassis design as was the case with Mantovani et. al. Using

a two-stage lattice approach a system of truss structures was used to represent the automotive

chassis and achieve an admissible stress level of 50MPa [119]. A more complex approach,

introduced by Yang et al. utilized a multi-step topology optimization frame work with adaptive

and varying design domains to automate the layout of spot welds within complex automotive

structures [120].

Topology optimization has also been utilized in the field of electronic cooling. Improvements to

the topology of fin density, as shown in Figure 22, keep electronic heat sources more isothermal

while decreasing total heat sink mass [121].

Page 52: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

35

Figure 22: Heat Sink with Increased Fin Density Topology. - [121]

One of the more radical works conducted by Alexander et al., demonstrated the application of

density-based topology optimization to the design of three-dimensional heat sinks under natural

convection [122]. The Multiphysics system coupled to this procedure was solved using stabilized

trilinear equal-order finite elements with an order of 20-330 million state degrees of freedom.

This allowed for the optimization of large scare problems and the generation of custom, novel

design iterations as depicted in Figure 23.

Figure 23:Optimized Heat Sink Designs with Constant Gr Value and Mesh Size of 329 x 640 x

320 Elements. - [122]

Page 53: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

36

2.5.3 Genetic Algorithms

As the computation tools for modeling and predicting the behavior of hydraulic-thermal systems

improve, so to do the optimization techniques involved with the design process. Of these

techniques, Genetic Algorithms (GA) are being established as one of the more advanced and

promising methods. GAs are a powerful tool, utilizing stochastic search and optimization

procedures to provide optimal solutions across a wide variety of applications [123]. This

optimization process follows the principles of organic evolution and does not require the rate of

objective functions to be determined, which can be a major benefit when dealing with

computational simulations [124]. The general procedure of this optimization method, as shown

in Figure 24 starts with the randomized generation of an initial population. After the performance

of population is evaluated based on the designated objective function and the high and low

performing models are identified. A ‘Survival of the Fittest’ approach is then implemented as the

poor performing members of the initial population are then discarded and new models are

formed by applying variations or ‘Mutations’ to the surviving models. A new population is then

generated for evaluation and this process is iteratively continued until an optimal model is

produced.

Figure 24: Basic Genetic Optimization Workflow. - [125]

This procedure has already begun to be implemented in the design optimization of electrical

cooling systems. An early evaluation of this method was conducted by Fonseca and Fleming

found that GAs ability to cope with discontinuous and noisy functions made them an ideal tool

for multi-objective optimization, predicting a promising future with engineering applications

[126]. Husain and Kim utilized a Response Surface approximation surrogate analysis working in

conjunction with an evolutionary algorithm to optimize the channel depth and fin width of their

Page 54: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

37

heat sink design with the objective functions related to require pressure and resulting heat

transfer. This achieve low thermal resistances of 0.09 – 0.08°C/W [127]. A different approach

was taken by Xie, Sunden and Wang, by employing a GA procedure to optimize the structure of

a compact plate-finned heat exchanger on the basis of minimizing total volume and annual cost.

When ignoring pressure drop constraint a 49% volume reduction was achieved, however when

pressure was a accounted only a 30% reduction was produced [128]. A non-dominated sorting

GA was applied by Sanate and Hajabdollahi for the multi-objective optimization of a common

plate fin heat exchanger. Several design variables, fin pitch, height, offset length, cold stream

flow length, no-flow length and hot stream flow length were used to provide an optimal thermal

performance at minimal annual cost. Further sensitivity analysis on the resulting, or Pareto,

models provided more insight into the functional relationship between the design parameters and

the multi-objective function through the optimization process [129].

More recent work has begun to pair the structure of Gas to not just size variable optimization but

to the more complex task of topology optimization. Wu, Ozpineci and Ayers sought to achieve

an optimized liquid cooled heat sink through a two stage GA based design process with FEA

simulations via COMSOL. The first stage, shown in Figure 25, implemented the conventional:

Initialization, Evaluation, Selection, Mutation and Reproduction iterative process, while the

second stage applied more refined perturbation mutations, as shown in Figure 26. With a 15%

improvement in heat transfer, it should be noted that findings of the study indicated designs

achieved through this method may only be attainable through 3D printing procedures due to their

complex nature [125].

Figure 25: First Stage Conventional Genetic Optimization.

Page 55: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

38

Figure 26: Second Stage Perturbation Genetic Optimization. - [125]

An innovative design process achieved by Bornoff et. al incorporated the fin topology layout

into a generative optimization procedure for an automotive audio amplifier. Mass reductions of

18% were achieve while maintaining thermal performance [130]. Wu et al. also utilized the

principles of genetic algorithms, coupled with FEA simulations to produce a complex 3-D

printed heat sink model, shown in Figure 27, specifically optimized for air cooling a 50kW

inverter [131].

Figure 27: Two Stage Genetically Optimized Air-Cooled Heat Sink. - [132]

Page 56: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

39

It is clear that GAs can provide a computationally effective solution to both single and multi-

objective optimization procedures. The automatic nature in which they operate makes them an

ideal design tool in the continual effort to improve the cost, size and performance of integrated

cooling structures.

Page 57: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

40

Chapter 3

Methodology and Construction of Optimization Program for Liquid Heat Sink Topologies

As discussed in Chapters 1 and 2, the growing thermal requirements of power electronics are

beginning to push to bounds of conventional heat sink technologies. If automotive designers

wish to keep up with the rapid increases in power density and electrical performance more

modern approaches need to be taken in order to achieve innovative solutions to the issue of

electronics cooling. With the availability of advanced production techniques, like additive

manufacturing, topology optimization is an ideal area of focus for those looking to achieve the

highest level of heat transfer within a heavily constrained design space, as is the case with on-

board EV applications.

In this chapter we present the creation of a novel design approach, utilizing the optimization

techniques of genetic algorithms, to develop compact high performing heat sink structures. Code

generated in MATLAB is used to form an iterative design loop, using arrays of binary data

points to represent heat sink geometries. Using the Ansys AAS toolbox, the MATALB code

uploads generated designs into the ANSYS workspace, taking advantage of its powerful

simulation capabilities, to evaluate potential heat sink geometries. This iterative optimization

process learns as it cycles through groups of design candidates, producing a final heat sink

geometry, specifically tailored to the thermal profile of the associated electrical system.

The layout of this chapter, depicted in Figure 28, is as follows: Section 3.1 discusses the

formation of the Genetic Optimization Logic and its associated functions, constructed in

MATLAB and implemented on simple two-dimensional flow geometries simulated in ANSYS

Fluent. Section 3.2 provides an overview of ANSYS Icepak and its unique abilities to model and

simulated electronic thermal assemblies and goes on to detail the integration of ANSYS Icepak

into the programming structure and the required changes to optimization procedure to account

for three-dimension liquid heat sinks topologies. Section 3.3 concludes this chapter with an

overview of the final genetic optimization process and the potential of this custom procedure.

Page 58: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

41

Figure 28: Content Breakdown for Chapter 3.

3.1 Stage I: Genetic Optimization Logic

The optimal heat sink topologies discussed in this chapter are attained via binary genetic

optimization [133], [134]. By applying this method to principles of topology optimization,

design of a structure can be broken down into a set of 2D pixels or 3D voxels, which are

represented by bit arrays of equal size [135]. The binary values within these arrays are assigned

to different materials, such that any alteration, manipulation or morphology carried out on the

data matrix directly controls the allocation of said materials throughout the physical design.

While the majority the of functions and operations responsible for this custom procedure, laid

out in shown in Figure 29, are implemented in MATLAB, ANSYS is used to simulate each heat

sink design and evaluate the corresponding cooling performance for an applied heat load.

Page 59: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

42

Figure 29: Structure of Genetic Optimization Process.

The following sections detail how each stage of the proposed genetic optimization loop were

formulated into MATALB functions and brought together to construct the foundational genetic

logic of the optimization program. This initial version of the program, largely drawn from [136],

dealt only with two-dimensional matrices (MB), representing the workspace of simple planar

Page 60: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

43

flow geometries. This approach kept simulation run times low and simplified geometry

visualization, as shown in Figure 30, allowing for better characterization of the initial logic.

Figure 30: Visualization of Structural Bit Array.

3.1.1 Constructing Model and Defining Workspace

When first defining the model for optimization, reducing the complexity of the overall system

and simplifying the associated geometry is the key to achieving simulation efficiency without

sacrificing system accuracy [137]. In this stage, designers must attempt to simplify the heat sink

model into the components that have direct influence on the performance variables in question.

As with conventional topology optimization [135], the physical bounds of the heat sink must be

defined within the simulation space as well as the “active” and “reserved” regions in relation to

the design process, as depicted in Figure 31a. In the case of planar geometries, the reserved

regions commonly take the form of inlet and outlet regions as well as heat sources acting as

pseudo devices or electronic ‘chips’. However, when dealing with more complex systems and

three-dimensional models the formation of this workspace and the allocation of these active and

passive design regions becomes more complex, as will be discussed in a later section.

Once these regions are defined, the optimization workspace are subdivided into a grid of

structural pixel elements much like Figure 31b. The size and shape of these elements are

controlled by the designer, however in this work, only square elements are used to discretize the

active workspace. This is done to simplify the grid indexing, as described in the following

section. It is important to note that the geometric nature of the workspace elements may be

constrained by both computational cost and manufacturability [17]. Increased discretization leads

Page 61: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

44

to smaller element size, which in turn can increase the cost or complexity of the manufacturing

techniques associated with creating the generated features [138].

Figure 31: Defining the Design Optimization Workspace.

This initial Partitioned Workspace (MP), model acts as a form of blank geometry for the

optimization to work with. Being a part of the programs Initialization Stage, this step is only

executed once and ideally is carried out in a modelling software prior to the activation of the

optimization program.

3.1.2 Grid Indexing

The geometric information of a candidate heat sink geometry must be input to Ansys. This is

achieved as follows. Once the designer constructs the initial heat sink model, the Genetic Logic

extracts the geometric information of the structural elements (ΔxPixel, ΔyPixel) as well as the

optimization workspace (XWS, YWS) to create an array of equal size, as shown in Figure 32. The

dimensions of the array are dictated by via (Eq. 3.1) and (Eq. 3.2). The cells within this array are

sequentially indexed, forming a Global Array (MG) of values, each representing an element

within physical model.

Page 62: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

45

Figure 32: Created Global Array.

𝒏 = 𝒀𝒘𝒔

𝜟𝒚𝑷𝒊𝒙𝒆𝒍

(3. 1)

𝒎 = 𝑿𝒘𝒔

𝜟𝒙𝑷𝒊𝒙𝒆𝒍

(3. 2)

The partitioned workspace (MP), developed in the previous section, is also indexed in a similar

fashion as shown in Figure 33. This is done by first counting the number of active elements to

form a size vector (xS) and then mapping that vector back to the positioning of the elements

within MP. This process assigns an element label to each of the structural cells within MP based

on their associated coordinates and positioning within the physical workspace.

Figure 33: Assigning Element Labels to Partitioned Workspace.

Page 63: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

46

The formation of the Global Array (MG) and the indexing of the Partitioned Workspace (MP)

allows the program to tie the element label of the structural cells to a value within MG as depicted

in Figure 34. This is a vital step moving forward with the genetic design process. It allows the

solution vectors generated by the Genetic Logic to be easily converted to Bit Matrices (MB)

consistent with the MP structure.

Figure 34: Linking Element Labels to Global Array.

3.1.3 Mutate Seed Design

Before the automated design process can start, an initial group of ‘Seed Designs’ must be

generated. The designer must input an initial geometric heat sink design for the program to

build from. This first model may, for example, represent the best engineering design, achieve by

the user and the desired structure to be optimized. For example, in the case of electronic heat

sinks, this could take the form of micro channeled heat exchangers [79], [80]. This base Seed

Design acts as a good ‘starting point’ for the optimization process and gives an initial

measurement of performance for future comparison.

The program then creates an initial group or ‘Population’ of design candidates, by inducing

random changes in the Seed Design as shown in Figure 35. Both the quality of the starting design

and the size of the genetic Population can have significant influence on the performance of the

overall optimization process.

Page 64: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

47

Figure 35: Seed Mutation.

These design candidates, or ‘Individuals’ take the form of solution vectors (xFE) containing the

values of elements within the workspace to be allocated as ‘Fluid Blocks’. By relating xFE back

to MG, as shown in Figure 36, these vectors are converted to the structural bit arrays presented

earlier.

Figure 36: Utilizing Solution Vector and Global Array to Form Structural Bit Matrix for New

Individual.

Page 65: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

48

3.1.4 Validation

Due to the randomized nature of this genetic inspired optimization process, it is important to

verify the geometric feasibility of generated Individuals and refine them if necessary. Each

design generated by this Seed Mutation stage undergoes a validation check, to ensure that the

generated structure is a feasible solution for the given topology problem. For the case of liquid

heat exchangers an ‘Invalid’ design may take the form of candidates with blocked flow channels,

unconnected to the defined Inlet and Outlet boundaries, as demonstrated by the design shown in

Figure 37. Treating each MB as a binary image, these bodies are easily identified via ‘Blob

Analysis’ and with extensive image processing capabilities within MATLAB, many options exist

for modifying these invalid designs.

Figure 37: Validating New Design Candidate with Blob Analysis and Image Morphology.

Once a blocked channel is identified, the designated functions deal with the Invalid aspects of the

designs. For this initial study the program first identifies isolated liquid bodies causing design

issues and applies the following conditions:

1. If the body is found to connect to only one boundary (either the Inlet or Outlet), it is dilated

horizontally until it either meets the missing boundary, forming a full channel, or connects

to an existing channel network.

2. If the body is found to have no boundary connections, the ‘Floating Body’ is removed from

the binary structure.

Page 66: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

49

This Initialization loop of Seed Design Mutation and Validation continues to produce design

candidate until an initial Population of Seed Designs is created, similar to the one demonstrated

in Figure 38. This set of design options marks the end of the programs initialization function and

move into the main optimization process, as discussed in the following sections.

Figure 38: 2D Example Population of Initial Design Candidates.

3.1.5 Evaluation

Once a full population of initial design candidates is established, each Individual is sequentially

evaluated. To test the performance of each potential heat sink design stored in the xFE solution

vectors, this program utilizes the modeling and simulation capabilities of the ANSYS

Workbench software suite. Specifically, the CFD based program Fluent is used to model the flow

and heat transfer associated with these active cooling systems. Establishing valid CAD

geometries and accurately predicting the temperature gradients of each design within a

generation is essential for identifying candidates with optimal design characteristics. By relating

the fluid elements stored in the solution vectors to the layout of the physical workspace, the

program can associate the resulting thermal performance to the data structures within the genetic

logic.

To accomplish this, the optimization program creates a Script File for each potential design

candidate, based on the topology information provided by the associated xFE vector and updates

Page 67: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

50

the corresponding structural model in ANSYS Design Modeler (DM). A Simulate function,

executed through the AAS Toolbox, sets the boundary conditions and materials properties to

Fluent. This file structure is demonstrated in Figure 39.

Figure 39: File Structure of Evaluation Stage.

Once the model and simulate space is set for the given design, ANSYS calculates the device

temperatures (TD) along with the pressure drop across the inlet and outlet flow faces (ΔPInlet-Outlet)

through CFD Multiphysics simulation, as shown in Figure 40.

Figure 40: Evaluation Stages. a) Candidate Solution Vector. b) Representative Bit Array.

c) ANSYS CFD Model.

The results obtained from Fluent are used to assign a Fitness Score to each Individual,

represented by (Eq. 3.3). The Fitness scoring is controlled by user defined weighting factors a

Page 68: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

51

and b, dictating the importance of minimizing the device temperatures to that of the pressure

drop.

𝑭𝒊𝒕𝒏𝒆𝒔𝒔 = 𝒂(𝑻𝑫𝒆𝒗𝒊𝒄𝒆𝒔) + 𝒃(∆𝑷𝒊𝒏𝒍𝒆𝒕−𝒐𝒖𝒕𝒍𝒆𝒕) (3. 3)

The population is then reorganized, ranking the Individuals from best (lowest fitness) to worst

(highest fitness). A ‘Breeding Probability’ is assigned to each candidate based on their rank as

demonstrated in Figure 40. This percentage value dictates how likely each candidate is to be

chosen as a ‘Parent’ for the new generation of designs, as discussed in the following section.

Figure 41: Evaluation Process. a) Generated Bit Arrays. b) Converted to ANSYS CAD

Structures and Simulated in FLUENT. c) Ranked and Assigned Breeding Probability.

