Design and Optimization of Lithium-Ion Batteries for

Electric-Vehicle Applications

by

Nansi Xue

A dissertation submitted in partial fulfillment

of the requirements for the degree of
Doctor of Philosophy
(Aerospace Engineering)
in the University of Michigan
2014

Doctoral Committee:
Associate Professor Joaquim R. R. A. Martins, Chair
Professor Wei Lu
Professor Anna G. Stefanopoulou
Professor Margaret S. Wooldridge
Nansi Xue

2014
To my parents, who have given me so much.

ii
 A C K N O W L E D G M E N T S

Many people have helped me in my pursuit of the doctorate degree.

First and foremost, I would like to thank my advisor, Professor Joaquim
R.R.A. Martins for his continued guidance and support throughout my
Ph.D. You gave me a chance when I was running out of options. Your
insight and dedication have always been a source of inspiration for me.
And your self-deprecating humor certainly made long group meetings
much more bearable.

Special thanks go to Professor Wei Shyy and Professor Werner Dahm,

who got me interested in pursuing a graduate degree and helped me
across the finish line. I would also like to acknowledge my committee,
Professor Wei Lu, Professor Anna Stefanopoulou and Professor Mar-
garet Wooldridge, for their help and input that went into this thesis.

My research work would not have gone so smoothly if not for my fellow
MDOLab-mates, especially Wenbo Du, John Hwang and Peter Lyu.
All the best in your pursuit of academic, career and personal success.
To my housemates, Lynn and Yunyuan, thank you for all the road
trips, golf sessions, pig-outs and sports days. You are my closest to a
family on this side of the Pacific Ocean, and families dont say goodbyes.

Lastly, I owe everything to my parents, and no amount of words can

express my gratitude to you. As I move foward to the next chapter of my
life, I will continue to give it all and live my life to the fullest, just like
you have taught me.

iii
 TABLE OF CONTENTS

Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

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

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv

Chapter

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Hybrid/Electric Vehicle Designs . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Hybrid/Electric Aircraft Designs . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1 General Aviation Aircraft . . . . . . . . . . . . . . . . . . . . . 11
1.3.2 Unmanned Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Energy Storage Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.1 Lithium-Ion Batteries . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.2 Future Batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.3 Lithium-Ion Battery Recycling . . . . . . . . . . . . . . . . . . . 17
1.5 Objectives and Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Cell Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Physics-Based Cell Model . . . . . . . . . . . . . . . . . . . . . 21
2.2.2 Numerical Treatment of the Cell Model . . . . . . . . . . . . . . 24
2.2.3 Equivalent-Circuit Model . . . . . . . . . . . . . . . . . . . . . 26
2.3 Vehicle Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Optimization Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.1 Gradient-Free Optimizers . . . . . . . . . . . . . . . . . . . . . 29
2.4.2 Gradient-Based Optimizers . . . . . . . . . . . . . . . . . . . . 31

iv
 2.4.3 Derivative Computation . . . . . . . . . . . . . . . . . . . . . . 33
2.4.4 Pareto Optimality . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.5 Hybrid Optimization Method . . . . . . . . . . . . . . . . . . . 36
3 Single-Cell Design Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Power vs. Energy . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.2 Sensitivity at Optimal Designs . . . . . . . . . . . . . . . . . . . 46
3.3.3 Separator Thickness and Electrode Particle Size . . . . . . . . . . 47
3.3.4 Electrode Thickness and Porosity . . . . . . . . . . . . . . . . . 48
3.3.5 Conductivity and Diffusivity . . . . . . . . . . . . . . . . . . . . 50
3.3.6 Practical Battery Optimization . . . . . . . . . . . . . . . . . . . 54
3.3.7 Optimizer Performance . . . . . . . . . . . . . . . . . . . . . . . 55
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4 Design Optimization of a Battery Pack for Plug-in Hybrid Vehicles . . . . . . 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.1 Battery Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.1 Discharge Profile . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.2 Optimization Results . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.3 Driving Cycle Test . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5 Multi-Cell Design Optimization for Electric Vehicle Battery Packs . . . . . . . 77
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 Simplified Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 Theoretical Multi-cell Battery Analysis . . . . . . . . . . . . . . . . . . . 85
5.5 Optimization Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.5.1 Energy Cell Optimization . . . . . . . . . . . . . . . . . . . . . 89
5.5.2 Uniform-cell Battery Pack Optimization . . . . . . . . . . . . . . 90
5.5.3 Power Cell Optimization . . . . . . . . . . . . . . . . . . . . . . 90
5.5.4 Multi-cell versus Uniform-cell Optimization . . . . . . . . . . . 92
5.5.5 Practical Design Considerations . . . . . . . . . . . . . . . . . . 94
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6 Conclusions and Recommendations . . . . . . . . . . . . . . . . . . . . . . . . 98
6.1 Concluding Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2 Realistic Cell Considerations . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2.1 Volume Change in Solid Phase . . . . . . . . . . . . . . . . . . . 100
6.2.2 Modeling of Additives . . . . . . . . . . . . . . . . . . . . . . . 100
6.2.3 Solid-Electrolyte Interface Formation . . . . . . . . . . . . . . . 101

v
 6.2.4 Intercalation-Induced Stress . . . . . . . . . . . . . . . . . . . . 101
6.3 Recommendations for Future Work . . . . . . . . . . . . . . . . . . . . . 102
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

vi
 LIST OF FIGURES

1.1 Total oil consumption by sector from 1971 to 2008 [1]. Other includes agri-
culture, commercial and public services, residential oil consumption. . . . . . 1
1.2 The fuel portion of direct operating costs of major North American airlines
has increased significantly due to rising fuel costs [2] . . . . . . . . . . . . . . 3
1.3 Fuel mass fraction as a function of range for different types of battery. The
range of a proposed electric aircraft with the same fuel mass fraction and pow-
ered by lithium-air battery is about 4300 nautical miles shorter than that of the
conventional B737-800. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Energy density of representative energy storage systems . . . . . . . . . . . . 14

2.1 Structure of a lithium-ion insertion cell shows three separate regions and two
different phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Original energy density function is see-saw shaped. The over-estimate of en-
ergy density increases as the final cell voltage decreases and is only reset when
the number of time step is reduced. The error can be reduced using Lagrangian
interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Thevenin equivalent circuit representation of a battery cell . . . . . . . . . . . 26
2.4 Optimization process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.5 Solution history using ALPSO to solve a 2D problem. Left: initial distribution
of particles. Right: converged optimization where all particles are at the global
optimum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6 Solution history using Sparse Nonlinear OPTimizer (SNOPT) package with
finite-difference derivative approximation on a 2D problem. Left: initial start-
ing location of the optimization. Right: final iteration with the iteration history
shown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.7 An optimization that is initiated from (x1 , x2 ) = (1.5, 1.5) converges to the
local optimum at and fails to find the global optimum . . . . . . . . . . . . . . 33
2.8 Error in the derivatives computed using finite-difference and complex-step ap-
proximations for varying step sizes. The complex-step approach is free of
cancellation errors that dominates finite-differences at small step sizes. . . . . 34
2.9 Derivatives of energy density with respect to cycling rate obtained using the
finite-difference method show much more scattering compared to the complex-
step approximation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

vii
2.10 The eXtended Design Structure Matrix (XDSM) for the hybrid optimization
process [3]. The numbers represent the steps in the optimization process.
The optimization control (Step 0) first initiates the gradient-free optimization
(Loop 14), which provides a rudimentary optimal solution and determines
the optimal integer design variables. The results from the gradient-free opti-
mizion is then used to initialize the gradient-based optimization (Step 5). The
gradient-based optimizater (Loop 69) refines the continuous design variables
for the final optimal solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.1 Contour plot of energy density shows monotonically decreasing energy den-
sity as cycling rate and particle size increases. Maximum energy density oc-
curs at minimum cycling rate and particle size. . . . . . . . . . . . . . . . . . 44
3.2 Contour plot of power density overlaid on top of energy density (light grey
lines) shows how power requirements restrict the design space. The energy
density increases when moving from the top right corner (high cycling rate,
large particles) of the design space towards the bottom left (low cycling rate,
small particles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 Pareto front of optimal energy density vs. specific power. Ragone plots at four
discharge rates are also shown, which represent the variation of energy density
with discharge rates for a particular cell. . . . . . . . . . . . . . . . . . . . . . 46
3.4 Variations of electrode thickness and porosity of optimal cell designs with
respect to cell power requirement . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5 Active material mass ratio and charge capacity ratio for optimal cell designs . . 50
3.6 Diffusivity variations at optimal cell designs. Diffusivities need to be high at
optimal cell designs to facilitate ion movement. . . . . . . . . . . . . . . . . . 51
3.7 Close-up view of diffusivity distributions at optimal cell designs. While the
diffusivity values need to be high, they do not converge to any specific values. . 52
3.8 Variation of energy density as a function of normalized diffusivity . . . . . . . 53
3.9 Number of iterations and optimization time versus cell power requirement . . . 55

4.1 Layout of a battery comprised of uniform cells . . . . . . . . . . . . . . . . . 60

4.2 Discharge profile for a lithium-ion cell undergoing 1 C constant current dis-
charge (main) followed by a 10-second peak power pulse at the end of the
discharge (insert) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Discharge profiles of the 10-second peak current phase. The secant method is
used to determine the maximum current such that the cell potential is exactly
at the minimum voltage at the end of the discharge (insert) . . . . . . . . . . . 64
4.4 Iteration history of an optimization to minimize battery cost showing the evo-
lution of: a) electrode thicknesses, b) cutoff-voltage and no of layers, c) elec-
trode porosities, and d) cost and normalized inequality constraint values . . . . 66
4.5 Contour plots of objective functions on the plane spanning the three optimal
design points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Federal driving cycle speed profiles and the corresponding battery power re-
quirement: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

viii
4.7 Comparison of the voltage and SOC profiles of the initial design and minimum-
mass optimal battery pack discharged through the simulated US06 driving cycle 73
4.8 Comparison of battery performance between initial and optimal designs using
driving cycle data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.1 Pareto front of optimal single-cell designs on a normal scale, comparing the
single-cell design with pack cell design . . . . . . . . . . . . . . . . . . . . . 78
5.2 Comparison of cell internal resistance between the initial PHEV pack design
and the optimized pack design . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.3 Layout of a battery pack with multiple cell designs. The power and energy
packs are connected in series so as to not exceed the voltage limits of other
electrical components. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 Top: variation of power density as functions of energy density. Bottom: mass
for a 30kWh and 120kW battery pack as functions of energy density . . . . . . 86
5.5 Mass of energy cells as a function of battery pack energy and power require-
ment. The specifications of some commercially available electric vehicle (EV)
battery packs are shown as well. . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.6 Mass of multi-cell pack designs as function of pack energy and power require-
ments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.7 Difference between multi-cell and uniform-cell pack designs . . . . . . . . . . 89
5.8 Energy density of energy cells as a function of galvanostatic cycling rate, with
the power-law curve fit shown as well. . . . . . . . . . . . . . . . . . . . . . . 90
5.9 Mass of uniform-cell battery pack designs . . . . . . . . . . . . . . . . . . . . 91
5.10 Power versus energy function of the power cells; the power to energy function
is only weakly quadratic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.11 Difference between uniform-cell battery pack and pure energy cell mass . . . . 93
5.12 Mass of multi-cell battery pack designs . . . . . . . . . . . . . . . . . . . . . 94
5.13 Fraction of pack mass that is power cell . . . . . . . . . . . . . . . . . . . . . 95
5.14 Difference between uniform-cell and multi-cell battery pack mass . . . . . . . 96
5.15 Difference between uniform-cell and multi-cell battery pack mass with mini-
mum 50 Wh/kg energy density . . . . . . . . . . . . . . . . . . . . . . . . . . 97

ix
 LIST OF TABLES

1.1 Net importers of crude oil (2008 Data) [1] . . . . . . . . . . . . . . . . . . . . 2

1.2 Comparison of flying cost for an electric aircraft vs. a piston-engined aircraft.
Electricity cost obtained from Michigan Public Commission Service [4] . . . . 4
1.3 Battery mass required to provide the kinetic and gravitational potential energy
for a Boeing 737-800 to reach cruise altitude and speed [5] . . . . . . . . . . . 9
1.4 Results of the top two teams of the NASA Green Flight Challenge [6, 7] . . . . 12
1.5 Performance comparison of future battery systems [8] . . . . . . . . . . . . . 17

3.1 Lithium-ion cell material properties and fixed parameters [9] . . . . . . . . . . 41

3.2 Design variables and their bounds for single cell optimization . . . . . . . . . 42
3.3 Design variables and their sensitivities at optimal designs. Sensitivity is cal-
culated as percentage change of objective function due to a percentage change
in design variable. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.1 Design variables are the morphological parameters and battery layout vari-
ables that can be easily altered by battery designer. . . . . . . . . . . . . . . . 61
4.2 Conversion of pack-level requirements to cell-level constraints . . . . . . . . . 62
4.3 Battery cell properties of initial designs . . . . . . . . . . . . . . . . . . . . . 65
4.4 Preliminary designs after gradient-free optimization. Results shown are the
best-available ones due to stochastic nature of the augmented Lagrangian par-
ticle swarm optimization (ALPSO) algorithm. . . . . . . . . . . . . . . . . . . 67
4.5 Refined optimal designs obtained using gradient-based optimizations . . . . . 68
4.6 Properties of the vehicle used to complete the driving cycle . . . . . . . . . . . 71
4.7 All electric driving range for various battery designs . . . . . . . . . . . . . . 73

x
 LIST OF ABBREVIATIONS

ALPSO augmented Lagrangian particle swarm optimization

DOH degree of hybridization

EV electric vehicle

GHG greenhouse gases

ICE internal combustion engine

KKT KarushKuhnTucker

MCMB mesocarbon microbead

MEA more electric aircraft

OCV open circuit voltage

PHEV plug-in hybrid electric vehicle

SEI solid electrolyte interface

SOC state of charge

SQP sequential quadratic programming

SNOPT Sparse Nonlinear OPTimizer

UAV unmanned aerial vehicle

XDSM eXtended Design Structure Matrix

xi
 LIST OF SYMBOLS

Electrochemistry variables
a interfacial surface area
A cross-section area of an electrochemical cell
b material unit cost
c salt concentration in electrolyte
cs salt concentration in solid matrix
D diffusion coefficient of electrolyte
Ds diffusion coefficient of solid matrix
f mean molar activity coefficient of electrolyte
F Faradays constant
in transfer current density at the surface of active material
io exchange current density
i2 current density in electrolyte
l thickness
m number of electrochemical cell in parallel
n number of electrochemical cell in series
Q charge capacity
R universal gas constant
T temperature
U surface overpotential
a , c anodic and cathodic transfer quotient
 volume fraction
ionic conductivity in electrolyte
ionic conductivity in solid matrix
1 potential in solid matrix
2 potential in electrolyte
material density
Subscripts
o initial state value
s value in solid matrix
+ positive electrode
negative electrode

Optimization variables
A Jacobian of constraints w.r.t. design variables
cj inequality constraints

xii
ck equality constraints
g gradient vector of objective w.r.t. design variables
pi best position of the ith particle
pg global best position
r1 , r2 random numbers between 0 and 1
s solution to the quadratic subproblem in SQP
V velocity of particle in design space
w0 inertia weight in ALPSO
w1 , w2 confidence parameters in ALPSO
W estimate of second-order derivatives in SQP
x position of particle in design space
t time step value in ALPSO, normally taken to be 1

xiii
 ABSTRACT

Design and Optimization of Lithium-Ion Batteries for Electric-Vehicle Applications

by

Nansi Xue

Chair: Joaquim R. R. A. Martins

Development of alternate energy storage systems for transportation use has been
driven by a combination of environmental preservation, fossil fuel price volatil-
ity and energy security concerns. Lithium-ion battery has emerged as a favored
choice, however its energy density is still orders of magnitude lower than the fos-
sil fuel. There is significant room for improvement in the battery cell and electric
vehicle system designs. The objective of this thesis is to automate the design
optimization of the lithium-ion battery pack. To achieve this goal three separate
optimization problems were formulated to provide guidelines on the cell parame-
ters at optimal solutions. The single cell design optimization is able to quantify the
variations of morphological parameters as a constant active mass ratio; the plug-
in hybrid vehicle battery design demonstrates an automated design process that
considers realistic performance constraints; the multi-cell design approach mini-
mizes the battery pack mass by utilizing separate cell designs to satisfy different
constraints. The usefulness of the current framework can be further enhanced by
considering various aging mechansims and to perform a design-control coupled
multidisciplinary optimization.

xiv
 CHAPTER 1

Introduction

1.1 Motivation
Peoples mobility has been significantly enhanced in the last two centuries by the invention
of new means of transportation, such as automobiles and airplanes. Due to their higher
speeds, these inventions have shortened the travel time across both continents and oceans.
While these means of transportation allow us to reach all corners of the world, they are
energy intensive and depend primarily on fossil fuels. In the past half century or so, hu-
mans demand for fossil fuels has steadily climbed, as both the larger population and their
economic prosperity has increased.
Million tons

Figure 1.1: Total oil consumption by sector from 1971 to 2008 [1]. Other includes agri-
culture, commercial and public services, residential oil consumption.

The rise in fuel demand is shown in Figure 1.1. In 2008, 3502 million tons of oil were

1
consumed globally, and a significant portion of it (61.4%) was used for transport [1]. In
1971, transport only accounted for 45.3% of world oil consumption. For countries without
their own reserve, or countries that do not produce enough oil, importing oil is the only
option to sustain domestic demand. Dependence on foreign oil is most evident in the United
States, where foreign oil accounted for more than one quarter of the worlds crude oil
import in 2008 [1]. Such dependence on foreign oil makes a country vulnerable to volatility
in foreign oil supplies, potentially leading to international crisis such as the oil embargo of
1973.

Table 1.1: Net importers of crude oil (2008 Data) [1]

Net importer Weight (106 tons)

United States 564
Japan 199
P. R. China 175
India 128
Korea 116
Germany 106
Italy 88
France 83
Spain 61
Netherlands 57
Others 514
Total 2090

Consumption of such large quantities of fossil fuels for transportation releases an equally
large amount of greenhouse gases (GHG) as well. GHG have been blamed as the main
cause of anthropogenic global warming. In 2011, transportation accounted for 28% of US
primary energy consumption, 93% of which came from petroleum [10]. This directly trans-
lates to 28% of GHG emissions [11]. The ability to control the amount and the sources of
energy used for transportation can result in a significant reduction in the amount of GHG
released into the atmosphere as well.
The difficulties in controlling the GHG emissions and the over-dependence of fossil
fuels play major roles in shaping the future of transportation. The impacts of these factors
are most evident in the air transport industry. In the past 30 years, great strides have been
made to make commercial airliners much more efficient by lowering the energy used per
passenger mile traveled and by increasing the aircraft utilization factor [12]. However, the
improvement in efficiency has been offset by the rising fuel costs. The crude oil prices have
risen from US$20 per barrel from 1985 to the peak price of US$140 in 2008. The seven-

2
 2001 2008
Fuel Fuel

13.4% Depreciation 34.2%

Labor Depreciation
36.2% 6.0%
5.5% Rental 4.5%
Rental
3.0%
Labor 21.5%

38.9% 36.9%

Other Other
Figure 1.2: The fuel portion of direct operating costs of major North American airlines has
increased significantly due to rising fuel costs [2]

fold increase in oil prices drastically increased the fuel portion of overall aircraft operating
costs. As shown in Figure 1.2, fuel cost has increased from 13.4% in 2001 to 34.2% in
2008 for all major North American airlines. The net profit margins for airline industry
globally were negative for six of the eight years in the corresponding time span [13]. The
total fuel-related expenses for 2013 is expected to be about USD $213-billion, while the
total profit is forecast to be about USD$ 11.7-billion (September 2013 forecast [14]), thus
making the profit margin slim and extremely sensitive to fuel cost fluctuations.
While the air transport industry currently produces only 2% of anthropogenic CO2 [15],
the growth of air industry makes it one of the fastest growing sources of GHG. The problem
is further compounded by the altitude effect of GHG. The aircraft emissions of NOx at
cruise altitude increase the production of ozone in the upper troposphere. The net effect is a
higher radiative forcing than if the NOx were emitted at lower altitudes. This increase in the
contribution towards climate warming has been estimated to be 24 times the contribution
of NOx emissions at sea level [15].
The initiative to move away from using fossil fuels as the energy source for transport
use, therefore, arises from the need to address the following concerns:

Energy security: reduce dependence on foreign oil and to sustain development while fac-
ing decreasing available resources.

Environmental conservation: sustain development without negatively impacting the en-

vironment.

Revenue protection: maintain profitability and reduce the operating costs by insulating

3
 against fluctuating fuel prices.
To address these issues, various green technologies, such as EVs, battery technology, and
alternative propulsion systems have gained prominence. The development has been most
obvious in the automotive industry, due to the need to improve vehicle fuel efficiency and
to satisfy increasingly stringent emission standards. Spurred by the feasibility of hydro-
gen fuel cells and development of higher energy density batteries, EVs have been demon-
strated as possible successors of traditional vehicles operating with an internal combustion
engine (ICE). Various energy carriers are available to power EV of different architecture.
Section 1.2 will explain various types of EVs, while Section 1.4 will discuss more about
various energy storage systems.
One of the main advantages of electric-powered vehicles is the significantly lower op-
erating costs compared to ICE powered vehicles. Table 1.2 shows an example comparing
the cost of flying a piston-engine general aviation aircraft and a theoretical electric aircraft
of the same design.

Table 1.2: Comparison of flying cost for an electric aircraft vs. a piston-engined aircraft.
Electricity cost obtained from Michigan Public Commission Service [4]

Description Value
1 gallon of 100LL fuel 35.3 kWh
100LL fuel cost $6.63/gal (DTW price)
Unit cost of 100LL fuel 18.78 cents/kWh
MI electricity cost (09/2013) 7.0518.37 cents/kWh
Electric drive system efficiency 90%
Aircraft engine at achieves .45 brake s.f.c. 30%
Decrease in flying cost due to electrification 38 lower

Electric powertrains are much more efficient than a piston-engines, converting up to

90% of the energy to useful shaft power. The cost of electricity varies depending on the
local supplier, but it is either comparable or significantly lower than the cost per unit en-
ergy of aviation fuel. This results in 38 times lower operating cost from a purely energy
perspective. With ever increasing fuel prices, the economic benefit of electrification of
vehicles is only going to increase as well.
While EVs do not produce any in-situ pollutants, the electricity powering the vehicle is
a formed of processed energy that has to be produced off-site. In terms of emissions pro-
duced, an EV is only as clean as the method used to produce the electricity. In the US, 67%
of electricity is generated by fossil fuels, 37% of which is produced from coal [16], the dirti-
est of all electricity production methods. While the electricity grid production methods vary

4
across the country, studies have shown that the amount of emissions from the least clean
electricity grid is comparable to the best non-hybrid vehicles, while the emissions from
the clean grid is much less than the amount produced from hybrid vehicles [17]. There-
fore, in addition to financial benefits, EVs can reduce transport-related pollution as well in
countries where most of the energy comes from nuclear or renewable energy sources.
Large transport electrical aircraft are not feasible with current battery and electric propul-
sion technology. The current generation batteries cannot be used to power large airliners
due to the low energy density. Therefore the development of electric aircraft has been
restricted to small general aviation aircraft and partial electrification of pneumatic and hy-
draulic systems. While the development of electric flyers have been limited by technology,
the benefits of such aircraft should be obvious. Even with the assumption that the future
electric aircraft have the same level of emissions as current aircraft, the simple act of trans-
ferring emission sources for aircraft at altitude to ground-based power plant will help to
reduce the net effect of GHG emissions on the climate.

1.2 Hybrid/Electric Vehicle Designs

EVs have existed for more than a century by now. In 1899, a Belgian electric vehicle
powered by lead-acid battery was able to reach 30 m/s [18]. However, the lack of progress
in batteries hindered the development of EVs and it was not until recently that electric and
hybrid vehicles re-emerged. EVs are powered entirely by electric propulsion systems, while
hybrid vehicles have two or more power sourcesnormally an ICE coupled to an electric
motor/generator powered by an electric energy storage system. A useful way to define the
powertrain characteristics of such vehicles is to use degree of hybridization (DOH), which
is defined as [19]:

electric motor power

DOH = (1.1)
electric motor power + IC engine power

Depending on the DOH of the vehicle, a hybrid vehicle can be classified into the fol-
lowing groups:

Mild hybrid: vehicles which rely on secondary energy storage systems to assist ICE. A
moderately-sized battery is normally used as the second power source. The battery
has limited discharge range, low power output, and it offers slight fuel economy
improvement. This type of vehicle requires little modification to the existing vehicles
and incur the lowest incremental cost among the hybrid options. Vehicles belonging
to this category of hybrids include Toyota Prius and Honda Insights.

5
Plug-in hybrid: vehicles with all-electric driving range. An ICE or turbine is available for
extended range or to recharge the battery. They use a large battery pack with high
power output that can be charged directly from the grid. Such vehicles offer signif-
icant fuel savings and reduced GHG emissions for short commutes. However, large
battery packs incur significant additional vehicle costs and weight. GMs Chevrolet
Volt and Ford C-Max Energi are two of the commercially available plug-in hybrid
electric vehicle (PHEV)s.

Electric vehicle: vehicles with only all-electric driving capability. These use an extremely
large battery pack, and can only be recharged with electricity from the grid. These
vehicles have zero in-situ emissions, but they are currently either much more expen-
sive than conventional vehicles or have very limited range. Nissan Leaf and Model S
are both EVs that contain large lithium-ion battery packs to provide all of the onboard
energy.

While the system design of an all-electric vehicle is straightforward, there are various
ways to configure the drivetrain components of a hybrid vehicle. In a serial configuration,
the electric motor is the only component connected directly to the drive-train. The decou-
pling of the engine from the wheels means it can always operate at an optimum torque and
speed regime. It performs best for low-speed, high-torque applications, such as buses or
other urban work vehicles. However, it is less efficient, as mechanical energy from the
ICE needs to be converted to electrical energy in the generator and then converted back to
mechanical energy again.
The parallel configuration allows wheel to be driven by either the electric motor, the
ICE, or both. The benefit of this system is redundancy, which is important for both civilian
and military vehicles. However, direct connection between the engine and the wheel means
that the ICE may not operate at its most efficient regime, thereby limiting its efficiency.
Alternatively, a power-split configuration can be employed in which neither the ICE
nor the electric motor are directly connected to the drivetrain. A planetary gear is used
to transfer power from either the ICE or the motor to power the vehicle. Such a system
offers increased efficiency and reduced emissions over the previous two systems. However,
design complexity due to the coupling of the various sub-systems adds to the cost and
control strategies required.
Hybrid or electric vehicles offer many advantages over the ICE-powered vehicles. The
additional drivetrain components enable various operating modes to be engaged to maxi-
mize vehicle efficiency. Some of the benefits of hybrid and electric vehicles are [20]:

Idle-off: the average vehicle spends 20% idling, so turning off the engine at idle can sig-

6
 nificantly reduce fuel consumption by 58%. A 35 kW electric motor can spin the
engine up to idle speed in less than .5 seconds, thus enabling a smooth transition.

Regenerative braking: the electric motor and energy storage system can be used to re-
capture some of the energy that would otherwise be lost during braking. 510% fuel
savings can be expected, though the benefit is a function of electrical component
sizes, and requires a brake-by-wire system and an additional clutch between engine
and motor.

Engine downsizing: a smaller engine is usually more efficient for a given load, as it has
lower frictional, heat, and pumping losses. Hybrid systems can be used to augment
engine power during peak demand, thus allowing a smaller engine without loss of
performance. Benefit of downsizing is proportional to electrical component sizes.
A 10-20 kW electric motor and corresponding energy storage system coupled with
a downsized engine can provide 515% fuel savings over an ICE of similar peak
power.

Improved engine efficiency: a hybrid system can keep the engine at higher loads and min-
imize operation at less efficient modes. For example, the vehicle can be powered by
the electric motor alone at low speeds and loads, and highway driving can be pow-
ered by the ICE at lower speed. In addition, hybrid systems allow integration of
innovative engine designs, such as the Atkinson cycle gasoline engine.

Electrical accessories: most accessories (air conditioning compressor, water pump, power-
steering pump) are currently driven directly by mechanical connections to engines.
This creates inefficiency, as accessory speed varies with engine speeds. Hybrids
allows accessories to be powered by electrical energy storage systems directly, al-
lowing their operation to be independent from the ICE.

