Embed Size (px)
Citation preview
Modeling, Simulation and Optimization of Fuel Cell/Battery
Hybrid Powertrains
By
Piyush Bubna
A thesis submitted to Faculty of the University of Delaware in partial fulfillment of the requirements for the degree of Master of Science in Mechanical Engineering
Summer 2010
Copyright 2010 Piyush Bubna All Rights Reserved
Modeling, Simulation and Optimization of Fuel Cell/Battery
Hybrid Powertrains
By
Piyush Bubna
Approved: ____________________________________________________
Ajay K. Prasad, Ph.D.
Professor in charge of the thesis on behalf of the advisory committee
Approved: ____________________________________________________
Suresh G. Advani, Ph.D.
Professor in charge of the thesis on behalf of the advisory committee
Approved: ____________________________________________________
Anette M. Karlsson, Ph.D.
Chair of the Department of Mechanical Engineering
Approved: ____________________________________________________
Michael J. Chajes, Ph.D.
Dean of the College of Engineering
Approved: ____________________________________________________
Debra Hess Norris, M.S.
Vice Provost for Graduate and Professional Education
iii
ACKNOWLEDGEMENTS
This thesis has developed with the help and contributions of many individuals and
it is a pleasure to extend my appreciation and gratitude to everyone involved. First and
foremost, would be my advisors - Dr. Ajay Prasad and Dr. Suresh Advani. Their
encouragement, advice and continuous guidance has helped me in completing this thesis
as well as the challenging research behind it.
I would also like to thank my fellow labmates: Doug, Sudhaker, Srikanth, Darren,
Adam, Mike, Glenn, Erik, Manish, Krishnan and Feng Yuan. My sincere appreciation
goes to Doug for his valuable help during the entire course of this project.
I thank my family (in India) for their unending love and for always placing my
dreams and interests before anything else. Finally, I dedicate this work to Aparna who
has been a great support and motivator throughout this work.
iv
TABLE OF CONTENTS
LIST OF TABLES.........................................................................................................vi
LIST OF FIGURES ..................................................................................................... vii
ABSTRACT...................................................................................................................xi
Chapter
1. INTRODUCTION
1.1. Introduction to Hybrid Vehicles.... .........................................................13
1.2 Fuel Cells ................................................................................................16
1.3 Battery.....................................................................................................18
1.4 Ultracapacitor..........................................................................................19
1.5 The University of Delaware Fuel Cell Bus.............................................20
1.6 Organization of the Thesis ......................................................................24
2. LFM SIMULATOR
2.1 Introduction.............................................................................................26
2.2 LFM ........................................................................................................27
2.3 LFM Subsystems ....................................................................................28
2.3.1 Fuel Cell....................................................................................28
2.3.2 Battery.......................................................................................36
2.3.3 Ultracapacitors ..........................................................................39
2.3.4 Hybrid Controller......................................................................40
2.3.5 Power Combiner .......................................................................42
2.3.6 Accessory Load.........................................................................42
2.3.7 Motor.........................................................................................43
2.3.8 Transmission.............................................................................43
2.3.9 Wheels/Chassis .........................................................................43
2.4 Validation................................................................................................44
2.5 Summary.................................................................................................53
3. PREDICTION-BASED OPTIMAL POWER MANAGEMENT
3.1 Introduction.............................................................................................54
3.2 Power Management Strategy ..................................................................57
3.3 Methodology and Algorithm...................................................................60
v
3.4 Simulation Results ..................................................................................63
3.5 Validation................................................................................................73
3.6 Summary.................................................................................................76
4. REDUCED BATTERY STRESS THROUGH BLENDED ENERGY STORAGE
4.1 Introduction.............................................................................................78
4.2 Blended ESS topology and energy management ....................................80
4.2.1 Battery-only ESS ......................................................................80