Page 69: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

52

3.1.6 Selection, Crossover and Mutation

Here we discuss the operation of the Genetic Algorithm to generate new solutions from previous

simulation results and candidate heat sink designs. In order to continually improve the genetic

pool of design candidates, ranked solutions are ‘Bred’ together in order to generate new

Individuals or ‘Child’ designs. Parents are randomly selected for Breeding, such that the higher

performing solutions are more likely to be chosen, as discussed in the previous section. Here,

designers can specify the desired ratio of Parents to a Child. Each Parent is split into a series of

‘Chromosomes’ which consist of a specified number of adjacent pixels. These strings of pixels

or ‘Genes’ from the associate design matrix, MB carry binary values of the Parent design,

indicating either a ‘fluid’ or ‘solid’ element being transferred to the new Child design. Each

parent has an equal probability of passing along a Chromosome string to the Child. This

Breeding process is repeated until a full Binary Design Matrix (MB) is constructed for the Child.

Once all Genes have been allocated and a new design is formed, the Child undergoes

“Mutation”. For this process, each element in the new design matrix has a small probability of

changing state. The role of this random Mutation stage is to induce a certain level of randomness

and “Genetic Diversity” throughout the duration of the design process. However, due to the

randomized nature of this gene-splitting process, each Child also undergoes the same validation

process as outlined in Section 3.1.4. This avoids the creation of blocked flow channels and

invalid geometric regions that may be formed by this breeding process. The entire formation,

mutation and validation of these Child designs is represented in Figure 42 below

Figure 42: Generating Child Design.

This process of Selection, Crossover, Mutation and Validation repeats until a new Population of

design candidates, similar to the one shown in Figure 38 is formed. In order to avoid objective

regression, the highest-ranking design from the evaluated group, or the ‘Elite Individual’ is

Page 70: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

53

carried over into the next Population. The designer can also choose to include a percentage of

completely randomized designs into each new Population. This can further increase the Genetic

Diversity of the design gene pool, but at the cost of computational efficiency [139]. Once a new

Population if formed the program then cycles back to the Evaluation Stage, repeating all the

functions within the Optimization Loop continuously until an optimal design is achieved, as

discussed in the following section.

3.1.7 Convergence and End Process

The Evaluation, Crossover and Reproduction stages are repeated until the convergence

conditions of the GA algorithm are met. Here the designer specifies both a maximum number of

iterations and a convergence condition. Figure 43 shows a typical convergence plot for an

optimization process, providing the designer with the best fitness evaluated during each

generational iteration and the corresponding pressure and temperature values of the design.

Figure 43: Convergence Window with Objective Tracking.

Many options exist for convergence criteria and are easily interchangeable. Designers decide

which suits the desired goals of the specific design optimization. Common end conditions can

include targeting desired temperatures or setting minimum changes in design improvements

between iterations [16]. Some designers may choose to assign a maximum computation cost to

the process (maximum iterations) with no stop condition in order to avoid a premature

convergence with a local optimal instead of a global [25]. Once the parameters defined in this

Page 71: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

54

stage are met, the highest-ranking Individual in the final generation is output as the optimized

heat sink design.

3.1.8 Preliminary Results

In order to assess the functionality of the initial Genetic Optimization Logic outlined in the

previous sections, several preliminary optimization trails were run on a 60mm by 60mm

aluminum heat sink design space, similar to Figure 31 . The corresponding physical layout and

operating conditions are as follows: left side inlet, right side outlet, 0.1 m/s flow rate and a single

centralized SiC heat load or ‘Chip ‘dissipating ~50 Watts. The mutation rate was set to 5%, with

two parents per child each donating full width chromosome strings. Figure 44 presents the

evolution of the constructed design process applied to a mini channel starting model.

Page 72: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

55

Figure 44: Genetic Optimization Process on a 2mm Grid Mini-Channel Design. a) Convergence

Plot. b) Intermediate Designs. c) Final Thermal-Flow Contours.

Page 73: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

56

This example demonstrates an improved performance score from 472 to 325 over the course of

70 genetic iterations, with the final design achieving a maximum device temperature of ~322 K

with a corresponding pressure drop of 15 Pa. Analyzing the progression of the design evolution,

the Genetic Logic appears to first eliminate central fluid channels in close proximity to the heat

source. In turn this surrounds the chip with a large area of aluminum, allowing the heat to spread

away from the load due to the higher conductivity of the metal. This initial transformation leads

to significantly lower operating temperatures. The program then proceeds to slowly grow the

outer fluid channels back inwards, altering the structure of the central aluminum body. This

shape morphing is seen to gradually allow more fluid flow, reducing the system pressure yet

creating higher velocity channels that quickly remove heat from the outside edges of the

aluminum.

The allocation of solid pixels around the heat load and the formation of this central aluminum

heat spreader is found to be a reoccurring trend with the genetic design process. Figure 45

depicts how a different starting design, with similar physics yet a finer grid and larger population

size arrives to a comparable solution. The shape and volume of the metal structure tends to vary,

but the principle design traits remain the same. A central aluminum block is formed around the

chip with surrounding coolant channels.

Page 74: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

57

Figure 45:Genetic Optimization Process on 1mm Grid Central Channel Design. a) Convergence

Plot. b) Intermediate Designs. c) Final Thermal-Flow Contours.

Page 75: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

58

The continual reappearance through various trials indicates that this arrangement represents a

Parento-optimal solution for the given design problem. Another point of interest is that this

centralize metal heat sink shape is comparable to other genetically inspired designs found in

literature [140], like the ones depicted in Figure 46.

Figure 46: Additive Topologically Optimized Heat Sink. a) Smoothed 3D View. b) 2D Profile. -

[140]

It is also noted that final designs provided by the program can be somewhat unstructured. Post

processing contours can also be useful to clearly indicate critical bodies within the design

structure in the event that user refinements are required to achieve a more simplified,

manufacturable final product.

The preliminary optimization trials also indicate certain system variables have a noticeable

impact on the performance of the genetic optimization procedure. Smaller cell sizes applied to

the workspace section of the model were found to lead to faster convergence and better fitness.

This is assumed to be a result of the increased discretization allowing for a wider variety of

potential design solutions and the finer grid formation producing thinner channels to increase the

rate of heat removal. The user specified population size was seen to represent a clear

fundamental trade-off between performance and computational efficiency. Faster convergence

and better fitness result from larger populations of design candidates at the cost of processing

time. Increasing the size of design options for each iteration raises the chance a Parento-optimal

result will be generated during a given generation. However, this comes at a high computational

cost, drastically increasing the potential of ‘inferior designs’ being generated and evaluated

during each iteration, which add little benefit to the optimization process. These traits help

characterize the constructed design process and identify key parameters that can be fine-tuned in

order to achieve the maximum design improvements.

Page 76: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

59

3.2 Stage II: Integrating Three-Dimensional Liquid Heat Sink Topologies

The coding developed in the previous section laid the framework for a novel optimization

process base on the principles of Genetic Algorithms. However, the initial models used to

represent potential heat sinks designs do not accurately capture the complexities of three-

dimensional flow or the intricacies of advance electronic systems. In order to develop a program

better suited to the design of real electronic heat sinks, a more powerful simulation tool was

needed to pair with the existing genetic loop. The AAS Toolbox and scripting communication

established by the existing code bridged the gap from ANSYS to MATLAB, allowing automated

access to any application within the Workbench. This allows the Genetic Logic to access a

variety of FEM and CFD based simulations tools, including those more equipped to handle

electronic systems.

This move towards more advanced, real world models, lead to the integration of ANSYS Icepak,

a software application specific to the thermal and fluid flow analysis of electronic assemblies. A

variety of new modeling techniques and unique design capabilities were made possible by the

inclusion of this new simulation tool, discussed in the following section. However, moving the

optimization process from simple two-dimensional geometries to more complex three-

dimensional structures required several alterations to the existing code, which is detailed in

Section 3.2.2.

3.2.1 ANSYS Icepak

Modern power electronic devices have compact packaging and rigid temperature constraints that

require the design of superior thermal management systems to ensure reliable performance and

avoid product degradation or failure. Accurate computational modeling is the key to achieving

and validating these thermal designs without the costly process of repeated experimental

prototyping.

ANSYS Icepak is a CFD based simulation tool within the ANSYS workbench, specific to

electronic devices and electromechanical systems. Usually tied to a CAD modeling tool, as seen

in Figure 47, Icepak allows for the thermal and flow analysis of complex design characteristics,

unique to compact electronics.

Page 77: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

60

Figure 47: Typical Icepak Workflow.

Utilizing a FLUENT based CFD solver, Icepak calculates Conduction, Convection and Radiation

heat transfer with efficient accuracy. The automatic Unstructured Hex-Dominant meshing allows

the software to capture small geometric details such as thin solder layers, devices packages and

other small components. Non-conformal meshing tools allow designers to separately mesh and

store critical areas within system designs, making it ideal for design optimization and running

iterative simulations. Due to the unique features it offers, Icepak is becoming an increasingly

populator tool for thermal and electrical engineers for design validation [141], [142]. A typical

progression of an Icepak design analysis is present in Figure 48 with a simple half bridge

converted constructed on a ceramic DBC module and a ruffled fin heat sink.

Page 78: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

61

Figure 48: Example Icepak Project on Half-Bridge DBC Converter Module.

Icepak also contains more advanced modelling capabilities for complex objects such as

semiconductor packaging [Figure 49], or ECAD importing for trace layers and copper vias

[Figure 50]. Key design components, such as these, can heavily influence the thermal profile of

Page 79: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

62

the respective system. Accurately capturing the heat transfer throughout the design structure is

essential for the genetic algorithm to optimally allocate liquid channels. For these reasons,

ANSYS Icepak was chosen as a simulation tool for evaluation of the heat sink topologies

generated by the genetic optimization process.

Figure 49: Icepak Semiconductor Package Design.

Figure 50: Icepak ECAD Import Structure.

Page 80: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

63

3.2.2 Changes to Modeling Procedure and Genetic Functions

While much of the code established in Section 3.1 remained valid, several key areas of the

genetic logic had to be adjusted in order to compensate for the three-dimensional models and the

incorporation of Icepak evaluations into the programming structure.

3.2.2.1 Initial Modeling

The main area of focus for the new process centers on the initial stage of Model Construction

and Workspace Initialization, presented earlier in Section 3.1.1. With the move towards more

complex geometries, the modeling procedure in turn is more complex:

1. Starting with an existing Electrical Layout , the users adds three main additional regions:

Optimization Workspace, Heat Sink Shell and Inlet/Outlet Bodies [Figure 51].

Figure 51: Workflow of Model Construction.

2. The Optimization Workspace is then partitioned into a series of three-dimensional voxels

[Figure 52], with user defined dimensions corresponding the desired height, width, and

depth of the liquid channels

Page 81: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

64

Figure 52: Partitioning of Optimization Workspace with Three-Dimensional Voxels.

3. Each voxel is assigned a sequential label, dictated by its positioning within the

Workspace [Figure 53].

Figure 53: Indexing Workspace Elements.

4. Design Modeler Scripts are used to active Boolean functions [Figure 54a], which select

the ‘Liquid’ voxels to be combined with the Inlet/Outlet bodies to form the fluid domain

[Figure 54b], while the remaining voxels are combined with the Heat Sink Shell to form

the solid domain [Figure 54c]

Page 82: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

65

Figure 54: Formation of Heat Sink Topology. a) Allocation of Structural Elements via Design

Modeler Boolean Functions. b) Forming Fluid Domain. c) Forming Solid Domain. d) Initial Seed

Model.

5. The user then transfers the CAD structure into an Icepak project, setting the desired

operating conditions, material properties and meshing parameters in the local GUI

[Figure 55]

Figure 55: Setting Simulation Conditions in Icepak GUI

6. Lastly, the user selects the desired performance values in Icepak they are seeking to

optimization [Figure 56a], defining them as Output Variables, with the final Workbench

structure taking the form of Figure 56b.

Page 83: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

66

Figure 56: Defining Output Parameters. a) Icepak GUI. b) Variables Exported from Workbench

Project as CSV File.

Unlike the Stage I program, this modeling procedure is carried out ahead of the optimization

process. This allows designers the flexibility to carry out intricate studies on the initial model,

such as mesh independency, sensitivity analysis or even experimental validation. Once a suitable

architecture is achieved, it is exported and saved as a Workbench (.WBPZ) file.

3.2.2.2 Grid Indexing

Upon activation, the genetic programming launches ANSYS in server mode and loads the

corresponding Workbench Project file, constructed in the previous section. The voxel and

workspace data is extracted in a similar fashion as demonstrated in Section 3.1.2, to construct a

three-dimensional Global Array of indexed values, depicted in Figure 57.

Page 84: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

67

Figure 57: Formation of Three-Dimensional Global Array.

3.2.2.3 Mutate Seed Design

The Seed Mutation process, outlined in Section 3.1.3, is carried out on the 3D model. Binary

voxel values within the optimization workspace are randomly altered at a low probability to form

a starting Population of Design Candidates, like the one shown in Figure 58.

Figure 58: Mutating the Three-Dimensional Workspace to Achieve New Fluid Domain.

3.2.2.4 Validation

The same binary validation methods presented in Section 3.1.4 are applied to the new three-

dimension liquid topologies to ensure that no blocked flow channels or floating solid/liquid

bodies are generated by the mutation process.

3.2.2.5 Evaluation

With the integration of Icepak into the design Evaluation, small alterations were made to the file

structure of this stage. As was the case in Section 3.1.5, new design candidates generated by the

Page 85: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

68

Genetic Logic take the form of solution vectors (xFE) containing the element values to be

allocated as fluid blocks. When evaluating a new Individual, the MATLAB code uses the

information within the solution vector to write a corresponding DM script. The script structure

first clears the previous topology by deleting the existing Boolean functions [Figure 59a,b],

mentioned in Section 3.2.2.1, responsible for forming the Fluid and Solid Domains. The file then

implements a new set of Boolean functions, selecting the fluid elements provided by xFE [Figure

59c] and combining then with the Inlet and Outlet bodies to form the new Fluid Domain. As

before, a second Boolean combines the remaining voxels with the Heat Sink Shell to form the

Solid Domain [Figure 59d].

Page 86: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

69

Figure 59: Operations for Generating New ANSYS Models. a) Starting with Previous Design. b)

Clearing Liquid and Solid Boolean Functions. c) Selecting New Fluid Elements. d) Reapplies

Boolean Functions to Generate New Design Topology.

Since the Workbench project and Icepak conditions are defined prior to the start of the genetic

program [Section 3.2.2.1] and activated during the Indexing Stage [Section 3.2.2.2] there is no

Page 87: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

70

need to resend the simulation parameters to ANSYS. Once the new topology is successfully

uploaded and formed in DM, the MATLAB code simply sends a command to update the entire

workbench process, with new heat sink geometry. With the Icepak parameters already set, each

Design Candidate will be automatically meshed and evaluated under the same conditions.

Once Simulated, the output parameters set during the Initial Modeling Stage [Section 3.2.2.1] are

exported in a CSV document and read into the MATALB coding structure. The performance of

each design is evaluated on the basis of a new Fitness Function, defined by (Eq. 3.4).

𝑭𝒊𝒕𝒏𝒆𝒔𝒔 = 𝒂 (𝑻𝑫𝒆𝒗𝒊𝒄𝒆𝒔

𝑻𝑫,𝑩𝒂𝒔𝒆) + 𝒃 (

∆𝑷𝒊𝒏𝒍𝒆𝒕−𝒐𝒖𝒕𝒍𝒆𝒕

∆𝑷𝑩𝒂𝒔𝒆) (3. 4)

The Temperature and Pressure values of the initial seed design (TD,Base, ΔPBase) are integrated into

the exiting Fitness Function (Eq 1.3) to normalize the output variables of each Design Candidate.

This allows the program to measure and track the design improvements induced by the genetic

process using the starting model as a base reference.

3.2.2.6 Selection, Crossover and Mutation

Once an entire Population of Design Candidates is evaluated and ranked, a new Generation of

Child designs is formed, following the procedure laid out previously in Section 3.1.6.

3.2.2.7 Convergence and End Process

Criteria for convergence of the new optimization program follow the same principles previously

described in Section 3.1.7. With the changes to the fitness scoring [Section 3.2.2.5], the new

convergence tracking window takes the form of Figure 60. Both pressure and temperature values

for the highest performing design in each Generation are still displayed for user convenience.

However, the fitness scores are presented as a factional value in comparison with the starting

design performance score of 1.

Page 88: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

71

Figure 60: Stage II Convergence Window with Objective Tracking.

Unlike the Stage I program, once optimization convergence is met, the Stage II program outputs

a full three-dimensional model in STEP format. This allows designers to continue analysis or

refinements on the resulting geometry, which make for easy prototyping and experimental

validation of the optimized structure.