Hybrid/electric vehicles tend to be more expensive due to the additional drivetrain com-
ponents. Lithium-ion battery packs are especially costly, and can account for up to 25%
of the total vehicle cost in an all-electric vehicle, such as Nissan Leaf. EVs are a rela-
tively new technology that just established its foothold in the mass market. As its design
becomes more refined and gains wider acceptance, the volume of the battery production
should increase accordingly, decreasing the cost [21].
The development of hybrid/electric vehicles in the past two decades have been the re-
sults of better batteries and tighter integration of electric drive systems with the vehicles.
However, hybrid vehicles also face competition from improvements in ICEs. ICEs may be
inefficient, viewed as inherently dirty, and exacerbate dependence on foreign oil. However,

7
they also offer long driving range, are quick to refuel. Recent improvements both in re-
duced emissions and increased fuel economy limit the incremental improvements offered
by alternative propulsion systems [20]. More sophisticated control strategies and additional
electric components could improve the hybrid efficiency and at the same time provide the
smoothness and adequate performance required by drivers. Optimizing all the interactions
in a hybrid system would demand a great deal of engineering design and software devel-
opment, but the benefits of improved efficiency could prove to be worthwhile in the long
term [20].

1.3 Hybrid/Electric Aircraft Designs

While the development of electric aircraft has been encouraged by the recent progress in
battery energy density, the idea has been proposed for close to a century. Patents for electric
airplane propulsion systems have been filed as early as 1924 [22]. The first manned electric
flight was achieved by Heino Brditschka in 1973, when he successfully flew in an electric
variant of the HB-3 motor-glider powered by Ni-Cd batteries. While the flight only lasted
15 minutes, it demonstrated the feasibility of electric aircraft.
There are numerous advantages to using electric propulsion in aircraft. Electric propul-
sion introduces the possibility of using multiple small electric motors instead of large en-
gines, thereby creating a distributed propulsion system [23]. Such a system would increase
safety through redundancy. In addition, the smaller cross-section areas of multiple propul-
sion units would enable embedding them into the airframe, providing additional aerody-
namic benefits, such as boundary layer ingestion, and reducing aircraft weight [24, 25].
The biggest issue hindering the development of electric aircraft is undoubtedly the en-
ergy density limitation of batteries. The energy density of current batteries is still orders
of magnitude lower than that of jet fuel (0.54 MJ/kg for lithium-ion batteries versus 43.02
MJ/kg for Jet-A fuel). While ground-based vehicles can manage the increased weight
due to electrification without drastic reduction in performance, the aircraft is much more
sensitive to mass increase. To demonstrate how the additional battery mass affects the per-
formance of the aircraft, a simplified equation based on conservation of energy is derived
for electric aircraft to examine how battery energy density affects its range. The energy
required for flying can be approximated by:

Etot = Egrav + Eke + Erange

(1.2)
Mbatt Ebatt = M ghcruise + 21 M Vcruise
2
+ 12 Vcruise
2
CD Sref R

8
Table 1.3: Battery mass required to provide the kinetic and gravitational potential energy
for a Boeing 737-800 to reach cruise altitude and speed [5]

B737-800 specifications Values

Maximum landing weight 66349 kg
Maximum fuel capacity 26020 L
Maximum fuel mass 21076 kg
Eing area 124.58 m2
Cruise altitude 10668 m
Cruise speed (IAS) 230 m/s
Cruise drag coefficient 0.03
Energy required to reach cruise altitude 8.70 x 103 MJ
Lithium-ion battery density 150 Wh/kg
Battery mass required to reach cruise altitude 1.61 x 104 kg

i.e., the total energy required by the aircraft is the sum of the gravitational potential energy
at the cruising altitude (Egrav ), the kinetic energy at cruise speed (Eke ), and the energy
required to overcome drag during steady level flight (Erange ). This is also the total amount
of energy provided by the battery of mass Mbatt with an energy density of Ebatt .
Using Equation (1.2) and the data for a Boeing 737-800 aircraft, we perform the calcu-
lation listed in Table 1.3, which shows the battery required to reach the start of cruise.
Table 1.3 shows that the minimum battery mass required to reach cruise altitude and
speed (assuming perfect propulsion efficiency) is already more than 2/3 of the maximum
fuel mass that a Boeing 737-800 can carry. This suggests that even when taking into ac-
count the increased efficiency of the electric propulsion systems, the current generation
batteries alone are unable to provide all the energy needed by a airliner for extended oper-
ation.
The battery mass as a fraction of the maximum landing mass for the Boeing 737 is
plotted as a function of the achievable range of the aircraft in Figure 1.3. We show the
variations of the range with respect to the energy density for the energy density of current
lithium-ion batteries as well as the theoretical values of future battery systems. It is clear
that in order for electric passenger airliners to be viable, the battery energy density has to
be much higher than the current state of the art. However, even with the theoretical energy
density of Li-air battery systems (which has the same energy density as Jet-A fuel), the pro-
posed electric Boeing 737 is still unable to match the maximum range of the conventional
737. An electric aircraft with the same fuel mass and powered by lithium-air battery can
only achieve 1/4 of the maximum range of the as the conventional B737-800. This is due to

9
the fact that unlike a conventional aircraft, which become lighter as fuel is consumed, the
battery mass remains constant during the flight. The additional mass of the battery requires
the aircraft to produce more lift in order to maintain cruise condition, which in turn leads
to reduced range.

100

current Li-ion (1466 NM)

battery mass/max landing mass

10-1 next-gen Li-ion B737-800 max range

(5765 NM)

Zn-air

Li-S
10-2

Li-air

10-3 0
10 101 102 103 104
range, nm

Figure 1.3: Fuel mass fraction as a function of range for different types of battery. The
range of a proposed electric aircraft with the same fuel mass fraction and powered by
lithium-air battery is about 4300 nautical miles shorter than that of the conventional B737-
800.

One of the main issues challenging the wisdom of developing of an electric aircraft is
the lack of energy recapture during flight. A hybrid system on a ground-based vehicle is
able to regenerate energy during braking that would otherwise be lost in a conventional
vehicle. However, there is no such advantage in an electric aircraft. The amount of energy
that can be recuperated by allowing propellers or fans to wind-mill during descent is small,
and since this happens only at the end of the flight, it would not be useful for extending the
range of the aircraft. The problem is further compounded by the improvement in aircraft
efficiency in terms of energy used per passenger mile traveled [12]. Unlike an automobile,
the flight path and cruise conditions of an airliner are predetermined, and therefore the
engine can be designed to be operate at optimum efficiency around cruise condition. Any
off-design flight conditions can be accounted for in the design phase by utilizing a multi-
point design method to maximize the aircraft efficiency over a range of flight plans [26].
Therefore, the improvement in aircraft efficiency from the energy perspective is likely to

10
be even lower compared to the gains in ground-based electric vehicles.
Currently, the electrification of aircraft is limited to small general aviation aircraft
and unmanned aerial vehicle (UAV)s. The electrification of large passenger aircraft is
non-existent due to the lack of a suitable high-energy density charge carrier. The airline
industry is currently promoting more electric aircraft (MEA) to improve aircraft perfor-
mance and reliability. This results from the growing power requirements due to additional
avionics systems, increased use of electro-mechanical actuators, and increased use of info-
entertainment systems. MEA aims to replace onboard hydraulic, pneumatic and mechani-
cal systems with the electrical equivalents in an effort to reduce weight, system complexity,
and maintenance cost [27, 28].
To realize the full benefit afforded by an electric propulsion system, a complete air-
craft redesign that takes advantage of future technologies should be considered. Boeing
considered a 737 equivalent hybrid concept aircraft as on of the NASA N+3 studies [29].
The aircraft has strut-braced high aspect ratio wings and uses geared turboprop engines for
propulsion. It is estimated that the battery density needs to be more than 750Wh/kg in order
for the hybrid system to be viable. NASA developed a hybrid blended-wing body aircraft
concept that takes advantage of distributed propulsion systems [23, 30]. Benefits of such
a distributed propulsion concept include boundary layer ingestion, low noise level due to
lower fan pressure ratio, and lower wing structural weight due to better weight distribution.
While these aircraft designs demonstrate the possible benefits and improvements in future
transport aircraft, they will remain as concepts until technologies such as high-capacity
batteries and superconducting motors become viable.

1.3.1 General Aviation Aircraft

Electric-powered general aviation aircraft concepts have been developed as battery technol-
ogy has improved. Numerous electric general aviation aircraft are already available on the
market. In addition, various hybrid aircraft demonstrators have been produced as well. Just
like in road vehicles, there can be different configurations of hybrid-electric systems. The
DA36 E-Start motor glider, is the worlds first aircraft with a serial configuration propulsion
system and was showcased in 2011 [31]. Flight Designa German companycoupled a
40 hp motor with a 115 hp Rotax 914 aircraft engine in a parallel configuration for a light-
sport aircraft [32]. Electric aircraft are generally characterized by high aspect ratio wings,
lightweight construction, low cruising speed, and limited endurance and range. The perfor-
mance of the aircraft is limited by the size and energy capacity of the battery pack, which
can easily make up to 1/3 of the empty aircraft weight [33]. However, given the low op-

11
 Table 1.4: Results of the top two teams of the NASA Green Flight Challenge [6, 7]

Team Pipistrel e-Genius

Empty mass (kg) 632 kg -
Battery mass (kg) 520 kg -
Energy used (kWh) 65.4 34.7
Distance (miles) 195.9 193.7
Speed (mph) 107.4 105.7
ePMPG 403.5 375.7
Noise at 250 feet (dB) 71.1 59.5

erational cost of electric flyingas highlighted in Table 1.2there is potential for electric
aircraft as a mode of short-range transportation in the near future. The concept of utilizing
electric aircraft as an on-demand vehicle has been explored [34, 35, 36]. Studies show that
these aircraft will have significantly lower operating cost compared to existing general avi-
ation aircraft, and they can be used for trips that are unprofitable for airliners and take too
long in road vehicles. However they are likely to remain as low-range variants with limited
payload capacity until battery technology improves dramatically. Careful integration of the
propulsion system with the airframe that can represent significant variation from existing
airframe is needed to maximize performance.
NASA organized a Green Flight Challenge in 2011 to demonstrate the feasibility of
long distance sustainable flight. The entry aircraft was required to fly 200 miles at 100 mph
while using less than one gallon of gasoline (or equivalent energy) per passenger. The top
two winning aircraft were both electric-powered, with the winning aircraft achieving 403.5
equivalent passenger MPG. Both aircraft are much quieter than a typical piston-engined
aircraft. The noise produced by a typical general aviation aircraft is about 92 dB at 200 feet
away, or more than 20 dB higher than the noise produced by the electric aircrafts at 250
feet away.
The main benefits of electric general aviation aircraft are lower operating cost, im-
proved efficiency, and reduced noise levels. However such aircraft requires significant
redesign from existing airframes in order to be practical and take full advantage of electric
propulsion. The winning aircraft of the NASA Green Flight Challenge, for instance, has a
battery pack that weighs nearly as much as the empty weight of the aircraft [7] as shown
in Table 1.4, yet it is still limited to a range of about 400 km. The degradation and loss of
battery capacity is also a major issue that needs to be addressed.

12
1.3.2 Unmanned Aircraft
The propulsion systems and fuel mass for small UAVs can exceed 60% of the vehicle
mass [37], rendering them more sensitive to propulsion system mass change. In addition,
the decreased aerodynamic efficiency of these vehicles at lower Reynolds number and de-
creased efficiency of power/propulsion systems at smaller scale makes the efficiency of the
power systems very critical to UAV design.
Electric propulsion is a better option than ICE for reconnaissance and surveillance
UAVs due to lower required maintenance and lower noise. However, any benefits of a
hybrid or electric system must be weighed against the loss in payload due to increased
energy storage system mass. For small UAVs, two important criteria often determine the
performance: loiter time (related to active operation time), and rate of climb (related to
vehicle survivability and safety). These two criteria are at odds at each other in an electric
UAV, as one maximizes the energy, while the other maximizes the power. Optimizing for
either objective results in the other being zero [38]. Various research groups have examined
innovative ways to design electric or hybrid UAVs, as they offer increased loiter time and
range compared to an electric-powered one and reduced acoustic and thermal signatures
over a gasoline-powered one [39]. Multiple propulsion systems also allow for more cre-
ative designs, such as the tail-dragger UAV proposed by Aksugur and Inalham [40]. Such
designs can achieve two hours of flight endurance and required only three minutes for a
vertical take-off and landing. A comprehensive review of hybrid propulsion systems for
small UAV is given by Hung et al. [41].
The design of an electric UAV is a multidisciplinary problem that includes aerody-
namic, structural, electric and performance analyses, and naturally lends itself to multidis-
ciplinary design optimization [42]. A fine balance between the onboard energy availability
and achieving the specific operational goals is needed for the best possible design. Fu-
ture design of an electric aircraft can potentially take advantage of the structural rigidity of
batteries and use the energy storage system as part of load-bearing structures [43, 44, 45].
Combining structure and energy functions into a single material could offer improvements
in system performance that would otherwise be impossible through separate individual
system optimizations. This is a long-term challenge that requires development of new
procedures to examine the multi-functional efficiency of such system. Problems such as
adequate load transfer from the structure through the energy storage materials and safety
concerns of battery performance under mechanical stress must be addressed as well [44].

13
 101
Gasoline
Methanol H2 (l)
Zinc-Air battery Natural gas (250 bar)
100 H2 (g) (700 bar)

volumetric energy density, kWh/l

Lithium-ion battery
NiMH battery
10-1
Lead-acid battery

Natural gas
10-2
H2 (g)

10-3 -2
10 10-1 100 101 102
gravimetric energy density, kWh/kg

Figure 1.4: Energy density of representative energy storage systems

1.4 Energy Storage Systems

One of the most crucial aspects of a hybrid/electric vehicle design is the onboard energy
storage devices. There are various options available for energy storage. Depending on
the DOH of the vehicle, the energy storage system can be either battery, fuel-cell, super-
capacitor, or flywheel. Less common choices such as pneumatic power are available as
well [46]. Figure 1.4 shows the comparison of energy density between various systems.
Among the various energy storage systems, batteries in particular have been developed
to power a diverse range of applications due to their ease of use and availability of ex-
isting electric infrastructure. The concept of the battery is simple, but the energy density
of batteries have not been able to keep pace with the progress in electronics that follow
Moores law. This is mainly due to the lack of suitable electrode and electrolyte materials,
and the difficulty in ensuring compatibility at the interfaces. Nevertheless, there have been
some breakthroughs in new materials recently, most notably the development of various
lithium-ion technologies that are the focus of this thesis.
Much progress has been made in the numerical modeling of battery systems to examine
the physical phenomena occurring within the electrochemical cell, with varying levels of
fidelity and computational cost. Single particle models represent each electrode as a single
spherical particle [47, 48]. Liaw et al. [49, 50] developed an equivalent circuit model
and subsequently used it to examine the impact of variation in cell properties on overall

14
energy capacity. Solving the simplified algebraic equations of the equivalent circuit model
enables real-time estimation of battery state of charge (SOC) and health [51]. Newman et
al. [52] developed a pseudo-2D model that uses porous electrode and concentrated solution
theories [53]. This is a macroscopic cell model that treats the electrode as a homogeneous
continuum. A review of various models for predicting cycling performance was written
by Santhanagopalan et al. [54]. Subsequently, various authors have studied the effects
of microstructural variations on transport properties and cell performance using micro-
scale models [55, 56, 57]. Additional work has been done to describe various degradation
mechanisms and side reactions occurring within the cell [58, 59, 60, 61, 62] to account for
differences in performance between ideal electrochemical cells and practical results.
Currently, there is yet to be one alternative energy storage system that is a clear-cut
choice to replace fossil fuels. Despite the improvement in new electrode materials, battery
energy densities are still orders of magnitude lower than that of fossil fuels. Hydrogen fuel
cells can satisfy both the high energy density and the zero GHG emission requirement, but
they have yet to be economically viable, as the electrode requires precious metals and the
infrastructure cost of hydrogen fuel stations is extremely high [63]. Continued research and
development on both the material science and the systems engineering fronts is crucial if
we are to shift away from fossil fuels and towards a more sustainable energy future.

1.4.1 Lithium-Ion Batteries

Lithium-ion batteries are a family of rechargeable batteries that shuttle lithium ions be-
tween the two electrodes during cycling. These batteries have emerged as the preferred
energy storage device for EV applications due to their relatively high energy density com-
pared to other batteries. Lithium-ion batteries are economically more viable than hydrogen
fuel cells as the associated infrastructurea network of charging stationshas a much
lower cost compared to hydrogen refuel stations. Lithium is favored as the anode material
since it is the most electropositive (standard electrode potential = 3.04 V [64]) and it is
also the lightest metal (equivalent mass = 6.94 g/mol [64]), both of which are essential
for high energy density. The lithium-based rechargeable battery was first demonstrated in
the 1970s by using lithium metal as the negative electrode and titanium sulphide as the
positive electrode [65]. Such a battery system was found to have poor cycling behavior as
dendritic growth due to lithium plating upon repeated cycling poses the potential hazards
of short-circuiting and explosion. Lithium metal was subsequently substituted by a second
insertion material as the negative electrode to avoid lithium plating problems [66]. The
lithium-ion battery operates by reversibly incorporating lithium into the active material via

15
an intercalation process, during which the ions are reversibly removed or inserted into a
porous host without significantly changing its structure. The family of compounds of the
form Lix MO2 (where M is Co, Ni, or Mn) was proposed [67, 68] in the 1980s and has
since gained wide-spread acceptance as the active material in cathodes.
The current generation of lithium-ion batteries consists of a cathode made of a metal
oxide with either a layered structure, such as lithium cobalt oxide, or a tunneled structure,
such as lithium manganese oxide. The negative electrode is usually a graphitic carbon.
Given its high energy density (5 times greater than that of lead-acid, and twice that of
Ni-MH), it has become the standard power source for a variety of electrical devices, from
personal electronic devices to vehicles and satellites. In addition, lithium-ion batteries have
low self-discharge rate, long cycle life, and a wide operating temperature range [69]. How-
ever, lithium-ion batteries are also more expensive than other battery types and require
complicated power management units to prevent degradation or thermal runaway due to
abusive use. Lack of overcharge or discharge tolerance has resulted in large battery packs
with limited useful capacity in order to extend battery cycle life for EV operations. Perma-
nent capacity loss also occurs at elevated temperatures. The high initial costs and restriction
on useful capacities resulting in driver range anxiety are two of the biggest obstacles in EV
acceptance. To circumvent these problems, lithium-ion battery costs need to be lowered
through increased production volume [21] and more sustainable production methods [18].
Useful capacity can be improved with more refined battery design and higher energy den-
sity materials.

1.4.2 Future Batteries

The development of next-generation battery systems looks promising, as researchers are
exploring multiple ways to increase the energy and power capabilities. Table 1.5 highlights
some of the more promising battery chemistries and the theoretical capacities of such bat-
tery systems.
Much attention has been focused on increasing the energy density of lithium batteries
with new electrode materials. Sulfur has been identified as a potential cathode material
as it has the possibility of increasing the energy density by 10 fold over current lithium-
ion batteries. Moreover, sulfur is an element that is naturally abundant, non-toxic and
inexpensive to obtain [70, 71]. However, development of sulfur-based cathode has been
plagued by low active material utilization, poor cycling, and low Coulombic efficiency [72].
Stable cycling behavior of lithium sulfide cathode with a poly-(vinylpyrrolidone) binder
was recently demonstrated [73].

16
 Table 1.5: Performance comparison of future battery systems [8]

Theoretical Theoretical specific

Battery type voltage (V) capacity (mAh/g) energy (Wh/kg)
Conventional lithium-ion 3.80 155 387
Li-S 2.20 1,672 2,567
Li-air (non-aqueous) 3.00 3,862 11,248
Al-air 2.70 2980 8,100
Zn-air 1.65 820 1,086

On the anode side, silicon has emerged as a viable replacement for carbon-based inser-
tion materials. It has problems of its own as well, such as excessive volume expansion [74],
and unstable electrolyte interphase growth on silicon surface [75]. Recent progress has
been made by immersing silicon-based anode in a conducting polymer hydrogel [76],
thereby creating a three-dimensional network that provides porous volume for expansion,
as well as a continuous electrically conductive network.
Lithium-air batteries, which couple a lithium anode with an air cathode, have extremely
high theoretical energy capacity that is comparable to that of gasoline. The first lithium-air
battery was demonstrated in 1996 [77]. However, desirable rechargeable behavior has yet
to be achieved. In order to achieve the desired performance for an Li-air battery, design-
ers need to master both the lithium and oxygen electrodes and overcome a multitude of
scientific and technical challenges [78].
One of the key areas of battery development has been application of nano-technology.
Nano-materials improve battery performance by increasing the interfacial surface area and
shortening the diffusion path for ions. They can also alter the reaction pathway in the
electrode, increasing capacity and life cycle in general [79]. However, batteries based on
nanomaterials need to overcome poor packing density and low energy efficiency before
becoming viable.

1.4.3 Lithium-Ion Battery Recycling

Since EVs are estimated to make up 7% of the global transportation market by [80], the
availability of lithium and other rare metals required for manufacturing batteries and the
disposal of lithium-ion batteries will become more critical factors in the life cycle analysis
of EVs. Unlike fossil fuel price fluctuations that only affect the running cost of vehicles,
raw material prices directly affect the production cost of EVs.
Battery recycling can lead to over 50% saving of natural resources in terms of decreased

17
mineral ore dependency and reduced energy consumption [81]. What is surprising is that
lithium recovery is not the impetus for battery recycling. While lithium is the energy car-
rier in the battery, it makes up only 57% of the total battery mass [82]. The average
lithium cost associated with battery production is less than 3% of the total battery produc-
tion cost [83]. The driving force behind battery recycling is the recovery of more precious
materials, such as cobalt and nickel. Lithium in batteries is 100% recoverable and technol-
ogy is not the limiting factor in recycling process. However, it is typically not recovered as
recycled lithium costs approximately five times more than brine-based mining process [83].
Current battery production accounts for only 1/4 of the total lithium consumption, and it is
expected to increase to 40% by 2020. With price of lithium carbonate tripling in the past
decade [84], lithium recycling could prove to be an option to hedge against fluctuations in
raw material prices in future.

1.5 Objectives and Outline

Using alternative energy storage systems such as lithium-ion battery to replace fossil fuels
grows out of the need to reduce emissions, provide energy security, and improve vehicle
efficiency. Given that the most critical factor in determining the performance of the hy-
brid/electric vehicles and aircraft is the energy density of the energy storage system, it is
imperative to maximize the amount of energy available while satisfying the mass, volume
and cost constraints for practical applications. There are three ways to maximize the energy
content of a battery:

1. Maximize the chemical potential difference between the two electrodes, such that the
battery gives the most energy per electron.

2. Maximize the charge capacity of the electrode, such that the mass of reactant per unit
of electron is as small as possible.

3. Minimize the non-energy contributing components of the battery.

The first two criteria are inherently chemical properties that depend on the materials
used to construct the electrodes. To a large extent, the cell performance is determined by
the properties of the material used. Dramatic improvements in the raw battery energy ca-
pacity can be achieved through the discovery of new electrode materials such as silicon
and sulphur mentioned in the previous section. The third point relates the design and con-
struction of battery to its performance of energy and power, and the problem is there is still

18
a significant gap between theoretical and realizable energy density [85, 86]. The electro-
chemical cell is a multi-scale, multi-phase system that involves multiple processes occuring
simultaeneously. An optimal cell design has to account for the various processes occuring
within a cell and find a balance point that minimizes the cumulative effects of these factors.
It is difficult to determine the best possible design in a highly nonlinear system such as the
electrochemical cell using simple parameteric analysis.
What has been missing is an efficient numerical tool and a comprehensive numerical
study to optimize the battery performance with respect to various cell parameters and sub-
ject to system-specific constraints. The objective of this thesis, therefore, is to devise a
numerical framework that automates battery design and to apply it in the design and analy-
sis of EV systems. By coupling a detailed electrochemical model that simulates all relevant
transport processes within a battery cell with nonlinear optimizers, a mathematical frame-
work for the complete and automatic battery optimization is demonstrated. The details of
the framework are presented in Chapter 2. Using this framework, three representative bat-
tery design problems representing different phases of battery pack designs are investigated.
Chapter 3 presents the design of a single electrochemical cell to maximize its energy
density subject to a power constraint. This chapter presents a design problem at the basic
cell level and the main objective is to quantify the variations of design variables with respect
to different cell power requirement. The design space encompasses all relevant variables,
including both morphological variables (electrode thickness, porosity, and particle size)
and transport parameters (conductivity and diffusivity). The optimal designs form a Pareto
front of maximum cell energy density at constrained power requirements.
The optimization framework is expanded in Chapter 4 to find the optimal design of a
battery pack for PHEV operation that minimizes the pack mass, volume or cost. The main
motivation is to provide a numerical framework for automated design and analysis of a re-
alistic PHEV battery design problem. The battery pack is made of uniform electrochemical
cells that have to simultaneously satisfy performance requirements of energy and power to
provide adequate vehicle performance. Additional voltage and current bounds as safety re-
quirements are considered in the design problem formulation as well. Actual federal-testing
driving cycles are used to simulate the real-life performance of the optimized battery pack
designs.
Chapter 5 examines the possible advantages of a battery pack with multiple cell designs
over one with uniform cells. This chapter is motivated by the discrepancies between opti-
mal single cell design and the optimal battery pack cell design in the prveious two chapters.
The contribution is to bridge the gap between an optimal cell design intended to perform
best at its specific design point and the need of a battery pack to satisfy a range of discharge

19
conditions that depends on the driver throttle input and road conditions. A multi-cell de-
sign scheme that considers a two-level design strategy that is outlined in this chapter. The
individual cells are optimized based on the battery pack requirement and the combination
of the individual cells to best satisfy the battery pack are optimized as well. The situations
where the multi-cell design approach outperforms the conventional uniform-cell design are
specified.
Finally, the inadequacies and potential for improvements of the numerical framework
are addressed in the concluding chapter. Additional details such as cycling stability and cell
degradation have not been included in the current framework and qualitative analyses are
provided to address these issues. While the battery optimization problems are demonstrated
with one type of lithium-ion cell chemistry, generalization of the numerical framework to
include other materials is discussed here as well.

20
 CHAPTER 2

Methodology

2.1 Introduction
An optimization framework requires three key components: a useful problem formulation,
an efficient optimization algorithm, and an accurate model to evaluate the objective func-
tion and constraints. This chapter describes these three components in detail.

2.2 Cell Model

Battery cell cycling behavior is typically nonlinear, since a number of factors influence
properties, such as the cell voltage and power output. Typically, as a battery discharges,
the open circuit voltage (OCV) of the cell is SOC-dependent. In addition, factors such as
activation polarization, concentration polarization, and internal resistance all contribute to
the potential drop in the cell [9]. Activation and concentration polarizations are related
to the charge and mass transfer kinetics, respectively, and internal resistance is the sum of
ionic, electrical, and interfacial resistances. To accurately model the cycling behavior of the
cell, a cell model has to account for all the major internal processes occurring between and
within the electrodes. Alternatively, an algebraic model that represents cell components
with simplified circuit components can be used as well to speed up computation. In the
following two sections, both types of models are be described.

2.2.1 Physics-Based Cell Model

A physics-based model that accounts for fundamental transport laws in a pseudo 2-dimensional
battery cell is used in this work [52, 87]. Specifically, we model the lithium ion transport
in the direction perpendicular to the current collectors in a full lithium-ion sandwich cell,

21
 Negative Positive

Current collector (+)

Current collector (-)
electrode electrode

Separator
Electrolyte

Solid materials

Figure 2.1: Structure of a lithium-ion insertion cell shows three separate regions and two
different phases

as shown in Figure 2.1. The cell model incorporates an homogenous electrode formu-
lation [52, 88] with concentrated solution theory [53]. The state variables solved in the
model are the ion concentration in electrolyte, c, and in the solid matrix, cs , the current
density in the electrolyte, i2 , the interfacial current density at the solid matrix surface, in ,
and the potentials in the electrolyte and solid phases, 2 and 1 , respectively. These state
variables in turn provide the cell properties that are used to evaluate the cell energy density
and power output.
The governing equations with the associated transport parameters are defined in all
three regions of electrochemical cell: the positive electrode, the separator, and the negative
electrode. The porous electrodes consist of solid active materials and liquid electrolyte.
One key aspect of the pseudo-2D model is that the exact geometric details and microstruc-
trual effects of the electrode are ignored, and instead the electrode is modeled as a con-
tinuum medium [52] that is a mixture of both phases. This avoids the additional work of
explicitly defining solid and liquid phases and their boundaries. The influence of porosity is
instead accounted for using Bruggemans relation for spherical particles [89]. This relates
the effective transport property to the bulk property and the porosity in the form of:

Keff = 1.5 K (2.1)

where K in this case is an arbitrary transport coefficient such as diffusivity or conductivity.