4.2.2 Blended ESS .............................................................................82
4.3 Simulation Results ..................................................................................85
4.3.1 Simulation Results with 48-cell Ucap.......................................85
4.3.2 Simulation Results with 36-cell Ucap.......................................88
4.4 Summary.................................................................................................90
5. BATTERY THERMAL MODEL
5.1 Introduction.............................................................................................91
5.2 Altairnano Lithium-Titanate Cells..........................................................92
5.3 Battery tests.............................................................................................93
5.4 Battery Thermal Model & Simulations.................................................104
5.4.1 Mathematical Model ...............................................................104
5.4.2 Results.....................................................................................108
5.4.3 Thermal Simulations...............................................................116
5.5 Summary...............................................................................................119
6. SUMMARY AND FUTURE WORK
6.1 Summary...............................................................................................121
6.2 Future Work ..........................................................................................123
6.2.1 Powertrain model and simulation ...........................................123
6.2.2 Prediction-based power management stratey..........................124
6.2.3 Blended energy storage...........................................................124
6.2.4 Battery thernal modeling and simulation................................125
6.2.5 Intelligent driving....................................................................126
REFERENCES ...........................................................................................................127
vi
LIST OF TABLES Table 2.1 Quantitative Comparison corresponding to Drive Cycle 1...........................48
Table 2.2 Quantitative Comparison corresponding to Drive Cycle 2...........................52
Table 3.1 Comparison of prediction-based and baseline strategy for SC03.................65
Table 3.2 Comparison of prediction-based and baseline strategy for UDDS...............66
Table 3.3 Comparison of prediction-based and baseline strategy for test drive cycle .76
Table 4.1 Comparison of advanced technology battery and Ucap ...............................79
Table 4.2 Battery Description.......................................................................................81
Table 4.3 Ultracapacitor Description............................................................................83
vii
LIST OF FIGURES Figure 1.1: Schematic of parallel hybrid [1].................................................................13
Figure 1.2: Schematic of series hybrid [1]....................................................................14
Figure 1.3: Simplified schematic of hybrid powertrain ................................................15
Figure 1.4: A simplified fuel cell and its basic operation [2] .......................................17
Figure 1.5: Ragone plot comparing energy and power density of different power
sources [3].........................................................................................................18
Figure 1.6: University of Delaware Fuel Cell Hybrid Bus #1 ......................................21
Figure 2.1: Different subsystems of the LFM Simulink model ....................................26
Figure 2.2: Voltage vs current corresponding to the Ballard Mark 9 SSL 110-cell
stack employed in our Phase 1 bus ...................................................................29
Figure 2.3: Variation of gross stack power, net stack power, compressor power, and
rest of BOP load with stack current ..................................................................32
Figure 2.4: Variation of fuel cell system efficiency with net fuel cell power as
recorded by the Phase 1 bus and as predicted by the model .............................34
Figure 2.5: Variation of gross stack power, net stack power, compressor power, and
rest of BOP load with stack current corresponding to our Phase 2 bus
employing the dual stack as predicted by LFM................................................35
Figure 2.6: Variation of fuel cell system efficiency with net fuel cell power
corresponding to the Phase 2 dual-stack bus as predicted by LFM .................36
Figure 2.7: Aerial view of UD Express Route............................................................45
Figure 2.8: Drive Cycle profile of UD Express Route................................................46
Figure 2.9: Comparison of simulation output and vehicle data for Drive Cycle 1.....47
Figure 2.10: Aerial view of second test run................................................................49
Figure 2.11: Drive Cycle profile of second test run....................................................50
Figure 2.12: Comparison of simulation output and vehicle data for Drive Cycle 2...51
viii
Figure 3.1: Battery SOC drop and fuel cell net power corresponding to the baseline
and the predictive control strategy for SC03 (~2 hours, 46 miles)...................59