3.3 Overview of Genetic Optimization Process for Three-Dimensional Liquid Heat Sinks

By combining the custom GA structure, presented in Section 3.1, with the computational power

of ANSYS Icepak, detailed in Section 3.2.1 and the changes presented in Section 3.2.2, a final

optimization program was achieved, capable of analyzing three-dimension liquid cooled power

electronic systems and intelligently generating optimal design solutions. Dividing the

Optimization Workspace of the heat sink model into an array of voxels and tying these structural

elements to representative bit arrays allows the MATLAB code to quantify the improvements

made by the genetic process, learning the optimal layout of liquid and solid materials, specific to

a given electrical design. The general workflow of this process, demonstrated in Figure 61,

Page 89: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

72

targets the topologically structure of the main heat transfer area within electronic-thermal

systems.

Figure 61: Design Progression of Genetic Optimization Process.

The nature of the GA program allows binary strings associated with high ranking designs to

carry on through the iterative process, building on the design improvements of the previous

Generation. By cycling through Populations of Design Candidates, the coding structure produces

novel heat sink topologies that adapt to the specific thermal profile of the designated electrical

load, including any trace layers, thermal vias or bonding materials. In addition, the performance

of the three-dimension models should account for the materials of the associated simulations,

capturing any important physical occurrences, such as areas of induced turbulence, thermal

spreading through conductive metals or convective heat dissipation. A complete overview of the

main program code, constructed in MATLAB R2018a, and all associated functions responsible

for the various stages of the genetic optimization process are presented in the Appendix.

Further analysis on the influence of environmental parameters on the optimization procedure is

presented in the following chapter. Another important aspect to consider is the impact of genetic

variables, such as population size, breeding parameters or fitness scoring, on the performance

nature of the GA code.

Page 90: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

73

Chapter 4

Simulation Results & Discussion

The Binary Genetic Optimization Process, constructed in Chapter 3, was found to provided novel

designs when applied to the topological optimization of liquid channels within electronic heat

sinks. As populations of potential designs are evaluated following this “Survival of Fittest”

methodology, engineering designs run through the program can only be improved upon. This

coding structure follows a simple learning process, associating binary patterns with high fitness

scores as good design structures within the simulation environment. However, the nature of this

learning process is heavily influenced by user-controlled factors, such as fitness scoring,

mutation rates and ranking procedures. Furthermore, environmental conditions defined in the

simulation space, such as applied materials, ambient settings or inlet parameters, can have

considerable impact on the nature of the automated design process.

In this chapter we seek to investigate the influence of several key operating factors on the

effectiveness of the Genetic Optimization Process. Section 4.1 introduces a compact, electrical

Half-Bridge (HB) converter design, that acts as the base test model for the simulation test carried

out in this chapter. Section 4.2 presents a ‘Case Study’ example of the optimization process,

detailing each stage associated with initializing and optimizing the base model through the

genetic design procedure. Building on the results of the Case Study, Section 4.3 investigates the

influence of key fitness values and environmental inlet conditions on the design procedure and

resulting optimized structures.

4.1 Introducing the Test Model

In order to characterize the genetic programming, a base model is used to standardize the

geometric layout associated with the various simulation-based assessments. This is done to

ensure a constant structure is utilized across all the various conditions being tested. Moreover, by

applying a consistent electrical heat load and thermal profile, both the qualitative and

quantitative adaptations induced by different tests variables can be compared and analyzed

accordingly.

Page 91: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

74

The base model, shown in Figure 62. focused around a compact, Half-bridge (HB) converter

design that utilizes two GaN power transistors along with corresponding drivers and a variety of

other passive components. The chosen converter scheme was as a suitable electrical model due

to its flexibility and versatility within power electronic applications. HB modules can be

arranged as both buck or boost converters, or replicated to create full-bridge or three-phase

systems [143].

Figure 62: Base Model for Simulation Testing.

The electrical layout was tailored to a DBC design structure [Figure 63], employing AlN as a

thermal conductive dielectric layer with top copper traces, for electrical routing and a bottom

copper layer for heat spreading.

Figure 63: Electrical DBC Design.

The heat sink geometry is comprised of thin copper layers directly integrated into the DBC

structure [Figure 64a]. This style of design is meant to mimic highly conductive 3D printer heat

sinks or advanced integrated coolers, similar in style to those presented in Section 2.4.4. With

Page 92: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

75

bottom side inlet and outlet manifolds [Figure 64b] the resulting thermal management system

maximizes the active heat transfer area while maintaining a compact overall design.

Figure 64: Compact HB Heat Sink Design. a) Integrated Cooler Approach. b) Inlet/Outlet

Manifold Design.

The main area heat sources or HDD for this base model take the form of two GaN power

transistors, shown in Figure 65a. These 650V devices from GaN Systems allow for high

switching frequencies resulting in better power conversion efficiencies and lower thermal losses

in comparison to other integrated circuit technologies. The compact packaging allows for high

rate of heat removal from the bottom trace layouts, seen in Figure 65b. Manufacturer provided

data specifies the junction-to-case thermal resistance values of the internal semiconductor

packaging [Table 3] providing accurate temperature modelling for these heat loads.

Page 93: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

76

Figure 65: GaN Power Transistors. a) GaN Systems GS66508B Schematic. b) Corresponding

CAD Model.

Table 3: GaN Transistors Manufacturer Provided Thermal Characteristics.

Parameter Value Units

Thermal Resistance (junction-to-case) - Bottom Side 0.5 °C/W

Thermal Resistance (junction-to-top) 8.5 °C/W

Thermal Resistance (junction-to-ambient) 24 °C/W

Maximum Soldering Temperatures (MSL3 rated) 260 °C

Another important aspect of this base model is the nature of the heat flux pattern from the

transistors to the integrated heat sink system. The footprint of the GaN devices on the ceramic

PCB, as shown in Figure 66, creates an irregular pattern for heat flow. This nonuniform

arrangement is an ideal candidate for the genetic optimization process, providing an unorthodox

thermal profile for program to capture and adapt to.

Page 94: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

77

Figure 66: Footprint of Heat Sources on PCB.

Furthermore, the positioning and orientation of these devices on the PCB, along with the layout

of the copper traces makes it difficult to reduce the model with conventional methods, such as

symmetry reduction or shape simplification. Thus, Icepak is the practical tool for this design,

encompassing important physical aspects, which would otherwise be difficult or infeasible to

incorporate.

4.2 Case Study

With the base model established in the previous section, the functionality of the genetic program

is demonstrated, under constant conditions, on a standardized geometry in order to better

quantify its behavior and effectiveness as a design optimization tool. This section presents a

detailed description of the genetic optimization process constructed in Chapter 3, applied to the

cooling structure of the integrated HB converter module presented in Section 4.1. All user

defined steps associated with preparing the model, defining the simulation environment and

initializing the optimization process are described. Intermediate designs generated by the

iterative program are presented, along with detailed comparisons of the improvements introduced

between the starting and ending geometries.

Page 95: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

78

4.2.1 Preparing the Seed Model

Before any testing can be carried out on an electrical cooling structure, designers must attempt to

reduce the complexity of the given model down to the main components responsible for design

performance. Removing trivial components, simplifying flow regimes or decreasing the overall

geometric size of an electronic thermal system can significantly reduce the computational cost

associated with the simulation portion of the optimization process. For this purpose, several

important changes were made to the base converter module, shown in Figure 62, in order to

create a design more suited to the optimization process:

1. Passive components, off-board connectors and unnecessary signal traces were eliminated

from the electrical design to avoid meshing objects with little to no impact on thermal

performance [Figure 67].

Figure 67: Electrical Simplification of HB Converter System.

2. The lower half of the integrated cooling structure, which included the inlet/outlet

connections and associated manifolds shown in Figure 64, was removed from the overall

design structure Figure 68. This action greatly reduced the complexity of the fluid

domain without sacrificing the accuracy of the pressure drop calculations across the heat

sink.

Page 96: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

79

Figure 68: Simplification of Fluid Domain. a) Elimination of Manifold Sections. b) Comparison

of Old vs New Inlet/Outlet Connections.

Once the design simplifications were carried out on the base model, an active workspace was

defined for the program to optimize. For the given system, this took the form of the main area for

active heat transfer [Figure 69a] within the integrated cooling structure. By partitioning this

region into a series of voxels with predetermined physical dimensions [Figure 69b] a three-

dimensional array of structural elements was formed, representing the optimization workspace

[Figure 69c]. The dimension of these voxels were chosen to correspond with reasonable copper

etching techniques as well as ensure minimal channel dimensions were suitably sized to avoided

blockages cause by debris particles [76].

Page 97: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

80

Figure 69: Generating Optimization Workspace for HB Convert Model. a) Defining Active and

Passive Regions. b) Sizing Structural Voxel Elements. c) Partitioning Active Region into Array

of Workspace Elements.

The final step in preparing the optimization model was to generate a starting seed design for

initializing the genetic process. This was done by forming the optimization workspace into a

standard crossflow channel arrangement, shown in Figure 70.

Figure 70: Forming Base Optimization Model into Starting Seed Design.

Page 98: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

81

4.2.2 Defining the Simulation Environment

With a defined optimization model and a starting seed design [Section 4.2.1] the initial structure

could be brought into the Icepak simulation space in order to set the meshing parameters and

operating characteristics of the simplified HB system [Figure 71]. Standard flow rate conditions

were applied to the inlet boundary, assuming an insulated environment with no natural

convection to the local surroundings. The GaN power transistors were modeled as Icepak

network blocks, integrating the junction-to-case thermal resistances detailed in Table 3.

Assuming an operating point of 2kW and a conversion efficiency of 98% the junction power of

each device was set to dissipate 40 Watts of thermal energy. A 0.1mm thick layer of Pb-50/Sn-50

Solder was applied as a bonding layer between the GaN devices and the corresponding power

traces. The remaining materials properties associated with the major components of the

integrated HB system applied in Icepak and can be seen in Table 4.

Figure 71: Icepak Modeling Environment & Example Surface Mesh.

Page 99: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

82

Table 4: Icepak Material Assignments.

Icepak Component/Body

Description Applied Material

Density (ρ)

Thermal Conductivity

(kth)

Specific Heat

Capacity (cp)

Transistor Semiconductor Packaging

Transistor elements, dissipating heat to simulate electrical losses in energy conversion

Gallium Nitride

(Modelled as

Network Block)

6150 kg/m3 130 W/mK 490 J/kgK

Transistor Pads Source, Drain and Gate pads located on backside of transistors. Main path for heat transfer between semiconductor junction and integrated PCB cooling structure

Copper (Pure) 8933 kg/m3 387.6 W/mK 385 J/kgK

Device Bonding Material Layer of electrically conductive metal used to bond transistor pads to top power/signal traces

Solder -

Pb50/Sn50

8890 kg/m3 46 W/mK 213 J/kgK

DBC Traces (Top & Bottom)

Thin layers of conductive material representing the power & signal traces of the electrical design

Copper (Pure) 8933 kg/m3 387.6 W/mK 385 J/kgK

Ceramic Substrate Dielectric body within the PBC used to isolate the energized components from the liquid cooling structure

Aluminum

Nitride

3300 kg/m3 200 W/mK 712 J/kgK

Heat Sink Body Solid elements associated with the generated cooling structure

Copper 8933 kg/m3 360 W/mK 385 J/kgK

Fluid Domain Fluid elements associated with the generated cooling structure

Water (@ 320K) 989 kg/m3 0.637 W/mK 4177 J/kgK

Cabinet Fluid filled cavity surrounding the structural model. Represents the ambient environment of the given system.

Air 1.1616 kg/m3 0.0261 W/mK 1005 J/KgK

Once the simulate environment for the convert module was fully defined, the entire ANSYS

project, along with the associated optimization model, mesh data and operating parameters, was

stored as a .wbpj file. This allows the full project to be archived and activated when called by the

optimization program for design evaluations.

Page 100: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

83

4.2.3 Initializing Genetic Logic

Having defined the parameters of the optimization model within the simulation environment, the

main genetic program is then activated. The primary code, constructed in MATLAB, controls the

progression of the GA workflow [Figure 29] and utilizes a variety of functions that

corresponding to for different stages of the optimization process [See Appendix].

There are a number of user defined factors that must be specified before the program can begin

the iterative design process, the most important of which, is the objective fitness function. This

equation, represented in (3.4), represents the programs interpretation of the designer’s

optimization goals. Objective weighting values, integrating different performance variables or

applying exponential scoring techniques are just some ways users can control the nature of the

genetic program. It is important to properly communicate the purpose of the given design

process for it to produce suitable optimized topologies. In the case of this trial the given fitness

function is represented by (4.1), putting more emphasis on the reduction of transistor junction

temperatures and less on the resulting hydraulic system pressure.

𝑭𝒊𝒕𝒏𝒆𝒔𝒔 = 𝟎. 𝟕 (𝑻𝑫𝒆𝒗𝒊𝒄𝒆𝒔

𝑻𝑫,𝑩𝒂𝒔𝒆) + 𝟎. 𝟑 (

∆𝑷𝒊𝒏𝒍𝒆𝒕−𝒐𝒖𝒕𝒍𝒆𝒕

∆𝑷𝑩𝒂𝒔𝒆) (4. 1)

Once a scoring mechanism is implemented the next important aspect to consider is optimization

convergence. These user defined conditions dictate when the program will classify the iterative

process as complete and provide a final optimized heat sink design. In order to avoid premature

convergence and set a standardized stop condition for further comparisons, the convergence was

defined by a maximum iterative value of 15. This indicates the GA program will evaluate 15

generations of design candidates, at which time it will provide the highest performing model as

the final optimized design.

Another significant variable to consider when initializing the program is the desired Population

size associated with the genetic logic. This specifies the number of design candidates the

program will evaluate and reproduce during each iteration. Increasing this value can have a

significant impact on the computational cost of the evaluation stage, as the program must

simulated each individual design. However, the greater the pool of design candidates the better

chance the program will generate an improved model. Given the mesh size associated with the

HB converter model, a Population size of 10 designs per Generation was chosen.

Page 101: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

84

Once the central parameters of the genetic logic were established the remaining genetic

variables, shown in Table 5,were applied and the optimization process was initiated.

Table 5: Genetic Variables for Case Study Trial.

User Defined Variable Associated Stage Description Applied Value

Mutation Probability Selection, Crossover & Mutation

Probability that a binary value within a newly generated design candidate will invert

5%

Seed Mutation Probability Mutating Seed Design Probability that a binary value within the initial seed design will invert when generating the initial population of design candidates

50%

Number of Breeding Parents

Selection, Crossover & Mutation

Number of parent designs, randomly chose from the previous generation, to be bred together in order to form a new child design

2 Parents Per Child

Gene String Size (Crossover Point)

Selection, Crossover & Mutation

Number of binary values to pull from a randomly chose parent design to donate to a newly forming child design

17 Genes Per Parent

Elite Individuals Repopulation Number of highest performing design candidates within an evaluated generation to be retained and brought into the next population of candidates

1 Individuals Per Generation

Random Individuals Repopulation Number of randomly generated design candidates to include in each new generation

2 Individuals Per Generation

The optimization process begins by launching ANSYS Workbench in server mode, activating the

MATLAB AAS toolbox to enable communication and then uploading the previously established

project file, containing the starting design structure, the Icepak simulation environment and the

defined output performance variables. The program then pauses, allowing the user to inspect the

contents of the project file and make any necessary modifications. After reviewing all geometric

and simulation-based conditions are properly defined, the user then triggers the main GA

process, allowing the program to iteratively optimize the given geometry on the basis of the

provided simulation environment and the desired design objectives.

4.2.4 Design Optimization

The HB integrated converter module, presented in Section 4.1 and simulated under the

conditions detailed in Section 4.2.2, was put through the genetic optimization process for 15

design iterations. Although this program was implemented on a 14 core Dell Precision 5820

Page 102: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

85

Tower with an Intel Xeon 2.5GHz CPU and 32 GB of installed RAM, ANSYS only allows 4

cores to be freely utilized with their Icepak products. Utilizing additional cores would improve

computational speed by require the purchasing of additional licenses from ANSYS Inc. The

result was a total computational run time of 31 hours 18 minutes, generating four increasingly

improved heat sinks designs before reaching the maximum iteration count. The final model

produced by the optimization program achieved a 13.2% improvement in overall fitness

performance as compared to the starting design.

The progression of the design improvements can be seen in the associated objective tracking

window, shown in Figure 72. From the presented data, it can be seen that the genetic

optimization program was responsible for a 138 Pa drop in system pressure with a junction

temperature reduction of ~1.3°C for the GaN power transistors. The random nature of the GA

breeding process significantly reduced both temperature and pressure values within the first

generations. Between the 2nd and 4th iterations, the program began to balance the two objectives,

arriving at a Parento-Optimal solution by the 5th generation of the design process.

Figure 72: Objective Tracking Window for HB Converter Case Study.

Page 103: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

86

The progression of the physical design structure is depicted in Figure 73, representing each new

heat sink topology by its corresponding fluid domain. The genetic process appears to retain the

general structure of the original crossflow channel arrangement. However, by integrating more

liquid channels into the multi-level workspace and inducing small changes in the top-level flow

layout, the program is able to significantly reduce system pressure while increasing the spread of

thermal energy throughout the heat sink structure.