The main forces driving ion mass transfer in a cell are 1) the gradient in electrochemical
potential, which causes migration, 2) the concentration gradient, which causes diffusion,

22
and 3) the electrolyte bulk velocity, which causes convection. A material balance on the
salt in the electrolyte is then given by:

to i2 + i2 to
 
c dlnco
 = D 1 c + vo + aj (2.2)
t dlnc z+ + F

where the terms on the right hand side represent the diffusion, the migration, the convec-
tion, and the reaction rate of the anion in the solution. The electrolyte velocity is usually
negligible [90], and the convection term can be ignored. The anion reaction term is zero in
the absence of the side reactions, leaving diffusion and migration as the main mass transfer
mechanisms. Initial conditions are prescribed for salt concentrations at every location in
the cell.
Ion movement is accounted for in both the liquid and solid phases. In the solid phase,
only the cations and electrons are mobile in the host matrix. If intercalation-induced stress
and anisotropic diffusion are ignored, the transport of ions in the solid phase is given by:

i1 to
 
cs dlnco
= Ds 1 cs + cs vo (2.3)
t dlncs z+ + F

If one further assumes that the conductive additives make the solid material a good elec-
tronic conductor and assumes small volume fraction change and uniform spherical particle
shape, Equation (2.3) can be further simplified to a diffusion equation in radial coordinates:
 
cs 1 2 cs
= 2 Ds r (2.4)
t r r r

where r is the radius of the solid particles. Note that in this treatment, there is no dis-
tinction between different types of solids. Any solid material property such as the solid
diffusion coefficient, Ds , has a bulk value that is representative of the conglomerate of var-
ious solids. In this case, Ds is approximated as a constant, and therefore, Equation (2.3) is
a linear partial differential equation. Instead of adding a second dimension to solve for the
diffusion in the solid phase, the flux at the solid surface can be approximated by Duhamels
superposition integral [91, 52] to simplify computation.
The potential in the electrolyte is given by a modified Ohms Law:
 
i2 2RT o

dlnf
2 = + 1 t+ 1 lnc (2.5)
F dlnc

given that the potential is a relative value rather than an absolute one, 2 is set to zero at
the positive electrode-current collector interface as a boundary condition.

23
 For galvanostatic discharge or charge of the electrochemical cell, the total current in the
cell is constant and specified as a boundary condition. Therefore, the potential in the solid
phase is given by the Ohms Law of the following form:

I i2 = i1 = 1 (2.6)

where I is the sum of current densities in both the solid and electrolyte phases in the
electrode. This value should equal to the liquid phase current density in the separator as
well. The cell voltage is given by the solid phase potential difference between the two ends
of the cell.
Finally the system is closed by accounting for the reaction rates at the solid-liquid
interface via the ButlerVolmer equation:

a F (1 2 U ) c F (1 2 U )
    
in = io exp exp (2.7)
RT RT

which relates the local electrochemical reaction rate (in ), or the net pore-wall flux due to
reaction at the solid surface, to the local concentration and potential. By electroneutrality,
the current balance requires the flux term to be equal to the divergence of the current in the
electrolyte [88]:
i2 = ain (2.8)

where a is the interfacial surface area. The governing equations (2.2)(2.8) are imple-
mented by Newman and his collaborators in a Fortran-based code called dualfoil [52,
87], which is modified for the work done in this thesis. The equations are solved on a one-
dimensional grid traversing from the negative to positive current collectors for the state
variables c, cs , i2 , in , 2 and 1 .

2.2.2 Numerical Treatment of the Cell Model

The model presented in Section 2.2.1 simulates the transient discharge process in an elec-
trochemical cell. Some numerical treatment is required to ensure the smoothness and con-
tinuity of the cell output with respect to the design parameters to ensure accurate derivative
output for the optimizers. Most notably, the discharge termination condition results in the
see-saw pattern of energy density function with respect to design variables such as cycling
rates (Figure 2.2).
The termination criterion for the galvnostatic discharge is determined by the cell voltage
a state variablereaching the cutoff value. Given that it is difficult to determine the cell

24
voltage a priori, the termination condition is instead given as the last time step that results in
the cell voltage decreasing to below the cutoff voltage. This results in a final cell voltage in
the original function that does not precisely match the cutoff voltage. Furthermore, the fi-
nal cell voltage does not vary continuously with respect to the design variables, resulting in
the see-saw pattern of the energy density function. While simulation with such termination
condition is sufficient for estimating the cell performance, it is inadequate for gradient-
based optimization, which assumed continuous smooth functions. The random termination
condition results in erroneous derivatives with respect to the design variables. The inex-
actness of the cell voltage at termination also causes other numerical discontinuities in cell
properties that depend on cell voltage values.
114
original function
2nd -order interp.
113.95
specific energy, Wh/kg

113.9

113.85

113.8 0.880 0.881 0.882 0.883 0.884

cycling rate, C

Figure 2.2: Original energy density function is see-saw shaped. The over-estimate of en-
ergy density increases as the final cell voltage decreases and is only reset when the number
of time step is reduced. The error can be reduced using Lagrangian interpolation

Figure 2.2 shows how the energy density of the original cell model varies with the
discharge cycling rate. As the cycling rate increases, the useful energy density extracted
from the cell should decrease. However, this is not the case from the original objective
function. As the cycling rate increases from 0.88 C to 0.884 C in Figure 2.2, the time step
termination condition results in the final cell voltage being increasingly less than the cutoff
voltage. This in turn causes an overestimate of the energy density due to the additional
discharge below the cutoff voltage. The error is only corrected when the number of time
step is reduced upon further increase in cycling rate, e.g., when the cycling rate is further
increased to beyond 0.884 C. The reduction in time step causes the final cell voltage to
reset to the cutoff voltage, and hence the loss of the final time step reduces the cell energy

25
density. This issue is addressed by using dynamic step sizing and higher order interpolation.
Dynamic step sizing reduces the final time step sizes as the cell voltage approaches the
cutoff voltage, so as to reduce the cell voltage over-estimate. Lagrangian interpolation [92]
is also applied to approximate the objective function such that it gives an accurate estimate
of the cell property at the exact discharge termination condition. To ensure continuity of
the derivatives, second order Lagrangian interpolation was used.

2.2.3 Equivalent-Circuit Model

A detailed model is useful to understand how design parameters affect the performance
of the cell. However, it is too computationally intensive to perform control optimization,
or to estimate battery performance and state of health in real-time. A simplified model is
required that can offer a relative high degree of accuracy in terms of cell discharge char-
acteristics with a reasonable computational cost. An equivalent circuit model represents
the various characteristics of a battery using simplified circuit components. When accurate
experimental data on cell characteristics are available, it can offer an accurate prediction of
cell behavior.
There are various types of equivalent circuit representations for a battery cell [93].
Figure 2.3 shows a Thevenin model that takes into account all major factors contributing
the potential drop of the cell voltage. It consists of a parallel RC circuit connected in series
to a resistor. The resistor R1 represents the internal resistance in the cell, R2 represents the
polarization resistance, and the capacitor C1 describes the transient response due to mass
transfer kinetics.

C1
R1

+
Voc R2 Vcell

Figure 2.3: Thevenin equivalent circuit representation of a battery cell

26
 The equivalent circuit model can be further simplified such that a single resistance
term is used to describe the total Ohmic drop in the cell. The behavior of the cell can be
represented by the following equation:

Vcell = Voc IRint

(2.9)
Pcell = Voc I Rint I 2

The resistance term Rint and the OCV can be both functions of SOC to reflect how the
internal impedance of the cell varies as the battery discharges. The current can be calculated
by solving a quadratic equation in terms of the power output of the cell.

2.3 Vehicle Model

A simplified vehicle model is used to compute the battery power and energy needed to
follow various driving cycles. The power required to propel the vehicle is defined to be:

Pveh = Pclimb + Pacc + Pdrag + Proll + Pmis (2.10)

where the total power required is the sum of the power required to overcome aerodynamic
forces (Pdrag ), rolling resistance(Proll ), power miscellaneous systems (Pmis ), and to achieve
the required acceleration (Pacc ) and rate of climb (Pclimb ) of the driving cycle. The mis-
cellaneous power refers to the power required for various auxiliary systems not related to
drivetrain. The total power required can be expressed in terms of vehicle parameters:
 
dV 1
Pveh = M V g sin + + air Cd,aero AV 3 + Cd,roll M gV + Pmis (2.11)
dt 2

where M is the combined mass of vehicle plus passengers, V is the instantaneous vehicle
velocity, A the frontal area of the vehicle, Cd,aero and Cd,roll are the aerodynamic drag and
rolling resistance coefficients, respectively.

2.4 Optimization Technique

Numerical optimization is the selection of the design variables that best satisfies certain
criteria. It differs from traditional engineering analysis by addressing the question of is this
the best possible instead of is this good enough. The advantage of optimization process
lies in the ability to quickly generate multiple alternative designs and to rigoriously analyze
them using a set of prioritized criteria. The optimization process is iterative, repeatedly

27
sampling the design space to locate the optimal design point, as shown in Figure 2.4. The
convergence conditions can be either the local derivatives in the form of the KarushKuhn
Tucker (KKT) conditions [94], or the improvement in objective functions over successive
iterations, depending on the type of optimizers.

start Optimizer

Update variables

no
Function evaluations

convergence?

yes

Stop

Figure 2.4: Optimization process

A general nonlinear optimization problem can be written as:

minimize f (x), f : Rn R

xmin x xmax
(2.12)
subject to cj (x) = 0 j = 1, ..., m

ck (x) 0, k = 1, ..., m

where f (x) is the objective function to be optimized with respect to the bounded variables
x, subject to equality constraints c(x) and inequality constraints c(x). Both gradient-
free and gradient-based optimizers can be employed to solve the battery pack optimization
problem. The previous section described the cell models used to compute the objective
function and the relevant constraints. This section will explain the optimizers and how they
determine the search directions and convergence criteria in the process.

28
2.4.1 Gradient-Free Optimizers
Gradient-free optimizers, as the name suggests, depend only on function evaluations to
locate the optimal solutions, and these class of optimizers are used to solve the battery
pack optimization problem described in Chapter 4. The battery pack design problem is a
nonlinear mixed-integer problem with nonlinear constraints. The optimizer has to traverse
a discontinuous constrained design space and handle nonlinear problems efficiently. For
this reason gradient-free optimizer, which is effective in solving noisy, non-differentiable
and multi-modal problems, is selected for optimization problems with discontinuity in
the design space. There are many different types of gradient-free optimizers available,
and performance comparisons have been made of these optimizers on noisy optimization
problems [95, 96]. For this work we use the ALPSO method implemented within the py-
Opt framework [97, 98]. The ALPSO algorithm is a stochastic population-based method
that employs a group of candidate solutionsknown as particlesto search for the opti-
mum [97]. These particles move about in the design space by updating their positions and
velocities according to the following equations:

xik+1 = xik + Vk+1

i
t
(2.13)
Vk+1 = w0 Vk + w1 r1 (pik xik ) + w2 r2 (pgk xik )
i i

where xik is the position of the ith particle at iteration k, V ik is its velocity vector, t is
the time step size, pik is the position with the best objective function for particle i (particle
best), and pgk is the global position with the best objective function for all the particles up to
the kth iteration (global best). The movement of the particles are hence governed by their
own movement history as well as the collective influence of the entire swarm. The weights,
w0 , w1 , and w2 are assigned to each component of the velocity update. They are bound by
the following relations to ensure stability and to guarantee convergence [97]:

0 < w1 + w2 < 4
w1 +w2
(2.14)
2
1 < w0 < 1

The constraints are included by introducing explicit Lagrangian multiplier estimates for
each constraint into the objective function. This approach transforms the constrained prob-
lem into an unconstrained one, while preserving the feasibility of the solution by including
the KKT conditions in the optimization problem formulation. The Lagrange function is
further augmented with a quadratic function to preserve the stationary properties of the

29
solution, making the final function of the form [99, 100]:
m
X m
X
Ji (xik , , rp ) = f (xik ) + j j (xik ) + rp,j j2 (xik ) (2.15)
j=1 j=1

where f (xik ) is the objective function of the ith particle at kth iteration, j and j are the
Lagrange multiplier and quadratic function extension for the jth constraint respectively,
and rp is the quadratic penalty factor.

5 5
20

20
1 1

20

20
10

10
0 0
0

0
1 1
x2

x2
2 5 40 2 5

40
3 3
-18

-18
10 10
-15

-15
-5 -5

-5 -5
44 44
-10

-10
3 2 1 0 1 3 2 1 0 1
0

0
x1 x1

Figure 2.5: Solution history using ALPSO to solve a 2D problem. Left: initial distribution
of particles. Right: converged optimization where all particles are at the global optimum

Figure 2.5 shows the iteration process of ALPSO applied to a simple 2-variable opti-
mization problem. The optimization is initiated by distributing 20 particles randomly in
the design space. At each iteration, the objective function values at all the particle posi-
tions are evaluated, and the position of the particle with the best objective function value
(marked by the red diamond in Figure 2.5) is broadcast to every other particle. This pro-
vides a spatial step in the optimization, represented by the last term on the right-hand side
of Equation (2.13). At the same time, each particle keeps track of its own best position
throughout its iteration history. This ensures that past iteration information is not lost, and
it is represented by the second term on the right-hand side in Equation (2.13). Lastly, a
random component is added to each particles velocity to prevent premature convergence
to a local minimum. However, given the stochastic nature of this optimization algorithm
and lack of enforcement of mathematical optimality conditions, it is essential that multiple
optimization runs are conducted to fully explore the design space.
Gradient-free optimizers are apt at solving non-smooth optimization problems and can
handle discrete variables relatively easily. However, there is less confidence in the final so-

30
lution because of the absence of mathematical optimality conditions. The optimization pro-
cess with gradient-free optimizers is computationally expensive, requiring a large number
of function evaluations. This is problematic if the objective function evaluation is costly,
or if there is a large number of variables. Therefore, gradient-free optimization algorithms
are only suitable for problems with small number of design variables, which is the case of
the optimization problem solved in Chapter 4.

2.4.2 Gradient-Based Optimizers

The performance and properties of a single electrochemical cell is determined by mor-
phological parameters, such as the electrode porosity and thickness, as well as transport
parameters, such as diffusivity and conductivity. The output of the cell should physically
vary smoothly with these continuous variables, though the effects can be nonlinear. In this
case, gradient-based optimization methods are preferable, as they are more efficient for
solving nonlinear optimization problems that have a continuous and smooth design space
that is convex near the optima. Gradient-based optimizers make use of both function eval-
uations and local derivatives to search for the optimal solutions in the design space.
There are various types of gradient-based optimizers available. One of the most effi-
cient ways to solve nonlinear constrained optimization problems is by approximating the
nonlinear problem as a quadratic subproblem at each iteration, also known as the sequential
quadratic programming (SQP) method. SQP is an active-set method that assumes the de-
sign space near the optimum to be convex. It uses the optimal solution for each quadratic
subproblem to define a new iterate point. This method can be viewed as the application of
Newtons method to the KKT optimality conditions. An effective SQP algorithm has to be
able to compute a step from the quadratic subproblem and to generate an estimate of the
optimal active set. From the many different SQP implementations available, we use the
SNOPT package [101], which is a line-search SQP implementation.
SQP solves the following subproblem:

minimize fk + fkT p + 21 pT 2xx Lk p

subject to ci (xk )T p + ci (xk ) = 0 (2.16)
ci (xk )T p + ci (xk ) 0

where p is the step size from the current iteration point that minimizes the quadratic sub-
problem, fk is the gradient of the objective function with respect to the design variables,
and 2xx Lk is the estimate of the second-order derivatives of the Lagrangian function. The
second-order derivative can be estimated by using a quasi-Newton approximation, such

31
BroydenFletcherGoldfarbShanno (BFGS) [94]. The matrices ci (xk ) and ci (xk ) are
the Jacobian of the equality and inequality constraints with respect to the design variables.
At each iteration, the solution to the quadratic subproblem is obtained using a quasi-Newton
approach.

5 5
20

20
1 1

20

20
10

10
0 0

0
1 1
x2

x2
2 5 2 5

40

40
3 3
-18

-18
10 10
-15

-15
-5 -5

-5 -5
44 44
-10

-10
3 2 1 0 1 3 2 1 0 1
0

0
x1 x1

Figure 2.6: Solution history using SNOPT package with finite-difference derivative ap-
proximation on a 2D problem. Left: initial starting location of the optimization. Right:
final iteration with the iteration history shown

Figure 2.6 shows the SQP applied to the same 2-variable problem as shown in Fig-
ure 2.5. Unlike ALPSO, SQP only requires one initial point. At each iteration, the opti-
mizer evaluates both the objective function and the local derivatives to determine the search
direction and step size that solves Equation (2.13). As shown in the figure, the optimizer
converges quickly to the optimum solution in less than ten iterations. The iteration history
shown in Figure 2.6 indicates that the optimizer has over-predicted the intermediate solu-
tion on two occasions, resulting in the intermediate solution being at the bounds. Given
that this sample problem only has two variables, the quadratic approximation used in SQP
should provide an exact approximation of the design space. The solution overshoot is most
likely due to errors in the derivatives supplied by the finite-difference approximation. How-
ever, in each case, the optimizer is able to backtrack its steps and successfully determine a
new search direction.
One of the issues with gradient-based optimization is the possibility of converging to
local optima instead of the global one. Given that one of the convergence criteria is the local
derivatives defined by the KKT conditions, it is foreseeable that the optimizer converges
to a local optimal solution when the problem is multi-modal. Figure 2.6 shows that there
exists a local minimum at (x1 , x2 ) = (0, 0). An optimization initiated on the top right

32
 5

20
1

20
10
0

0
1
x2

2 5

40
3
-18

10
-15

44 -5 -5
-10
3 2 1 0 1
0 x1
Figure 2.7: An optimization that is initiated from (x1 , x2 ) = (1.5, 1.5) converges to the
local optimum at and fails to find the global optimum

corner of the design space can converge to the local minimum and fails to find the global
minimum on the bottom right corner, as shown in Figure 2.7. One way to increase the
chance of finding the global optimum is to initiate multiple optimization runs at random
locations.

2.4.3 Derivative Computation

One of the keys to taking full advantage of gradient-based optimization is to use efficient
and accurate method for computing derivatives derivatives. A range of derivative computa-
tion methods are available, depending on the nature of the optimization problem, the com-
plexity of the derivative computation, and the number of design variables [102]. Given that
the battery pack design problem has a relatively small number of variables, the complex-
step approximation method [103] is used to obtain the derivatives due to its accuracy and
the ease of implementation. The complex-step method is similar to the more common
finite-difference approach, except that a small step is taken in the imaginary axis instead of
the real axis. Taking the Taylor series expansion about a point in the real axis with a purely
complex step:

df x2 d2 f ix3 d3 f
f (x + ix) = f (x) + ix + + (2.17)
dx 2 dx2 3! dx3

33
A second-order approximation of the first-order derivative can be isolated from Equa-
tion (2.17):
df Im [f (x + ix)]
+ O x2


= (2.18)
dx x
Equation (2.18) shows that the complex step derivative does not involve a subtraction
operation, and hence it is not subject to cancellation errors resulting from finite-precision
arithmetic, which is a problem with finite-difference derivative approaches. A step size
study is often needed for finite-difference approximations to determine the step size that
minimizes the cumulative effects of both approximation and cancellation error. However,
this is not a problem with the complex-step method, as its step size can be made arbitrarily
small. A comparison of the derivatives computed using the complex-step method and a
forward-difference approach is shown in Figure 2.8. The normalized errors of the two ap-
proaches are plotted against the step size. Figure 2.8 shows how the cancellation error due
to loss of significant digits dominates at small step size for the forward difference approach,
whereas the error of the complex-step approximation remains at machine-zero (1016 for
double-precision arithmetic). The lack of cancellation error means that the complex-step
method does not require a step-size study to determine the optimal step size to minimize
the total error, and hence expedites the optimization process and saves computational time.

101

10-1

10-3

10-5
Normalized error

f(x) = ex
10-7 q
sin3 x +cos3 x
10-9

10-11 one-sided finite difference

complex-step
10-13

10-15

10-17
10-1 10-3 10-5 10-7 10-9 10-11 10-13 10-15 10-17 10-19
x

Figure 2.8: Error in the derivatives computed using finite-difference and complex-step ap-
proximations for varying step sizes. The complex-step approach is free of cancellation
errors that dominates finite-differences at small step sizes.

34
 The difference between the derivatives calculated using finite differences and the com-
plex step applied to the physics-based electrochemical model is shown in Figure 2.9. Note
that derivatives obtained from both the finite-difference and complex-step methods follow
the same trend, but there is much more scattering in the derivatives obtained using the
finite-difference method.
600
complex step
finite difference
500
derivative of energy wrt cycling rate

400

300

200

100

0
0.0 0.2 0.4 0.6 0.8 1.0
normalized cycling rate

Figure 2.9: Derivatives of energy density with respect to cycling rate obtained using the
finite-difference method show much more scattering compared to the complex-step ap-
proximation.

While the complex-step approximation offers the advantage of increased accuracy, it

only works in cases where the design variable or objective function of interest are not
already complex in the original function. The application of the complex-step to computa-
tional models requires the modification to the original source code, which can be non-trivial
in certain instances. The increased accuracy in the function evaluation also comes with
additional computational cost: changing all real variable operations to complex variable
operations increases the computational time by 3 to 4 times. The increased accuracy of the
derivatives computed by the complex-step approximation method as shown in Figures 2.8
and 2.9, however, make up for the additional computational cost in ensuring the optimizers
locate the true optimal solutions.

35
2.4.4 Pareto Optimality
The optimization of a single objective function with respect to the design variables is
straight-forward as it will be the solution of the optimization problem. It is often the case,
however, that there are multiple competing objectives of interest. For instance, one may
want to maximize the energy density of a battery cell while at the same time maximize
the power output. Alternatively, one may want to minimize the mass, volume and cost of
a battery pack. Given that the decision maker may not be the person performing the op-
timization, it is often difficult to determine a priori the compromise between the various
objectives for the desired solution. Such multi-criteria decision making scenarios require
a Pareto front, or Pareto optimal points to examine the relations and trade-offs between
multiple objectives. By definition, a point is Pareto optimal if and only if none of the ob-
jectives can be improved without penalizing at least one of the others [104]. A Pareto front
therefore contains solutions that are not dominated by any other points in the design space,
i.e., the solutions on the Pareto front have the best objective functions at their respective
constraints [105, 106]. This front offers valuable insight into multi-objective optimization
in numerous engineering applications [105, 107], and it is used to examine the relation
between energy and power of optimal cell designs in Chapters 3 and 5.

2.4.5 Hybrid Optimization Method

The advantages and the disadvantages of gradient-free and gradient-based optimizers were
illustrated in Sections 2.4.1 and 2.4.2. While the different groups of optimizers work well
on separate classes of problems, it is possible to employ a hybrid optimization scheme
that combines the two types of optimizers, as shown in Figure 2.10 in the form of an
XDSM diagram. The hybrid process is governed by the optimization control, which uses
the rudimentary optimal results from the gradient-free optimization as the starting point
of the gradient-based optimization. The main reason to use the hybrid approach is to take
advantage of the different niches of the two classes of optimizers to facilitate an efficient
optimization process. The gradient-free optimizer minimizes the chance of getting trapped
in a local minimum in the nonlinear design space, and the gradient-based optimizer pin-
points the exactly location of the optimum with respect to the continuous variables. The
hybrid optimization process is used to solve the battery pack design problem discussed in
Chapter 4. While the design at the cell level involves continuous variables, the battery lay-
out (number of cells in series and parallel) at the pack level requires integer values, thus
making the battery pack design a mixed-integer optimization problem. For the pack design
problem, the hybrid optimization scheme can take advantage of the independence between

36
the cell design and the pack design.

(0) (0)
xR , xN

0, 5:
0
Optimization (0)
1:xR , xN
(0)
6: xN , xR 8: xN
Control

1, 4 2 :
0
xN 5: xN , xR Gradient-free 2 : xR 3 : xN
Optimization

6, 9 7 :
Gradient-
xR 7 : xR
based
Optimization

y 2, 7: Cell 3, 8: y
model

3, 8:
Battery
4 : f, c 9 : f, c, J
pack
analysis

Figure 2.10: The XDSM for the hybrid optimization process [3]. The numbers repre-
sent the steps in the optimization process. The optimization control (Step 0) first initiates
the gradient-free optimization (Loop 14), which provides a rudimentary optimal solution
and determines the optimal integer design variables. The results from the gradient-free
optimizion is then used to initialize the gradient-based optimization (Step 5). The gradient-
based optimizater (Loop 69) refines the continuous design variables for the final optimal
solution.

37
 CHAPTER 3

Single-Cell Design Optimization

3.1 Introduction
An important aspect in lithium-ion cell design is to maximize the energy density to meet
the demand of mass or volume sensitive applications. Conventional cell designs have relied
on ad-hoc rulessuch as using thicker electrodes for high energy densityor parametric
analysis of one or two variables to guide the design process. Both experimental analy-
ses and numerical simulations have been applied to examine how various properties of an
electrochemical cell affect its performance. Choi et al. [108] investigated the particle size
effects on performance of a LiCoO2 . Li et al. [109] studied carbon black and carbon nan-
otubes as conductive additives, and how they improve energy capacity and cycling stability.
Doyle et al. [110] used numerical simulation to study the trade-off between conductivity
and transference number for lithium-polymer batteries. While these efforts increased the
understanding of how individual components affect cell performance, they are unable to
fully maximize the cell performance, since the coupled interactions of the various parame-
ters were not been considered. To identify optimal cell designs, the design process has to
include all relevant design variables, and the coupling between various parameters during
cycling needs to be considered.
Several authors have carried out numerical optimizations of lithium-ion cells. New-
man [111] optimized the geometric design of a cell with respect to the positive electrode
thickness and porosity using a simplified reaction zone model. By confining the lithium-ion
intercalation to a narrow zone in the positive electrode, he obtained an analytical solution
and related the dependence of energy capacity to the following dimensionless parameter:

U tdis
Trz = (3.1)
q+ L2s

where U is the open-circuit potential, the electrolyte conductivity, tdis the discharge time,

38
q+ the capacity density of the active material, and Ls the separator thickness. Values for
optimal electrode thickness and porosity were found by fixing the discharge time tdis . Sub-
sequent optimization efforts involved design of an iron phosphate lithium-ion cell while
maintaining constant capacity ratio and porosity in the negative electrode [112], as well as
coupling lithium-ion batteries with capacitors for hybrid electric vehicle operation [113]. In
an optimization study with respect to cycling rate, particle size, diffusivity and conductivity
using a surrogate model formulation, Du et al. [114] quantified the cell performance as a
function of ratio of discharge time to diffusion time. Ramadesigan et al. [115] found that
optimizing the spatial distribution of porosity can lower the internal resistance by 1533%
compared to a uniform distribution. Golmon et al. [116] went one step further and varied
both the spatial porosity and the particle size distribution using a multi-scale model. They
found that when both porosity and particle size distribution are varied, the improvement in
energy capacity is less than 2% compared to a cell with optimized constant porosity and
particle size. More recently De et al. [117] used the control vector parameterization method
to sequentially optimize the electrode thicknesses and porosities with a reformulated model
for improved computational efficiency.
The application of numerical optimization to battery design is still in its nascent stage.
The high computational cost of a physics-based model and the lack of a complete math-
ematical description of the processes occurring in a battery system have hindered the im-
plementation of optimization schemes in battery design. One way to reduce the compu-
tational cost of battery optimization is to speed up the cell model simulations. Specif-
ically, approximations have been derived to accelerate the solid phase diffusion compu-
tation [118, 119, 55]. Additional improvements in computational efficiency are achieved
by improving the model calculation in the direction normal to the electrode/separator inter-
face [120, 121]. Reduction in computational cost can be realized by minimizing the number
of function evaluations as well. Surrogate model analyses proposed by Du et al. [114, 122]
reduces the design space by removing design variables that do not affect objective func-
tion. An efficient optimizer that requires relatively few function evaluations to converge to
the optimum can be used as well. Various optimization methods can be used to tackle the
battery optimization problem [98, 114, 123, 124, 125, 126]. Given the nonlinear nature of
the battery problem and the computational cost (each function evaluation is on the order
of minutes), a gradient-based optimization scheme, which makes use of derivatives to de-
termine the search directions, is well suited to handle such a problem due to its numerical
efficiency and accuracy [124].
In optimizing the battery design, both morphology and material transport properties
need to be considered as variables in the design process. The design space should in-

39
clude all relevant parameters of a full lithium-ion sandwich cell with insertion compounds
for both electrodes. With this goal in mind, a numerical framework that couples an opti-
mization algorithm with a detailed electrochemistry model to optimize the cell design is
presented here. The objectives are as follows:

1. Maximize the energy density of a lithium-ion cell subject to a power requirement.

2. Determine electrode morphological designs at optimal cell designs under various

discharge rates.