Figure 3.2: Longer drive cycles formed by repeating standard cycles ......................64
Figure 3.3: Deviation in battery SOC drop and fuel cell net power corresponding to
inaccuracy in prediction for the SC03 (~2 hours, 46 miles) ............................67
Figure 3.4: Deviation in battery SOC drop and fuel cell net power corresponding to
inaccuracy in prediction for the UDDS (~2 hours, 45 miles) ..........................68
Figure 3.5: Fuel savings and final battery SOC for varying degree of inaccurate
predictions and for variable drive lengths for the SC03 driving schedule........69
Figure 3.6: Fuel savings and final battery SOC for varying degree of inaccurate
predictions and for variable drive lengths for the UDDS driving schedule......70
Figure 3.7: Possible SOC profiles corresponding to the condition maxavgP Pη< ..........72
Figure 3.8: Aerial view of the trajectory traced by the fuel cell hybrid bus ...............73
Figure 3.9: Profile of the test drive cycle....................................................................74
Figure 3.10: Battery SOC drop and fuel cell net power corresponding to baseline and
predictive control strategy ................................................................................75
Figure 4.1: Topology of a fuel cell/battery hybrid......................................................81
Figure 4.2: Topology of fuel cell/battery/ultracapacitor series hybrid .......................83
Figure 4.3: Simulated battery C-rate frequency distribution for a battery only, as well
as blended ESS (48-cell Ucap) at different threshold levels.............................86
Figure 4.4: Simulated energy throughput for a battery only, as well as blended ESS
(48-cell Ucap) at different threshold levels ......................................................87
Figure 4.5: SOC swing of 48-cell Ucap module at 0 kW and 30 kW threshold power
corresponding to UD Express Route ................................................................88
Figure 4.6: Simulated battery C-rate frequency distribution for a battery only, as well
as blended ESS (36-cell Ucap) at different threshold levels.............................89
ix
Figure 4.7: Simulated energy throughput for a battery only, as well as blended ESS
(36-cell Ucap) at different threshold levels ......................................................89
Figure 5.1: An Altairnano Lithium-Titanate cell (2.3 V, 50 Ah, 25x25x1.2 cm).......93
Figure 5.2: Schematic of the 5 cell stack and thermistor locations ............................94
Figure 5.3: Snapshot of the battery pack of five cells used for experiments ..............95
Figure 5.4: Temperature distribution on the surface of the cell recorded by IR Camera
at different time instants during charging at 400 A .........................................96
Figure 5.5: Temperature distribution on the surface of the cell recorded by IR Camera
at the end of 15 minutes of charging time with 100 A of current.....................97
Figure 5.6: Temperature readings at different locations of the battery pack during 200
A charge/discharge cycles.................................................................................99
Figure 5.7: Temperature readings at different locations of the battery pack during 100
A charge/discharge cycles...............................................................................100
Figure 5.8: Experimentally measured variation of open circuit voltage (Voc) with
temperature for LiTi cell .................................................................................101
Figure 5.9: Time trace of water and cell surface temperature during calorimetric test
for measuring specific heat capacity of LiTi cell............................................102
Figure 5.10: Rate temperature drop due to heat loss to the environment as a function
of water temperature ......................................................................................103
Figure 5.11: Schematic diagram of current flow in parallel electrodes of a cell ......105
Figure 5.12: Distribution of Vp .................................................................................109
Figure 5.13: Distribution of Vn .................................................................................111
Figure 5.14: Distribution of the potential difference (Vp-Vn) .................................113
Figure 5.15: Distribution of current density J...........................................................115
Figure 5.16: Variation of heat generation rate with the height of the cell ................116
Figure 5.17: 3D model of half LiTi cell created in Gambit ......................................117
x
Figure 5.18: Temperature distribution on the cell surface after 5 minutes of charge at
400A obtained from FLUENT simulation (above) IR imaging (below) .......118
xi
ABSTRACT
Fuel cells have emerged as one of the most promising candidates for fuel-efficient
and emission-free vehicle power generation. Fuel cells are typically paired with
reversible energy storage devices such as batteries or ultracapacitors to create hybrid
electric powertrains. The electrification of the propulsion system and the presence of
multiple onboard power sources require optimization of the hybrid system design in order
to achieve good performance, high fuel economy, and enhanced component life at low
cost. The overall goal of this research is to develop accurate vehicle models and conduct
simulations to explore and demonstrate improvements in a fuel cell/battery hybrid bus.