Page 104: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

87

Figure 73: Design Progression of Case Study Optimization Process.

The unique footprint of the transistor pads results in the formation of several hotspots on the

PCB. As seen in Figure 76a, these areas experience concentrated heat flux as thermal energy

transfers from the GaN devices into the integrated cooling structure. The optimization process

Page 105: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

88

appears to target these areas, enhancing heat transfer and significantly reducing the local

temperature gradients on the PCB, as seen in Figure 74b.

Figure 74: Comparing PCB Temperatures Contours. a) Starting Seed Design. b) Final Optimized

Design.

The noticeable reduction in lateral heat spreading within the PCB structure is believed to be a

result of the increased liquid channel density generated by the optimization program. By

allocating layers of fluid channels throughout the three-dimensional workspace the pressure

experienced by the inlet region [Figure 75a] as well as the corresponding gradients across the

flow regime are greatly improved [Figure 75b].

Figure 75: Comparing Fluid Domain Pressure Contours. a) Starting Seed Design. b) Final

Optimized Design.

Page 106: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

89

Furthermore, comparing the cross-sections of the HB system [Figure 76] shows how the

arrangement of these liquid channels, generated by the GA program, enhance the thermal

dissipation throughout the heat sink structure. Utilizing the high conductivity of the metal heat

sink material, the fluid network is distributed throughout the body of the integrated system,

targeting hotspots and reducing temperature gradients. The end result is a much more uniform

thermal distribution throughout the designated cooling structure.

Figure 76: Comparing Cross-Sectional Temperature Contours of Heat Sink Cooling Structure.

a) Starting Seed Design. b) Final Optimized Design.

The random nature of the genetic design process combined with the thermal modeling

capabilities of ANSYS Icepak and freedom of a three-dimensional optimization workspace

proves to be an innovative approach to electronic heat sink design. Applying the principles of

topology optimization to the layout of fluid channels within pressurized cooling structures, the

GA can adapt to the specific needs of a given electrical system, improving on traditional

engineering designs and producing unconventional geometric solutions.

4.3 Optimization Testing

The functionality of the proposed optimization process, demonstrated through the initial case

study [Section 4.2], proves its ability to adapt the flow structure within the defined workspace, to

improve heat sink performance for the designated thermal load. However, there are a large range

of variables associated with these simulation environments as well as the underlying genetic

logic that can influence the operation of the design process as well as the corresponding

Page 107: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

90

generated structures. In order to investigate these parameters, several design comparison studies

were carried out, isolating specific variables within the optimization process.

4.3.1 Weighting Fitness Function

The scoring function presented in (Eq. 3.3) represents the programs interpretation of the thermal

performance associated with each simulated geometry. Moreover, this formula governs the

ranking procedure for each population of potential design candidates, dictating the logics

understanding of ‘Good’ and ‘Bad’ performing designs. Thus, the weighting values applied to

each design objective can directly influence the intelligence of the genetic optimization process.

In order to study the impact of these variables, two different optimization procedures were run

with the same starting seed model and simulation conditions presented in the Section 4.2 case

study. However, each design procedure was performed with varying objective weightings, such

that the resulting topologies generated by each optimization trial could be compared. The

parameters of the fitness testing were as follows:

• Design 1: Optimized with objective weightings of a = 1.0, b = 0.0 such that no

considerations were given to the pressure drop of the heat sink liquid domain and fitness

scoring was only on the basis of reduced device temperatures

• Design 2: Optimized with objective weightings of a = 0.7, b = 0.3 such that the scoring

was biased to the reduction of device temperatures while some consideration was given

to the corresponding system pressure

The design progression and final structures generated by each optimization trial is presented in

Figure 77.

Page 108: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

91

Figure 77: Convergence and Optimized Fluid Topologies for Fitness Testing. a) Temperature

Dependent Fitness Scoring (a=1, b=0). b) Temperature Biased Fitness Scoring (a=0.7, b=0.3).

Each optimal model produced by the designated genetic optimization trails were tested under

constant conditions to investigate the impact of these fitness weighted values on the performance

characteristics of the resulting geometries. Device temperatures and system pressure for each

design was assessed and compared to that of the starting seed model as depicted by the graph in

Figure 78.

Page 109: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

92

Figure 78: Performance Comparisons of Fitness Testing Models.

From the resulting data it is clear that the weighted fitness values can control the nature of the

proposed optimization process. By applying a completely temperature-based scoring system, the

Design 1 process [Figure 77a] achieves small improvements in the flow geometry over the

course of its design cycle, reducing final device temperatures by approximately 3°C, while the

Design 2 model is found better balance the pressure and temperature improvements over the

course of its design cycle [Figure 77b]. The allocation of solid metal elements in the Design 1

heat sink, centered around the transistor pads creates a chaotic fluid regime with high

corresponding pressure requirements when compared to the starting seed model. The more

balanced fitness equation applied to the Design 2 process better ingrates the pressure aspects of

the design performance. The resulting Design 2 model, shown in Figure 77b, utilizes a more

uniform, multilevel, crossflow arrangement, allowing for improved pressure gradients across the

fluid domain while still achieving significant reductions in the device temperatures.

The results presented in Figure 77 clearly indicate that the fitness scoring can be an effect means

of altering the nature of the genetic optimization process and resulting geometries. Designers

can utilize the objective weighting values and tailor the programs evaluation scoring to either

target specific design goals or aim to produce more balanced, optimized models.

0

500

1000

1500

2000

2500

46.5

47

47.5

48

48.5

49

49.5

50

50.5

51

51.5

Seed Design 1 Design 2

Pre

ssu

re (

Pa)

Tem

pe

ratu

re (

°C)

Fitness Function Weighting

Device Temperature System Pressure

Page 110: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

93

4.3.2 Inlet Temperature Variation

The inlet conditions of the fluid domain, especially that of the available coolant temperature

TFluid,In, can vary depending on the systems location and function within an EV cooling loop.

However, designers have to apply constant values when assessing their thermal models and

commonly assume higher inlet temperatures to simulate ‘worst case’ conditions. Thus, it is

important to investigate the potential impact of designer specified conditions, such as TFluid,In on

the functionality of the genetic optimization program. Similar to the fitness tests of Section 4.3.1,

three different optimization procedures were run with the same standardized seed model and

simulation conditions. Each trial employed a different temperature with respect to the fluid inlet

boundary so that the resulting designs could be analyzed. The applied conditions were as

follows:

• Design I: Optimized with an applied inlet temperature of 0°C

• Design II: Optimized with an applied inlet temperature of 15°C

• Design III: Optimized with an applied inlet temperature of 50°C

The design progression and final structures generated by each optimization trial is presented in

Figure 79.

Page 111: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

94

Figure 79: Convergence and Optimized Fluid Topologies for Inlet Temperature Testing. a) 0°C

Inlet Fluid. b) 15°C Inlet Fluid. c) 50°C Inlet Fluid.

Page 112: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

95

Following the comparative nature of Section 4.3.1, each optimal model produced by the inlet

temperature optimization trails were tested under constant conditions to investigate the impact of

this environmental condition on the performance of the genetic program. Device temperatures

and system pressure were calculated and compared to the starting seed model as depicted by the

graph in Figure 80.

Figure 80: Performance Comparisons of Inlet Temperature Testing Models.

Altering the physics of the fluid domain associated with these optimization trials had noticeable

influence on the resulting structures. From Figure 79, it appears that the trials run with lower

inlet temperatures allocated more solid elements into the fluid domain, specifically in proximity

to the area of high heat flux. It is assumed that this is a result of the program utilizing the high

temperature differentials offered by the inlet fluid, to easily achieve design improvements when

combined with the heat spreading induced by the thermal conductive metal elements. However,

when comparing the results of the Design I and II models to the Design III model [Figure 80], it

is clear that the optimization trials associated with higher TFluid,In conditions produced better

performing overall designs. It is assumed that the lower thermal gradients offered by the high

inlet temperature environments cause the genetic program to target the convective heat transfer

offered by the fluid regime as a means of achieving better design improvements and higher

fitness scores. By spreading the flow area across the defined workspace and increasing the

0

50

100

150

200

250

300

350

400

49

49.2

49.4

49.6

49.8

50

50.2

50.4

50.6

50.8

51

51.2

Seed Design I Design II Design III

Pre

ssu

re (

Pa)

Tem

pe

ratu

re (

°C)

Inlet Temperature Testing

Device Temperature System Pressure

Page 113: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

96

surface area of the fluid domain the program was able to attain higher rates of heat removal.

Moreover, comparing the fluid domains of the respective optimized models, Design I [Figure

79a] is found to have significantly less flow blockages that the Design II and Design III

structures [Figure 79b-c], allowing it to achieve the lowest comparative pressure drop.

The results of this testing indicate a motivation for designers to continue the practice of applying

‘worst case’ conditions to the inlet boundaries, even in the case of the proposed optimization

process. This forces the genetic logic to better identify patterns that increase relative convective

heat transfer and not rely on thermal spreading techniques which fall short when high liquid

temperatures area applied.

4.3.3 Flow Rate Variation

Some environmental conditions associated with the simulations of these liquid heat sinks are

predefined by the desired application. A common example of this is the coolant flow rate QFluid,

which is applied to the inlet boundary of the fluid domain. This rate is usually a predetermined

value, dictated by the pumping conditions of the associated automotive cooling system.

However, the available flow rate of the fluid domain is a significate system parameter that can

impact the thermal performance of any heat sink design. Thus, it is important to assess the

impact of this variable on the programs ability to a generate optimal heat sink topologies. The

standard conditions of Section 4.2 were once again applied to three different optimization trials,

each defined with a different coolant flow rate as follows:

• Design A: Optimized with an applied inlet flow rate of 0.25 LPM

• Design B: Optimized with an applied inlet flow rate of 0.5 LPM

• Design C: Optimized with an applied inlet flow rate of 1 LPM

The design progression and final structures generated by each optimization trial is presented in

Figure 81.

Page 114: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

97

Figure 81: Convergence and Optimized Fluid Topologies for Flowrate Testing a) Inlet Flow 0.25

LPM. b) Inlet Flow 0.5 LPM. c) Inlet Flow 1.0 LPM.

Once again, each optimized model was simulated under constant conditions to compare their

corresponding design characteristics in relation to the starting seed design model Figure 82.

Page 115: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

98

Figure 82: Performance Comparisons for Flowrate Testing Models.

The results of the flow rate testing appear to reinforce the findings of the Section 4.3.2, such that

the models produced under the lower flow rate environments outperform those optimized with

higher flow rates [Figure 82]. The resulting structures, shown in Figure 81, exhibit similar design

traits to those produced during the inlet temperature testing. The low levels of convective heat

transfer offered by the Design A flow rate, cause the optimization program to spread the fluid

domain throughout the workspace, increasing surface area and heat removal while

simultaneously reducing system pressure. The higher flow rates, applied to the Design B and

Design C simulation environments offered better initial heat remove, such that the optimization

program positioned more solid elements close to the transistors source areas, as a means of

thermal conductive heat spreading. Moreover, it is assumed that applying high coolant flow rates

to the simulation environments result in high initial system pressures, such that the optimization

program can attain more dramatic improvements in fitness scoring by focusing on reducing

pressure gradients across corresponding heat sink designs in leu of improving waste heat

removal.

The results of this section support the notion that applying extreme or ‘worst case’ conditions to

the simulation environment causes the genetic optimization program to identify better design

traits within the defined workspace.

0

20

40

60

80

100

120

140

51

51.5

52

52.5

53

53.5

54

Seed Design A Design B Design C

Pre

ssu

re (

Pa)

Tem

pe

ratu

re (

°C)

Inlet Flow Testing

Device Temperature System Pressure

Page 116: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

99

Chapter 5

Conclusion

As stated in Section 1.2, this thesis investigates the construction of a design optimization

program, utilizing a genetic binary methodology to produce optimal three-dimensional heat sink

structures for use with high power EV electronics. Approaching the layout of the flow regimes

within these heat sink designs as a topological optimization problem, the workflow of the

genetically inspired process captures the thermal characteristics of a given electrical system to

effectively tailor the corresponding heat sink structure, maximizing waste heat removal and other

performance objectives. The resulting program is an effective tool for achieving significant

improvements in the design and prototyping of compact liquid heat exchangers for high output

power electronics.

5.1 Contributions

A review of the existing literature predicts a continued increase in the energy density

requirements of power electronics, specifically in the case of EV/HEV technologies. A variety of

innovative methods for manufacturing compact, high performing cooling systems are being

explored in attempt to address the growing issue of proper thermal management for these

electronic devices [Section 2.1 - 2.4]. With greater constraints on weight, volume, performance

and reliability, the EV industry is being to beginning experience a need for more creative

methods of design optimization [Section 2.5]. Most techniques seek to achieve design

improvements by finding optimal combinations of pre-defined heat sink variables, such as liquid

flow rates, fin density, or channel dimensions. The presented work breaks away from these

conventional methods in the following ways:

1. The prosed heat exchanger design optimization process is designed to allow for

completely unconstrained optimization of three-dimensional topologies, allowing it to

freely search the defined workspace for optimal design solutions

2. The proposed heat exchanger design optimization process utilizes the intelligence of

binary genetic optimization to systematically identify high performing design

characteristics and build on them, evolving towards unique, optimal solutions

Page 117: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

100

3. The proposed heat exchanger design optimization process integrated a custom validation

process, employing the principles of binary image analysis and morphology, ensures the

genetic optimization produces feasible structures that are practical solutions to the desired

application

4. Utilizing the industry leading simulation capabilities of ANSYS Workbench attains

accurate performance-based evaluations in specific used-defined environmental

conditions for each potential design candidate

5. Incorporating ANSYS Icepak allows designers to capture detailed thermal models and

physical aspects of these complex, Multiphysics systems

6. The workflow of the proposed heat exchanger design optimization process integrates

seamlessly into the prototyping design process, allowing for easy post processing and

continued analysis of the resulting geometries

5.2 Prototyping Tool

The study presented in Section 4.2, for the optimization of a liquid cooled half-bridge converter

system, demonstrates the optimization approaches capability to attain novel design

improvements on a starting, engineered seed model. By incorporating this genetic optimization

into the prototyping process, electronic-thermal designers can customize their models to meet

specific performance objectives. The results of Section 4.3.1 demonstrate the influence of the

user-defined fitness function and how weighted values can alter the evaluation scoring and

corresponding optimal designs generated by the program. Sections 4.3.2 - 4.3.3 compare the

environmental impact of the simulation conditions on the resulting optimization structures. It is

clear that the GA not only adapts to the physics of the given design problem but also captures the

influence of the boundary conditions associated with the thermal modelling and performance

evaluations.

Operating within the ANSYS simulation space, the optimized models produced by the GA can

work seamlessly with other Workbench applications, allowing for continued design analysis or

post processing. Figure 83 demonstrates the potential benefits of this aspect, applied to the scope

of the Half-Bridge model presented in Chapter 4. By re-integrating the optimized heat sink

Page 118: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

101

geometries, such as the one in Figure 73, back into the complete system model, continued

analysis on passive device temperatures and PCB thermal expansion can be performed.

Page 119: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

102

Figure 83: Workflow of Genetic Optimization is Prototyping Process. a) Starting Design

Temperature Profile. b) Optimized Design Temperature Profile. c) Thermal Deformation of

Optimized Design.

Page 120: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

103

Flexible to a variety of heat sink patterns as well as adaptable to any liquid cooled PCB layout,

the presented Genetic Design Optimization program represents a powerful tool to address the

continual design challenges required to design better thermal management systems for high

output power electronics.

5.3 Closing Remarks and Future Works

The initial results of the optimization platform show great promise in the GA’s ability to adapt to

the physics and environment of automotive liquid heat sinks. An important step to strengthen the

programs validity will be to integrate this design optimization into an experimental prototyping

process. Experimentally validating and comparing both the starting seed designs and the final

optimized designs associated with this process will better verify the programs ability to induced

quantifiable design improvements for these liquid cooled components. Furthermore, applying the

proposed design optimization methodology and utilizing the genetic optimization process for

different PCB designs or cooling schemes would further establish the program as a flexibility

design tool.

Having bridged the logic gap from the GA code to ANSYS Workbench, continued work could

also seek to integrated different simulation tools into the design evaluation stages. Additional

performance variables, such as structural reliability, thermal fatigue or semiconductor efficiency

could be incorporated into the fitness scoring of each potential design. ANSYS has a large

library of powerful CFD and FEA based applications that could aid in the electronic thermal

design process, creating a more inclusive multi-objective optimization tool.