3. Quantify the relative sensitivity of the design variables on cell performance.

3.2 Problem Formulation

A numerical optimization framework coupling the SQP method (Section 2.4.2) with the
macroscopic cell model (Section 2.2.1) is used to maximize the gravimetric energy density
of an ideal lithium-ion cell subject to power requirement. The materials of the cell and
their respective properties are listed in Table 3.1. The cell used in this problem consists of
a lithium-ion cell with spinel manganese oxide LiMn2 O4 for the cathode and mesocarbon
microbead (MCMB) graphite for the anode. Manganese oxide is used as the cathode ma-
terial, since it is relatively inexpensive, easily disposable, and exhibits good capacity re-
tention rate. Carbon is used as the anode active material as it experiences small volume
change, which is essential in capacity retention. The validation of lithium-ion cell with
spinel manganese cathode and graphite anode using the cell model outlined in Section 2.2.1
is provided by Doyle et al [127].
The cell performance is obtained through galvanostatic discharge of the cell until a
cutoff voltage of 3.0 V is reached. This cutoff value is selected as further discharge of in a
real cell could lead to irreversible damage to the battery. The energy density of the cell is
calculated using the trapezoidal approximation to the integral over discharge time:
Z tcut n
1 X Vk+1 + Vk
Ecell = Vcell (t)Idt I (tk+1 tk ) (3.2)
Mcell 0 k=0
2

where I denotes the cell current, which is constant. k is the time step number and n is the
number of time steps needed to reach the cutoff voltage. The cell mass, Mcell , in turn is
given as follows:
Mcell = M+ + Msep + M + Mmisc (3.3)

40
 Table 3.1: Lithium-ion cell material properties and fixed parameters [9]

Parameter Value
Cathode material Spinel Mn2 O4
Cathode initial stoichiometric parameter (y in Liy Mn2 O4 ) 0.2
Coulombic capacity of cathode material 148 mAh/g
Density of cathode material 4280 kg/m3
Volume fraction of inert filler in cathode 0.1
Anode material MCMB 2528 graphite
Anode initial stoichiometric parameter (x in Lix C6 ) 0.9
Coulombic capacity of anode 372 mAh/g
Density of anode material 2260 kg/m3
Volume fraction of inert material in anode 0.05
Electrolyte material LiPF6 in EC:DMC
Initial salt concentration 1000 mol/m3
Inert filler material PVDF
Ambient temperature 298 K

where M+ and M are the mass of the positive and negative electrode active materials
respectively, Msep is the mass of the electrolyte, and Mmisc is the mass of all the other non-
energy contributing materials such as current collectors. The mass of each component is a
function of its thickness and porosities of various materials that make up its composition.
Additional components, such as casings, are not considered; hence the energy densities
obtained will be that for a jelly roll.
To maximize the energy density, the 12 design variables listed in Table 3.2 are selected.
While cycling rate is an operating condition and not a cell design variable, it is necessary
to allow it to vary freely within the design space in order to meet the power requirement.
The simulation technique and the subsequent optimizations are not confined to any partic-
ular operating condition. Allowing cycling rate to vary freely enables us to obtain optimal
cell designs to meet different application requirements. The remaining 11 design vari-
ables consist of 5 design variables for each of the electrodes, and the separator thickness.
Among the 5 electrode variables, porosity, thickness, and to some extent, particle size,
can be controlled during the manufacturing process, while the diffusivity and conductivity
are material properties. However, there are substantial variations in the values reported in
the literature for transport coefficients. These variations are due to experimental uncer-
tainty, differences in electrode microstructure, manufacturing process, and experimental
technique [9, 128, 129]. The bounds on the particle size, diffusivity and conductivity there-
fore represent the minimum and maximum reported values from published research. These

41
 Table 3.2: Design variables and their bounds for single cell optimization

Design variable Lower bound Upper bound

Cycling rate (C) 0.1 10.0
Separator thickness (m) 10.0 100.0
Cathode particle size (m) 0.2 20.0
Cathode thickness (m) 40.0 250.0
Cathode porosity 0.1 0.6
2 16 11
Cathode diffusivity (m /s) 10 10
Cathode conductivity (S/m) 1.0 100.0
Anode particle size ((m) 0.5 50.0
Anode thickness (m) 40.0 250.0
Anode porosity 0.1 0.6
2 16 11
Anode diffusivity (m /s) 10 10
Anode conductivity (S/m) 1.0 100.0

parameters are treated as bounded variables instead of given exact values as it is not the
intent of this framework to optimize the design of a particular lithium-ion cell subject to
a specific pre-fabrication treatment and manufacturing process. Rather, these variables are
intended to provide guidelines on what combination of properties and cell designs would
produce batteries of optimized performance, such as maximum achievable cell capacity.
To achieve optimality, the cell has to balance energy capacity with the ionic transport re-
quirement, while at the same time minimize Ohmic losses. Among the 12 design variables,
the cycling rate controls the current density, and hence galvanostatic boundary conditions.
Low cycling rate is desired to maximize energy density, as it reduces Ohmic losses. Sep-
arator thickness controls the amount of electrolyte in the separator region. Out of the five
electrode variables, thickness and porosity determine the amount of active materials avail-
able for lithium-ion intercalation process and the amount of electrolyte available for ionic
transport. The electrode thicknesses and porosities, together with separator thickness, de-
termine the mass of the cell and hence its energy density. Diffusivity and conductivity are
transport parameters that affect the concentration overpotential in the cell, while the parti-
cle size affects the energy density by changing diffusion path length and interfacial surface
area.
The range of the conductivities for both electrodes is high compared to the conduc-
tivities of the pure active materials in each electrode. These values reflect the increased
electronic conductivity after the addition of carbon additives to the solid matrix. The cell
model does not account for the effects of carbon additives on effective transport explicitly.
However, using the high conductivity allows for theoretical cells that have performance

42
more representative of what can be obtained in practice.
Nine different optimizations at each constrained power requirement were carried out.
The optimizations are initiated from random design points to reduce the possibility of op-
timization converging to local optima. Due to the random nature of the initial points and
the convergence tolerance of the optimization, there are some differences between the re-
sults in terms of number of iterations required and the optimal solutions. The results from
the various optimization runs are collectively presented in the form of box plots. The box
represents the interquartile range, with the line in the center representing the median. The
whiskers extend from the boxes to show the full range of the data excluding the outliers,
which are marked with the + symbols.

3.3 Results

3.3.1 Power vs. Energy

In the absence of power requirement on the cell, the maximum energy density would
achieved by using a very small discharge current such that there is minimal Ohmic drop
in the cell. The cell voltage would be as close to the OCV as possible, and the resulting
cell design would be one with maximum thickness, minimum porosity due to low ionic
transport rate requirement. In this case, most (if not all) of the design variables listed in Ta-
ble 3.2 would go to their respective bounds. Figure 3.1 shows a simple cell design problem
where cycling rate and positive electrode particle size are the only two variables. Maximum
energy density occurs in the lower left corner of the design space, resulting in the simple
optimal solution where both the cycling rate and the positive electrode particle size are at
their lower bounds.
A meaningful cell design problem, therefore, should consider how the cell designs
are affected by the power requirements. Figure 3.2 shows the power density contours (in
color) on top of the energy density contours (in gray) for the same 2-variable problem. As
the power requirement increases, the design space that is able to satisfy the requirement
becomes smaller. In this case the power density is mostly a function of the cycling rate.
The graphical representation of the design space and objective function clearly shows that
the cycling rate has to increase as the power increases while the cell design with maximum
energy density has the smallest particle size regardless of the power requirement.
The optimization problem is expanded to the full design space of 12 design variables.
The potential for using gradient-based optimization framework in cell design is demon-
strated by maximizing its gravimetric energy density at various specific power require-

43
 4.0 65
100
3.5

75
3.0

90

85
105
2.5
cycling rate, C

95
2.0 110

1.5

1.0
maximum energy density 113
0.5
115
5 10 15 20
cathode particle size, m

Figure 3.1: Contour plot of energy density shows monotonically decreasing energy density
as cycling rate and particle size increases. Maximum energy density occurs at minimum
cycling rate and particle size.

ments. The constrained power requirement increases from 50 W/kg/m2 at increment of

50 W/kg/m2 until 1500 W/kg/m2 , at which point a further increase in power is only possi-
ble by expanding the upper bound of the cycling rate. The competing effects of increasing
power and decreasing energy at higher power form a Pareto front. In this case the Pareto
front is formed from the set of points with the maximum achievable cell energy capacity
for each required power level.
Figure 3.3 shows the optimal energy density with respect to various power require-
ments. As expected, the maximum achievable energy density decreases as the power re-
quirement increases. For all nine optimization runs at each specified power requirement,
the final optimal energy densities converged to within 0.1% of one another. The Pareto front
is contrasted with the Ragone plots. The Ragone plot compares the performance of the en-
ergy storage device at different power ratings by plotting its available energy density against
power density [130]. While the Pareto front consists of cells optimized for each power re-
quirement, the Ragone plot shows the variation of energy density versus power density for
a specific cell. Four representative Ragone plots are shown in Figure 3.3. The 0.2-C cell
corresponds to the power requirement of 50 W/kg/m2 , while the 2-C is at 400 W/kg/m2 ,
4-C is at 800 W/kg/m2 , and the 9-C is at 1500 W/kg/m2 . Each Ragone plot intersects the
optimal Pareto front at only one point, indicating that each cell design is optimized for only

44
 4.0
600
3.5

3.0 500

2.5
400
cycling rate, C
2.0
300
1.5
200
1.0

0.5 100

5 10 15 20
cathode particle size, m

Figure 3.2: Contour plot of power density overlaid on top of energy density (light grey
lines) shows how power requirements restrict the design space. The energy density in-
creases when moving from the top right corner (high cycling rate, large particles) of the
design space towards the bottom left (low cycling rate, small particles)

a specific power requirement and no single cell is the best design for all power require-
ments. The 0.2-C cell has the highest energy density at 50 W/kg/m2 . However, its energy
capacity decreases rapidly if its discharge rate is increased. All four Ragone curves exhibit
the typical knee, indicating the rapid deterioration of cell performance away from its spec-
ified design condition. Comparison between the Ragone plots and the optimal Pareto front
demonstrates the importance of a properly designed cell for a specific operating condition.
Any deviation of cell design away from its intended purpose will lead to rapid loss of en-
ergy capacity. Thereofore, it is important that the cell is properly designed using numerical
algorithms so as to achieve maximum performance at its intended discharge rate.
Some representative energy and power density values from the literature [131, 132, 133]
are also plotted in Figure 3.3 for comparison. The energy and power densities obtained
from the optimization results are for an idealized lithium-ion battery jelly roll only, hence
it is expected that their values are significantly higher than those achieved experimen-
tally [131, 132]. Comparison of jelly-roll energy density with published simulation re-
sults [133] shows that the maximum achieved energy density shown here is about 4%
higher.

45
 103

Bauerlein [1999]
specific power density(W/kg)

Horiba [2011]
Pareto front

102
0.2-C cell Ragone

2-C cell Ragone

4-C cell Ragone
9-C cell Ragone
Howard [2001]
100 120 140 160 180 200 220 240
gravimetric energy density (W-hr/kg)
Figure 3.3: Pareto front of optimal energy density vs. specific power. Ragone plots at
four discharge rates are also shown, which represent the variation of energy density with
discharge rates for a particular cell.

3.3.2 Sensitivity at Optimal Designs

Next, the design variables of the optimal cell designs for various power requirements are
quantified in terms of their numerical values and effects on optimal designs. Table 3.3
shows the sensitivities of energy density with respect to the variables at the optima. Note
that since there is an optimal design point for each cell power requirement, it would be too
cumbersome to show the sensitivity of each variable for all the optimum points. Therefore
the average sensitivity across all optima is listed in Table 3.3 to give a qualitative compari-
son of the relative effects of variables on energy density. The sensitivity studies show that
the morphological parameters such as cell thickness and particle sizes have much greater
effect on energy density than transport parameters, such as diffusion and conduction coeffi-
cients. On average, the effects of transport parameters are an order of magnitude lower than
those of other parameters. A more complete treatment of global sensitivity with respect to
cathode variables is provided by Du et al. [122]. The relative effects of various design pa-
rameters will be elaborated on when discussing the transport parameters at optimal designs.

46
Table 3.3: Design variables and their sensitivities at optimal designs. Sensitivity is cal-
culated as percentage change of objective function due to a percentage change in design
variable.

Design variables Optimized design values Average sensitivity (%)

Cycling rate (C) 0.29.4 2.3
Separator thickness (m) 10.0 0.25
Cathode particle size (m) 0.2 0.34
Cathode thickness (m) 133.4250.0 0.34
Cathode porosity 0.130.54 0.30
Cathode diffusivity (m2 /s) 1.09 1012 1011 0.028
Cathode conductivity (S/m) 80.3100.0 0.043
Anode particle size (m) 0.5 0.054
Anode thickness (m) 67.3163.7 1.3
Anode porosity 0.140.33 0.37
Anode diffusivity (m2 /s) 1.66 1012 1011 0.029
Anode conductivity (S/m) 1.3100.0 0.029

3.3.3 Separator Thickness and Electrode Particle Size

Table 3.3 also shows the ranges of design variables at the optima for all cell power re-
quirements. While some design variables change with cell power requirements, the separa-
tor thickness and electrode particle sizes invariably converge to their respective minimum
bounds. While these parameters go to the bound in an idealized cell design, their effects on
practical cell designs cannot be ignored.
Separator thickness converges to its minimum bound for an optimal cell design. The
separator does not contribute to the physics model of an ideal cell, since no electrochemical
reactions occur there. The purpose of the ions in the separator is solely to transport charges
from one electrode to the other. Therefore, the separator thickness only affects the weight
of the cell, which is to be minimized in order to maximize the energy density. In practical
cell design, there needs to be a minimum separator thickness to prevent short-circuiting due
to dendritic growths from the electrodes [134].
Smaller particle sizes are preferred as they reduce diffusion path length for lithium ions
in the electrodes, and also increase interfacial surface area. A large interfacial surface area
is important for ensuring adequate rate of ion transport between the two phases. Reduction
in particle size results in more interfacial surface area at a smaller porosity, and hence the
electrode is able to pack more active material without sacrificing the mass transfer capabil-
ity. This results in a cell design with higher energy density. Increased surface area has the
additional benefit of reducing the current density and subsequently the transport overpo-

47
tential in the cell. Various authors have shown improved the electrochemical performance
of smaller particles in positive electrodes [135, 136]. Cells made with nanoscale particles
demonstrate high energy capacities and fast charge and discharge kinetics of lithium ions
compared to cells with micro-sized particles. On the other hand, the cycling stability of
electrodes with small particles is unfavorable due to the polarization overpotential associ-
ated with small particles and the possible increase in manganese dissolution [61, 137].
The particle size is not uniform in the electrode, but exhibits a distribution of size and
aspect ratio [138]. The effect of particle size distribution was examined by Darling and
Newman [139]. The authors showed that the electrodes with a particle-size distribution
have larger solid-phase resistance and longer relaxation time than an electrode with uniform
particle size. Particles of different sizes also have varying packing density compared to
uniform particles and the discrepancies affect the porosity of the electrode. It is possible
to optimize the porosity in the electrode to maximize utilization by changing the particle
size distribution [140]. However, such refined control over spatial distribution within an
electrode is difficult to achieve in practice and the gain in energy density is insignificant
compared to an optimally designed cell with uniform porosity distribution [116].

3.3.4 Electrode Thickness and Porosity

The electrode morphological variables at optimal solutions are examined next. Figure 3.4
plots the optimal values for electrode thickness and porosity as the discharge rate is varied.
When the discharge rate increases, a thinner, more porous electrode is preferred to meet
the transport rate requirements. The combination of decreased electrode thickness and
increased porosity reduces the amount of active materials in the cell and hence reduces the
energy capacity of the cell.
Figure 3.4 also shows that the optimal cell design requires a thicker and more porous
positive electrode paired with a thinner, less porous negative electrode. This is due to the
difference in volumetric Coulombic capacities between the two electrodes (633 mAh/cm3
vs. 841 mAh/cm3 ). The positive electrode is therefore thicker than the negative one to
balance the charge capacity. Higher porosity is required in a thick electrode to provide
adequate rate of ion transport, which in turn causes the electrode to be even thicker. This
shows that to achieve optimal energy density, it is essential to that the thick electrode should
either have very high transport coefficient or be made of small particles in order to meet
mass transfer requirements and to maximize the packing efficiency of active materials.
The changes in electrode thickness and porosity can be quantified by plotting the mass

48
 250

200

thickness (m)
150 cathode

100 anode
50 50 350 650 950 1250 1550

0.5
cathode
0.4
porosity

0.3 anode
0.2
0.1 50 350 650 950 1250 1550
specific power density (W/kg)

Figure 3.4: Variations of electrode thickness and porosity of optimal cell designs with
respect to cell power requirement

ratio of positive to negative active materials, given by:

L+ + +
M = (3.4)
L 

where L is the electrode thickness,  is the mass fraction of active materials, is the solid
density, and the + and signs denote positive and negative electrodes, respectively. The op-
timal mass ratios range from 2.77 to 2.85, as shown in Figure 3.5, which is consistent with
the optimal mass ratio of 2.8 reported by Tarascon and Guyomard [141]. The optimization
results also show that the optimal mass ratio for a well-designed cell is not constant but is
a linear function of the cell power, although the overall variation in the mass ratio is less
than 3%.
The small variation in optimal mass ratio is attributed to the fixed cutoff voltage of 3 V,
which results in a slightly different SOC at the end of simulation. The positive to negative
electrode charge capacity ratio is also plotted in Figure 3.5, and is given by:

q+ y
C = M (3.5)
q x

where q denotes the Coulombic capacity of active material in mAh/g, y and x represent
the change in stiochiometric lithium coefficients in the cathode and anode, respectively.

49
 1.02
2.86

1.01
2.84

capacity ratio, Q +/Q

mass ratio, m +/m
2.82 1.00

2.80 0.99

2.78
0.98
0 200 400 600 800 1000 1200 1400 1600
specific power density (W/kg)

Figure 3.5: Active material mass ratio and charge capacity ratio for optimal cell designs

Note there is balance between positive and negative charge ratio for all optimized cells.
This is the result of charge conservation. If the SOC remaining in the cell at the cutoff volt-
age is taken into account, the actual capacity ratio varies slightly from 0.99 to 1.02. The
slight imbalance in capacity ratios maximizes the SOC range during which the cell voltage
is above the cutoff voltage, and subsequently maximizes the amount of useful energy ex-
tracted during galvanostatic discharge. In practical cell designs, the negative electrode often
has a larger capacity to ensure full utilization of the positive electrode and to compensate
for loss of cyclable lithium due to side reactions such as solid electrolyte interface (SEI)
layer formation. Such side reactions are not included in the current model and their effects
on cell behavior are ignored. The proximity of the capacity ratios to unity for optimal cell
designs at all power requirements validates the approach of requiring charge balance in the
electrodes for preliminary cell design.

3.3.5 Conductivity and Diffusivity

The results from the various optimization runs initialized from different starting points
indicate that the diffusivity and conductivity values do not converge to specific values at
the optimal designs. Initialization from different starting points result in different final
design points as well. The conductivity value range used in this study reflects cathode
materials already doped with conduction-enhancing carbon additives [142]. While higher

50
values are favored, in this range the cell performance is relatively insensitive to change in
conductivity.

10-11
10-12
diffusivity (m2 /s)

10-13
10-14
10-15
cathode
10-16 50 350 650 950 1250 1550
-11
10
10-12
diffusivity (m2 /s)

10-13
10-14
10-15
anode
-16
10 50 350 650 950 1250 1550
specific power density (W/kg)

Figure 3.6: Diffusivity variations at optimal cell designs. Diffusivities need to be high at
optimal cell designs to facilitate ion movement.

The diffusivity ranges at optimal cell designs are plotted in Figures 3.6, which shows the
variations of diffusivity over the entire diffusivity range, and 3.7, and presents the close-up
view that spans only one order of magnitude. One can see that for all cases the diffusivity
values of the optimal designs converge close to the upper bound, with values ranging from
1012 to 1011 m2 /s. Given that the diffusivity bound spans 5 orders of magnitude, the optimal
diffusivity ranges are small. Du et al. [114] have shown that the performance of the lithium
manganese batteries is related to a dimensionless time , which is defined as the ratio
between the discharge time tdis , and the diffusion time, tdif , i.e.,

tdis r2
= ; tdif = (3.6)
tdif Ds

where r and Ds are the solid particle radius and bulk diffusivity in the positive electrode,

51
 10-11
diffusivity (m2 /s)

cathode
10-12
50 350 650 950 1250 1550
-11
10
diffusivity (m2 /s)

anode
10-12
50 350 650 950 1250 1550
specific power density (W/kg)

Figure 3.7: Close-up view of diffusivity distributions at optimal cell designs. While the
diffusivity values need to be high, they do not converge to any specific values.

respectively. Cell designs for which is greater than unity demonstrate very little increase
in energy capacity upon further increase in diffusivity. All optimal designs converged to
the smallest particle sizes given by the lower bound, and hence this results in very large
, even at high cycling rates such that the cells are diffusion-limited only when the bulk
diffusivity is close to the lower bound of 1016 m2 /s.
To examine the effects of diffusivity on energy density, a one-dimensional sweep of
diffusivity is carried out at the optimal design points, with all other design parameters
fixed. The differences between the energy densities and the maximum achievable values
are plotted as functions of bulk diffusivities and dimensionless time in Figure 3.8. Effects
of diffusivity on energy density at four representative discharge rates are shown. The values
of dimensionless time, , at the optimal cell designs for all four cells are much greater than
unity. As the value of decreases as diffusivity decreases, the energy density of the cell
decreases as well, but the reduction is not significant. As shown in the right-side plots in
Figure 3.8, the decrease in energy density is more than 1% only when anode is close to the

52
 101
100

105
105
deviation from maximum energy density (%)

deviation from maximum energy density (%)

1010 0.2C cell
1010 2C cell
4C cell
9C cell
1015 -14 1015 -16
10 10-13 10-12 10-11 10 10-15 10-14 10-13 10-12 10-11
cathode diffusivity, m2 /s anode diffusivity, m2 /s
101
100
105
105

1010 0.2C cell

10 10
2C cell
4C cell
9C cell
1015 2 1015 -1 0
10 103 104 105 106 10 10 101 102 103 104 105 106
cathode anode
Figure 3.8: Variation of energy density as a function of normalized diffusivity

value of unity, which corresponds to a decrease in diffusivity of three orders of magnitude

from its optimal values. Figure 3.8 also indicates that the effects of diffusivity on energy
density increases as cell power requirement increases. This is to be expected, as higher cell
power requires higher effective diffusion rate, which is directly proportional to the bulk
diffusivity.
That an optimal cell design should have diffusivity as large as possible is a trivial solu-
tion. However, the objective of the analysis done here is to quantify the effects of diffusivity
on cell performance. Results for Table 3.3 and Figure 3.8 show that cells with suboptimal
diffusivity suffer only small decreases in energy density if the optimality of other design
variables can be guaranteed.
In the current cell model, the bulk material properties are treated as constants, and
the effective properties are only functions of porosity via Bruggemans equation [89]. In

53
reality, the bulk properties are functions of the electrode material compositions and particle
sizes as well. Zhu et al. [143] showed that the aggregation of carbon additives to active
material as function of particle size, carbon-to-active-material mass ratio and temperature.
Decreasing particle size was shown to reduce the amount of carbon particles attached to
active materials, hence limiting the conductivity. A more refined treatment should include
the dependence of material properties on other cell variables, which would lead to more
realistic electrode designs.

3.3.6 Practical Battery Optimization

Results shown in the above sections have provided the theoretical guidelines to optimize the
battery performance. Maximizing the energy density of a lithium-ion battery cell requires
a combination of optimal engineering design provided via numerical optimization as well
as precise manufacturing control over the design parameters. Optimal mass ratio has often
been cited in the experimental as well as simulation studies as key to maximize the battery
capacity [144, 127]. It is mainly a function of the charge capacity of the active materials,
however it is also dependent on other aspects such as the OCV curve, the remaining capac-
ity at the cut-off voltage, and the amount of lithium lost to the SEI layer. Consideration of
the additional factors causes the mass ratio to deviate from the optimal one given by the
charge capacity balance and the precise value can be provided via optimization results.
Morphological change due to external pressure loads on battery needs to be accounted
for to ensure the right electrode porosity distribution to meet the mass transfer requirement.
Compression on electrodes due to either battery assembly or mechanical loading can alter
the porosity composition by as much as 40% [145]. Therefore it is important to simulate
the exterior environment of the battery pack to ensure that the post-assembly electrode
properties are as close to the theoretical design guideline as possible.
The quality control of the electrode materials is a key aspect in providing high energy
density. Optimization results have shown that keeping particle size small is important to
minimize diffusion path length. At the same time the uniformity of the particle sizes is
critical in ensuring high Coulombic capacity [139] as well. Refined manufacturing process
to minimize contact resistance at the current collector/ electrode interfaces also plays a key
role to minimize the overpotential that cannot be accounted for the transport-based cell
model [127].

54
3.3.7 Optimizer Performance
The CPU time and the number of iterations taken to reach convergence are plotted in Fig-
ure 3.9. Due to the randomized initial starting points, the number of iterations taken to reach
convergence differs significantly. The average CPU time decreases as the cell power re-
quirement increases. Optimization for the 50 W/kg/m2 cell, which corresponds to a cycling
rate of 0.2 C, required on average 62 hours to reach convergence; while the 1500 W/kg/m2
(9 C) cell required about 26 hours. This is because a low powered cell requires more time
steps to reach the cutoff voltage, and hence requires longer CPU time per function evalua-
tion.
70
100
60
90
50 80
mean optimization time, hr

no. of iterations
40 70

60
30
50
20
40
10 30

0 50 350 650 950 1250 20

1550
specific power density (W/kg/m2 )

Figure 3.9: Number of iterations and optimization time versus cell power requirement

3.4 Summary
In this chapter, a new numerical framework is developed to address the inadequacy of
using ad-hoc rules and parametric analysis to design battery cells. The objective is to au-
tomate cell designs and provide mathematically optimal designs. The framework couples
the physics-based cell model explained in Section 2.2.1 with a gradient-based optimiza-
tion algorithm (Section 2.4.2). In the present study the framework is applied to maximize
the gravimetric energy density of a lithium-ion cell with manganese dioxide cathode and

55
graphite anode under different discharge scenarios. A complete design of the cell that takes
both morphological and transport parameters is shown here, and the numerical relations of
how the optimal electrode parameters vary with the power requirement are quantified. The
gradient-based optimization framework is able to efficiently find the optimal cell designs
for each power requirements.
All optimal cell designs have the smallest possible particle size and separator thick-
ness. This is to minimize diffusion path length and to reduce the non-energy contributing
component mass. The electrode thickness decreases and the porosity increases as the cell
power increases, in order to satisfy increasing mass transfer requirement. The decrease in
electrode thickness and increase in porosity at higher discharge rates can be quantified via
the active material mass ratio and the charge capacity ratio at optimal designs. The charge
capacity ratios remain at unity for all optimal cells and the computed optimal electrode
mass ratio of 2.8 is consistent with the ones reported in literature. To maximize the use-
ful energy density, there are slight variations in the optimal mass ratios as the discharge
rate changes. This is to maximize the range of SOC that the cell voltage remains above
the cutoff voltage. Bulk transport parameters, such as diffusivity and conductivity, need to
be as high as possible, but their effects on energy density are limited. A non-dimensional
analysis of the discharge time versus diffusion time indicates that their effects on energy
density is only significant if the dimensionless ratio greater than one.
The current framework is able to obtain mathematically optimal designs and provide
the exact electrode morphological parameters needed to achieve maximum energy density.
Its practicality can be improved by including additional processes due to various capacity
fade mechanisms and micro-scale constraints. Factors such as SEI layer formation alter the
active material mass ratio as the charge capacity has to be balanced after, by accounting for
the consumption of cyclable lithium by various side reactions. The electrode dissolution
and lithium plating on charging are likely to limit the minimum particle sizes and separator
thicknesses as well. Accounting for these processes would introduce additional constraints
to the design problem and further restrict the feasible design space.