The first part of this thesis presents the features incorporated to improve a hybrid
powertrain simulation package called Light, Fast and Modifiable (LFM). The improved
LFM simulator was validated against test data acquired from various sensors onboard
UD’s Phase 1 fuel cell bus, and shown to be a reliable tool to simulate hybrid powertrain
performance which could be used to perform design and optimization studies of future
fuel cell hybrid systems.
This attribute of LFM was then demonstrated by optimizing the fuel cell/battery
hybrid power management by introducing a new prediction-based power management
strategy. Simulation results for this strategy showed significant improvements in fuel cell
system efficiency and reduction in hydrogen consumption compared to a conventional,
baseline strategy of charge sustenance. A stable power request which promotes fuel cell
durability was also realized with the help of this novel strategy. Finally, the benefits
predicted from simulation studies were confirmed through implementation of the
xii
proposed strategy in the Phase 1 fuel cell/battery hybrid bus. It was concluded that the
prediction-based strategy will lead to energy savings for transit applications.
The validated LFM tool was next used to evaluate one approach to reducing
battery stress by adding an ultracapacitor module, and thereby enhancing battery lifetime.
Simulation of the energy storage performance showed a substantial reduction in battery
current-load and energy throughput for the blended storage system, which are two of the
contributing factors towards battery degradation. These results have opened up new
research directions in which powertrain simulations can help in further evaluation of the
blended storage concept and assess its feasibility and usefulness in electric-drive vehicles.
Finally, the thermal behavior of the Altairnano LiTi battery, the future battery of
UD fuel cell buses, was investigated. Preliminary experiments were conducted to
understand the thermal behavior of batteries under typical operating conditions. A model
was developed to predict the temperature during charging and discharging of the battery.
The findings of this work should prove useful in designing effective and efficient battery
thermal management systems.
13
1 Introduction
1.1 Introduction to Hybrid Vehicles
Conventional internal combustion engine (ICE) vehicles rely on a single power
source typically fueled with gasoline, to drive a complex transmission mechanism.
Although the technology has evolved continuously over the past 100 years, it suffers
from a number of disadvantages. These include low energy efficiency, excessive harmful
chemical emissions, high noise level, high degree of complexity, and complete
dependence on a single fuel source.
In contrast, hybrid vehicles use two or more distinctly different power sources to
propel the vehicle. The term commonly used to refer to such vehicles is hybrid electric
vehicles (HEV), which employ a combination of an internal combustion engine with an
electric propulsion system. The electric propulsion system of an HEV mainly comprises
an electrical storage (battery or ultracapacitor) and an electric motor. The ICE and the
electric propulsion system can be combined in several ways. Two common ways are
shown below.
Figure 1.1 Schematic of parallel hybrid [1]
14
The schematic in figure 1.1 shows a parallel HEV where the engine and electric
propulsion work in parallel to turn the differential/wheels. The electrical storage provides
power which is converted to mechanical power/torque at the electric motor. On the other
hand, the engine can be decoupled from the differential/wheels and instead provide
electrical power by turning a generator (figure 1.2). This type of configuration is called a
series hybrid.