The constructed program pairs the computational abilities of ANSYS with a custom optimization

process, largely inspired by the workflow of traditional GA’s. With an abundance of existing

literature surrounding the study of genetic optimization, future works should seek to implement

more advanced principles into to the logic of the underlying code [144]–[146]. Adaptive

mutations, exponential fitness scoring and intelligent gene swapping are just some unique

improvements that could enhance the computational efficiency and overall effectiveness of the

proposed genetic optimization process.

The presented work and results clearly show that the genetically inspired optimization program

can provide effective solutions when applied to the design of compact heat sink topologies for

Page 121: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

104

EV power electronics. By targeting multiple design objectives and allowing the program to work

with un-constrained three-dimensional geometries, this optimization program can generate

significant improvements for designers across a wide range of applications. The automatic and

intuitive nature of the customized design process makes it a useful tool in the continual effort to

improve the cost, size and performance of integrated cooling structures for automotive power

electronics.

Page 122: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

105

Bibliography

[1] M. H. Rashid, Power Electronics Handbook Academic Press Series in Engineering. 2001.

[2] S. S. Kang, “Advanced Cooling for Power Electronics,” in 2012 7th International Conference on Integrated Power Electronics Systems (CIPS), 2012, pp. 1–8.

[3] A. K. Agarwal, “An overview of SiC power devices,” in 2010 International Conference on Power, Control and Embedded Systems, 2010, pp. 1–4.

[4] J. A. Cooper and A. Agarwal, “SiC power-switching devices-the second electronics revolution?,” Proc. IEEE, vol. 90, no. 6, pp. 956–968, Jun. 2002.

[5] F. Blaabjerg, Z. Chen, and S. B. Kjaer, “Power electronics as efficient interface in dispersed power generation systems,” IEEE Trans. Power Electron., vol. 19, no. 5, pp. 1184–1194, 2004.

[6] B. K. Bose, “Power Electronics And Motor Drives,” Power Electron. Mot. Drives, vol. 56, no. 2, pp. 649–729, 2006.

[7] J. I. Leon, S. Kouro, L. G. Franquelo, J. Rodriguez, and B. Wu, “The Essential Role and the Continuous Evolution of Modulation Techniques for Voltage-Source Inverters in the Past, Present, and Future Power Electronics,” IEEE Trans. Ind. Electron., vol. 63, no. 5, pp. 2688–2701, May 2016.

[8] C. C. Chan and K. T. Chau, “An overview of power electronics in electric vehicles,” IEEE Trans. Ind. Electron., vol. 44, no. 1, pp. 3–13, 1997.

[9] T. Franke, H. Glonner, D. Nowak, and F. Osterreicher, “Electrified power train - challenges and opportunities for the electrical industry,” 2005 Eur. Conf. Power Electron. Appl., pp. 12 pp.-P.12, 2005.

[10] A. Emadi, S. S. Williamson, and A. Khaligh, “Power Electronics Intensive Solutions for Advanced Vehicular Power Systems,” Power, vol. 21, no. 3, pp. 567–577, 2006.

[11] A. Emadi, Young Joo Lee, and K. Rajashekara, “Power Electronics and Motor Drives in Electric, Hybrid Electric, and Plug-In Hybrid Electric Vehicles,” IEEE Trans. Ind. Electron., vol. 55, no. 6, pp. 2237–2245, 2008.

[12] T. R. McNutt, A. R. Hefner, H. A. Mantooth, D. Berning, and S. Ryu, “Silicon Carbide Power MOSFET Model and Parameter Extraction Sequence,” IEEE Trans. Power Electron., vol. 22, no. 2, pp. 353–363, Mar. 2007.

[13] M. Willander, M. Friesel, Q. U. Wahab, and B. Straumal, “Silicon carbide and diamond for high temperature device applications,” J. Mater. Sci. Mater. Electron., vol. 17, no. 1, pp.

Page 123: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

106

1–25, 2006.

[14] J. Biela, M. Schweizer, S. Waffler, and J. W. Kolar, “SiC versus Si-Evaluation of Potentials for Performance Improvement of Inverter and DC- DC Converter Systems by SiC Power Semiconductors,” IEEE Trans. Ind. Electron., vol. 58, no. 7, pp. 2872–2882, Jul. 2011.

[15] T. Ayalew, “SiC Semiconductor Devices Technology , Modeling , and Simulation,” Vienna University of Technology, 2004.

[16] Y. Wang, X. Dai, G. Liu, Y. Wu, D. Li, and S. Jones, “Integrated Liquid Cooling Automotive IGBT Module for High Temperatures Coolant Application,” in Proceedings of PCIM Europe 2015; International Exhibition and Conference for Power Electronics, Intelligent Motion, Renewable Energy and Energy Management, 2015, pp. 1–7.

[17] S. Kinio and J. K. Mills, “Design of electrode topologies for dielectrophoresis through the use of genetic optimization with COMSOL Multiphysics,” 2015 IEEE Int. Conf. Mechatronics Autom. ICMA 2015, pp. 1019–1024, 2015.

[18] T. P. Chow and M. Ghezzo, “SiC Power Devices,” MRS Proc., vol. 423, p. 9, 1996.

[19] A. W. Scott, Cooling of electronic equipment. New York, NY: Wiley, 1974.

[20] A. Pesaran, “Battery Thermal Management in EVs and HEVs : Issues and Solutions,” Adv. Automot. Batter. Conf., no. January, p. 10, 2001.

[21] K. J. Kelly, T. Abraham, K. Bennion, D. Bharathan, S. Narumanchi, and M. O. Keefe, “Assessment of thermal control technologies for cooling electric vehicle power electronics,” Energy, no. January, pp. 1–17, 2008.

[22] K. Hromadka, J. Reboun, K. Rendl, V. Wirth, and A. Hamacek, “Comparison of the surface properties of power electronic substrates,” in 2015 38th International Spring Seminar on Electronics Technology (ISSE), 2015, pp. 146–150.

[23] K. S. Ong, C. F. Tan, K. C. Lai, and K. H. Tan, “Heat spreading and heat transfer coefficient with fin heat sink,” Appl. Therm. Eng., vol. 112, pp. 1638–1647, 2017.

[24] Y. Gai et al., “Cooling of Automotive Traction Motors: Schemes, Examples, and Computation Methods,” IEEE Trans. Ind. Electron., vol. 66, no. 3, pp. 1681–1692, Mar. 2019.

[25] S. Rogers, “Electrical and Electronics Technical Team Roadmap,” DoE Fuel Cell Technol. Off. / NREL, no. June, 2013.

[26] J. Schulz-harder, E. C. Gmbh, and I. G. D, “Review on Highly Integrated Solutions for Power Electronic Devices.”

[27] M. B. Kleiner, S. A. Kuehn, and K. Haberger, “High performance forced air cooling scheme

Page 124: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

107

employing microchannel heat exchangers,” IEEE Trans. components, Packag. Manuf. Technol. Part A, vol. 18, no. 4, pp. 795–804, 1995.

[28] A. Bouknadel, I. Rah, H. El Omari, and H. El Omari, “Comparative study of fin geometries for heat sinks in natural convection,” in 2014 International Renewable and Sustainable Energy Conference (IRSEC), 2014, pp. 723–728.

[29] A. Abdoli, G. Jimenez, and G. S. Dulikravich, “Thermo-fluid analysis of micro pin-fin array cooling configurations for high heat fluxes with a hot spot,” Int. J. Therm. Sci., vol. 90, pp. 290–297, 2015.

[30] A. M. E. Arefin, “Thermal analysis of modified pin fin heat sink for natural convection,” in 2016 5th International Conference on Informatics, Electronics and Vision (ICIEV), 2016, pp. 1–5.

[31] D. Christen, M. Stojadinovic, and J. Biela, “Energy efficient heat sink design: Natural vs. forced convection cooling,” in 2016 IEEE 17th Workshop on Control and Modeling for Power Electronics (COMPEL), 2016, pp. 1–8.

[32] K. Kitamura, A. Mitsuishi, T. Suzuki, and F. Kimura, “Fluid flow and heat transfer of natural convection induced around a vertical row of heated horizontal cylinders,” Int. J. Heat Mass Transf., vol. 92, pp. 414–429, 2016.

[33] D. B. Tuckerman and R. F. W. Pease, “High-performance heat sinking for VLSI,” IEEE Electron Device Lett., vol. 2, no. 5, pp. 126–129, 1981.

[34] C. Hilbert, S. Sommerfeldt, O. Gupta, and D. J. Herrell, “High performance micro-channel air cooling,” Sixth Annu. IEEE Proc. Semicond. Therm. Temp. Meas. Symp., no. Mcc, 1990.

[35] R. W. Knight, J. S. Goodling, and B. E. Gross, “Optimal thermal design of air cooled forced convection finned heat sinks,” Intersoc. Conf. Therm. Phenom. Electron. Syst., vol. 15, no. 5, pp. 206–212, 1992.

[36] K. Azar, R. S. McLeod, and R. E. Caron, “Narrow channel heat sink for cooling of high powered electronic components,” [1992 Proceedings] Eighth Annu. IEEE Semicond. Therm. Meas. Manag. Symp., pp. 12–19.

[37] B. Gromoll, “Advanced micro air-cooling systems for high density packaging,” Proc. 1994 IEEE/CHMT 10th Semicond. Therm. Meas. Manag. Symp., pp. 53–58, 1994.

[38] C. Marques and K. W. Kelly, “Fabrication and Performance of a Pin Fin Micro Heat Exchanger,” J. Heat Transfer, vol. 126, no. 3, p. 434, 2004.

[39] J. Garg, M. Arik, S. Weaver, T. Wetzel, and S. Saddoughi, “Meso Scale Pulsating Jets for Electronics Cooling,” J. Electron. Packag., vol. 127, no. 4, p. 503, 2005.

[40] Y. S. Chung, D. H. Lee, and P. M. Ligrani, “Jet Impingement Cooling of Chips Equipped

Page 125: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

108

With Cylindrical Pedestal Profile Fins,” J. Electron. Packag., vol. 127, no. June 2005, pp. 106–112, 2005.

[41] S. Y. Kim, M. H. Lee, and K. S. Lee, “Heat removal by aluminum-foam heat sinks in a multi-air jet impingement,” IEEE Trans. Components Packag. Technol., vol. 28, no. 1, pp. 142–148, 2005.

[42] S. S. Anandan and V. Ramalingam, “Thermal management of electronics: A review of literature,” Therm. Sci., vol. 12, no. 2, pp. 5–25, 2008.

[43] J. Schulz-Harder, “Efficient cooling of power electronics,” Power Electron. Syst. Appl. 2009. PESA 2009. 3rd Int. Conf., pp. 1–4, 2009.

[44] Z. A. Williams and J. A. Roux, “Thermal Management of a High Packing Density Array of Power Amplifiers Using Liquid Cooling,” J. Electron. Packag., vol. 129, no. 4, p. 488, 2007.

[45] M. C. Lu and C. C. Wang, “Effect of the inlet location on the performance of parallel-channel cold-plate,” IEEE Trans. Components Packag. Technol., vol. 29, no. 1, pp. 30–38, 2006.

[46] S. G. Kandlikar and C. N. Hayner, “Liquid cooled cold plates for industrial high-power electronic devices thermal design and manufacturing considerations,” Heat Transf. Eng., vol. 30, no. 12, pp. 918–930, 2009.

[47] I. Y. Kim and O. L. De Weck, “Variable chromosome length genetic algorithm for progressive refinement in topology optimization,” Struct. Multidiscip. Optim., vol. 29, no. 6, pp. 445–456, 2005.

[48] L. H. Olesen, F. Okkels, and H. Bruus, “A high-level programming-language implementation of topology optimization applied to steady-state Navier-Stokes flow,” Int. J. Numer. Methods Eng., vol. 65, no. 7, pp. 975–1001, 2006.

[49] T. Borrvall and J. Petersson, “Topology optimization of fluids in Stokes flow,” Int. J. Numer. Methods Fluids, vol. 41, no. 1, pp. 77–107, 2003.

[50] E. M. Sparrow, P. W. Chevalier, and J. P. Abraham, “The design of cold plates for the thermal management of electronic equipment,” Heat Transf. Eng., vol. 27, no. 7, pp. 6–16, 2006.

[51] J. H. Nam, K. J. Lee, S. Sohn, and C. J. Kim, “Multi-pass serpentine flow-fields to enhance under-rib convection in polymer electrolyte membrane fuel cells: Design and geometrical characterization,” J. Power Sources, vol. 188, no. 1, pp. 14–23, 2009.

[52] M. Fesanghary, E. Damangir, and I. Soleimani, “Design optimization of shell and tube heat exchangers using global sensitivity analysis and harmony search algorithm,” Appl. Therm. Eng., vol. 29, no. 5–6, pp. 1026–1031, 2009.

Page 126: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

109

[53] I. Aranzabal and et al., “Status and advances in Electric Vehicle ’ s power modules packaging technologies,” no. May, pp. 10–12, 2016.

[54] D. Ghosh, P. D. Maguire, and D. X. Zhu, “Design and CFD Simulation of a Battery Module for a Hybrid Electric Vehicle,” SAE Int., 2009.

[55] A. Jarrett and I. Y. Kim, “Design optimization of electric vehicle battery cooling plates for thermal performance,” J. Power Sources, vol. 196, no. 23, pp. 10359–10368, 2011.

[56] B. Agostini, M. Fabbri, J. E. Park, L. Wojtan, J. R. Thome, and B. Michel, “State of the art of high heat flux cooling technologies,” Heat Transf. Eng., vol. 28, no. 4, pp. 258–281, 2007.

[57] D. Chen, J. Jiang, G. H. Kim, C. Yang, and A. Pesaran, “Comparison of different cooling methods for lithium ion battery cells,” Appl. Therm. Eng., vol. 94, pp. 846–854, 2016.

[58] Y. Wang, S. Jones, A. Dai, and G. Liu, “Reliability enhancement by integrated liquid cooling in power IGBT modules for hybrid and electric vehicles,” Microelectron. Reliab., vol. 54, no. 9–10, pp. 1911–1915, 2014.

[59] A. Wintrich, U. Nicolai, W. Tursky, and T. Reimann, “Applications Manual Power Semiconductors,” vol. 6.

[60] X. Dai et al., “Reliability design of direct liquid cooled power semiconductor module for hybrid and electric vehicles,” Microelectron. Reliab., vol. 64, pp. 474–478, 2016.

[61] S. Klaka and R. Sittig, “Reduction of Thermomechanical Stress by applying a Low Temperature Joining Technique,” 6th Int. Symp. power Semicond. devices IC’s, no. 94, pp. 259–264, 1994.

[62] K. C. Otiaba, N. N. Ekere, R. S. Bhatti, S. Mallik, M. O. Alam, and E. H. Amalu, “Thermal interface materials for automotive electronic control unit: Trends, technology and R&D challenges,” Microelectron. Reliab., vol. 51, no. 12, pp. 2031–2043, 2011.

[63] M. Schneider-Ramelow, “Design and assembly of power semiconductors with double-sided water cooling,” 5th Int. Conf. Integr. Power Syst. (CIPS), 2008, pp. 1–7, 2008.

[64] G. L. Romero, J. M. Fusaro, and J. L. . J. Martinez, “Metal matrix composite power modules: improvements in reliability\nand package integration,” IAS ’95. Conf. Rec. 1995 IEEE Ind. Appl. Conf. Thirtieth IAS Annu. Meet., vol. 1, 1995.

[65] M. Ivanova, Y. Avenas, C. Schaeffer, J. B. Dezord, and J. Schulz-Harder, “Heat pipe integrated in direct bonded copper (DBC) technology for cooling of power electronics packaging,” IEEE Trans. Power Electron., vol. 21, no. 6, pp. 1541–1547, 2006.

[66] J. Weyant, S. Garner, M. Johnson, and M. Occhionero, “Heat pipe embedded AlSiC plates for high conductivity - Low CTE heat spreaders,” 2010 12th IEEE Intersoc. Conf. Therm. Thermomechanical Phenom. Electron. Syst. ITherm 2010, 2010.

Page 127: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

110

[67] T. Stockmeier, “From packaging to ‘un’-packaging - Trends in power semiconductor modules,” Proc. Int. Symp. Power Semicond. Devices ICs, pp. 12–19, 2008.

[68] A. Uhlemann and A. Herbrandt, “A new base plate concept on the basis of aluminium-copper clad materials,” PCIM Eur. Conf. Proc., no. May, pp. 680–685, 2012.

[69] L. Xu, M. Wang, Y. Zhou, Z. Qian, and S. Liu, “An optimal structural design to improve the reliability of Al2O3-DBC substrates under thermal cycling,” Microelectron. Reliab., vol. 56, pp. 101–108, 2016.

[70] Y. Peles, A. Koşar, C. Mishra, C. J. Kuo, and B. Schneider, “Forced convective heat transfer across a pin fin micro heat sink,” Int. J. Heat Mass Transf., vol. 48, no. 17, pp. 3615–3627, 2005.