56
 CHAPTER 4

Design Optimization of a Battery Pack for

Plug-in Hybrid Vehicles

4.1 Introduction
The previous chapter establishes the framework for achieving maximum energy density of a
single lithium-ion cell subject to power constraints. Such design methods may be sufficient
to compute the optimal battery designs for personal electronics, where the battery design
is simple, and the discharge rate is approximately constant. The battery design for EVs,
on the other hand, is much more involved, since the discharge rate of the battery cannot
be expected to be constant, and the battery pack often contains thousands of cells. Slight
variations in cell properties are multiplied in the pack property difference. In addition, the
battery pack is required to satisfy multiple objectives and constraints to meet the drivers
demand as well as safety requirements. These criteria include energy density, specific
power, voltage, energy efficiency, commercial availability of battery materials, and cost. A
properly designed EV battery pack has to take into account all these factors in the design
process.
There are different levels of hybrid vehicles with varying DOH, each requiring a purpose-
built battery pack. Mild hybrid battery packs are mainly used to assist the ICE and to cap-
ture energy through regenerative braking. The discharge depth is shallow and the battery
packs are small. Battery packs for PHEVs and EVs have to satisfy vehicle power demand
and provide adequate driving range between recharging. However, the energy density of
the most advanced batteries is currently still orders of magnitude lower than that of gaso-
line. As such, the battery packs on PHEVs tend to be heavy, bulky and expensive. This
in turn limits the all-electric driving range for these vehicles, and thus optimally designed
battery system is essential to maximize the potential benefits of an electric powertrain.
Various types of battery models have been developed to analyze battery performance
with respect to various parameters. While numerical models provide an understanding

57
of the physics of battery operation, optimization algorithms provide the means to maxi-
mize battery properties and performance in hybrid vehicle operations. Shahi et al. [146]
applied a multi-objective optimization approach for the hybridization of a PHEV subject
to Urban Dynamometer Driving Schedule (UDDS) and Winnipeg Weekday Duty Cycle
(WWDC) drive cycle requirements. Wu et al. [147] described a methodology to mini-
mize the drivetrain cost of a parallel PHEV by optimizing its component sizes. Hung and
Wu [148] developed an integrated optimization strategy in which both the component siz-
ing and control strategies are taken into consideration to maximize the energy capacity
stored, while minimizing the energy consumed for a given driving cycle. Optimization of
combined component sizing and control strategies were explored by Zou et al. [149] to
study the hybridization of a tracked vehicle, and by Kim and Peng [150] for the design of
fuel cell/battery hybrid vehicles. In all these efforts, however, EV system designers treated
the battery either as a black box with energy and power output, or used simple algebraic
equations to model the battery performance.
While power management, control strategies, and component sizing play key roles
in achieving greater overall vehicle efficiency, a detailed optimization of PHEV battery
packs has not been considered. Most of the earlier battery optimization efforts focused
on single-cell optimization, where the battery was optimized for maximum energy den-
sity [111, 151, 114, 115, 116]. More recent efforts have optimized the energy capac-
ity of battery cells with respect to different power capacities by varying the applied cur-
rent [126, 117]. Most optimization studies, however, ignored the multitude of hybrid ve-
hicle operational requirements. Specifically, the battery pack has to satisfy: (1) voltage
and current constraints for both safety reasons and to minimize power electronic cost, and
(2) energy and power requirements for performance. To address these issues, a numerical
framework is used to optimize the mass, volume and material costs of the battery pack,
while satisfying all the relevant requirements. By combining efficient numerical methods
with existing battery models, waste attributed to sub-optimal pack design can be reduced.
This type of analysis is especially important as EVs become more mainstream and higher
volume, where small variations from the optimal solution, which may only result in slight
overdesign (in terms of cost or volume), results in large penalties when compounded over
large quantity of vehicles. This type of analysis also provides a rapid, cost effective de-
sign tool in the early phases of vehicle development, giving realistic guidelines on what is
possible in terms of cost, size, and weight for a given battery chemistry.
In this chapter, optimization of a representative PHEV battery pack is performed us-
ing a hybrid optimization strategy explained in Section 2.4.5. Optimization is used to
demonstrate that the resulting battery pack is able to fulfill the various PHEV operation

58
requirements most efficiently, hence maximizing the potential gains of PHEV operation.
Finally, three federal test drive cycles are used to evaluate the realistic performance of the
optimized battery designs by comparing the all-electric driving ranges.

4.2 Problem Formulation

4.2.1 Battery Model

A battery pack model that consists of identical electrochemical cells is arranged in series
and parallel, as shown in Figure 4.1. There are n modules arranged in series, and each
module consists of m layers arranged in parallel. The cells are arranged in such a way that
they have to satisfy the safety (voltage and current) and performance (energy and power)
constraints. In reality, the cell properties differ slightly due to manufacturing imprecisions,
causing variations in cell capacities. As such, accurate battery pack modeling requires each
cell to be treated individually [50]. Charge equalization techniques are employed to balance
the cells to enhance battery life and to maintain total capacity as well [152, 153]. Variations
in cell properties also cause additional difficulties in estimating the battery SOC.
These additional details in the battery modeling are ignored, as they are not fundamen-
tal to the design philosophy outlined in this chapter. Accounting for these issues would
increase computational burden without shedding new insights into the pack design. How-
ever, using the physics-based cell model as the foundation for the pack model would easily
allow cell-to-cell variations and accurately reflect intrinsic cell imbalance due to variations
in the amount of active materials in the cells [50].
PHEV operations require large battery packs of high energy density cells to provide
adequate all-electric driving range. Currently, only lithium-ion batteries are able to fulfill
the requirements. There are various viable lithium-ion electrochemical cells of different
energy densities, costs and cycling stability. An energy density comparison of various
lithium-ion cells using dimensional analysis has been provided by Du et al. [154]. The
cells used for PHEV battery pack is identical to the ones used in the single-cell optimization
shown in Chapter 3. The fixed properties of the cell are listed in Table 3.1 and will not be
repeated here. In addition to the fixed parameters of Table 3.1, the cycling rate is kept
constant and the negative to positive electrode Coulombic capacity ratio is fixed to one
in the PHEV battery design. For an ideal cell, equality in the electrode charge capacities
results in optimal cell designs, as shown in Figure 3.5. In reality, commercially available
lithium-ion cells are made with negative to positive electrode charge capacity ratios larger
than one. This is to compensate for cyclable lithium loss due to SEI formation in the initial

59
 n modules

m layers

layer
module

Figure 4.1: Layout of a battery comprised of uniform cells

phase, and to prevent lithium plating from occurring during charging.

In addition to the fixed variables, six free variables (Table 4.1) are selected to deter-
mine the optimal cell design and pack configuration that best fulfills the pack requirements.
These six variables are selected because they represent design parameters that can be read-
ily manipulated by the battery manufacturer to tailor the battery pack properties. There are
other parameters, such as diffusivity and conductivity, as well as electrode particle size, that
also affect the performance of the cell. Results from the previous chapter [126] showed that
in the absence of degradation mechanisms and side reactions, these other variables should
go to the bounds at optimal cell designs. Their relative effects on energy density of the cell
were also shown to be less than the morphological parameters chosen for this study [114].
Therefore, these are treated as fixed parameters in this study. Among the six variables, the
electrode thicknesses and porosities are morphological parameters that balance the amount
of energy content in the cell with the ion transport requirement. A thicker electrode con-
tains more active material for the intercalation process, and hence higher energy capacity.
A more porous electrode allows higher rate of ion transport by increasing the effective
transport coefficient. However, this reduces the fraction of charge-storing active materi-
als and overall capacity of the cell. The competing effects of higher energy and higher
power results in an optimal cell design that must reach a compromise between energy and
power. The cutoff voltage determines the lowest SOC in the cell that can still fulfill the

60
power requirement. The peak power available is dependent on the capacity remaining in
the battery and must be accurately determined to avoid over-discharging the battery. While
the SOC and cell voltage required to calculate peak power can be conveniently extracted
from the cell model in the current framework, a more practical approach would involve a
multi-parameter, model-based method [155] to estimate the peak power. The last variable
is the number of cells connected in parallel within each module, which also determines the
maximum current of the pack. The number of modules connected in series is fixed by the
maximum voltage of the battery pack, which will be explained in greater detail at the end
of this section.

Table 4.1: Design variables are the morphological parameters and battery layout variables
that can be easily altered by battery designer.

Parameters Lower bound Upper bound

Cathode thickness (m) 40.0 250.0
Anode thickness (m) 40.0 250.0
Cathode porosity (m) 0.1 0.6
Anode porosity (m) 0.1 0.6
Cut-off voltage (V) 2.6 3.6
Number of layers 1 30

Three key properties of the battery pack are identified as the objective functions to be
minimized: mass, volume and material cost. Each objective can be given as linear functions
of the design variables:
PP
Mass = nmA ij i lj
j i
P
Volume = nmA lj (4.1)
j
PP
Cost = nmA bi ij i lj
j i

where n and m are the number of modules and layers in the battery pack, respectively, A is
the cross-section area of each cell, , and b are the volume fraction, the mass density and
the unit cost [156] of the constitutive materials, respectively, l is the thickness of the cell
component. The index j cycles over the cell components, namely the positive electrode,
separator, negative electrode, and the current collectors, while i cycles over the cell mate-
rials. Note that if the materials with higher density have higher unit prices, then the cost
becomes directly correlated with the mass, and reducing the battery pack mass reduces the
cost as well.

61
 Table 4.2: Conversion of pack-level requirements to cell-level constraints

Pack Cell
Vpack 400 V n Vcell,init 400 V
Voltage
Vpack 280V n Vcell,end 280 V
Current Ipack 420 A = m Icell 420 A
Energy Epack 12 kWh nm Ecell 12 kWh
Power Ppack 120 kW nm Pcell 120 kW
Charge balance Q+ = Q

Equations (4.1) show that the objective functions are linear with respect to the design
variables. Minimization of the objective functions without proper consideration of con-
straints results in the trivial solution where all design variables are at their lower bounds.
Therefore, a useful battery pack design requires satisfying appropriate design constraints.
These safety and performance requirements impose limits on how close to the lower bounds
the design variables can go. The constraints for the problem are listed in Table 4.2.
The energy of the battery pack is computed by galvanostatic discharge of the cells at
1 C cycling rate, while the maximum power is the average power available during a 10-
second maximum current pulse at the end of the 1 C discharge. The maximum power is
computed at the lowest SOC, as this is the point where cell voltage is the lowest. If the
battery can meet the power requirement at the end of discharge, it can meet the power re-
quirement throughout its operation. The 10-second current pulse requirement provides a
good estimate of the maximum power required for vehicle operation. Based on the con-
straints outlined in Table 4.2, the maximum voltage limits the number of battery modules
connected in series. The maximum voltage in a cell is the OCV at the fully charged state,
and the maximum pack voltage divided by this value provides the number of modules al-
lowed in the battery pack. In this study, the number of cells connected in series is fixed at
99. The minimum voltage is computed at the end of the 10-second maximum current pulse
and it sets the limit on the depth of discharge of the cells. Hence, the optimization problem
can be simplified by replacing the voltage constraints with the fixed number of modules in
the pack and the minimum cell voltage to terminate discharge.

62
4.3 Results

4.3.1 Discharge Profile

A typical discharge profile obtained from the cell-model simulation is shown in Figure 4.2,
where both the cell voltage and the current density profiles are plotted. The small insert in
Figure 4.2 shows the voltage and current profiles at the transition between the galvanostatic
discharge and peak power current pulse. The sudden jump in the cell current creates a
discontinuity in the voltage profile, and it is the result of a sudden increase in the current
density. The cell voltage decreases rapidly as the amount of charge is depleted at a greater
rate at the end of the discharge.
4.0 130

3.8 120
voltage profile
110
3.6

current density, A/m2

3.4 Last 20s of discharge 120
cell potential, V

100
3.4
3.2 100 90
3.2 3.0 80
80
2.8 52.5 60
3.0 52.7 52.9
70
current profile
2.80 10 20 30 40 50 60
time, min

Figure 4.2: Discharge profile for a lithium-ion cell undergoing 1 C constant current dis-
charge (main) followed by a 10-second peak power pulse at the end of the discharge (insert)

The secant method is used to obtain the maximum current during the 10-second pulse.
The maximum current is defined as the largest current possible while maintaining the min-
imum voltage at the end of the pulse. The iterative process for the maximum current com-
putation is shown in Figure 4.3. The final cell voltage decreases monotonically as the pulse
current is increased, and the negative correlation between the maximum current and the
final cell voltage is shown in the insert.

63
 3.2 1st iteration

3.1

cell potential, V
3.0 potential vs. current

final iteration
2.9 Vmin 2nd iteration

min. voltage
2.8

3160 3162 3164 3166 3168 3170

time, sec

Figure 4.3: Discharge profiles of the 10-second peak current phase. The secant method
is used to determine the maximum current such that the cell potential is exactly at the
minimum voltage at the end of the discharge (insert)

4.3.2 Optimization Results

As mentioned previously, three different optimization problems are solved to design the
PHEV battery pack: 1) battery mass minimization, 2) battery volume minimization, and 3)
battery material cost minimization. The six design variables are listed in Table 4.1 and the
four constraints are listed in Table 4.2. To compare the performance of the optimized cell
designs, three initial pack designs are selected. The first two designs follow the general ad
hoc guidelines for cell electrode design: the first one is a power cell with thin electrode
and high porosity, and the second one is an energy cell with thick electrode and low poros-
ity. The third one is based on the earlier work of single cell optimization subjected to a
power constraint [126]. The optimized cell selected for comparison is designed at the same
power-to-energy ratio as specified in the problem constraint shown in Table 4.2. The dif-
ference between the single cell and pack design philosophy is that single cell optimization
is performed at one discharge rate, while the power and energy requirement for the pack
design are performed at different discharge conditions, resulting in additional constraints
in the design problem. The specifications for the three initial designs and their properties
are listed in Table 4.3. All three initial designs satisfy the capacity balance requirement
imposed on the problem.

64
 Table 4.3: Battery cell properties of initial designs

Power cell Energy cell Optimized single cell

Cathode thickness (m) 129.9 189.8 141.7
Anode thickness (m) 80.0 120.0 70.0
Cathode porosity 0.4 0.2 0.442
Anode porosity 0.4 0.2 0.322
Cutoff voltage (V) 3.53 3.63 3.53
Number of layers 13 13 13
Mass (kg) 86.89 126.00 87.14
Volume (dm3 ) 29.34 39.62 29.52
Cost ($) 1398 1862 1401
Energy capacity [ 12kWh] 11.9 11.8 11.9
Maximum current [ 420A] 388.0 385.2 389.3
Peak power [ 120kW] 121.5 123.3 121.7

Analyses of initial pack designs show that the optimized single cell and power cell
designs have similar properties, with the power cell exhibiting slightly better performance.
The energy cell design is the worst among the three, as its thick electrodes and low porosity
result in an expensive and bulky cell without providing any energy density improvement
over the other two. All three initial pack designs require 13 layers of cells in parallel to
satisfy the pack constraints. This results in the energy cell designs being the heaviest, most
voluminous, and most expensive.
Given that the three separate objective functions are not linear combinations of one an-
other, the optimal design for one objective should not be the optimum for another. Table 4.3
shows that the power cell design performs the best for all three objectives, it is thus not ex-
pected to be the true optimal design. In addition, while the power requirement of the pack
is exceeded, the energy capacity requirement is not fully satisfied by any of the three initial
designs. Therefore, further improvement is possible and can be achieved using numerical
optimization.
The gradient-free optimization is first carried out to obtain an approximate estimate of
the optimal design and to determine the integer value of number of cell layers. A repre-
sentative iteration history of cost optimization is shown in Figure 4.4. Note that ALPSO
is a population-based optimization method such that at each iteration, there are multiple
design points existing simultaneously. The plots shown here contain only the design point
that best satisfies the design problem criteria at each iteration, evaluated by the Lagrange
function for each particle [97].
Subplots a, b and c in Figure 4.4 show how the six design variables change during

65
 3.70 16
200 15
180 3.65
14
thickness (m) 160 3.60 cutoff voltage

cutoff volt. (V)

13

no. of layers
140 cathode 3.55 12
120 anode 11
3.50
100 3.45 10
80 9
3.40 no. layers
60 (a) (b) 8
0 10 20 30 40 50 3.350 10 20 30 40 507
iteration no. iteration no.
1800 140
0.40 cathode 1700 peak power 130
anode 1600 max current 120
0.35 1500 energy 110

feasibility, %
cost (US$)
porosity

0.30 1400 100

1300 90
0.25 1200 80
1100 cost 70
0.20 (c) 1000 (d) 60
0 10 20 30 40 50 9000 10 20 30 40 5050
iteration no. iteration no.
Figure 4.4: Iteration history of an optimization to minimize battery cost showing the evolu-
tion of: a) electrode thicknesses, b) cutoff-voltage and no of layers, c) electrode porosities,
and d) cost and normalized inequality constraint values

the iteration, while subplot 4.4d shows the evolution of the objective function (solid line)
and the normalized constraint values (dashed lines) during the optimization process. The
normalized constraint value is given as a percentage, with 100% and above indicating that
the constraint is satisfied. The charge balance constraint is an equality constraint that has
to be satisfied exactly at all times and hence is not plotted. The optimization history shown
in Figure 4.4 can be broken down into three phases. The initial design has large number
of layers and this resulted in a high battery cost and extremely high current. The first six
iterations reduce the number of layers and hence lower the current to below the maximum
allowed value. However, this results in energy and power requirements not being met.
Further adjustment of the design variables in the next five iterations results in the cell design
converging towards the final cell design, at which point the energy and power constraints
are fulfilled exactly. Given the competing effects of energy and power in the battery cell,
this result is expected, as the optimal cell design would be one that satisfies but does not
exceed both requirements. The maximum current is well below the maximum allowed
level, indicating that it is not an active constraint.
Multiple optimizations runs of ALPSO were performed for each optimization problem.

66
The best results are listed in Table 4.4. Optimizing for mass and material costs results in
very similar optimal cell designs. In fact, the optimal design obtained for cost minimization
has the lowest battery pack mass as well. Given that reduction in battery mass naturally
leads to less materials and hence lower cost, this result is not surprising. While the differ-
ence between the minimal-mass weight and minimal-cost weight is within the convergence
tolerance of the optimizer, it also indicates that ALPSO is unable to locate the true op-
tima in this situation, such that the final result found for mass minimization problem is
sub-optimal.

Objective function
Properties
Mass Volume Cost
Mass (kg) 74.97 75.67 74.90
Volume (dm3 ) 24.91 24.02 24.48
Cost ($) 1153 1169 1147
Cathode thickness (m) 169.8 141.9 168.8
Anode thickness (m) 104.7 86.4 99.8
Cathode porosity 0.321 0.272 0.305
Anode porosity 0.314 0.252 0.269
Cutoff voltage (V) 3.52 3.53 3.53
Number of layers 9 10 9
Energy capacity [ 12kWh] 12.0 12.0 12.0
Maximum current [ 420A] 398.4 395.0 396.6
Peak power [ 120kW] 120.1 119.9 120.1
Table 4.4: Preliminary designs after gradient-free optimization. Results shown are the
best-available ones due to stochastic nature of the ALPSO algorithm.

A comparison between Tables 4.3 and 4.4 shows decreases of 13.4%, 18.1%, and 17.9%
in mass, volume and cost respectively from initial to optimal designs. Compared with
the best initial cell designs, all three optimal designs have thicker electrodes and lower
porosities. Therefore, each cell has higher energy density, while still satisfying the power
requirement. The energy and power requirements are satisfied with far fewer cells (9 or 10
layers vs. 13), and hence better overall pack properties.
There are some differences between the three optimized designs. The mass and cost
minimization problems produce optimal cell designs that have thick electrodes, such that
the cell has as much active material for lithium ion intercalation as possible. This is to
maximize the energy density of the cell, and to reduce the total amount of materials needed.
While the cost minimization problem results in a battery pack lighter than the one obtained
from the mass minimization problem, the difference is within the convergence tolerance
of ALPSO. Volume minimization, on the other hand, produces an optimal cell design that

67
has thinner electrodes with lower porosities. This results in a cell design that has lower
gravimetric energy density, but higher volumetric energy density compared to the designs
for the other two problems.

Table 4.5: Refined optimal designs obtained using gradient-based optimizations

Objective function
Properties
Mass Volume Cost
Mass (kg) 74.77 79.53 75.06
3
Volume (dm ) 24.49 23.85 24.37
Cost ($) 1146 1241 1146
Cathode thickness (m) 169.2 111.7 169.4
Anode thickness (m) 99.4 64.3 97.5
Cathode porosity 0.310 0.220 0.300
Anode porosity 0.270 0.150 0.245
Cutoff voltage (V) 3.53 3.55 3.54
No. of layers 9 12 9
Energy capacity [ 12kWh] 12.0 12.0 12.0
Maximum current [ 420A] 395.9 392.5 394.4
Peak power [ 120kW] 119.9 120.1 119.8

The optimal cell designs are further improved by using SNOPT to refine the continuous
cell variables. The gradient-based optimization is initiated at the optimal designs obtained
from ALPSO. The number of layers in each module, which is a discrete variable, is fixed
at the values around the optimal number of layers obtained via ALPSO, while the gradient-
based optimizer fine-tunes the continuous variables. The results obtained using SNOPT
are listed in Table 4.5. Again the mass minimization and cost minimization solutions are
almost identical, and the material cost difference between the two is less than one dollar.
The volume minimization problem produces a battery pack with much thinner electrodes
and lower porosities. Comparing the SNOPT and ALPSO volume minimization results,
the cathode and anode thicknesses are 21.3% and 25.6% thinner, respectively, while the
porosities are 19.1% and 40.5% lower. The number of cells required increases by 20%. The
low electrode porosity means the cell design is less adequate for handling high discharge
rates. However, this is alleviated by having a larger number of cells connected in parallel,
and hence a smaller current density through each cell. The SNOPT optimal design has
higher mass and material cost, but lower volume. While utilizing larger number of cells
with lower energy density may seem counter-intuitive, it demonstrates the ability of the
optimizer to drive the design to achieve the objective, which in this particular instance is to
minimize the volume of the pack.

68
 Results in Table 4.5 show that improvements from gradient-free optimization to gradient-
based optimization results are less than 1% for all three optimization problems. This is
mainly due to the flatness of the design space near the optimum, and there is little gain
in pinpointing the exact location of the optimal designs, as evidenced by the 8% differ-
ence in negative electrode porosity between ALPSO and SNOPT results. Such differences,
however, may become more important as more details about cell modeling are included.
For instance, manganese dissolution rate has been shown to correlate to the interfacial sur-
face area in the porous electrode [157]. Increase in cathode porosity increases interfacial
surface area and could potentially cause accelerated cell degradation. Inclusion of addi-
tional degradation mechanisms would likely add nonlinearity to the design space, further
restricting the feasible design regions.
One common problem with gradient-based optimization is that it often converges to
local optimum solutions instead of global best ones. To show that the solutions are indeed
global optima, and that the design space near the solutions does not have local optima,
the contour plots of the objective functions near the optimal regions are plotted. The region
plotted is a 2-dimensional plane defined by the three optimal design points of mass, volume
and cost, respectively. The plane is obtained by projecting the design space onto the plane
spanned by the coordinates of the three optimal design points. These three points are
transformed into the non-dimensional coordinates (0 , 0 ), (1 , 0 ), (0 , 1 ), respectively, on
the plane. The shaded area on Figure 4.5 indicates the region where the energy, power,
voltage and current requirements are all satisfied. The blue curve indicates the narrow band
that fulfills the charge balance requirement, while the black lines show the design space
that satisfies the integer requirement imposed on the number of layers.
In the plots shown in Figure 4.5, one can see that the objective functions vary smoothly
in the feasible design space. The mass and cost contours vary monotonically in the feasible
design space, while there is a local volume maximum for the volume contour plot. All three
optimal design points are located on the boundary of the feasible space, again confirming
that the constraints are active in this design problem. The location of the optimal designs
are further restricted by the charge capacity equality constraints and the integer requirement
on the number of cell layers. Based on the information available in Figure 4.5, it is clear that
the optimal designs are indeed the best possible designs in the 2-dimensional plane shown
here. Such information, together with the smoothness of the objective functions, gives us
confidence about the global optimality of the results. The similarity between mass and cost
optimal designs can also be explained from the contour plots in Figure 4.5. Comparison
of the mass and cost contours reveal that the two objectives are very similar to one another
in the given plane. The gradients for both objectives point in the same x-direction, and in

69
 1.2 1.2
number of layers (a) number of layers (b)
integer requirement integer requirement

79.8
1.0
cost opt 1.0
cost opt

24.0
76.0
0.8 0.8
continuous feasible region
cost opt mass opt continuous feasible region

cost opt mass opt

23.1
0.6 0.6

79.0
77.0

79.0

23.7
charge balance
0.4 0.4 charge balance
73.0

75.0

24.0
0.2 0.2

0.0 mass opt vol opt 0.0 mass opt vol opt
74.0

78.0

24.3

23.4
24.3
24.6

24.6
0.20.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 0.20.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
vol opt mass opt vol opt mass opt
1.2
number of layers (c)
cost opt integer requirement
1.0
1120

continuous feasible region

0.8 Grey area represents the feasible design space that sat-
cost opt mass opt

isfies the energy, power, and current constraints. The

0.6
black lines represent integer constraint on the num-
0.4
ber of layers and blue curves represent the part of the
charge balance design space satisfying charge capacity balance. The
1160

0.2 backgrounds are overlay with the contour plots of a)

mass in kg, b) volume in liters, and c) cost in US$.
0.0 mass opt vol opt
1200

1220
1140

1180

1240

0.20.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2

vol opt mass opt

Figure 4.5: Contour plots of objective functions on the plane spanning the three optimal
design points

both cases the best objective functions are on the left boundary of the feasible space. The
similarities of the objectives result in mass and cost optimal designs being very close to
each other in the actual design space.
The optimal cell design for mass minimization problem is compared with the optimal
single-cell design results from the previous work [126], in which the optimal cell has the
maximum energy density at constrained discharge rates. Comparison to the optimal cell
at 1C discharge rate shows the PHEV pack cell design has thinner electrodes and higher
porosities. This is due to the additional peak power requirement at the end of discharge,
which imposes a higher ion transport requirement. The resulting cell design, while not
optimal in terms of energy density, is able to meet both the energy and power requirements
simultaneously, which is more important, given the variations in power demand under nor-
mal driving conditions.