Figure 1.2 Schematic of series hybrid [1]
A hybrid propulsion system can overcome many of the problems associated with
the conventional ICE vehicle. An ICE is a one-way power source, meaning that it can
burn fuel to produce the required drive power, but it cannot run the reverse reaction and
convert the vehicle’s kinetic energy back to fuel during deceleration. In city driving
conditions, roughly 10 to 20% of the drive system’s energy is lost in braking. The energy
loss is expected to be even higher for transit buses due to frequent starts and stops. On the
other hand, the electric propulsion system is bidirectional and has the ability to convert
the vehicle's kinetic energy into battery-replenishing electric energy, rather than wasting
it as heat energy as is the case for conventional friction brakes. Furthermore, an electric
motor operates at a higher efficiency than an ICE, and its efficiency is more or less
15
constant over most of its operating range. Also, many HEVs reduce idling emissions by
shutting down the ICE when the vehicle is stationary, and restarting it when needed
(start-stop system). A hybrid-electric vehicle produces less emissions from its ICE than a
comparably-sized ICE vehicle, because the HEV employs a much smaller ICE which is
typically operated close to its maximum efficiency point, further improving fuel
economy.
The advantages of hybridization also apply to fuel cell powered vehicles. A fuel
cell engine, which uses hydrogen as the fuel, exhibits highly varying efficiencies
depending upon its operating point (current draw) and also does not perform well under
rapidly changing power demands. Power assistance from a reversible energy storage
system such as a battery or ultracapacitor alleviates the problem and enhances
performance. Fuel cells like IC engines are a one-way power source, and therefore,
reversible energy storage is required to accomplish regenerative braking.
Figure 1.3 Simplified schematic of the hybrid power train
16
1.2 Fuel Cells
Fuel cells are one of the most promising candidates for fuel-efficient and
emission-free vehicle propulsive power. The increasing popularity of fuel cells is due to
its high efficiency (2-3 times greater than ICE) and no harmful emissions. In particular,
proton exchange membrane (PEM) fuel cells have received much attention for
automotive applications due to their low operating temperature, rapid start-up, high
power density, and high efficiency.
Fuel cells are electrochemical cells which convert the source fuel into electrical
power along with byproducts of the reaction. They generate electricity through reactions
between a fuel and an oxidant within the membrane electrode assembly (MEA), which
consists of two electrodes separated by an electrolyte. The fuel cell produces a voltage
and a current when it is supplied by reactants that flow into the cell, while removing the
reaction products from the cell. Fuel cells can operate virtually continuously as long as
the necessary flows are maintained. Figure 1.4 illustrates the process.
17
Figure 1.4 A simplified fuel cell and its basic operation [2]
A single fuel cell can produce a typical voltage of only about 0.6-0.7 V.
Therefore, several fuel cells must be joined together in series to create a “stack” in order
to produce a useful voltage. Automotive stacks comprise over a hundred cells in series
producing up to 100 kW of power. It should be noted that a fuel cell cannot function on
18
its own and requires a significant balance-of-plant (BOP) to handle fuel and oxidant
supply, cooling, humidification, power conditioning, and for control and management
tasks. In addition to the BOP, fuel cell powered vehicles need a reversible energy storage
device such as a battery or ultracapacitor to absorb energy from regenerative braking and
meet transient power demands.
1.3 Battery
A battery is a combination of one or more electrochemical cells, used to convert
stored chemical energy into electrical energy or vice versa. The battery pack in a hybrid
vehicle is one of the most important parts of the propulsion system. Batteries, unlike ICEs
and fuel cells, can supply as well as absorb energy.
Figure 1.5 Ragone plot comparing energy and power density of different power sources
[3].
Batteries have a higher power density than ultracapacitors, but lower energy density than
the fuel cell (figure 1.5). Therefore in fuel cell hybrid vehicles, batteries are typically
19
used to meet transient power demands and absorb energy during regenerative braking,
while the fuel cell continues to charge the battery and extends the range of the vehicle.
1.4 Ultracapacitors
Ultracapacitors are electric double-layer capacitors that have an unusually high energy
density when compared to common capacitors, typically about a thousand times greater
than a high-capacity electrolytic capacitor. In a conventional capacitor, energy is stored
by the removal of charge carriers, typically electrons, from one metal plate by depositing
them on another. The total energy stored is the product of the amount of charge stored
and the potential between the plates. The amount of charge stored is a function of the size
and the material properties of the plates. The potential between the plates is limited by the
dielectric breakdown of the material separating the plates.