[71] S. Belhardj, S. Mimouni, A. Saidane, and M. Benzohra, “Using microchannels to cool microprocessors: A transmission-line-matrix study,” Microelectronics J., vol. 34, no. 4, pp. 247–253, 2003.

[72] D. Y. Lee and K. Vafai, “Comparative analysis of jet impingement and microchannel cooling for high heat flux applications,” Int. J. Heat Mass Transf., vol. 42, no. 9, pp. 1555–1568, 1999.

[73] C. Harris, M. Despa, and K. Kelly, “Design and fabrication of a cross flow micro heat exchanger,” J. Microelectromechanical Syst., vol. 9, no. 4, pp. 502–508, 2000.

[74] T. Steiner and R. Sittig, “IGBT module setup with integrated micro-heat sinks,” 12th Int. Symp. Power Semicond. Devices ICs. Proc. (Cat. No.00CH37094), pp. 209–212, 2000.

[75] W. Qu and I. Mudawar, “Thermal design methodology for high-heat-flux single-phase,” Inter Soc. Conf. Therm. Phenom., vol. 26, no. 3, pp. 347–359, 2002.

[76] M. Reeves, J. Moreno, P. Beucher, S.-J. Loong, and D. Brown, “Investigation on the Impact on Thermal Performances of New Pin and Fin Geometries Applied to Liquid Cooling of Power Electronics,” vol. 3, no. 3, p. 35601, 2009.

[77] H. Y. Zhang, D. Pinjala, Y. K. Joshi, T. N. Wong, and K. C. Toh, “Thermal modeling and design of liquid cooled heat sinks assembled with flip chip ball grid array packages,” Proc. - Electron. Components Technol. Conf., pp. 431–437, 2003.

[78] S. J. Kim, “Methods for thermal optimization of microchannel heat sinks,” Heat Transf. Eng., vol. 25, no. 1, pp. 37–49, 2004.

[79] V. Sajith, D. Haridas, C. B. Sobhan, and G. R. C. Reddy, “Convective heat transfer studies in macro and mini channels using digital interferometry,” Int. J. Therm. Sci., vol. 50, no. 3, pp. 239–249, 2011.

[80] A. Husain, M. Ariz, N. Z. H. Al-Rawahi, and M. Z. Ansari, “Thermal performance analysis of

Page 128: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

111

a hybrid micro-channel, -pillar and -jet impingement heat sink,” Appl. Therm. Eng., vol. 102, pp. 989–1000, 2016.

[81] T. Dixit and I. Ghosh, “Review of micro- and mini-channel heat sinks and heat exchangers for single phase fluids,” Renew. Sustain. Energy Rev., vol. 41, pp. 1298–1311, 2015.

[82] L. Hosain, R. Bel, and A. Daneryd, “Heat transfer by liquid jets impinging on a hot flat surface,” Appl. Energy, vol. 164, pp. 934–943, 2016.

[83] G. Nasif and R. Balachandar, “Conjugate jet impingement heat transfer investigation,” 13th Int. Conf. Heat Transf. Fluid Mech. Thermodyn., pp. 554–559, 2018.

[84] J. Johannes et al., “Direct Single Impinging Jet Cooling of a MOSFET Power Electronic Module,” vol. 33, no. 5, pp. 4224–4237, 2018.

[85] S. Kong, F. Application, and P. Data, “( 12 ) United States Patent,” 2011.

[86] P. R. Parida, S. V. Ekkad, and K. Ngo, “Impingement-based high performance cooling configurations for automotive power converters,” Int. J. Heat Mass Transf., vol. 55, no. 4, pp. 834–847, 2012.

[87] E. Lai, M. A. Moss, B. L. Button, and K. Jambunathan, “A review of heat transfer data for single circular jet impingement,” Int. J. Heat Fluid Flow, vol. 13, no. 2, pp. 106–115, 1992.

[88] S. Garimella, “Nozzle-geometry effects in liquid jet impingement heat transfer,” Int. J. Heat Mass Transf., vol. 39, no. 14, pp. 2915–2923, 1996.

[89] K. Oliphant, B. W. Webb, and M. Q. McQuay, “An experimental comparison of liquid jet array and spray impingement cooling in the non-boiling regime,” Exp. Therm. Fluid Sci., vol. 18, no. 1, pp. 1–10, 1998.

[90] R. G. Mertens et al., “Spray Cooling of IGBT Devices,” J. Electron. Packag., vol. 129, no. 3, p. 316, 2007.

[91] A. Bhunia, S. Chandrasekaran, and C. L. Chen, “Performance improvement of a power conversion module by liquid micro-jet impingement cooling,” IEEE Trans. Components Packag. Technol., vol. 30, no. 2, pp. 309–316, 2007.

[92] L. J. Turek, D. P. Rini, and B. a Saarloos, “Evaporative Spray Cooling of Power Electronics Using High Temperature Coolant,” 11th IEEE Intersoc. Conf. Therm. Thermomechanical Phenom. Electron. Syst., pp. 346–351, 2008.

[93] J. Barrau, M. Omri, D. Chemisana, J. Rosell, M. Ibañez, and L. Tadrist, “Numerical study of a hybrid jet impingement/micro-channel cooling scheme,” Appl. Therm. Eng., vol. 33–34, no. 1, pp. 237–245, 2012.

[94] M. Hadziabdic and K. Hanjalic, “Vortical structures and heat transfer in a round impinging

Page 129: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

112

jet,” vol. 596, pp. 221–260, 2008.

[95] A. Morozumi, H. Hokazono, Y. Nishimura, Y. Ikeda, Y. Nabetani, and Y. Takahashi, “Direct liquid cooling module with high reliability solder joining technology for automotive applications,” Proc. Int. Symp. Power Semicond. Devices ICs, pp. 109–112, 2013.

[96] K. Gould, S. Q. Cai, C. Neft, and S. Member, “Liquid Jet Impingement Cooling of a Silicon Carbide Power Conversion Module for Vehicle Applications,” vol. 30, no. 6, pp. 2975–2984, 2015.

[97] K. Olesen, R. Bredtmann, and R. Eisele, “‘ShowerPower’ New Cooling Concept for Automotive Applications,” Automot. Power Electron., no. June, pp. 1–9, 2006.

[98] E. G. Colgan et al., “A practical implementation of silicon microchannel coolers for high power chips,” IEEE Trans. Components Packag. Technol., vol. 30, no. 2, pp. 218–225, 2007.

[99] X. et al. Tang, “Hybrid substrate - A future mterial for power semiconductor devices,” PCIM Eur., no. May, pp. 20–22, 2014.

[100] K. W. Jung et al., “Embedded cooling with 3D manifold for vehicle power electronics application: Single-phase thermal-fluid performance,” Int. J. Heat Mass Transf., vol. 130, pp. 1108–1119, 2019.

[101] S. Chen, X. Peng, N. Bao, and A. Garg, “A comprehensive analysis and optimization process for an integrated liquid cooling plate for a prismatic lithium-ion battery module,” Appl. Therm. Eng., vol. 156, no. January, pp. 324–339, 2019.

[102] R. Van Erp, G. Kampitsis, and E. Matioli, “A manifold microchannel heat sink for ultra-high power density liquid-cooled converters,” Conf. Proc. - IEEE Appl. Power Electron. Conf. Expo. - APEC, vol. 2019-March, pp. 1383–1389, 2019.

[103] M. März, A. Schletz, B. Eckardt, S. Egelkraut, and H. Rauh, “Power Electronics System Integration for Electric and Hybrid Vehicles,” Integr. Power Syst. (CIPS), 2010 6th Int. Conf., pp. 16–18, 2010.

[104] B. C. Charboneau et al., “Double-sided liquid cooling for power semiconductor devices using embedded power packaging,” IEEE Trans. Ind. Appl., vol. 44, no. 5, pp. 1645–1655, 2008.

[105] C. Wang, L. Zheng, L. Han, H. Fang, and J. Xu, “Thermal performance investigation of three-dimensional structure unit in double-sided cooling IGBT module,” in 2014 15th International Conference on Electronic Packaging Technology, 2014, pp. 622–625.

[106] P. Ning, Z. Liang, F. Wang, and L. Marlino, “Power module and cooling system thermal performance evaluation for HEV application,” Conf. Proc. - IEEE Appl. Power Electron. Conf. Expo. - APEC, vol. 2, no. 3, pp. 2134–2139, 2012.

Page 130: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

113

[107] C. Gillot, C. Schaeffer, C. Massit, and L. Meysenc, “Double-sided cooling for high power IGBT modules using flip chip technology,” IEEE Trans. Components Packag. Technol., vol. 24, no. 4, pp. 698–704, 2001.

[108] H. R. Chang, J. Bu, G. Kong, and R. Labayen, “300A 650V 70 um thin IGBTs with double-sided cooling,” Proc. Int. Symp. Power Semicond. Devices ICs, pp. 320–323, 2011.

[109] T. Sturgeon, J. Van Biesebroeck, and G. Gereffi, “Value chains, networks and clusters: Reframing the global automotive industry,” J. Econ. Geogr., vol. 8, no. 3, pp. 297–321, 2008.

[110] X. Wei and Y. Joshi, “Optimization study of stacked micro-channel heat sinks for micro-electronic cooling,” Intersoc. Conf. Therm. Thermomechanical Phenom. Electron. Syst. ITHERM, vol. 2002-Janua, no. 1, pp. 441–448, 2002.

[111] R. W. Knight, J. S. Goodling, D. J. Hall, and R. C. Jaeger, “Heat Sink Optimization with Application to Microchannels,” IEEE Trans. Components, Hybrids, Manuf. Technol., vol. 15, no. 5, pp. 832–842, 1992.

[112] T. M. Ritzer and P. G. Lau, “Economic optimization of heat sink design,” AIP Conf. Proc., vol. 316, no. 1994, pp. 177–180, 1994.

[113] S. Lee, “Optimum Design and Selection of Heat Sinks,” IEEE Trans. Components Packag. Manuf. Technol. Part A, vol. 18, no. 4, pp. 812–817, 1995.

[114] T. Y. Lee and Tien-Yu Lee, “Design optimization of an integrated liquid-cooled IGBT power module using CFD technique,” IEEE Trans. Components Packag. Technol., vol. 23, no. 1, pp. 55–60, 2000.

[115] J. Li and G. P. Peterson, “Geometric optimization of a micro heat sink with liquid flow,” IEEE Trans. Components Packag. Technol., vol. 29, no. 1, pp. 145–154, 2006.

[116] A. Husain and K.-Y. Kim, “Shape optimization of microchannel heat sink for microelectronics cooling,” IEEE Trans. Components Packag. Technol., vol. 31, no. 2, pp. 322–330, 2008.

[117] C. C. Wang, C. I. Hung, and W. H. Chen, “Design of heat sink for improving the performance of thermoelectric generator using two-stage optimization,” Energy, vol. 39, no. 1, pp. 236–245, 2012.

[118] O. Sigmund, “On the Design of Compliant Mechanisms Using Topology Optimization,” Mech. Struct. Mach., vol. 25, no. 4, pp. 493–524, 1997.

[119] S. Mantovani, G. A. Campo, A. Ferrari, and M. Cavazzuti, “Optimization methodology for automotive chassis design by truss frame: A preliminary investigation using the lattice approach,” Adv. Transdiscipl. Eng., vol. 7, no. September, pp. 984–992, 2018.

Page 131: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

114

[120] S. Yang, L. Yan, and C. Qi, “An adaptive multi-step varying-domain topology optimization method for spot weld design of automotive structures,” Struct. Multidiscip. Optim., vol. 59, no. 1, pp. 291–310, Jan. 2019.

[121] S. S. S. Kang, “Advanced Cooling for Power Electronics,” Int. Conf. Intergrated Power Electron. Syst. CIPS 2012, 6-8 March 2012, Nuremberg, Ger., vol. 9, p. 8, 2012.

[122] J. Alexandersen, O. Sigmund, and N. Aage, “Large scale three-dimensional topology optimisation of heat sinks cooled by natural convection,” Int. J. Heat Mass Transf., vol. 100, pp. 876–891, 2016.

[123] M. Gen and R. Cheng, Genetic Algorithms for Engineering Optimization. John Wiley & Sons, Inc, 2006.

[124] M. Mitchell, An introduction to genetic algorithms, vol. 32, no. 6. 1998.

[125] T. Wu, B. Ozpineci, and C. Ayers, “Genetic algorithm design of a 3D printed heat sink,” 2016 IEEE Appl. Power Electron. Conf. Expo., pp. 3529–3536, 2016.

[126] C. M. Fonseca and P. J. Fleming, “Genetic Algorithms for Multiobjective Optimization: Formulation, Discussion and Generalization, Genetic Algorithms,” Proc. Fifth Int. Conf., no. July, pp. 416–423, 1993.

[127] A. Husain and K. Y. Kim, “Optimization of a microchannel heat sink with temperature dependent fluid properties,” Appl. Therm. Eng., vol. 28, no. 8–9, pp. 1101–1107, 2008.

[128] G. N. Xie, B. Sunden, and Q. W. Wang, “Optimization of compact heat exchangers by a genetic algorithm,” Appl. Therm. Eng., vol. 28, no. 8–9, pp. 895–906, 2008.

[129] S. Sanaye and H. Hajabdollahi, “Thermal-economic multi-objective optimization of plate fin heat exchanger using genetic algorithm,” Appl. Energy, vol. 87, no. 6, pp. 1893–1902, 2010.

[130] R. Bornoff, B. Subat, and J. Wilson, “Generative heatsink design for an automotive audio amplifier,” in 2018 34th Thermal Measurement, Modeling Management Symposium (SEMI-THERM), 2018, pp. 218–223.

[131] T. Wu, B. Ozpineci, M. Chinthavali, Z. Wang, S. Debnath, and S. Campbell, “Design and Optimization of 3D Printed Air-Cooled Heat Sinks Based on Genetic Algorithms,” pp. 650–655, 2017.

[132] T. Wu, B. Ozpineci, M. Chinthavali, Z. Wang, S. Debnath, and S. Campbell, “Design and Optimization of 3D Printed Air-Cooled Heat Sinks Based on Genetic Algorithms,” 2017 IEEE Transp. Electrif. Conf. Expo, pp. 650–655, 2017.

[133] M. Mitchell, An introduction to genetic algorithms. MIT press, 1998.

Page 132: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

115

[134] C. García-Martínez, F. J. Rodriguez, and M. Lozano, “Genetic algorithms,” Handb. Heuristics, vol. 1–2, no. 1, pp. 431–464, 2018.

[135] M. P. Bendsøe and O. Sigmund, Topology optimization: theory, methods, and applications. Springer, 2004.

[136] A. J. Michalak and J. K. Mills, “Genetic Optimization of Thermal Management Systems for EV Power Electronics via ANSYS Multiphysics,” in 2019 IEEE International Conference on Mechatronics and Automation (ICMA), 2019, pp. 2401–2406.

[137] T. Matsumori, T. Kondoh, A. Kawamoto, and T. Nomura, “Topology optimization for fluid-thermal interaction problems under constant input power,” Struct. Multidiscip. Optim., vol. 47, no. 4, pp. 571–581, 2013.

[138] Y. Altintas, Manufacturing automation: metal cutting mechanics, machine tool vibrations, and CNC design. Cambridge university press, 2012.

[139] G. Chen, C. P. Low, and Z. Yang, “Preserving and exploiting genetic diversity in evolutionary programming algorithms,” IEEE Trans. Evol. Comput., vol. 13, no. 3, pp. 661–673, 2009.

[140] R. Bornoff and J. Parry, “An additive design heatsink geometry topology identification and optimisation algorithm,” in 2015 31st Thermal Measurement, Modeling Management Symposium (SEMI-THERM), 2015, pp. 303–308.

[141] K. Gupta et al., “Thermal Management Strategies for a High-Frequency, Bi-Directional, On-Board Electric Vehicle Charger,” Proc. 17th Intersoc. Conf. Therm. Thermomechanical Phenom. Electron. Syst. ITherm 2018, pp. 935–943, 2018.

[142] D. Guirguis, M. Nasr, S. K. Murray, H. Matsumoto, O. Trescases, and C. H. Amon, “Thermal Management Within Multi-Disciplinary System Design of a Rubik’s-Cube-sized 2kW Power Inverter,” Proc. 17th Intersoc. Conf. Therm. Thermomechanical Phenom. Electron. Syst. ITherm 2018, pp. 921–926, 2018.

[143] Jih-Sheng Lai and Fang Zheng Peng, “Multilevel converters-a new breed of power converters,” IEEE Trans. Ind. Appl., vol. 32, no. 3, pp. 509–517, May 1996.

[144] Q. Al-Tashi, S. J. Abdul Kadir, H. M. Rais, S. Mirjalili, and H. Alhussian, “Binary Optimization Using Hybrid Grey Wolf Optimization for Feature Selection,” IEEE Access, vol. 7, pp. 39496–39508, 2019.