70
 Table 4.6: Properties of the vehicle used to complete the driving cycle

Properties Values
Vehicle mass (kg) 1500
Passenger mass (kg) 150
Rolling resistance coeff. 0.01
Drag coefficient 0.30
2
Frontal area (m ) 2.0
Motor efficiency 0.85
Drivetrain efficiency 0.8
Generator efficiency 0.85
Regenerative braking factor 0.1
Miscellaneous power (kW) 1.0

4.3.3 Driving Cycle Test

While the optimal designs obtained using the aforementioned optimization framework
show better overall properties than the initial design, it is important to show that they
translate to actual performance advantage in the operation of the PHEV. In this case, it
is useful to compare the performance of the battery packs by simulating discharge for stan-
dard federal testing driving cycles. A standard sedan with the properties listed in Table 4.6
is used to compute the battery power required to complete the driving cycles. The required
power is computed using the vehicle model shown in Section 2.3, and taking the motor and
drivetrain efficiency factors, d and m , into account as well:

Pveh
Pbatt = (4.2)
d m

While the drivetrain and motor efficiencies change with the power and velocity of the ve-
hicle, they are assumed to be constant in this case.
The battery packs are discharged through the three standard federal driving cycles:
UDDS, SC03 and US06, whose speed and corresponding power profiles are shown in Fig-
ure 4.6. The UDDS and SC03 cycles both mimic city driving conditions. The SC03 cycle
is the more aggressive. US06 simulates highway driving condition. The highway accel-
eration requirements impose higher power demands on the battery, with power demands
peaking at 100 kW. The battery is discharged from an initial SOC of approximately 0.8
until the minimum voltage of 280 V is reached and the total distance covered for each of
the driving cycle is calculated. Calculations have shown that the minimum voltage of 280
V corresponds to a SOC level of approximately 5%. In reality, the depth of the discharge

71
 30
25 (a) UDDS
40

speed, m/s
30

Power, kW
20
15 20
10 10
5
00 0
5 10 15 20
25
(b) 50
20 SC03 40
speed, m/s

Power, kW
15 30
10 20
5 10
00 0
2 4 6 8 10
40 100
35 (c)
30 US06 80
speed, m/s

Power, kW
25 60
20 40
15
10 20
5 0
00 2 4 6 8 10
time, min

Figure 4.6: Federal driving cycle speed profiles and the corresponding battery power re-
quirement:

of EV battery packs is much less than that, hence the all-electric range obtained here will
be higher than what can be expected from actual EVs with a similar battery pack.
Both the initial design and optimal battery packs are discharged through the simulated
driving cycles. The voltage and SOC profiles of the minimum mass optimal battery pack
and the initial design discharged through the simulated US06 driving cycle are shown in
Figure 4.7. The open-circuit voltage (OCV) curve is plotted on the same figure for com-
parison. The SOC profile shows that the optimal battery pack lasts longer than the initial
design battery pack, resulting in 3.5% longer electric range. The initial battery design con-
tains more cells in parallel compared to the optimal battery pack, and therefore each cell
is subjected to a smaller current density. This results in higher battery pack voltage in the
initial design, and less energy lost per cell due to internal resistance. However, the higher
energy density of the optimal battery pack cells still results better performance overall.
The all electric ranges for various battery pack designs subject to the three driving cy-
cles are listed in Table 4.7. For all three driving cycles, the optimal battery pack designs
outperform the initial battery pack designs in terms of all-electric range. The most improve-
ment is in the SC03 cycle discharge, for which the electric ranges of minimum-volume and
minimum-cost battery packs are almost 10% better than the initial design.
While the electric driving range of the optimal battery pack designs show significant
improvement over the initial design, a better measure of the battery performance is the

72
 0.8 400

0.7 380
OCV
0.6
360
0.5
voltage 340

voltage, V
0.4
SOC

init. range: 39.7 km 320

0.3 opt. range: 41.1 km SOC
300
0.2

0.1 280
optimal
initial
0.00 5 10 15 20 25 30 260
time, min

Figure 4.7: Comparison of the voltage and SOC profiles of the initial design and minimum-
mass optimal battery pack discharged through the simulated US06 driving cycle

Table 4.7: All electric driving range for various battery designs

Battery design
Drive cycle Initial Min. mass Min. volume Min. cost
UDDS 53.7 km 57.2 km 58.1 km 58.1 km
SC03 48.6 km 52.5 km 53.4 km 53.4 km
US06 39.7 km 41.1 km 42.1 km 41.8 km

73
 0.8
0.7 mass opt
0.6 init

km/kg
0.5
0.4
0.3 (a)
0.2 UDDS SC03 US06
2.5
vol opt
2.0 init
km/l 1.5
1.0
(b)
0.5 UDDS SC03 US06
0.06
0.05 cost opt
0.04 init
km/$

0.03
0.02
0.01 (c)
0.00 UDDS SC03 US06
Figure 4.8: Comparison of battery performance between initial and optimal designs using
driving cycle data

distance traveled per unit of battery performance metric, where the metric is either mass,
volume or material cost. Figure 4.8 compares the performance of the optimal battery de-
signs with the best initial design. While the initial and optimal designs have similar all-
electric drive range, the optimal designs clearly have better performance per unit of battery
performance metric. On average, the optimal battery packs show 23.1% improvement in
distance per unit of battery mass, 32.8% improvement in distance per unit volume and
31.4% improvement in distance per unit cost.
The results in Table 4.7 also show that the minimum-volume battery pack has the
longest electric range among the optimal battery pack designs. The disparity between the
electric driving ranges of the various optimal battery packs demonstrates that while the bat-
tery packs satisfy similar energy and power requirements, the performance is dependent on
the driving cycle, or control variables governing the discharge of the battery. This points to
the possible advantage of performing a design-control coupled optimization. In addition,
only the all-electric operating mode rather than the overall performance of the PHEV drive
is considered. The problem definition is limited to minimizing the properties of the bat-
tery pack, and this allows decoupling of the battery from other drivetrain components. To
truly optimize the operation of the PHEV, the hybrid mode, during which both engine and
electric power are in use, and the driving conditions that favor the hybrid mode need to be
considered as well. The degree of hybridization in the PHEV, which is the ratio of electric

74
motor power to the total drivetrain power, is shown to affect the optimality of the drivetrain
components performance [19]. Therefore, the overall hybridization scheme of a PHEV is
defined by the battery, electric motor, and internal combustion engine collectively [146].
Inclusion of various components also necessitates the coupling of power control strategies
with design parameters to determine the optimal performance. Kim and Peng [150] showed
that the power management and design optimization improved fuel economy by 17% rel-
ative to power management only optimization. Consideration of the engine and electric
motor sizes could allow for battery design that best meets the peak power demand during
climbing and acceleration, while keeping the engine operation at maximum efficiency.

4.4 Summary
This chapter extends the battery optimization framework shown in Chapter 3 to the de-
sign of a PHEV battery pack. Given the nonlinear and mixed-integer nature of the pack
design problem, a hybrid optimization approach that takes advantage of type of optimizer
is developed. A gradient-free optimizer is first used to obtain an approximate estimate of
the optimal design and to obtain the optimal integer design variables. The design is then
further refined using the gradient-based optimization. Most EV system optimization tends
to treat the battery as a black box or a simplified circuit model, thus ignoring the transport
laws governing the behavior of the electrochemical cells. The framework developed here
provides detailed battery design in the context of EV operation. It enables optimization of
the battery pack layout, as well as the morphological design parameters of the individual
cells.
Three optimization problems of minimizing the battery pack mass, volume, and cost are
demonstrated using the optimization framework. Comparisons between the initial designs
using general guidelines and the optimal designs show overall improvements of 13.9%,
18.7% and 18.0% in battery mass, volume, and cost, respectively, thus validating the ne-
cessity of a numerical approach to the battery pack design. The optimal designs also per-
form better in real drive cycle simulations. The improvements in battery pack properties
can be translated to 23.1% increase in distance traveled per unit mass, 32.8% increase in
distance per unit volume, and 31.4% increase in distance per unit cost. The electrochemical
cell in this case is assumed to be ideal, so additional capacity fade mechanisms should be
considered in the future.
The design methodology shown here is useful in obtaining the best possible initial
designs, from which further refinement can be carried out by accounting for additional
details, such as degradation mechanisms and manufacturing constraints. The driving cycle

75
comparison shows that the three optimal battery pack designs, while satisfying the same
set of constraints, have slightly different all-electric driving ranges. The minimum volume
battery pack has the furthest driving range for all three federal driving cycles. This is due to
the difference between design discharge condition and the actual driving cycle conditions.
Therefore, it could be useful to perform a cell design optimization that uses the driving
cycle power requirement as the discharge profile. Multi-point design schemes [26] can be
employed to ensure the obtained optimal solution is not specific cycle beating.

76
 CHAPTER 5

Multi-Cell Design Optimization for Electric

Vehicle Battery Packs

5.1 Introduction
The previous chapter demonstrates the design optimization of a battery packs for PHEV op-
eration. The optimization algorithm is able to minimize the battery pack mass, volume or
cost subject to various constraints. The problem formulation follows the conventional phi-
losophy of using uniform cells for battery pack design and altering the cell power-to-energy
ratio for different battery packs. However, given that energy and power are competing ob-
jectives, cell designs that have to satisfy both are sub-optimal in both aspects. Therefore,
it is unclear whether a uniform-cell pack design is the best way to fulfill both requirements
in an EV application.
The Pareto front in Figure 5.1 shows that the power-to-energy function of optimal cell
designs is nonlinear functionas the discharge current is increased, the power capability
of the optimal cell design is increasing at a rate faster than it is losing energy capacity.
There are two reasons affecting the power to energy relation of optimal cell designs. As
the power requirement is reduced, the rate of ion transport required in the electrode is
reduced as well. The optimal electrode is one that contains more active materials and less
electrolyte. Therefore, the lower-power optimal cell design gains additional energy density
compared to the higher-power optimal cells. In addition, Equation (2.9) shows that the
power lost to the internal resistance is proportional to the square of the current. As the cell
power is reduced, the amount of power lost to overcome the internal resistance is reduced
as a quadratic function of current density, thus further increasing the energy density of the
optimal cells.
The Ragone plots shown on Figures 3.3 and 5.1 indicate that the energy density of a
lower-power optimal cell design decreases rapidly as the power requirement is increased
beyond its design point, while a higher-power optimal cell is unable to increase its energy

77
 1600
4-C cell Ragone
1400

1200
specific power density(W/kg)

1000
2-C cell Ragone
800

600 Pareto front

PHEV batt cell Ragone

400
9-C cell Ragone
200 0.2-C cell Ragone
0
100 120 140 160 180 200 220 240
gravimetric energy density (W-hr/kg)
Figure 5.1: Pareto front of optimal single-cell designs on a normal scale, comparing the
single-cell design with pack cell design

density significantly as the cell discharge power is decreased, resulting in cell discharge
characteristics deviating significantly from the Pareto front. There is not one cell that per-
forms the best across a wide range of power spectrum. The battery pack in an EV is ex-
pected to operate across a wide range of discharge scenarios as road conditions and driver
throttle input varies. The requirement for both high energy density and adequate power
capability results in a cell design that is non-optimal from a single-cell discharge perspec-
tive, as shown by the energy and power densities of the minimal-mass pack cell design
plotted in Figure 5.1. The concave shape of the power-energy curve and the non-optimality
of the pack cell design suggest that it may be advantageous to utilize cell designs that are
either power-specific or energy-specific, such that they lie on the power-energy Pareto front
instead of a cell design that is a compromise.
Another reason for utilizing multi-cell pack design comes about when we consider
the variation of internal resistance with current and SOC. Figure 5.2 compares the total
internal resistance variations of cells undergoing the UDDS federal driving cycle discharge
introduced in Chapter 4. The black points are the internal resistance of the initial power-cell

78
 designs, and the green points are the internal resistance of the minimal-mass optimal cell
designs. The initial cell design has much higher porosity in both electrodes, resulting in
lower energy density, and hence higher battery pack mass due to lower resistance to mass
transfer. However, its internal resistance is consistently lower than that of the optimal cell
designs. The difference is most obvious during peak power demands, such as when SOC is
approximately 0.55, where the internal resistance of the optimal cell designs are more than
double of that of the initial cell designs. Calculation shows that the 1.2% of nominal energy
capacity is used to overcome the internal resistance for the initial cell designs, while in the
optimal designs, it is 3.8%.
400 2.0
350
1.5

power, kW/m2
voltage, V

300
1.0
250
0.5
200
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
25 SOC
opt design resistance
internal resistance, m/m2

20 init design resistance

15
10
5
00.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
SOC
Figure 5.2: Comparison of cell internal resistance between the initial PHEV pack design
and the optimized pack design

Figure 5.2 shows that while the optimally-designed uniform-cell pack is able to achieve
the lowest battery pack propertiessuch as mass, volume, or costit is not the most effi-
cient way of utilizing the stored energy capacity. It can be inferred that while the a single
type of cell is optimal for a constant discharge rate, it is possible that a multi-cell design can
potentially minimize energy lost to overcome internal resistance during a varied discharge
scenario, such as during a typical driving cycle. The high-energy cells can be used to pro-

79
vide constant power at low discharge requirement, while high-power cells can be used to
meet peak power demand and capture energy from regenerative braking. To fully utilize
such a system, a complex control strategy that is SOC and power dependent is required.
Various multi-chemistry energy storage systems have been proposed by combining dif-
ferent batteries, fuel cells with lithium-ion batteries [150], or lithium-ion batteries with
super-capacitors [158, 159]. The choice between batteries, fuel cells and capacitors is one
between cost, energy density, power, and cycling stability. As stated in Chapter 1, lithium-
ion cells have very high energy density and good power and cycling behavior; fuel cells
have the highest energy density among the three but highest infrastructure cost; super-
capacitors, on the other hand, have high power densities and very long cycling stability.
The problem with super-capacitors, though, is that its energy density is orders of magni-
tude lower than that of lithium-ion battery [160]. While capacitors can supplement the
instantaneous peak power, their small energy density makes them unsuitable for large bat-
tery packs, such as those required in a PHEV or EV. Most of the lithium-ion/capacitor
packs that have been designed are for mild hybrid vehicles to provide moderate gain in
efficiency, rather than for full-scale vehicle operation, such as in an EV. In addition, its
advantage in power density has been diminished as modern purpose-built batteries are able
to achieve power densities comparable to that of capacitors [161]. Having two different
energy storage systems also adds to the complexity and costs of the already expensive EV
architecture, as additional componentssuch as DC-to-AC convertersare needed to en-
able compatibility between the various systems.
Figure 5.1 shows that even within a single lithium-ion chemistry, altering design variables
such as electrode thickness and porositycan produce cells with orders of magnitude more
power and power density. Therefore, it is foreseeable that a battery pack with multiple bat-
tery cells of the same chemistry can be produced to satisfy the varying discharge require-
ments.
The research done on modeling and optimization of a single lithium-ion cell and a
uniform-cell battery pack was explored in the previous two chapters. The idea of a battery
pack with multiple cell designs has been previously proposed [69, 19]. However, there are
no studies on establishing the designs of such multi-cell systems. Multi-cell batteries have
been used in medical devices, such as pace-makers [162, 163], where a low-power circuit
is used to control and monitor the device, while a second high-power output circuit delivers
the electrical pulses. In this case, the power requirements of the constant-discharge low-
power control circuit and pulse-discharge output circuit are known and independent of one
another. Specialized cells with multiple electrodes tailored for such discharge conditions
have been created and subsequently patented [164, 165].

80
 Various research groups and companies have developed multi-cell battery systems for
automobiles as well. Earlier efforts focused on developing a dual-battery system that con-
sists of a main unit and reserve pack for emergency vehicle start [166]. This concept has
been extended by Tesla to a full EV hybrid-battery pack [167], in which a second battery
pack is used only when the extended range mode is selected, or the SOC of the main battery
pack is critically low. General Electric demonstrated an electric bus fitted with a dual bat-
tery system in 2010 as well [168], where a high-energy density sodium battery is coupled
with a high-power lithium battery to fulfill the EV performance requirement.
What has is missing is a systematic study and quantification of the advantages of such
concepts over traditional uniform-cell designs. The objective of this chapter is to char-
acterize the performance of a multi-cell battery pack design with respect to cell design
parameters, and to compare that with a uniform-cell battery pack design. The conditions
under which a multi-cell design is advantageous are established and the improvements of
such designs are quantified as well.

5.2 Problem Formulation

In this section a new battery pack design that contains multiple cell designs is proposed. In
this case, a full EV that is driven purely by electricity is considered as the baseline vehicle.
The battery pack for an EV has to satisfy a minimum of two criteria, namely the energy and
power requirements. Figure 5.3 illustrates a simple multi-cell battery configuration, where
the energy and power requirements are fulfilled by separate high-energy density and high-
power cells. To simplify subsequent computation, the power requirement is solely satisfied
by the power cells, while the energy requirement is satisfied by the combined energy of both
types of cells. The number of cells in series for each pack is the same, and the battery packs
are connected in parallel. This configuration ensures that each battery pack has the same
maximum voltage, such that the two battery packs can be used individually or together to
power the EV.
To simplify the analyses in the subsequent sections, only one type of battery chemistry
is considered in the multi-cell battery design. The cell designs are varied to produce cells
of different power-to-energy ratio. The design variables in consideration here are the same
as those listed in Table 4.1, namely: battery layout, electrode thicknesses, and electrode
porosities.

81
 energy cells

power cells

Figure 5.3: Layout of a battery pack with multiple cell designs. The power and energy
packs are connected in series so as to not exceed the voltage limits of other electrical com-
ponents.

5.3 Simplified Analysis

To compare the performance of a multi-cell battery pack with a uniform-cell battery pack,
and to examine how the power and energy ratio of the different cells affect the battery
pack properties, the equations governing the design methodology have to be derived. The
battery pack is analyzed in the criterion space, where the energy and power of cells are
the parameters in consideration. These parameters in turn depend on cell variables, such
as those highlighted in Table 3.2, whose effects can be modeled via a cell model. For the
work shown here, the mass of the battery pack is the objective that is to be minimized.
The requirement of the multi-cell battery pack follows the specifications listed Ta-
ble 4.2. This is to ensure consistency in pack design for comparison. For a given battery
pack with energy requirement Ereq and power requirement Preq , the mass of the power

82
cells, Mpow , is given by:

Preq
Mpow = (5.1)
p
where p is the maximum achievable power density of the power cells. The peak power
requirement in this case is calculated from a 10-second maximum current pulse at the
end of discharge. It is very unlikely that the battery will be completely drained from the
10-second pulse discharge. Any energy capacity remaining from the 10-second pulse dis-
charge contributes towards the energy requirement. Furthermore, the maximum power
attainable is dependent on how much energy is used to contribute towards the energy re-
quirement. Hence, the power density of the power cell can be given as a function of its
energy density:
p = F(upow ), 0 upow umax (5.2)

where upow is the energy density of the power cell and umax corresponds to the nominal
energy density of the cell, i.e., when the power requirement is zero. Hence umax uegy .
The total mass of the power cells can be rewritten in terms of upow :

Preq
Mpow = (5.3)
F(upow )

The energy contribution of the power cells, Epow , is:

Preq
Epow = upow Mpow = upow (5.4)
F(upow )

Since the combined energy from the power and energy cells have to meet the pack require-
ment, the mass of the energy cells can be obtained from the additional energy required to
meet the pack energy demand:

Ereq Epow
Megy = uegy
h i
1 Preq
= uegy
Ereq upow F (upow )
(5.5)

Hence the mass of the battery pack can be expressed solely as a function of the energy

83
density of the power cells as follows:

Mpack = Megy + Mpow h i

Preq 1 Preq
= F (upow )
+ E req upow F (upow )
 uegy  (5.6)
Preq upow Ereq
= F (upow ) 1 uegy + uegy

The energy cells can be assumed to undergo discharge at a constant rate that is inversely
proportional to the energy capacity of the pack. There is an optimal cell design that attains
the maximum energy density at that particular discharge rate. The maximum achievable
energy density is assumed for the energy cells. Given that the optimal energy density is a
constant for a particular discharge rate, Equation (5.6) is only a function of the power cell
energy density, upow . How the battery pack mass varies with the power cell energy density
in turn depends on the function F(upow ).
The value of upow for which the minimum pack mass is obtained can be calculated by
taking the derivative:

uegy upow dF
 
dMpack Preq
= 1+ (5.7)
dupow uegy F(upow ) F(upow ) dupow

By equating the derivatives in Equation (5.7) to zero, the energy density of the power cell
that gives minimum battery mass can be obtained as:

upow uegy dF
=1 (5.8)
F(upow ) dupow

For a uniform-cell design, a single type of cell has to provide both the power and the energy
requirements. Given the problem definition in Section 5.2, the uniform-cell design becomes
a multi-cell pack design but with power cells only. Hence, the required energy density of
the cell can be obtained from Equation (5.5):

Preq
Ereq upow F (upow )
= 0
upow Ereq (5.9)
= F (upow )
= Preq

The equations above allow the battery designer to examine how the performance of
the cells affect the battery pack design using a reduced-order model in the criterion space,
where the battery pack property is a function of power and energy density only. The de-
sign of a multi-cell pack involves judiciously selecting the right combination of power and
energy cells, such that the battery pack property is optimized. The cell designs can be

84
subsequently determined by interpolating between the optimal cell designs at the required
energy and power.

5.4 Theoretical Multi-cell Battery Analysis

Equation (5.6) shows that the multi-cell battery pack mass is dependent on the energy
density of the power cell as well as its power versus energy relation, which is a function
of the batterys material properties and cell parameters. The analysis of battery pack mass
variation can be simplified by assuming certain generic functions for F(upow ). This enables
optimal battery pack mass to be derived analytically. The situations in which the multi-cell
battery pack has better properties than the uniform-cell design can be readily identified as
well. The dependence of maximum power density on energy density can be assumed to
obey a simple power-law function of the form:

F (u) = a b uc (5.10)

The power coefficient c is given three representative values of 0.5, 1.0, and 2.0 to simulate
the power to energy functions. These values allows the analytical functions to cover a wide
range of battery discharge characteristics. In addition, a quadratic function is added for
comparison as well. Reasonable values for a and b that reflect realistic bounds on energy
and power densities are selected to complete the functions.
Figure 5.4 plots how the battery pack mass varies with the energy density of the power
cells. The energy capacity of the battery pack is 30 kWh, and the required power is 150 kW.
The battery pack mass variations with different power to energy curves, and the comparison
with the mass of a uniform-cell battery pack are plotted on the bottom half of Figure 5.4.
For all cases, the theoretical cells are assumed to have the same maximum energy density
of 210 Wh/kg and maximum power density of 10 kW/kg.
The battery pack masses for all four curves converge to the same point when the power
density of the cells is maximum. This is the result of the power to energy curves having the
same y axis intersection when the energy density is zero. One can also see that if the
relationship between the energy and power is linear, there is no difference in battery pack
mass as the energy density of the power cell is varied. Therefore, there is no advantage in
terms of mass between the multi-cell and uniform-cell pack designs.
For a battery with a convex power to energy function, i.e., F(u) = a b u0.5 , the
uniform-cell pack design is lighter than the multi-cell pack design. However, Figure 5.1
shows that the power to energy relations of optimal cell designs can be concave. In this

85
 10

power density of P cell, kW/kg

c=0.5
8 c=1
6 c=2
quadratic
4 F(x) = a - bxc
2
00 50 100 150 200
energy density of P cell, Wh/kg
210
200 Ereq = 30kWh
battery pack mass, kg

190 Preq = 150kW

180 single-cell design
170
160
150
0 50 100 150 200
energy density of P cell, Wh/kg

Figure 5.4: Top: variation of power density as functions of energy density. Bottom: mass
for a 30kWh and 120kW battery pack as functions of energy density

case, it is more advantageous to have cell designs that are on the extreme ends of the
function, such that the power and energy requirements are satisfied entirely by different
types of cells. The lightest battery pack consists of energy cells and power cells with
maximum power density and zero energy density.
Figure 5.4 shows the differences between of multi-cell and uniform-cell designs for a
particular battery pack. The battery characteristic that favors the multi-cell pack design has
been identified. To examine how the advantages vary with the battery pack specifications,
a grid is used to compare battery pack masses of 648 kWh and 70250 kW (Figure 5.5).
The energy and power ranges encompass battery packs of a diverse group of commercially
available EVs, and the subsequent analysis can be used to examine how the multi-cell de-
sign affect the battery pack mass. On the same plot, the mass of cells needed to satisfy only
the energy requirement for different packs is shown. The energy-only battery mass pro-
vides the minimum bounds on the battery pack mass. As expected, the mass of energy cells
increases with the pack energy requirement, and is independent of the power requirement.
The mass of the multi-cell battery pack is calculated using Equation (5.9). Analyses
concentrate on the cases in which a multi-cell design outperforms a uniform-cell design.
For situations when the uniform-cell design is optimal, the battery pack design simply
follows the design methodology outlined in Chapter 4.

86
 250
225
Fisker Karma
Model S 195
200 Nissan Leaf
Chevy Volt 165
Battery pack power, kW

energy pack mass, kg

135
150
105

75
100
45

10 15 20 25 30 35 40 45 15
Battery pack energy, kWh

Figure 5.5: Mass of energy cells as a function of battery pack energy and power require-
ment. The specifications of some commercially available EV battery packs are shown as
well.

Figure 5.6 shows the variation of the optimal battery pack mass when the multi-cell
design approach yields the lighter battery pack. Note that there are two different power-
to-energy functions for which the multi-cell approach is better, and they both result in the
identical multi-cell design of maximum-power and zero-energy power cells. These results
indicate that the mass is mostly a function of the energy capacity requirement. This is
because the amount of cells required to satisfy the power requirement is much lower than
that required for energy capacity.
The uniform-cell battery packs for the two cases when the multi-cell approach produces
lighter packs are computed as well. The mass difference between multi-cell and uniform-
cell designs are plotted in Figure 5.7. The quadratic power to energy relation results in
a uniform-cell pack design that is heavier compared to those with power-law power-to-
energy function, and thus the weight-saving advantage of the multi-cell approach is greater
as well. The maximum reduction in mass is about 20% for the quadratic power-to-energy
function battery, while for the concave power-law function it is about 14% .
Figure 5.7 also indicates that the fraction of mass reduction achieved using the multi-
cell approach is proportional to the power-to-energy ratio of the battery pack in both sce-
narios. This result can be inferred from the bottom graph in Figure 5.4: As the power to

87
 250 320

280

200 240
Battery pack power, kW

multi-cell, kg
200

150 160

120

100 80

10 15 20 25 30 35 40 45 40
Battery pack energy, kWh

Figure 5.6: Mass of multi-cell pack designs as function of pack energy and power require-
ments

energy ratio increases, there is a greater compromise in the uniform-cell design to satisfy
both requirement. For the multi-cell approach, increasing power at fixed energy require-
ment results in a linear increase in battery mass, as the only change involves stacking more
power cells into the battery pack. This results in a greater mass difference between the
uniform-cell and multi-cell designs. It is also worth noting that given the different shapes
of the power to energy function, the advantage of using a multi-cell design at higher power-
to-energy ratio is increasing at a faster rate for the quadratic power to energy function, while
it is increasing at a decreasing rate for the power-law power-to-energy function.

5.5 Optimization Results

To examine how the multi-cell battery pack compares with the uniform-cell designs in
reality, a lithium-ion cell with manganese spinel cathode and MCMB anode is used. The
fixed properties of the cell are listed in Table 3.1. The multi-cell design process requires
optimization of the individual cell designs as well as optimization of the cell combinations.
Numerical optimization is performed to obtain the following:

Maximum energy density cells at different discharge rates

88
 F(x) = a-bx0.5 F(x) = ax2 +bx+c
250 250
20

7.5
7.5

6.0
6.0

9.0
9.0
200 200 16

single-cell multi-cell, %
Battery pack power, kW

4.5
4.5
12
150 150
3.0 3.0 8

100 100 4

2.0 2.0 0
10 15 20 25 30 35 40 45 10 15 20 25 30 35 40 45
Battery pack energy, kWh Battery pack energy, kWh

Figure 5.7: Difference between multi-cell and uniform-cell pack designs

Minimum mass uniform-cell battery packs

Maximum power to energy curve for the power cells by minimizing the cell mass at
different power-to-energy ratios

Minimum mass multi-cell battery packs by determining the power cell energy density
that achieves optimal pack mass.

A brief explanation of each of the optimization process is provided in the following sec-
tions.

5.5.1 Energy Cell Optimization

The energy cells in a multi-cell pack are only required to meet energy capacity of the
battery pack, therefore its design process is relatively straight forward. The theoretical
analysis in the previous section assumes that energy cells for all battery packs have identical
energy density. This is a simplification, as batteries of different energy capacity discharge
at different cycling rate. Inverse proportionality is assumed between the discharge rate and
battery pack energy capacity. Using the generic PHEV specifications listed in Table 4.2 as
the baseline, a battery pack of 12 kWh is assumed to have a galvanostatic discharge rate of
1 C. Therefore, a 6 kWh battery pack is discharged at 2 C while a 48 kWh pack is discharged
at a rate of 0.25 C. The maximum energy density is then obtained as a function of the

89
cycling rate by using the single cell design optimization process outlined in Chapter 3. The
resulting energy density as a function the cycling rate is plotted in Figure 5.8. The mass
of energy cells for different battery packs can then be obtained by using the appropriate
energy density for at specific cycling rates.
230
optimized energy density
power-law curve-fit
220

210
energy density, Wh/kg

200
E = -110.8C0.2326 + 302.4
190

180

1700.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
cycling rate, C

Figure 5.8: Energy density of energy cells as a function of galvanostatic cycling rate, with
the power-law curve fit shown as well.

5.5.2 Uniform-cell Battery Pack Optimization

The mass of the optimal uniform-cell battery pack is obtained using the battery pack de-
sign methodology outlined in Chapter 4. The design constraints are relaxed by removing
the maximum current constraints due to the variations in power requirements. While the
number of cells in series is fixed at 99, the integer requirement on the number of layers is
removed to obtain the best possible cell designs. The mass of the battery pack is plotted
in Figure 5.9. The trends shown here are similar to those in Figure 5.12. The increase in
uniform-cell battery pack mass is mostly a function of energy capacity and weakly depen-
dent on the power requirement.

5.5.3 Power Cell Optimization

In the context of multi-cell battery design, the role of the power cells is to provide maxi-
mum amount of energy while maintaining certain end-of-discharge power output. Hence

90
 250 240

210

200 180

batt. pack mass (kg)

pack power, kW 150

150 120

90

100 60

10 15 20 25 30 35 40 45 30
pack energy, kWh
Figure 5.9: Mass of uniform-cell battery pack designs

the optimal cell designs are obtained by maximizing the energy density subject to power
constraints. The power and energy discharge conditions are the same as the ones used in
Chapter 4the energy density of the cell is computed from galvanostatic discharge, while
the maximum power output is the average power of the 10-second peak current pulse dis-
charge. The resulting power versus energy curve is shown in Figure 5.10.
To examine how the power-to-energy ratio varies in the cell, the power density and
energy density are plotted against the cutoff voltage at which the galvanostatic discharge
rate ends and the peak power pulse current begins. The cutoff voltage is used as a proxy to
estimate the fraction of charge capacity used for power or energy discharge. The resulting
plot shown in the top graph of Figure 5.10 indicates that the variations of the energy and
power with the cutoff voltage follow similar trend: As the cutoff voltage is increased, both
the rate of energy density decrease and the rate of power density increase are accelerated.
The trend shown here is more indicative of the material properties, such as the shape of the
OCV curve.
A comparison between the energy and maximum power density is shown in the bottom
half of Figure 5.10. The blue dots in the top graph of Figure 5.10 are all the iteration points
of the optimizations, while the red circles represent the Pareto optimal points that have the
best power density to energy density ratio. A best-fit curve of the Pareto optimal points
is used to determine the relationship between maximum power and energy density. The

91
 peak power density, kW/kg
8 (a) 200

energy density, Wh/kg

6 150
4 100
2 50
0
2.8 3.0 3.2 3.4 3.6 3.8 4.0
10 galvanostatic discharge cutoff voltage, V
(b)
peak power density, kW/kg

all data
8 Pareto front
6 quadratic
linear
4
2
0
0 50 100 150 200
energy density, Whr/kg
Figure 5.10: Power versus energy function of the power cells; the power to energy function
is only weakly quadratic

relationship of power to energy ratio obtained from optimization is only weakly quadratic
in this case. The quadratic curve fit of the Pareto front only differs slightly from the linear
interpolation. The proximity of the Pareto front to a linear interpolation suggests that the
improvement of multi-cell battery packs is small relative to uniform-cell battery packs.

5.5.4 Multi-cell versus Uniform-cell Optimization

The multi-cell battery pack mass can be obtained by combining the optimal energy cells
and the power cells for the required energy and power capacity. Using the curve fit of power
to energy relation of power cells, the power cell energy density that minimizes battery pack
mass is first obtained. The energy and power cell masses can then be obtained based on the
optimal power cell energy density. The difference between the uniform-cell battery pack
mass and the mass of cells required to satisfy energy constraint is shown in Figure 5.11.
As expected, the difference between the uniform-cell and energy-only battery pack is a
function of the power requirement of the battery pack. Note that the maximum difference

92
between the uniform-cell and energy-only battery pack mass is only about 15% for the grid
of battery packs given, and thus the possible mass saving of the multi-cell battery pack is
limited.
250 32

28

200 24
pack power, kW

pack mass, kg
20

150 16

12

100 8

10 15 20 25 30 35 40 45 4
pack energy, kWh
Figure 5.11: Difference between uniform-cell battery pack and pure energy cell mass

The combined energy cells and power cells form the multi-cell battery packs and the
total mass as a function of pack power and energy requirement is shown in Figure 5.12. The
battery mass is again mostly a function of the energy capacity, and it is only weakly de-
pendent on the power requirement. Given the shape of the optimal power to energy density
curve of the power cell (Figure 5.10), the resulting minimum-mass multi-cell battery packs
consist of pure energy and pure power cells, similar to the results presented in Section 5.4.
The mass fraction of the power-cells is shown in Figure 5.13. The power-cell mass
fraction increases as a function of the power-to-energy ratio of the battery pack, as more
power cells are needed to satisfy the power demand at higher power-to-energy ratios. The
power cells accounts for less than 15% of the total battery mass, and therefore its influence
on the total battery mass is limited.
The difference between the multi-cell and uniform-cell battery pack masses obtained
through cell model simulation is shown in Figure 5.14, as a percentage of the uniform-
cell battery pack mass. The results show that the multi-cell battery packs are lighter than
the uniform-cell battery packs, with a maximum reduction in mass of approximately 6%.
This is much less than the reduction in mass shown in Section 5.4, where the maximum

93
 250 240

210

200 180

batt. pack mass (kg)

pack power, kW 150

150 120

90

100 60

10 15 20 25 30 35 40 45 30
pack energy, kWh
Figure 5.12: Mass of multi-cell battery pack designs

reduction in mass is close to 20%. The decreased in improvement is due to the power-
to-energy curve of the power-cells. The theoretical power-to-energy curve is assumed to
have a concave shape, while the actual curve obtained via optimization is more close to the
linear profile. This reduces the advantages of using separate cells for different requirement.
Figure 5.14 also shows that the difference between the multi-cell and uniform-cell de-
signs is mostly a function of the power requirement. For battery packs of low power re-
quirement, there is little difference between the power and energy cells, regardless of the
pack energy capacity. The difference increases steadily as the pack power requirement in-
creases. This is in contrast with the results shown in Figure 5.7, where the mass difference
is a function of the power-to-energy ratio of the battery pack.

5.5.5 Practical Design Considerations

The results shown in the previous sections indicate that the optimal multi-cell pack design
consists of power-cells with maximum power density and zero energy density. In another
words, the power-cells are expected to completely discharge within 10 seconds. Such cells
have minimal thickness and high porosity to satisfy the high mass transfer rate due to the
peak current demand. However, their energy capacity is extremely limited. Given that
each current pulse is a deep discharge for the cell, the cells will be undergoing multiple

94
 250 14

12

200 10

power cell mass, %

pack power, kW 8

150 6

100 2

10 15 20 25 30 35 40 45 0
pack energy, kWh
Figure 5.13: Fraction of pack mass that is power cell

charge and discharge cycles within a typical drive cycle in order to meet the power peaks.
Such discharge behavior is likely to cause rapid degradation of the cells, thus negating any
weight-saving advantages in the battery pack design.
Hence, it is more practical to design a multi-cell battery pack requiring power-cells to
have some minimal energy capacities for practical considerations. In the following analy-
sis, the energy density of the power-cells is limited to a minimum of 50 Wh/kg. The battery
pack energy requirement are satisfied by both energy and power cells. Figure 5.15 shows
that the resulting battery packs are still lighter than the uniform-cell packs, but that the
weight reduction is diminished.

5.6 Summary
A new multi-cell battery pack design approach is explored in this chapter. Current EV
battery packs utilize a single type of cells to meet both power and energy requirements
of vehicle operations. While multi-cell or multi-chemistry energy storage systems have
been explored by various other authors, there still lacks a systematic study of the possible
benefits of such battery pack design. The objective of this chapter is to derive simplified
equations to analyze the performance of such batteries and to compare the multi-cell with
the uniform-cell battery packs.

95
 250 6

reduction in mass using multi-cell, %

200
4
pack power, kW
3
150

100 1

10 15 20 25 30 35 40 45 0
pack energy, kWh
Figure 5.14: Difference between uniform-cell and multi-cell battery pack mass

The multi-cell design separates the battery pack into power and energy cells, where
each cell type is responsible for a particular constraint. It is shown that the multi-cell
design approach depends on the power-to-energy relation of the cells in use. When the
power-to-energy function has a concave shape, the multi-cell design is more advantageous
than the uniform-cell approach. Multi-cell battery pack designs with manganese spinel-
MCMB lithium-ion cells are demonstrated, and they are found to be up to 6% lighter than
the uniform-cell pack designs.
There are practical hurdles to be overcome before the implementation of such pack de-
signs is feasible. While the multi-cell design may lead to better battery properties in terms
of mass or volume, its cost is likely to increase as multiple type of cells are installed within
the same battery pack, thereby increasing material and manufacturing cost. In addition,
using separate power and energy cells also lead to the possibility of completely discharging
the power cells before the energy cells are discharged. This leads to the EV being unable
to meet the power demand at low SOC, or to the power-cells requiring onboard charging
via the energy-cells or via a second energy source. However, the multi-cell concept can
be advantageous in the long run, as more specialized batteries are produced for various
applications. In this chapter, the multi-cell design consists of a single type of lithium-ion
chemistry with varying electrode designs to achieve different power to energy ratios. The
multi-cell concept can be readily extended to different lithium-ion chemistries such that the

96
 250 4.5

4.0

reduction in mass using multi-cell, %

3.5
200
3.0

pack power, kW 2.5

150 2.0

1.5

1.0
100
0.5

10 15 20 25 30 35 40 45 0.0
pack energy, kWh
Figure 5.15: Difference between uniform-cell and multi-cell battery pack mass with mini-
mum 50 Wh/kg energy density

optimal battery pack design consists of different cells for different discharge rate or cycling
stability.

97
 CHAPTER 6

Conclusions and Recommendations

6.1 Concluding Summary

This dissertation is a numerical optimization study directed towards the improvement of
battery designs. The objectives are to develop numerical frameworks that can automate
the optimization of battery designs for practical applications. The optimal solutions can
provide guidelines and devise new strategies to further enhance battery performance. Three
battery design problems were discussed in this work. Chapter 3 analyzes the optimization
of a single lithium-ion cell with respect to the electrode morphological parameters, as well
as the transport properties. The objective is to maximize the gravimetric energy density,
and to quantify the effects of various parameters on energy density. Chapter 4 extends
the framework to the more practical problem of the design of a PHEV battery pack. The
objective is to minimize the battery pack mass, volume and cost. Chapter 5 then combines
the design frameworks in the previous two chapters and explores the possibility of a multi-
cell design approach in battery pack design. The summary of the contributions and findings
are briefly recounted here.
A numerical framework coupling a detailed electrochemistry model with an optimizer
is able to efficiently determine the optimal cell designs suitable for various applications. A
complete design that includes all relevant parameters is considered in Chapter 3, where the
objective is to maximize the gravimetric energy density of a single lithium-ion cell sub-
ject to constrained power. For an ideal lithium-ion cell, the electrode particles should be
as small as possible to minimize diffusion-related overpotential, while the bulk transport
parameters need to be as large as possible. However, its relative effects EW diminished for
the optimal designs, as the small particle sizes significantly reduced the characteristic dif-
fusion time. The resulting cell is not diffusion limited in the solid phase. The optimization
results also enables the quantification of the decrease in thickness and increase in porosity
as the power requirement increases. The mathematical relation between the optimal elec-

98
trode parameters can be computed via the active mass ratio, and it is found to be almost
constant for all power constraints.
A battery pack designed for PHEV operation requires consideration of separate energy
and power constraints. Additional voltage and current limits have to be taken into account
as well. A hybrid optimization scheme utilizing both gradient-free and gradient-based
optimizers is used to tackle the pack design problem. This simplifies the design process
by considering multiple design constraints simultaneously. The optimization algorithm is
applied to minimize the mass, volume and cost for a generic PHEV battery pack. The
optimized designs have battery pack properties that are superior to ad hoc designs that
follow general guidelines. They can sustain longer all-electric driving ranges when subject
to federal drive cycle discharge simulations as well. An optimized battery pack provides
realistic bounds on the possible configurations for a given battery chemistry, and gives the
best possible starting points in a battery design cycle.
Given that the discharge of a cell at any off-design condition results in rapid loss of
either power or energy density away from the Pareto front, the optimal battery pack cells
obtained in Chapter 4 are a compromise between power and energy at different discharge
rates. A new design approach that uses multiple cells to satisfy power and energy require-
ments separately is explored. It is more favorable to utilize the multi-cell approach for cells
with concave power-to-energy relations. By utilizing cells at the extreme ends of the power
to energy function, the multi-cell battery is shown to be lighter than the single-cell pack
designs.
The methods and results presented in this dissertation provide contributions towards
battery and EV system design process. Single cell optimization quantifies the optimal
electrode parameters and sets guidelines in dealing with the interactions between various
design variables. The PHEV pack design methodology provides an efficient tool for esti-
mating the best cell designs and battery layout to meet vehicle requirement. Minimization
with respect to different objective function shows the importance of an objective-oriented
process to achieve the best possible design for particular function. Finally, the multi-cell
approach demonstrates how a more complex optimization process can be used to further
advance the design of battery packs.

6.2 Realistic Cell Considerations

For all three design problems shown here, the cell model assumes an ideal electrochem-
ical cell and ignores any cell degradation mechanisms that may result from any extreme
cell designs. These mechanisms are important when considering the overall life cycle per-

99
formance. While some of the capacity fade reactions have been discussed throughout the
dissertation, they are treated more systematically in this section. Phenomena not considered
in the current cell model that may be important for other types of battery cells are discussed
here as well. These processes may be discussed independently, but they are tightly coupled
and do not occur separately from each other.
For cell designs that include aging mechanisms, the evaluation of the cell performance
with respect to one discharge cycle would be insufficient. A cell design that has maximum
initial energy density or performance only to suffer catastrophic capacity loss or power
fade within a small number of cycles would be a poor design. Additional constraints that
set lower bounds on the end-of-life energy density or modifications to the objective function
that accounts for cumulative useful energy over the entire lifetime should be included.

6.2.1 Volume Change in Solid Phase

Volume fraction change in the graphite anode and manganese spinel cathode is small and
is neglected in the cell model, however it is not the case for some of the other materials,
such as silicon in anodes. Volume change up to factors of 34 can be expected during
intercalation. The large volume increase means that the ion transport equation in the solid
phase cannot be simplified to the diffusion equation (2.3).
Volume change in the electrode affects the cell performance in several ways. First,
volume change causes change in porosity. Reduction in porosity during intercalation in-
creases cell impedance, leading to loss of power. The effects of porosity change is further
enhanced at high cycling rates. Drastic volume change can also cause contact loss of active
material particles, resulting in capacity fade due to loss of cyclable lithium and available
intercalation sites [169].
Accounting for the effects of volume expansion in the cell model requires coupling of
a physics-based degradation model to determine how design variables, such as particle size
and volume fraction, contribute to contact loss. Factors such as particle packing density
and size influence the effect of volume change [170] and need to be considered as well.
Significant change in porosity requires modeling the porosity as a dynamic parameter that
changes with the current density and direction.

6.2.2 Modeling of Additives

The solid matrix in the electrode is made up of active material, conduction additives, and
polymer binders that hold everything in place. Therefore, the actual properties of the solid
material are functions of the solid matrix composition, and not constant in a homogeneous

100
electrode. The solid conductivity, for instance, depends on the interaction between the car-
bon additives and the active material, which is a function of both active material particle
size, and carbon additives to active material mass ratio [143]. Larger active material par-
ticles exert greater attraction forces on the carbon additives, while a higher mass fraction
increases the probability of additives attaching the active material.
Instead of treating conductivity as an independent variable whose effect is only mod-
ified by the porosity, conductivity can be modeled as a function the solid phase composi-
tion, where the mass fraction of active material, polymer binder, and carbon additives are
the independent variables. The cell optimization problem then becomes a balance between
maximizing the amount of energy-contributing energy material, the need to maintain suf-
ficient electronic conductivity using additives and to reduce concentration polarization by
ensuring adequate porosity in the electrode.

6.2.3 Solid-Electrolyte Interface Formation

One of the primary causes of lithium loss in the cell is the decomposition of electrolyte at
the negative electrode/electrolyte interface to form a passivated SEI layer during the initial
cycles [169]. This is due to the anode operating at voltages that are outside the electrochem-
ical stability window of electrolyte components. SEI formation is an irreversible process
that consumes cyclable lithium. However the SEI layer is also beneficial in preventing
further decomposition of solvent and at the same time allows lithium ions to pass through.
Ideally, a good cell design should encourage an SEI layer that is both chemically and
mechanically stable. The latter is especially important for electrodes experiencing large
volume change during cycling. To accurately model the formation, growth and subsequent
stabilization of SEI layer, physics-based interaction between electrode active material and
electrolyte and influence of SEI thickness on electrolyte diffusivity should be included in
the cell model [171]. Electrode design also has to take lithium loss due to SEI layer into
account when determining the optimal mass ratio.

6.2.4 Intercalation-Induced Stress

Factors such as volume change and SEI formation all affect the intercalation-induced stress
experienced by the solid particles. The amount of stress depends on the shape and size of
particles, and it is interrelated through the local surface flux, and ion concentration gra-
dient [56]. Multi-scale modeling has shown that lithium intercalation for large particles
is diffusion limited and that the resulting diffusion-stress coupling causes mismatching

101
strains and stresses [116]. High stress is also experienced in nanoscale particles due to
surface stress effect [172].
Stress and particle fracture inclusion in the cell design process requires modeling of
local material and transport properties, such as local particle geometry and flux density.
Simulation of the entire cell at the particle level requires three-dimensional modeling of all
electrode particles to replace the ButlerVolmer equation.

6.3 Recommendations for Future Work

The main focus of this dissertation was to maximize the gravimetric energy density of
lithium-ion cells, and to improve battery pack designs using numerical optimization. The
homogeneous electrode treatment and the lack of cell degradation considerations high-
lighted in the previous section limit the scope of the dissertation. Several aspects can be
undertaken to expand on the existing framework to improve the practical impact of the
optimization results.
First, detailed micro-structure modeling can be incorporated into the cell model to
provide high-fidelity cell design optimization. It has been shown that there is discrep-
ancy between the effective transport coefficients and the estimates using Bruggemans re-
lation [173], and porosity alone cannot fully account for the variations in the effective
transport coefficients in a porous material [174]. A second parameter, such as tortuosity,
may be required to provide more accurate predictions of the material properties.
To account for effects of microscale geometries on cell performance, a multi-scale cell
model is needed. However, the transient microscale simulation of the entire cell can be
prohibitively expensive for optimization purposes. A surrogate multi-scale approach can
be applied to provide microscale information to the cell-level model [174]. A response
surface is generated using a large number of microscale simulations in a representative
elementary volume. In this case different particle geometries and orientations have to be
considered and statistical analysis can be used to quantify their effects on the local mass
transfer rate within the electrode. Subsequent computations at the cell level will be based
on the metamodel that is much faster to execute.
A more complete way to perform multi-cell battery pack optimization than the one
presented in Chapter 5 is to couple the electrode design parameters, battery layout, and
control variables that determine the optimal discharge profiles for different cells. Sequential
optimization of the battery pack design followed by an optimal control strategy often leads
to non-optimal overall systems as the coupling between the battery pack design and control
parameters is not captured. Therefore, simultaneous optimization of the various disciplines

102
in a multidisciplinary optimization framework is required to obtain the true optimal system
design. Such formulation would also remove the necessity of having fixed energy and
power capacities as the constraints and simply focus on maximizing the electric driving
ranges of the battery pack. To prevent the optimal battery pack from becoming tailored
for a particular driving cycle, an aggregate objective function could be used to determine
a multi-point optimal design, such that the optimal solution is one that performs the best
across the different driving cycles.

103
 BIBLIOGRAPHY

[1] IEA. Key World Energy Statistics 2010. OECD/IEA, 2010.

[2] IATA. IATA economic briefing: Airline fuel and labour cost share,
February 2010. URL: http://www.iata.org/whatwedo/Documents/
economics/Airline_Labour_Cost_Share_Feb2010.pdf.

[3] Andrew B. Lambe and Joaquim R. R. A. Martins. Extensions to the design struc-
ture matrix for the description of multidisciplinary design, analysis, and optimiza-
tion processes. Structural and Multidisciplinary Optimization, 46:273284, August
2012. doi:10.1007/s00158-012-0763-y.

[4] Michigan Public Service Commission. Comparisons of average rates (in cents per
kwh) for MPSC-regulated electric utilities in michigan, October 2013. URL: www.
dleg.state.mi.us/mpsc/electric/download/rates1.pdf.

[5] Chris Brady. The Boeing 737 Technical Guide. Tech Pilot Services, 2012.

[6] The CAFE Foundation. The 2011 Green Flight Challenge sponsored by Google,
2011. URL: cafefoundation.org/v2/gfc_2011_results.html.

[7] Tine Tomazic, Vid Plevnik, Gregor Veble, Jure Tomazic, Franc Popit, Saso Kolar,
Radivoj Kikelj, Jacob W Langelaan, and Kirk Miles. Pipistrel taurus G4: on cre-
ation and evolution of the winning aeroplane of NASA Green Flight Challenge 2011.
Strojniski vestnik-Journal of Mechanical Engineering, 57(12):869878, 2011.

[8] Marzieh Barghamadi, Ajay Kapoor, and Cuie Wen. A review on Li-S batteries as a
high efficiency rechargeable lithium battery. Journal of The Electrochemical Society,
160(8):A1256A1263, 2013.

[9] Myounggu Park, Xiangchun Zhang, Myoungdo Chung, Gregory B Less, and
Ann Marie Sastry. A review of conduction phenomena in Li-ion batteries. Jour-
nal of Power Sources, 195(24):79047929, 2010.

[10] EIA. Annual energy review. Energy Information Administration, US Department of

Energy: Washington, DC, 2011. URL: www.eia.doe.gov/emeu/aer.

[11] EIA. Inventory of U.S. greenhouse gas emissions and sinks: 1990 2011. U.S.
Environmental Protection Agency, Washington, DC, 2013. URL: www.epa.gov/
climatechange/ghgemissions/usinventoryreport.html.

104
[12] IATA. Vision 2050: Shaping aviations future.
[13] IATA. Financial forecast: Losses expected to continue into 2010, September
2009. URL: www.iata.org/publications/economics/Documents/
Industry_Outlook_Sep09.pdf.
[14] IATA. Financial forecast - prospects for 2014 look good (in parts), Septem-
ber 2013. URL: www.iata.org/whatwedo/Documents/economics/
iata-economic-briefing-financial-forecast-september-2013.
pdf.
[15] Joyce E Penner. Aviation and the Global Atmosphere: Special Report of the IPCC
Working Groups I and III in Collaboration with the Scientific Assessment Panel to
the Montreal Protocol on Substances that Deplete the Ozone Layer. Cambridge
University Press, 1999.
[16] EIA. Monthly energy review september 2013. U.S. Energy Information Administra-
tion, 2013. URL: http://www.eia.gov/totalenergy/data/monthly/
pdf/sec7_5.pdf.
[17] Don Anair and Amine Mahmassani. State of charge: Electric vehicles global warm-
ing emissions and fuel-cost savings across the united states. Technical report, June
2012.
[18] Michel Armand and J-M Tarascon. Building better batteries. Nature,
451(7179):652657, 2008.
[19] James Larminie and John Lowry. Electric vehicle technology explained. Wiley
Online Library, 2012.
[20] John M German. Hybrid powered vehicles. SAE International, 2003.
[21] Andreas Dinger, Ripley Martin, Xavier Mosquet, Maximilian Rabl, Dimitrios Ri-
zoulis, Massimo Russo, and Georg Sticher. Batteries for electric cars: Challenges,
opportunities, and the outlook to 2020. Boston Consulting Group, 2010.
[22] Alphonsus L Drum. Electrically-propelled aircraft, 1924.
[23] James L Felder, Hyun Dae Kim, and Gerald Brown. Turboelectric distributed
propulsion engine cycle analysis for hybrid-wing-body aircraft. In 47th AIAA
Aerospace Sciences Meeting, number 2009-1132. AIAA, 2009.
[24] A Gibson, David Hall, Mark Waters, P Masson, B Schiltgen, and T Foster. The
potential and challenge of turboelectric propulsion for subsonic transport aircraft.
AIAA, 276:2010, 2010.
[25] Amir S Gohardani, Georgios Doulgeris, and Riti Singh. Challenges of future air-
craft propulsion: a review of distributed propulsion technology and its potential ap-
plication for the all electric commercial aircraft. Progress in Aerospace Sciences,
47(5):369391, 2011.

105
[26] Rhea P. Liem and Joaquim R. R. A. Martins. Multifidelity aerostructural optimiza-
tion of a long-range aircraft configuration with uncertain missions and operating
conditions. In Proceedings of the 54th AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics, and Materials Conference, Boston, MA, April 2013.
[27] Cesar A Luongo, Philippe J Masson, Taewoo Nam, Dimitri Mavris, Hyun D Kim,
Gerald V Brown, Mark Waters, and David Hall. Next generation more-electric air-
craft: a potential application for hts superconductors. Applied Superconductivity,
IEEE Transactions on, 19(3):10551068, 2009.
[28] JA Rosero, JA Ortega, E Aldabas, and LARL Romeral. Moving towards a more
electric aircraft. Aerospace and Electronic Systems Magazine, IEEE, 22(3):39,
2007.
[29] Marty K Bradley and Christopher K Droney. Subsonic ultra green aircraft research:
Phase I final report. 2011.
[30] Hyun Dae Kim. Distributed propulsion vehicles. In 27th International Congress of
the Aeronautical Sciences, 2010.
[31] Siemens. Worlds first serial hybrid electric aircraft to fly at Le Bour-
get, 2011. URL: http://www.siemens.com/press/pool/de/
pressemitteilungen/2011/corporate_communication/
AXX20110666e.pdf.
[32] Jason Paur. Hybrid power comes to aviation, July 2009. URL: www.wired.com/
autopia/2009/07/hybrid-aviation.
[33] Yuneec. Yuneec E430 electric aircraft specifications, 2013. URL: yuneeccouk.
site.securepod.com/Aircraft_specification.html.
[34] Yuri Gawdiak, Bruce J Holmes, Bruce K Sawhill, James Herriot, David Ballard,
Jeremy Eckhause, Dou Long, Robert Hemm, Jeremiah Creedon, Charles Murphy,
Terry Thompson, Fred Wieland, Mike Marcolini, Mark Moore, and Monica Alcabin.
Air transportation strategic trade space modeling and assessment through analysis
of on-demand air mobility with electric aircraft. In 12th AIAA Aviation Technology,
Integration, and Operations (ATIO) Conference and 14th AIAA/ISSM. AIAA.
[35] Mark D Moore. Concept of operations for highly autonomous electric zip aviation.
12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and
14th AIAA/ISSM, (AIAA 2012-5472), 2012.
[36] Michael D Patterson, Brian J German, and Mark D Moore. Performance analysis
and design of on-demand electric aircraft concepts. In 12th AIAA Aviation Technol-
ogy, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSM, number
AIAA 2012-5474. AIAA, 2012.
[37] Darryll J Pines and Felipe Bohorquez. Challenges facing future micro-air-vehicle
development. Journal of aircraft, 43(2):290305, 2006.

106
[38] Ohad Gur and Aviv Rosen. Optimizing electric propulsion systems for unmanned
aerial vehicles. Journal of aircraft, 46(4):13401353, 2009.

[39] Frederick G Harmon, Andrew A Frank, and Jean-Jacques Chattot. Conceptual de-
sign and simulation of a small hybrid-electric unmanned aerial vehicle. Journal of
aircraft, 43(5):14901498, 2006.

[40] Mirac Aksugur and Gokhan Inalhan. Design methodology of a hybrid propulsion
driven electric powered miniature tailsitter unmanned aerial vehicle. Journal of In-
telligent and Robotic Systems, 57(1-4):505529, 2010.

[41] JY Hung and LF Gonzalez. On parallel hybrid-electric propulsion system for un-
manned aerial vehicles. Progress in Aerospace Sciences, 51:117, 2012.

[42] Joaquim R. R. A. Martins and Andrew B. Lambe. Multidisciplinary design opti-

mization: A survey of architectures. AIAA Journal, 51(9):20492075, Sep 2013.
doi:10.2514/1.J051895.

[43] Ping Liu, Elena Sherman, and Alan Jacobsen. Design and fabrication of multifunc-
tional structural batteries. Journal of Power Sources, 189(1):646650, 2009.

[44] James P Thomas and Muhammad A Qidwai. The design and application of multi-
functional structure-battery materials systems. JOM Journal of the Minerals, Metals
and Materials Society, 57(3):1824, 2005.

[45] Ronald F Gibson. A review of recent research on mechanics of multifunctional

composite materials and structures. Composite structures, 92(12):27932810, 2010.

[46] K David Huang and Sheng-Chung Tzeng. Development of a hybrid pneumatic-

power vehicle. Applied Energy, 80(1):4759, 2005.

[47] Dong Zhang, Branko N Popov, and Ralph E White. Modeling lithium intercalation
of a single spinel particle under potentiodynamic control. Journal of The Electro-
chemical Society, 147(3):831838, 2000.