[145] N. Al-Madi, H. Faris, and S. Mirjalili, “Binary multi-verse optimization algorithm for global optimization and discrete problems,” Int. J. Mach. Learn. Cybern., Feb. 2019.

[146] S. Mirshekarian and G. A. Süer, “Experimental study of seeding in genetic algorithms with non-binary genetic representation,” J. Intell. Manuf., vol. 29, no. 7, pp. 1637–1646, Oct. 2018.

Page 133: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

116

Appendix: Details of Heat Sink Design Optimization

The following MATLAB code implements the heat sink optimization methodology

outlined in Chapter 3. The applied variables correspond with the case study conditions presented

in Chapter 4. Variables share names with those defined in the chapter wherever possible.

Page 134: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

117

Main Program Code

%%%%% Launch ANSYS in Server Mode and Pause for User to Define Simulation %%%%% Generate Model WorkSpace Grid, Apply suitable Meshing parameters and %%%%% their corresponding Physics

GeometryFile = 'HB GaN Chip Optimization Model.agdb';%'Optimization Model - Indexing.SLDPRT'; WorkingDirectory = 'C:\\Users\\Andrew\\Desktop\\Thesis\\Thesis Example Simulations\\Chapter

4\\Initial Testing'; %%Define Working Directory Path

initialize(GeometryFile,WorkingDirectory); pause

tic;

%%%%% Define workspace Size

deltaX = 42; %mm deltaY = 40; %mm deltaZ = 4; %mm

%%%%% Define Mesh Pixels/Voxels

x = 2; %mm y = 2; %mm z = 1; %mm

%%%%%Define GA Parameters

PopulationSize = 10; MutationProbability = 0.05; %Percentage SeedMutationProbability = 0.5; %Percentage ParentsPerChild = 2; %Size of Breeding Group

MaxGeneration = 15; %%Define Maximum Number of Iterations for Program MaxWait = 1500; %%Define Time to Wao

EliteIndividuals = 1; %Set Number of Elite Designs to Carry over during

Repopulation SeedDesigns = 1; %Set Number of User Defined Seed Designs RandomIndividuals = round(PopulationSize/5); %Set Number of Randomly

Generated Design to arry over % during Repopulation

%%%%%Set Remaining File Paths

SeedDesignFolder = fullfile(WorkingDirectory,'SeedDesigns'); %%Create Folder for Seed Designs GeomFolder = fullfile(WorkingDirectory,'GeometryFiles'); %%Create Folder for Geometry

files of %Best Individuals ScriptPath = fullfile(WorkingDirectory,'DesignModelerScript.js'); ResultsFile = 'Output.csv';

%%%%%Set Scoring Constants

weightSum = 0;

for count = 1:1:(PopulationSize-RandomIndividuals) weightSum = weightSum + 1/count; end

for count = 1:1:(PopulationSize-RandomIndividuals)

Page 135: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

118

weight = 1/count; BreedingProbability(count,:) = weight/weightSum;

end

%%%%%Begin Automated Process

mkdir(SeedDesignFolder); %%Create Folders mkdir(GeomFolder);

X_Elements = deltaX/x; %%Determine Number of WS Elements in X Direction Y_Elements = deltaY/y; %%Determine Number of WS Elements in Y Direction Z_Elements = deltaZ/z; %%Determine Number of WS Elements in Z Direction

Horizontal_Coordinate = Y_Elements; %%Flow Direction Vertical_Coordinate = Z_Elements; %%Up-Down Lateral_Coordinate = X_Elements; %%Width

ElementCount = X_Elements*Y_Elements*Z_Elements; %%Determine Total Number of WS Elements

%%%%%Remaining User Defined/Related Factors % X = []; % % for n = 1:2:42 % X = [X,((n*:1:40)]; % end SeedDesign(1,:) =

{[(1:1:20),(41:1:60),(81:1:100),(121:1:140),(161:1:180),(201:1:220),(241:1:260),(281:1:300),(321:

1:340),(361:1:380),(401:1:420),]};%{[(1:1:9),(19:1:27),(37:1:45),(55:1:63)]};

%%Input Seed Design(s) %SeedDesign(2,:) = {[4,5,6]};

CrossoverPoint = round(0.01*ElementCount); %Input Size of Chromosone String Defined as Specific

Number % of WS Bodies or as Percentage of Total Number of WS

Bodies

WorkSpaceMatrix = zeros(Horizontal_Coordinate,Lateral_Coordinate,Vertical_Coordinate);

%%Create empty WS Matrix

for i = 1:1:ElementCount

WorkSpaceMatrix(i) = i; %%Fill WS Matrix and Ensure this Grid Indexing Aligns

with that of the % physical matrix of Design Element % within the model end

WorkSpaceMatrix2 = rot90(fliplr(WorkSpaceMatrix),1); BinaryMatrix = zeros(size(WorkSpaceMatrix2)); ReferenceMatrix = reshape((1:1:ElementCount),size(WorkSpaceMatrix2));

WorkSpace2Reference = containers.Map(WorkSpaceMatrix2,ReferenceMatrix); %%%%%Create Monitor Plots

figure(1)

TempPlot = subplot(2,2,1); title('Temperature Plot'); ylabel('Chip Temp (C)'); xlabel('Generation'); hold on

PressurePlot = subplot(2,2,2); title('Pressure Plot');

Page 136: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

119

ylabel('Pa'); xlabel('Generation'); hold on

ConvergPlot = subplot(2,2,[3 4]); title('Convergence'); ylabel('Fitness'); xlabel('Generation'); hold on

%%%%%Create Initial Population of Seed Designs

for SD = 1:1:PopulationSize

if SD<=SeedDesigns SolutionVector = SeedDesign{SD,1}; ValidIndividual = ismember(WorkSpaceMatrix2,SolutionVector); else [NewIndividual] =

GenerateSeedDesign(WorkSpace2Reference,BinaryMatrix,ElementCount,SeedMutationProbability); [ValidIndividual,SolutionVector] =

ValidateSeedDesign(NewIndividual,WorkSpaceMatrix2,ReferenceMatrix,Horizontal_Coordinate,Vertical_

Coordinate,Lateral_Coordinate,ElementCount); %%Function to take in Generated Designs and Validate

them for Physics end

%Individual_Label = sprintf('%s/Individual_%d',GeomFolder,SD);

InitialPopulation(SD,:) = {SolutionVector}; %%Store Solution Vectors to Population

Vector

end

%%%%%Initialize Numerics and Population

NewPopulation = InitialPopulation; GenerationCount = 1;

%%%%% Begin Main GA Loop %%%%%%

while GenerationCount<=MaxGeneration

%%%% Evaluate Each Individual in Current Population for PopulationCount = 1:1:PopulationSize

CurrentIndividual = NewPopulation{PopulationCount};

callout = sprintf('Current Individual = %d Current Generation =

%d',PopulationCount,GenerationCount); disp(callout); %%Callout Status of Evaluation Process

[ScriptFile] = GenerateScriptFile(CurrentIndividual,ElementCount); %%Write DM Scipt

File Corresponding to Current %%Individuals

Solution Vector

if PopulationCount==1 NewFileLabel =

sprintf('%s/Design_%d_%d.js',GeomFolder,PopulationCount,GenerationCount); copyfile(ScriptFile,NewFileLabel); % Store Geometry Script File end

if GenerationCount==1 NewFileLabel =

sprintf('%s/Design_%d_%d.js',SeedDesignFolder,PopulationCount,GenerationCount); copyfile(ScriptFile,NewFileLabel); % Store Geometry Script File

Page 137: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

120

end

[Temperature,Pressure] = Simulate(ScriptPath,ResultsFile,MaxWait);

if (GenerationCount==1)&&(PopulationCount==1) Base_Temp = Temperature; Base_Press = Pressure; end

[FitnessValue] = FitnessFunction(Temperature,Pressure,Base_Temp,Base_Press); %%Fitness

Function

%%%%% Store Performance Values, Fitness Score and Solution Vectors of Each Individual in

current Population EvaluatedPopulation(PopulationCount,:) = [FitnessValue, {CurrentIndividual},

PopulationCount, Temperature, Pressure];

execwbcommand('setup1.Exit()'); %%Close/Reset Icepak Window

end

%%%%% Rank Population, Graph & Store Best Performing Design of Generation %%%%%

RankedPopulation = sortrows(EvaluatedPopulation,1); %%ReOrder from Lowest-to-Highest Based

on Fitness Score TrimmedPopulation = RankedPopulation([1:(PopulationSize-RandomIndividuals)],:); %%Remove

Lostest Performing Individuals

BestFitness = RankedPopulation{1,1};%%Check these values BestTemp = RankedPopulation{1,4};%%Check these indexed values BestPress = RankedPopulation{1,5};%%Check these indexed values

TopPerformers(GenerationCount,:) = BestFitness; TopTemp(GenerationCount,:) = BestTemp; TopPress(GenerationCount,:) = BestPress;

callout = sprintf('Top Fitness of Generation = %f At Individual =

%d',BestFitness,RankedPopulation{1,3}); disp(callout);

subplot(2,2,1) scatter(GenerationCount,BestTemp,'filled','b')

subplot(2,2,2) scatter(GenerationCount,BestPress,'filled','m')

subplot(2,2,[3 4]) scatter(GenerationCount,BestFitness,'filled','k')

%%%%% Repopulate %%%%%

for RepopulateCount = 1:1:PopulationSize

if RepopulateCount<=EliteIndividuals

SolutionVector = RankedPopulation{RepopulateCount,2};

elseif RepopulateCount>(PopulationSize-RandomIndividuals)

[NewIndividual] =

GenerateRandomDesign(WorkSpace2Reference,BinaryMatrix,ElementCount,MutationProbability); [ValidIndividual,SolutionVector] =

ValidateRandomDesign(NewChild,WorkSpaceMatrix2,ReferenceMatrix,Horizontal_Coordinate,Vertical_Coo

rdinate,Lateral_Coordinate,ElementCount);

Page 138: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

121

else [NewChild,BreedingGroup] =

CreateChild(TrimmedPopulation,ParentsPerChild,CrossoverPoint,BreedingProbability,MutationProbabil

ity,ElementCount); [ValidIndividual,SolutionVector] =

ValidateChildDesign(NewChild,WorkSpaceMatrix2,ReferenceMatrix,Horizontal_Coordinate,Vertical_Coor

dinate,Lateral_Coordinate,ElementCount); end

NewPopulation(RepopulateCount,:) = {SolutionVector}; %%Store Solution Vectors of

Newly Formed Childern to New Population Group

end

GenerationCount = GenerationCount+1;

end

Elapsed_Time = toc; toc;

%%%%% End Process and Callout Final Info

clear('orb'); %% Cant Save this Object - causes Errors save('Workspace Variables.mat'); %% Save All Variables saveas(gcf,'Convergence Plot.fig'); %% Save Convergence Plot

BestFitnessCount = length(unique(TopPerformers)); callout = sprintf('Number Of Different Top Performing Designs = %2d',BestFitnessCount); disp(callout);

Page 139: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

122

Initialization Function

function [] = initialize(GeometryFile,WorkingDirectory)

Command = '"C:\Program Files\ANSYS Inc\v192\Framework\bin\Win64\runwb2.exe" -I -E "with

open(''aaS_WbId.txt'',''w'') as file:file.write(Server.StartServer(''localhost'',0,9000,9200))"

&';

status = system(Command);

pause(30);

orb=initialize_orb(); %% Only need to be executed once per session

load_ansys_aas();

actwbserver('C:/Users/Andrew/aaS_WbID.txt');

execwbcommand('Open(FilePath="C:/Users/Andrew/Desktop/Thesis/Thesis Example Simulations/Chapter

4/Temp_Simulate.wbpj")');

execwbcommand('system1 = GetSystem(Name="Geom")');

execwbcommand('geometry1 = system1.GetContainer(ComponentName="Geometry")');

execwbcommand('geometry1.Edit()');

execwbcommand('system2 = GetSystem(Name="IPK")');

execwbcommand('setup1 = system2.GetContainer(ComponentName="Setup")');

execwbcommand('setup1.Edit(SystemCoordinate="B")');

Page 140: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

123

Seed Design Generation Function

function [NewIndividual] =

GenerateSeedDesign(WorkSpace2Reference,BinaryMatrix,ElementCount,MutationProbability)

NewBinary = BinaryMatrix;

for element = 1:1:ElementCount

x =rand;

if x<=MutationProbability

NewBinary(WorkSpace2Reference(element))=~NewBinary(WorkSpace2Reference(element));

end

end

NewIndividual = NewBinary;

Page 141: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

124

Seed Design Validation Function

function [ValidIndividual,SolutionVector] =

ValidateSeedDesign(NewIndividual,WorkSpaceMatrix2,ReferenceMatrix,Horizontal_Coordinate,Vertical_

Coordinate,Lateral_Coordinate,ElementCount)

IsValid = 0;

FluidElements = [];

while ~(IsValid == 1)

IsValid = 1;

IdentifyBlobs = bwconncomp(NewIndividual,6);

BlobNum = IdentifyBlobs.NumObjects;

BlobsROI = regionprops(IdentifyBlobs,'BoundingBox');

for num = 1:1:BlobNum

BlobBox = BlobsROI(num);

XBound = BlobBox.BoundingBox(1); %%Right-to-Left in Binary Matrix

YBound = BlobBox.BoundingBox(2); %%Up-Down in Binary Matrix

ZBound = BlobBox.BoundingBox(3); %%Across the Matrice Levels

BoundWidth = BlobBox.BoundingBox(4); %%Number of Pixels Across

BoundHeight = BlobBox.BoundingBox(5); %%Number of Pixels Up-Down

BoundDepth = BlobBox.BoundingBox(6); %%Number of Pixels Deep

if (BoundWidth~=Horizontal_Coordinate)

IsValid = 0;

SE_line = strel('line',3,180);

BlobBinary = ismember(ReferenceMatrix,IdentifyBlobs.PixelIdxList{num});

BlobBinary = imdilate(BlobBinary,SE_line);

NewIndividual = NewIndividual|BlobBinary;

end

end

end

ValidIndividual = NewIndividual;

WorkSpace2Individual = containers.Map(WorkSpaceMatrix2,NewIndividual);

FCount = 1;

for element = 1:1:ElementCount

if (WorkSpace2Individual(element)== 1)

FluidElements(FCount,:) = element;

FCount = FCount+1;

end

end

SolutionVector = FluidElements;

Page 142: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

125

Write Design Modeler Script Function

function [ScriptFile] = GenerateScriptFile(SolutionVector, ElementCount)

A = (1:1:ElementCount);

B = ismember(A,SolutionVector);

SolidVector = A(~B);

ScriptFile = sprintf('%s.js','DesignModelerScript');

ScriptWrite = fopen(ScriptFile,'w');

DocumentStart = '////This Document Clears Previous BOOLEAN Functions, \n ////then Selects Bodies

for Fluid and HeatSink BOOLEAN Combine';

fprintf(ScriptWrite,'%s\n\n' ,DocumentStart);

%%%%%Delete Pre-Existing Booleans

fprintf(ScriptWrite,'ag.selectedFeature = ag.gui.TreeviewFeature("FluidCombine", 0);\n');

fprintf(ScriptWrite,'ag.selectedFeature = ag.gui.TreeviewFeature("SolidCombine", 0);\n');

fprintf(ScriptWrite,'ag.gui.FeatureSuppression (701 - 700);\n\n');

fprintf(ScriptWrite,'ag.selectedFeature = ag.gui.TreeviewFeature("FluidCombine", 0);\n');

fprintf(ScriptWrite,'ag.gui.Delete(0);\n\n');

fprintf(ScriptWrite,'ag.selectedFeature = ag.gui.TreeviewFeature("SolidCombine", 0);\n');

fprintf(ScriptWrite,'ag.gui.Delete(0);\n\n');

%%%%%Combine Fluid Bodies With Boolean

FluidBoolean1 = 'var F_Boolean= ag.gui.CreateBoolean();';

FluidBoolean2 = 'F_Boolean.Name = "FluidCombine";';

FluidBoolean3 = 'F_Boolean.Operation = 0; ';

FluidBoolean4 = 'ag.listview.ActivateItem("Tool Bodies"); ';

FluidBoolean5 = 'agb.ClearSelections(); ';

fprintf(ScriptWrite,'%s\n%s\n%s\n%s\n%s\n\n',FluidBoolean1,FluidBoolean2,FluidBoolean3,FluidBoole

an4,FluidBoolean5);

fprintf(ScriptWrite,'F1 = selectNode("Fluid.1");\n');

fprintf(ScriptWrite,'ag.bodyPick;\n');

fprintf(ScriptWrite,'agb.AddSelect(agc.TypeBody, F1);\n\n');

fprintf(ScriptWrite,'F2 = selectNode("Fluid.2");\n');

fprintf(ScriptWrite,'ag.bodyPick;\n');

fprintf(ScriptWrite,'agb.AddSelect(agc.TypeBody, F2);\n\n');

for block = 1:1:(length(SolutionVector))