[48] Gang Ning and Branko N Popov. Cycle life modeling of lithium-ion batteries. Jour-
nal of The Electrochemical Society, 151(10):A1584A1591, 2004.

[49] Bor Yann Liaw, Ganesan Nagasubramanian, Rudolph G Jungst, and Daniel H
Doughty. Modeling of lithium ion cells - a simple equivalent-circuit model approach.
Solid state ionics, 175(1):835839, 2004.

[50] Matthieu Dubarry, Nicolas Vuillaume, and Bor Yann Liaw. From single cell model to
battery pack simulation for Li-ion batteries. Journal of Power Sources, 186(2):500
507, 2009.

[51] BS Bhangu, P Bentley, DA Stone, and CM Bingham. Nonlinear observers for pre-
dicting state-of-charge and state-of-health of lead-acid batteries for hybrid-electric
vehicles. Vehicular Technology, IEEE Transactions on, 54(3):783794, 2005.

107
[52] Marc Doyle, Thomas F Fuller, and John Newman. Modeling of galvanostatic charge
and discharge of the lithium/polymer/insertion cell. Journal of the Electrochemical
Society, 140(6):15261533, 1993.

[53] John Newman and Karen E Thomas-Alyea. Electrochemical systems. Wiley-

Interscience, 2004.

[54] Shriram Santhanagopalan, Qingzhi Guo, Premanand Ramadass, and Ralph E White.
Review of models for predicting the cycling performance of lithium ion batteries.
Journal of Power Sources, 156(2):620628, 2006.

[55] CY Wang, WB Gu, and BY Liaw. Micro-macroscopic coupled modeling of batter-

ies and fuel cells i. model development. Journal of The Electrochemical Society,
145(10):34073417, 1998.

[56] Xiangchun Zhang, Wei Shyy, and Ann Marie Sastry. Numerical simulation of
intercalation-induced stress in Li-ion battery electrode particles. Journal of the Elec-
trochemical Society, 154(10):A910A916, 2007.

[57] Amit Gupta, Jeong Hun Seo, Xiangchun Zhang, Wenbo Du, Ann Marie Sastry, and
Wei Shyy. Effective transport properties of LiMn2 O4 electrode via particle-scale
modeling. Journal of The Electrochemical Society, 158(5):A487A497, 2011.

[58] Pankaj Arora, Ralph E White, and Marc Doyle. Capacity fade mechanisms and
side reactions in lithium-ion batteries. Journal of the Electrochemical Society,
145(10):36473667, 1998.

[59] P Ramadass, Bala Haran, Parthasarathy M Gomadam, Ralph White, and Branko N
Popov. Development of first principles capacity fade model for Li-ion cells. Journal
of the Electrochemical Society, 151(2):A196A203, 2004.

[60] John Christensen and John Newman. Stress generation and fracture in lithium inser-
tion materials. Journal of Solid State Electrochemistry, 10(5):293319, 2006.

[61] Jonghyun Park, Jeong Hun Seo, Gregory Plett, Wei Lu, and Ann Marie Sastry.
Numerical simulation of the effect of the dissolution of LiMn2 O4 particles on Li-
ion battery performance. Electrochemical and Solid-State Letters, 14(2):A14A18,
2011.

[62] Bor Yann Liaw, Rudolph G Jungst, Ganesan Nagasubramanian, Herbert L Case, and
Daniel H Doughty. Modeling capacity fade in lithium-ion cells. Journal of Power
Sources, 140(1):157161, 2005.

[63] Steve Siler. Pump it up: We refuel a hydrogen fuel-cell vehicle,

November 2008. URL: www.caranddriver.com/features/
pump-it-up-we-refuel-a-hydrogen-fuel-cell-vehicle.

[64] William M Haynes, David R Lide, and Thomas J Bruno. CRC Handbook of Chem-
istry and Physics 2012-2013. CRC press, 2012.

108
[65] M Stanley Whittingham. Electrical energy storage and intercalation chemistry. Sci-
ence, 192(4244):11261127, 1976.

[66] DW Murphy, FJ Di Salvo, JN Carides, and JV Waszczak. Topochemical reactions

of rutile related structures with lithium. Materials Research Bulletin, 13(12):1395
1402, 1978.

[67] K Mizushima, PC Jones, PJ Wiseman, and JB Goodenough. Lix CoO2 (0 x-1):

A new cathode material for batteries of high energy density. Materials Research
Bulletin, 15(6):783789, 1980.

[68] MM Thackeray, WIF David, PG Bruce, and JB Goodenough. Lithium insertion into
manganese spinels. Materials Research Bulletin, 18(4):461472, 1983.

[69] David Linden and Thomas B Reddy. Handbook of batteries. Mcgraw-Hill, 2002.

[70] J Wang, SY Chew, ZW Zhao, S Ashraf, David Wexler, J Chen, SH Ng, SL Chou, and
HK Liu. Sulfurmesoporous carbon composites in conjunction with a novel ionic
liquid electrolyte for lithium rechargeable batteries. Carbon, 46(2):229235, 2008.

[71] Byoung Ho Jeon, Jin Hee Yeon, and In Jae Chung. Preparation and electrical proper-
ties of lithiumsulfur-composite polymer batteries. Journal of materials processing
technology, 143:9397, 2003.

[72] Yuriy V Mikhaylik and James R Akridge. Polysulfide shuttle study in the li/s battery
system. Journal of the Electrochemical Society, 151(11):A1969A1976, 2004.

[73] Qianfan WeiaSeh, Zhi andZhang, Weiyang Li, Guangyuan Zheng, Hongbin Yao,
and Yi Cui. Stable cycling of lithium sulfide cathodes through strong affinity with a
bifunctional binder. Chemical Science, 4(9):36733677, 2013.

[74] Uday Kasavajjula, Chunsheng Wang, and A John Appleby. Nano-and bulk-silicon-
based insertion anodes for lithium-ion secondary cells. Journal of Power Sources,
163(2):10031039, 2007.

[75] Doron Aurbach. Review of selected electrodesolution interactions which determine

the performance of li and li ion batteries. Journal of Power Sources, 89(2):206218,
2000.

[76] Hui Wu, Guihua Yu, Lijia Pan, Nian Liu, Matthew T McDowell, Zhenan Bao, and
Yi Cui. Stable Li-ion battery anodes by in-situ polymerization of conducting hydro-
gel to conformally coat silicon nanoparticles. Nature communications, 4, 2013.

[77] KM Abraham and Z Jiang. A polymer electrolyte-based rechargeable

lithium/oxygen battery. Journal of The Electrochemical Society, 143(1):15, 1996.

[78] G Girishkumar, B McCloskey, AC Luntz, S Swanson, and W Wilcke. Lithium-

air battery: Promise and challenges. The Journal of Physical Chemistry Letters,
1(14):21932203, 2010.

109
[79] P Poizot, S Laruelle, S Grugeon, L Dupont, and JM Tarascon. Nano-sized transition-
metal oxides as negative-electrode materials for lithium-ion batteries. Nature,
407(6803):496499, 2000.

[80] Dave Hurst and John Gartner. Electric vehicle market forecasts - global forecasts for
light duty hybrid, plug-in hybrid, and battery electric vehicles: 2013-2020. Technical
report.

[81] Jo Dewulf, Geert Van der Vorst, Kim Denturck, Herman Van Langenhove, Wouter
Ghyoot, Jan Tytgat, and Kurt Vandeputte. Recycling rechargeable lithium ion bat-
teries: Critical analysis of natural resource savings. Resources, Conservation and
Recycling, 54(4):229234, 2010.

[82] Jinqiu Xu, HR Thomas, Rob W Francis, Ken R Lum, Jingwei Wang, and Bo Liang.
A review of processes and technologies for the recycling of lithium-ion secondary
batteries. Journal of Power Sources, 177(2):512527, 2008.

[83] Aswin Kumar. The lithium battery recycling challenge. Waste Manage-
ment World, 12, August 2011. URL: www.waste-management-world.
com/articles/print/volume-12/issue-4/features/
the-lithium-battery-recycling-challenge.html.

[84] Thomas G. Goonan. Lithium use in batteries. Technical report.

[85] K Zaghib, J Shim, A Guerfi, P Charest, and KA Striebel. Effect of carbon source as
additives in lifepo4 as positive electrode for lithium-ion batteries. Electrochemical
and Solid-State Letters, 8(4):A207A210, 2005.

[86] Seung-Taek Myung, Myung Hun Cho, Hyeon Taik Hong, Tae Hyuk Kang, and
Chi-Su Kim. Electrochemical evaluation of mixed oxide electrode for li-ion sec-
ondary batteries: Li1.1 Mn1.9 O4 and LiNi0.8 Co0.15 Al0.05 O2 . Journal of power sources,
146(1):222225, 2005.

[87] Thomas F Fuller, Marc Doyle, and John Newman. Simulation and optimization of
the dual lithium ion insertion cell. Journal of the Electrochemical Society, 141(1):1
10, 1994.

[88] Karen Thomas, John Newman, and Robert Darling. Mathematical modeling of
lithium batteries. Advances in lithium-ion batteries, pages 345392, 2002.

[89] Von DAG Bruggeman. Berechnung verschiedener physikalischer konstanten von

heterogenen substanzen. i. dielektrizitatskonstanten und leitfahigkeiten der mis-
chkorper aus isotropen substanzen. Annalen der Physik, 416(7):636664, 1935.

[90] Richard Pollard and John Newman. Mathematical modeling of the lithium-
aluminum, iron sulfide battery i. galvanostatic discharge behavior. Journal of The
Electrochemical Society, 128(3):491502, 1981.

[91] Francis B Hildebrand. Advanced Calculus for Applications. Prentice-Hall, 1976.

110
 [92] Brian Bradie. A Friendly Introduction to Numerical Analysis: With C and MATLAB
Materials on Website. Person Prentice Hall, 2006.

[93] Hongwen He, Rui Xiong, and Jinxin Fan. Evaluation of lithium-ion battery equiv-
alent circuit models for state of charge estimation by an experimental approach.
Energies, 4(4):582598, 2011.

[94] Jorge Nocedal and Stephen J. Wright. Numerical Optimization. Springer, 2nd edi-
tion, 2006.

[95] Kathleen R Fowler, Jill P Reese, Chris E Kees, JE Dennis Jr, C Tim Kelley, Cass T
Miller, Charles Audet, Andrew J Booker, Gilles Couture, Robert W Darwin, et al.
Comparison of derivative-free optimization methods for groundwater supply and
hydraulic capture community problems. Advances in Water Resources, 31(5):743
757, 2008.

[96] Jorge J More and Stefan M Wild. Benchmarking derivative-free optimization algo-
rithms. SIAM Journal on Optimization, 20(1):172191, 2009.

[97] Peter W. Jansen and Ruben E. Perez. Constrained structural design optimization via
a parallel augmented Lagrangian particle swarm optimization approach. Computers
& Structures, 89:13521366, 2011. doi:10.1016/j.compstruc.2011.03.
011.

[98] Ruben E. Perez, Peter W. Jansen, and Joaquim R. R. A. Martins. pyOpt: a Python-
based object-oriented framework for nonlinear constrained optimization. Struc-
tural and Multidisciplinary Optimization, 45(1):101118, January 2012. doi:
10.1007/s00158-011-0666-3.

[99] Philip E Gill and Walter Murray. The computation of Lagrange-multiplier estimates
for constrained minimization. Mathematical Programming, 17(1):3260, 1979.

[100] Kai Sedlaczek and Peter Eberhard. Using augmented lagrangian particle swarm
optimization for constrained problems in engineering - using augmented Lagrangian
particle swarm optimization for constrained problems in engineering. Structural and
Multidisciplinary Optimization, 32(4):277286, 2006.

[101] P.E. Gill, W. Murray, and M.A. Saunders. SNOPT: An SQP algorithm for large-scale
constraint optimization. SIAM Journal of Optimization, 12(4):9791006, 2002.

[102] Joaquim R. R. A. Martins and John T. Hwang. Review and unification of methods
for computing derivatives of multidisciplinary computational models. AIAA Journal,
51(11):25822599, 2013. doi:10.2514/1.J052184.

[103] Joaquim R. R. A. Martins, Peter Sturdza, and Juan J. Alonso. The complex-step
derivative approximation. ACM Transactions on Mathematical Software, 29(3):245
262, 2003. doi:10.1145/838250.838251.

111
[104] Wolfram Stadler. Multicriteria Optimization in Engineering and in the Sciences,
volume 37. Springer, 1988.

[105] Wei Shyy, Young-Chang Cho, Wenbo Du, Amit Gupta, Chien-Chou Tseng, and
Ann Marie Sastry. Surrogate-based modeling and dimension reduction techniques
for multi-scale mechanics problems. Acta Mechanica Sinica, 27(6):845865, 2011.

[106] Nestor V Queipo, Raphael T Haftka, Wei Shyy, Tushar Goel, Rajkumar
Vaidyanathan, and P Kevin Tucker. Surrogate-based analysis and optimization.
Progress in Aerospace Sciences, 41(1):128, 2005.

[107] Tushar Goel, Rajkumar Vaidyanathan, Raphael T Haftka, Wei Shyy, Nestor V
Queipo, and Kevin Tucker. Response surface approximation of pareto optimal front
in multi-objective optimization. Computer Methods in Applied Mechanics and En-
gineering, 196(4):879893, 2007.

[108] W Choi and A Manthiram. Comparison of metal ion dissolutions from lithium ion
battery cathodes. Journal of The Electrochemical Society, 153(9):A1760A1764,
2006.

[109] XL Li, K Du, JM Huang, FY Kang, and WC Shen. Effect of carbon nanotubes on the
anode performance of natural graphite for lithium ion batteries. Journal of Physics
and Chemistry of Solids, 71(4):457459, 2010.

[110] Marc Doyle, Thomas F Fuller, and John Newman. The importance of the lithium ion
transference number in lithium/polymer cells. Electrochimica Acta, 39(13):2073
2081, 1994.

[111] John Newman. Optimization of porosity and thickness of a battery electrode by

means of a reaction-zone model. Journal of the Electrochemical Society, 142(1):97
101, 1995.

[112] Venkat Srinivasan and John Newman. Design and optimization of a natural
graphite/iron phosphate lithium-ion cell. Journal of the Electrochemical Society,
151(10):A1530A1538, 2004.

[113] Sarah G Stewart, Venkat Srinivasan, and John Newman. Modeling the perfor-
mance of lithium-ion batteries and capacitors during hybrid-electric-vehicle oper-
ation. Journal of The Electrochemical Society, 155(9):A664A671, 2008.

[114] Wenbo Du, Amit Gupta, Xiangchun Zhang, Ann Marie Sastry, and Wei Shyy. Effect
of cycling rate, particle size and transport properties on lithium-ion cathode perfor-
mance. International Journal of Heat and Mass Transfer, 53(17):35523561, 2010.

[115] Venkatasailanathan Ramadesigan, Ravi N Methekar, Folarin Latinwo, Richard D

Braatz, and Venkat R Subramanian. Optimal porosity distribution for minimized
ohmic drop across a porous electrode. Journal of The Electrochemical Society,
157(12):A1328A1334, 2010.

112
[116] Stephanie Golmon, Kurt Maute, and Martin L Dunn. Multiscale design optimization
of lithium ion batteries using adjoint sensitivity analysis. International Journal for
Numerical Methods in Engineering, 92(5):475494, 2012.

[117] De Sumitava, Paul WC Northrop, Venkatasailanathan Ramadesigan, and Venkat R

Subramanian. Model-based simultaneous optimization of multiple design parame-
ters for lithium-ion batteries for maximization of energy density. Journal of Power
Sources, 227:161170, 2013.

[118] Joel C Forman, Saeid Bashash, Jeffrey L Stein, and Hosam K Fathy. Reduction
of an electrochemistry-based Li-ion battery model via quasi-linearization and Pade
approximation. Journal of The Electrochemical Society, 158(2):A93A101, 2011.

[119] Venkatasailanathan Ramadesigan, Vijayasekaran Boovaragavan, J Carl Pirkle,

and Venkat R Subramanian. Efficient reformulation of solid-phase diffusion in
physics-based lithium-ion battery models. Journal of The Electrochemical Society,
157(7):A854A860, 2010.

[120] Long Cai and Ralph E White. Reduction of model order based on proper orthogonal
decomposition for lithium-ion battery simulations. Journal of The Electrochemical
Society, 156(3):A154A161, 2009.

[121] Paul WC Northrop, Venkatasailanathan Ramadesigan, Sumitava De, and Venkat R

Subramanian. Coordinate transformation, orthogonal collocation, model reformu-
lation and simulation of electrochemical-thermal behavior of lithium-ion battery
stacks. Journal of The Electrochemical Society, 158(12):A1461A1477, 2011.

[122] Wenbo Du, Nansi Xue, Amit Gupta, Ann M Sastry, Joaquim RRA Martins, and Wei
Shyy. Optimization of limn2o4 electrode properties in a gradient-and surrogate-
based framework. Acta Mechanica Sinica, 29(3):335347, 2013.

[123] Philip E. Gill, Walter Murray, and Michael A. Saunders. SNOPT: An SQP algorithm
for large-scale constrained optimization. SIAM Review, 47(1):99131, 2005. doi:
10.1137/S0036144504446096.

[124] Elizabeth D. Dolan and Jorge J. More. Benchmarking optimization soft-

ware with performance profiles. Mathematical Programming, 91:201213,
2002. 10.1007/s101070100263. URL: http://dx.doi.org/10.1007/
s101070100263.

[125] Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and TAMT Meyarivan. A fast and
elitist multiobjective genetic algorithm: Nsga-ii. Evolutionary Computation, IEEE
Transactions on, 6(2):182197, 2002.

[126] Nansi Xue, Wenbo Du, Amit Gupta, Wei Shyy, Ann Marie Sastry, and Joaquim R.
R. A. Martins. Optimization of a single lithium-ion battery cell with a gradient-based
algorithm. Journal of the Electrochemical Society, 160(8):A1071A1078, 2013.

113
[127] Marc Doyle, John Newman, Antoni S Gozdz, Caroline N Schmutz, and Jean-Marie
Tarascon. Comparison of modeling predictions with experimental data from plastic
lithium ion cells. Journal of the Electrochemical Society, 143(6):18901903, 1996.
[128] MD Levi and D Aurbach. Diffusion coefficients of lithium ions during intercala-
tion into graphite derived from the simultaneous measurements and modeling of
electrochemical impedance and potentiostatic intermittent titration characteristics of
thin graphite electrodes. The Journal of Physical Chemistry B, 101(23):46414647,
1997.
[129] JR Dygas, M Kopec, F Krok, D Lisovytskiy, and J Pielaszek. Conductivity and
dielectric relaxation phenomena in lithium manganese spinel. Solid State Ionics,
176(25):21532161, 2005.
[130] DV Ragone. Review of battery systems for electrically powered vehicles. Society of
Automotive Engineers, 1968.
[131] Peter Bauerlein, Rudolf Herr, Matthias Kloss, Jorg Kumpers, Matthias Maul, and
Eberhard Meissner. Advanced lithium ion cells with lithium manganese spinel. Jour-
nal of power sources, 81:585588, 1999.
[132] T Horiba, K Hironaka, T Matsumura, T Kai, M Koseki, and Y Muranaka. Man-
ganese type lithium ion battery for pure and hybrid electric vehicles. Journal of
power sources, 97:719721, 2001.
[133] Wilmont F Howard and Robert M Spotnitz. Theoretical evaluation of high-energy
lithium metal phosphate cathode materials in Li-ion batteries. Journal of Power
Sources, 165(2):887891, 2007.
[134] Charles Monroe and John Newman. Dendrite growth in lithium/polymer systems a
propagation model for liquid electrolytes under galvanostatic conditions. Journal of
the Electrochemical Society, 150(10):A1377A1384, 2003.
[135] Tetsuya Kawamura, Masami Makidera, Shigeto Okada, Kazumichi Koga, Norio
Miura, and Jun-ichi Yamaki. Effect of nano-size LiCoO2 cathode powders on Li-ion
cells. Journal of power sources, 146(1):2732, 2005.
[136] Thierry Drezen, Nam-Hee Kwon, Paul Bowen, Ivo Teerlinck, Motoshi Isono, and
Ivan Exnar. Effect of particle size on limnpo4 cathodes. Journal of Power Sources,
174(2):949953, 2007.
[137] Mohamed Aklalouch, Rosa M Rojas, Jose Maria Rojo, Ismael Saadoune, and
Jose Manuel Amarilla. The role of particle size on the electrochemical properties
at 25 and at 55 c of the licr0. 2ni0. 4mn1. 4o4 spinel as 5v-cathode materials for
lithium-ion batteries. Electrochimica acta, 54(28):75427550, 2009.
[138] Jane Yao, GX Wang, Jung-ho Ahn, HK Liu, and SX Dou. Electrochemical studies
of graphitized mesocarbon microbeads as an anode in lithium-ion cells. Journal of
power sources, 114(2):292297, 2003.

114
[139] Robert Darling and John Newman. Modeling a porous intercalation electrode
with two characteristic particle sizes. Journal of The Electrochemical Society,
144(12):42014208, 1997.

[140] Gowri S Nagarajan, JW Van Zee, and RM Spotnitz. A mathematical model for
intercalation electrode behavior - i. effect of particle-size distribution on discharge
capacity. Journal of The Electrochemical Society, 145(3):771779, 1998.

[141] JM Tarascon and D Guyomard. The Li1+x Mn2 O4 /C rocking-chair system: a review.
Electrochimica Acta, 38(9):12211231, 1993.

[142] Y-H Chen, C-W Wang, Gao Liu, X-Y Song, VS Battaglia, and Ann Marie Sastry.
Selection of conductive additives in li-ion battery cathodes a numerical study. Jour-
nal of The Electrochemical Society, 154(10):A978A986, 2007.

[143] M Zhu, J Park, and AM Sastry. Particle interaction and aggregation in cathode ma-
terial of li-ion batteries: a numerical study. Journal of The Electrochemical Society,
158(10):A1155A1159, 2011.

[144] Dominique Guyomard and Jean-Marie Tarascon. Rocking-chair or lithium-ion

rechargeable lithium batteries. Advanced Materials, 6(5):408412, 1994.

[145] C-W Wang, Y-B Yi, AM Sastry, J Shim, and KA Striebel. Particle compression and
conductivity in li-ion anodes with graphite additives. Journal of the Electrochemical
Society, 151(9):A1489A1498, 2004.

[146] Shashi K Shahi, G Gary Wang, Liqiang An, Eric Bibeau, and Zhila Pirmoradi. Using
the Pareto set pursuing multiobjective optimization approach for hybridization of a
plug-in hybrid electric vehicle. Journal of Mechanical Design, 134:09450316,
2012.

[147] Xiaolan Wu, Binggang Cao, Xueyan Li, Jun Xu, and Xiaolong Ren. Component
sizing optimization of plug-in hybrid electric vehicles. Applied Energy, 88(3):799
804, 2011.

[148] Yi-Hsuan Hung and Chien-Hsun Wu. An integrated optimization approach for a
hybrid energy system in electric vehicles. Applied Energy, 98:479490, 2012.

[149] Yuan Zou, Fengchun Sun, Xiaosong Hu, Lino Guzzella, and Huei Peng. Combined
optimal sizing and control for a hybrid tracked vehicle. Energies, 5(11):46974710,
2012.

[150] Min-Joong Kim and Huei Peng. Power management and design optimization of fuel
cell/battery hybrid vehicles. Journal of Power Sources, 165(2):819832, 2007.

[151] Christian Fellner and John Newman. High-power batteries for use in hybrid vehicles.
Journal of Power Sources, 85(2):229236, 2000.

115
[152] Yuang-Shung Lee and Ming-Wang Cheng. Intelligent control battery equalization
for series connected lithium-ion battery strings. Industrial Electronics, IEEE Trans-
actions on, 52(5):12971307, 2005.

[153] Stephen W Moore and Peter J Schneider. A review of cell equalization methods for
lithium ion and lithium polymer battery systems. SAE Publication, pages 010959,
2001.

[154] Wenbo Du, Nansi Xue, Ann M Sastry, Joaquim RRA Martins, and Wei Shyy. Energy
density comparison of Li-ion cathode materials using dimensional analysis. Journal
of The Electrochemical Society, 160(8):A1187A1193, 2013.

[155] Fengchun Sun, Rui Xiong, Hongwen He, Weiqing Li, and Johan Eric Emmanuel
Aussems. Model-based dynamic multi-parameter method for peak power estimation
of lithiumion batteries. Applied Energy, 96(0):378 386, 2012.

[156] Kevin G Gallagher, Dennis Dees, and Paul Nelson. Phev battery cost assessment,
2011.

[157] Dong H Jang, Young J Shin, and Seung M Oh. Dissolution of spinel oxides and
capacity losses in 4 V Li/Lix Mn2 O4 cells. Journal of The Electrochemical Society,
143(7):22042211, 1996.

[158] Antti Lajunen. Evaluation of the benefits of using dual-source energy storage in
hybrid electric vehicles. In Vehicle Power and Propulsion Conference (VPPC), 2010
IEEE, pages 16. IEEE, 2010.

[159] Joao P Trovao, Paulo G Pereirinha, Humberto M Jorge, and Carlos Henggeler An-
tunes. A multi-level energy management system for multi-source electric vehicles
an integrated rule-based meta-heuristic approach. Applied Energy, 105:304318,
2013.

[160] Andrew F Burke. Batteries and ultracapacitors for electric, hybrid, and fuel cell
vehicles. Proceedings of the IEEE, 95(4):806820, 2007.

[161] John Christensen, Venkat Srinivasan, and John Newman. Optimization of lithium
titanate electrodes for high-power cells. Journal of the Electrochemical Society,
153(3):A560A565, 2006.

[162] Theodore P Adams, Dennis A Brumwell, Joseph S Perttu, and Charles G Supino.
Dual battery power system for an implantable cardioverter defibrillator, Decem-
ber 13 1994. US Patent 5,372,605.

[163] Craig L Schmidt. High power implantable battery with improved safety and method
of manufacture, April 24 2007. US Patent 7,209,784.

[164] Hong Gan and Esther S Takeuchi. Electrochemical battery for conversion of low
rate energy into high rate energy by parallel discharging, August 22 2001. EP Patent
1,126,539.

116
[165] David M Spillman and Esther S Takeuchi. Electrochemical cell having multiplate
and jellyroll electrodes with differing discharge rate regions, December 26 2000. US
Patent 6,165,638.

[166] Thomas J Dougherty, Alan J Klebenow, Edward N Mrotek, David A Thuerk, and
Maurice C Michaud. Dual battery system, November 10 1992. US Patent 5,162,164.

[167] Sarah G Stewart, Scott Ira Kohn, Kurt Russell Kelty, and Jeffrey Brian Straubel.
Electric vehicle extended range hybrid battery pack system, March 6 2013. US
Patent App. 13/787,421.

[168] Todd Alhart. GE demonstrates electric bus of the future, Decem-

ber 2010. URL: www.genewscenter.com/Press-Releases/
GE-Demonstrates-Electric-Bus-of-the-Future-2d43.aspx#
downloads.

[169] J Vetter, P Novak, MR Wagner, C Veit, K-C Moller, JO Besenhard, M Winter,

M Wohlfahrt-Mehrens, C Vogler, and A Hammouche. Ageing mechanisms in
lithium-ion batteries. Journal of Power Sources, 147(1):269281, 2005.

[170] JO Besenhard, J Yang, and M Winter. Will advanced lithium-alloy anodes have a
chance in lithium-ion batteries? Journal of Power Sources, 68(1):8790, 1997.

[171] Xianke Lin, Jonghyun Park, Lin Liu, Yoonkoo Lee, AM Sastry, and Wei Lu. A
comprehensive capacity fade model and analysis for li-ion batteries. Journal of The
Electrochemical Society, 160(10):A1701A1710, 2013.

[172] Christopher M DeLuca, Kurt Maute, and Martin L Dunn. Effects of electrode parti-
cle morphology on stress generation in silicon during lithium insertion. Journal of
Power Sources, 196(22):96729681, 2011.

[173] Dirk Kehrwald, Paul R Shearing, Nigel P Brandon, Puneet K Sinha, and Stephen J
Harris. Local tortuosity inhomogeneities in a lithium battery composite electrode.
Journal of The Electrochemical Society, 158(12):A1393A1399, 2011.

[174] Wenbo Du. Multi-Scale Modeling, Surrogate-Based Analysis, and Optimization of

Lithium-Ion Batteries for Vehicle Applications. PhD thesis, University of Michigan,
2013.

117