VarLabel = sprintf('FBlock%d',block);

BodyLabel = sprintf('Body.%d',SolutionVector(block));

fprintf(ScriptWrite,'%s = selectNode("%s");\n',VarLabel,BodyLabel);

fprintf(ScriptWrite,'ag.bodyPick;\n');

fprintf(ScriptWrite,'agb.AddSelect(agc.TypeBody, %s);\n\n',VarLabel);

end

fprintf(ScriptWrite,'ag.listview.ItemValue = "Apply";\n\nag.b.Regen();\n');

%%%%%Combine Solid Bodies with Boolean

SolidBoolean1 = 'var S_Boolean= ag.gui.CreateBoolean();';

SolidBoolean2 = 'S_Boolean.Name = "SolidCombine";';

SolidBoolean3 = 'S_Boolean.Operation = 0; ';

SolidBoolean4 = 'ag.listview.ActivateItem("Tool Bodies"); ';

SolidBoolean5 = 'agb.ClearSelections(); ';

fprintf(ScriptWrite,'%s\n%s\n%s\n%s\n%s\n\n',SolidBoolean1,SolidBoolean2,SolidBoolean3,SolidBoole

an4,SolidBoolean5);

Page 143: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

126

fprintf(ScriptWrite,'HS = selectNode("Heat Sink");\n');

fprintf(ScriptWrite,'ag.bodyPick;\n');

fprintf(ScriptWrite,'agb.AddSelect(agc.TypeBody, HS);\n\n');

for block = 1:1:(length(SolidVector))

VarLabel = sprintf('SBlock%d',block);

BodyLabel = sprintf('Body.%d',SolidVector(block));

fprintf(ScriptWrite,'%s = selectNode("%s");\n',VarLabel,BodyLabel);

fprintf(ScriptWrite,'ag.bodyPick;\n');

fprintf(ScriptWrite,'agb.AddSelect(agc.TypeBody, %s);\n\n',VarLabel);

end

fprintf(ScriptWrite,'ag.listview.ItemValue = "Apply";\n\nag.b.Regen();\n\n');

%%%%%Define 'Select Node Function'

fprintf(ScriptWrite, 'function selectNode (target)\n{\n var DM = ag.wb.AppletList.Applet(

"AGApplet" ).App;\n var Nodes = DM.Script.ag.tree.Nodes;\n var count = Nodes.Count;\n

var name, current;\n for (var i =1; i <= count; i++)\n {\n current = Nodes(i);\n

name = current.Text.toLowerCase();\n if (name == target.toLowerCase())\n {\n

DM.Script.agTree_LeftClick(current, false);\n var obj = ag.listviewSelectedObject;\n

return obj;\n }\n }\n}');

Page 144: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

127

ANSYS Simulation Function

function [Temperature,Pressure] = Simulate(ScriptPath,ResultsFile,MaxWait)

WaitCount = 0;

execwbcommand('geometry1.SendCommand(Command

="(ag.wb.ScriptEngine.RunScript(''C:/Users/Andrew/Desktop/Thesis/Thesis Example

Simulations/Chapter 4/Initial Testing/DesignModelerScript.js''))")');

execwbcommand('boolean1 = CheckPartialUpdateAndRetainPartialUpdatePropertiesSetConsistently()');

execwbcommand('designPoint1 = Parameters.GetDesignPoint(Name="0")');

execwbcommand('backgroundSession1 = UpdateAllDesignPoints(DesignPoints=[designPoint1])');

execwbcommand('Parameters.ExportAllDesignPointsData(FilePath="C:/Users/Andrew/Desktop/Thesis/Thes

is Example Simulations/Chapter 4/Initial Testing/Output.csv")');

while exist(ResultsFile)==0

pause(1);

WaitCount = WaitCount+1;

if WaitCount>=MaxWait

disp('No Output File Created in Allowed Time, Suspectyed Simulation Error');

break;

end

end

%%%%% Check and Read Results File From WB

if isfile(ResultsFile)

[Temperature,Pressure] = CFD_Results(ResultsFile); %%Function to Read Contents of WB Results

File

if Pressure<=0 %% Check if Results are Currupt

Pressure = NaN; %% If so apply 'BAD' values to Performance Variables

Temperature = NaN;

end

fclose('all');

delete(ResultsFile);

else

Pressure = NaN;

Temperature = NaN;

end

Page 145: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

128

Read ANSYS Results Function

function [Temperature,Pressure] = CFD_Results(filename, startRow, endRow)

%IMPORTFILE Import numeric data from a text file as a matrix.

% OUTPUT = IMPORTFILE(FILENAME) Reads data from text file FILENAME for

% the default selection.

%

% OUTPUT = IMPORTFILE(FILENAME, STARTROW, ENDROW) Reads data from rows

% STARTROW through ENDROW of text file FILENAME.

%

% Example:

% Output = importfile('Output.csv', 8, 8);

%

% See also TEXTSCAN.

% Auto-generated by MATLAB on 2019/05/03 17:44:38

%% Initialize variables.

delimiter = ',';

if nargin<=2

startRow = 8;

endRow = inf;

end

%% Format for each line of text:

% column2: double (%f)

% column3: double (%f)

% For more information, see the TEXTSCAN documentation.

formatSpec = '%*s%f%f%[^\n\r]';

%% Open the text file.

fileID = fopen(filename,'r','n','UTF-8');

% Skip the BOM (Byte Order Mark).

fseek(fileID, 3, 'bof');

%% Read columns of data according to the format.

% This call is based on the structure of the file used to generate this

% code. If an error occurs for a different file, try regenerating the code

% from the Import Tool.

dataArray = textscan(fileID, formatSpec, endRow(1)-startRow(1)+1, 'Delimiter', delimiter,

'TextType', 'string', 'HeaderLines', startRow(1)-1, 'ReturnOnError', false, 'EndOfLine', '\r\n');

for block=2:length(startRow)

frewind(fileID);

dataArrayBlock = textscan(fileID, formatSpec, endRow(block)-startRow(block)+1, 'Delimiter',

delimiter, 'TextType', 'string', 'HeaderLines', startRow(block)-1, 'ReturnOnError', false,

'EndOfLine', '\r\n');

for col=1:length(dataArray)

dataArray{col} = [dataArray{col};dataArrayBlock{col}];

end

end

%% Close the text file.

fclose(fileID);

%% Post processing for unimportable data.

% No unimportable data rules were applied during the import, so no post

% processing code is included. To generate code which works for

% unimportable data, select unimportable cells in a file and regenerate the

% script.

%% Create output variable

Output = table(dataArray{1:end-1}, 'VariableNames', {'P1','P2'});

Temperature = table2array(Output(1,1));

Pressure = table2array(Output(1,2));

Page 146: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

129

ANSYS Simulation Function

function [FitnessScore] = FitnessFunction(Temperature,Pressure,Base_Temp,Base_Press)

a = 0.70;

b = 1-a;

FitnessScore = (a*(Temperature/Base_Temp)) + (b*(Pressure/Base_Press));

Page 147: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

130

Generate Child Design Function

function [NewChild,BreedingGroup] =

CreateChild(TrimmedPopulation,ParentsPerChild,CrossoverPoint,BreedingProbability,MutationProbabil

ity, ElementCount)

BreedingGroup = []; %%Create Empty Array for Breeding Group

BreedingParents = {};

ChildVector = [];

GeneStringCount = 1;

ChildGeneCount = 1;

Parent = 1;

while Parent<=ParentsPerChild

x = rand;

while x>BreedingProbability(1)

x = rand;

end

for count = 1:1:size(TrimmedPopulation,1)

if BreedingProbability(count) > x

Prob = BreedingProbability(count);

CurrentParent = (TrimmedPopulation(count,2));

end

end

if ~ismember(Prob,BreedingGroup)

BreedingGroup = [BreedingGroup Prob];

BreedingParents(Parent,:) = CurrentParent;

Parent = Parent+1;

else

Parent=Parent;

end

end

ChooseParent = randi(ParentsPerChild);

SelectedParent = cell2mat(BreedingParents(ChooseParent));

for CurrentGene = 1:1:ElementCount

if GeneStringCount>=CrossoverPoint

GeneStringCount = 1;

ChooseParent = randi(ParentsPerChild);

SelectedParent = cell2mat(BreedingParents(ChooseParent));

end

if ismember(CurrentGene,SelectedParent)

ChildVector(ChildGeneCount,:) = CurrentGene;

ChildGeneCount = ChildGeneCount+1;

end

end

%%%%% Mutate Child Design Genes

for element = 1:1:ElementCount

x = rand;

if x<=MutationProbability

if ismember(element,ChildVector)

NewChild(element,:) = NaN;

else

NewChild(element,:) = element;

end

else

Page 148: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

131

if ismember(element,ChildVector)

NewChild(element,:) = element;

else

NewChild(element,:) = NaN;

end

end

end

Page 149: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

132

Validate Child Design Function

function [ValidIndividual,SolutionVector] =

ValidateChildDesign(NewChild,WorkSpaceMatrix2,ReferenceMatrix,Horizontal_Coordinate,Vertical_Coor

dinate,Lateral_Coordinate,ElementCount)

FluidValid = 0;

SolidValid = 0;

FluidElements = [];

NewIndividual = ismember(WorkSpaceMatrix2,NewChild);

while ~(FluidValid == 1)

FluidValid = 1;

IdentifyBlobs = bwconncomp(NewIndividual,6);

BlobNum = IdentifyBlobs.NumObjects;

BlobsROI = regionprops(IdentifyBlobs,'BoundingBox');

for num = 1:1:BlobNum

BlobBox = BlobsROI(num);

XBound = BlobBox.BoundingBox(1); %%Right-to-Left in Binary Matrix

YBound = BlobBox.BoundingBox(2); %%Up-Down in Binary Matrix

ZBound = BlobBox.BoundingBox(3); %%Across the Matrice Levels

BoundWidth = BlobBox.BoundingBox(4); %%Number of Pixels Across

BoundHeight = BlobBox.BoundingBox(5); %%Number of Pixels Up-Down

BoundDepth = BlobBox.BoundingBox(6); %%Number of Pixels Deep

Xe = XBound-0.5+BoundWidth;

if (BoundWidth~=Horizontal_Coordinate)

FluidValid = 0;

if (XBound==0.5)||(Xe==Horizontal_Coordinate)

SE_line = strel('line',3,180);

BlobBinary = ismember(ReferenceMatrix,IdentifyBlobs.PixelIdxList{num});

BlobBinary = imdilate(BlobBinary,SE_line);

NewIndividual = NewIndividual|BlobBinary;

else

BlobBinary = ismember(ReferenceMatrix,IdentifyBlobs.PixelIdxList{num});

NewIndividual = NewIndividual - BlobBinary;

end

end

end

end

NewIndividual = not(NewIndividual); %%Invert Values of Logical array such that 1'

= Solid Elements and 0's = Fluid

while ~(SolidValid == 1)

SolidValid = 1;

IdentifyBlobs = bwconncomp(NewIndividual,6);

BlobNum = IdentifyBlobs.NumObjects;

BlobsROI = regionprops(IdentifyBlobs,'BoundingBox');

for num = 1:1:BlobNum

BlobBox = BlobsROI(num);

Xs = BlobBox.BoundingBox(1); %%Right-to-Left in Binary Matrix

Ys = BlobBox.BoundingBox(2)+0.5; %%Up-Down in Binary Matrix

Page 150: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

133

Zs = BlobBox.BoundingBox(3); %%Across the Matrice Levels

X_Width = BlobBox.BoundingBox(4); %%Number of Pixels Across

Y_Width = BlobBox.BoundingBox(5); %%Number of Pixels Up-Down

Z_Width = BlobBox.BoundingBox(6); %%Number of Pixels Deep

Ye = Ys-0.5+Y_Width;

Ze = Zs-0.5+Z_Width;

if ((Ys~=0.5)&&(Zs~=0.5)&&(Ye~=Lateral_Coordinate)&&(Ze~=Vertical_Coordinate))

SolidValid = 0;

BlobBinary = ismember(ReferenceMatrix,IdentifyBlobs.PixelIdxList{num});

NewIndividual = NewIndividual - BlobBinary;

end

end

end

NewIndividual = ~(NewIndividual);

ValidIndividual = NewIndividual;

WorkSpace2Individual = containers.Map(WorkSpaceMatrix2,NewIndividual);

FCount = 1;

for element = 1:1:ElementCount

if (WorkSpace2Individual(element)== 1)

FluidElements(FCount,:) = element;

FCount = FCount+1;

end

end

SolutionVector = FluidElements;

Page 151: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

134

Generate Random Design Candidate Function

function [NewIndividual] =

GenerateRandomDesign(WorkSpace2Reference,BinaryMatrix,ElementCount,MutationProbability)

NewBinary = BinaryMatrix;

for element = 1:1:ElementCount

x =rand;

if x<=MutationProbability

NewBinary(WorkSpace2Reference(element))=~NewBinary(WorkSpace2Reference(element));

end

end

NewIndividual = NewBinary;

Page 152: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

135

Validate Random Design Candidate Function

function [ValidIndividual,SolutionVector] =

ValidateRandomDesign(NewChild,WorkSpaceMatrix2,ReferenceMatrix,Horizontal_Coordinate,Vertical_Coo

rdinate,Lateral_Coordinate,ElementCount)

FluidValid = 0;

SolidValid = 0;

FluidElements = [];

NewIndividual = ismember(WorkSpaceMatrix2,NewChild);

while ~(FluidValid == 1)

FluidValid = 1;

IdentifyBlobs = bwconncomp(NewIndividual,6);

BlobNum = IdentifyBlobs.NumObjects;

BlobsROI = regionprops(IdentifyBlobs,'BoundingBox');

for num = 1:1:BlobNum

BlobBox = BlobsROI(num);

XBound = BlobBox.BoundingBox(1); %%Right-to-Left in Binary Matrix

YBound = BlobBox.BoundingBox(2); %%Up-Down in Binary Matrix

ZBound = BlobBox.BoundingBox(3); %%Across the Matrice Levels

BoundWidth = BlobBox.BoundingBox(4); %%Number of Pixels Across

BoundHeight = BlobBox.BoundingBox(5); %%Number of Pixels Up-Down

BoundDepth = BlobBox.BoundingBox(6); %%Number of Pixels Deep

if (BoundWidth~=Horizontal_Coordinate)

FluidValid = 0;

SE_line = strel('line',3,180);

BlobBinary = ismember(ReferenceMatrix,IdentifyBlobs.PixelIdxList{num});

BlobBinary = imdilate(BlobBinary,SE_line);

NewIndividual = NewIndividual|BlobBinary;

end

end

end

NewIndividual = not(NewIndividual); %%Invert Values of Logical array such that 1'

= Solid Elements and 0's = Fluid

while ~(SolidValid == 1)

SolidValid = 1;

IdentifyBlobs = bwconncomp(NewIndividual,6);

BlobNum = IdentifyBlobs.NumObjects;

BlobsROI = regionprops(IdentifyBlobs,'BoundingBox');

for num = 1:1:BlobNum

BlobBox = BlobsROI(num);

Xs = BlobBox.BoundingBox(1); %%Right-to-Left in Binary Matrix

Ys = BlobBox.BoundingBox(2)+0.5; %%Up-Down in Binary Matrix

Zs = BlobBox.BoundingBox(3); %%Across the Matrice Levels

X_Width = BlobBox.BoundingBox(4); %%Number of Pixels Across

Y_Width = BlobBox.BoundingBox(5); %%Number of Pixels Up-Down

Z_Width = BlobBox.BoundingBox(6); %%Number of Pixels Deep

Ye = Ys-0.5+Y_Width;

Ze = Zs-0.5+Z_Width;

Page 153: Generative Design Optimization of Thermal Management ... · proposes a novel design methodology that utilizes genetic algorithms to optimize the liquid topologies of compact heat

136

if (Ys~=0.5)&&(Zs~=0.5)&&(Ye~=Lateral_Coordinate)&&(Ze~=Vertical_Coordinate)

SolidValid = 0;

BlobBinary = ismember(ReferenceMatrix,IdentifyBlobs.PixelIdxList{num});

NewIndividual = NewIndividual - BlobBinary;

end

end

end

NewIndividual = ~(NewIndividual);

ValidIndividual = NewIndividual;

WorkSpace2Individual = containers.Map(WorkSpaceMatrix2,NewIndividual);

FCount = 1;

for element = 1:1:ElementCount

if (WorkSpace2Individual(element)== 1)

FluidElements(FCount,:) = element;

FCount = FCount+1;

end

end

SolutionVector = FluidElements;