Ultracapacitors do not have a conventional dielectric sandwiched between the two
plates. Instead, these plates are two layers of the same substrate which form an electrical
double layer, resulting in effective separation of charge despite a vanishingly thin
physical separation of the layers. The elimination of the bulky dielectric layer permits
compact packing of layers and with much larger surface area. The resulting devices
possess extraordinarily high capacitances resulting in higher storage capacities (Ah) than
conventional capacitors. Ultracapacitors possess higher power densities than batteries.
Batteries, which are based on the movement of charge carriers in a liquid electrolyte,
have relatively slow charge and discharge times. Capacitors, on the other hand, can be
charged or discharged at a rate that is typically limited by current heating of the
electrodes. However, batteries possess higher energy densities than ultracapacitors. Due
20
to these varying properties, the decision to use batteries or ultracapacitors is dictated by
the intended application. Due to their higher energy content, batteries perform better at
isolating the fuel cell from transient power demands for the entire duration of the drive
cycle. For this reason the University of Delaware fuel cell buses are fuel cell/battery
hybrids, and are described next.
1.5 The University of Delaware Fuel Cell Bus
The University of Delaware’s fuel cell hybrid vehicle served as the subject for the
research conducted during this thesis. This 22 ft bus was designed and built by EBus,
Inc. and can hold 22 seated and 10 standing passengers (Figure 1.5). It is powered by a
Ballard Mark 9 SSL 110-cell stack, rated for 19.4 kW gross power. The bus is driven by a
single three-phase AC induction motor that is rated for 130 kW peak and 75 kW
continuous power and speeds up to 5000 rpm. The motor is coupled to the rear drive
wheels through a single-speed chain drive and a differential, with gear ratio selected to
allow speeds of up to 45 mph while providing enough torque to climb a 20% grade fully
loaded.
The bus incorporates a series-hybrid powertrain that employs SAFT Nickel-
Cadmium (NiCd) liquid-cooled batteries in two 300 V strings. The strings are connected
in parallel because of the traction inverter voltage limitation, and the two together are
capable of meeting high power demands (~120 kW). Each string consists of 50
monoblocks, each containing 5 cells. The cells are rated for a nominal charge capacity of
100 Ah and total energy capacity of 60 kWh. This typically gives the vehicle an all-
electric range of 40 miles. The bus uses compressed hydrogen stored in twin composite
21
high-pressure tanks mounted on the roof of the bus. The tanks are rated for 350 bar and
have a total storage capacity of approximately 12.8 kg. This amount of hydrogen yields
an average range of about 140 miles. The plug-in feature of the bus permits the initial
portion of each route to be driven solely on battery power with the fuel cell switching on
when the battery state-of-charge falls below the chosen threshold value.
Figure 1.6 University of Delaware Fuel Cell Hybrid Bus #1
The fuel cell stack is fed with air by a scroll-type compressor at pressures of 83 to
124 kPa, depending on load, which is humidified by moisture from the cathode exhaust
air using membrane humidifiers. Hydrogen is supplied at a slightly higher pressure, and
hydrogen is recirculated from the stack outlet to the inlet using a rotary single-vane
pump, to ensure clearance of water from all parts of the anode. The stack is liquid cooled,
using a low-conductivity ethylene glycol/water mixture and a fan-cooled radiator which
Battery Pack
Hydrogen Tanks Radiator
Rear panel houses stack and balance of plant
22
is mounted on the roof of the bus as can be seen from Figure 1.6. After subtracting the
power requirements of the balance of plant, the fuel cell stack delivers a maximum net
power output of 14 kW.
Since the stack's operating voltage is typically between 65 and 75 V, a boost
converter is used to deliver power to the main DC bus at a nominal battery voltage of 300
V (which can range from 250 V to 370 V during normal vehicle operation), controlling
the amount of power drawn from the fuel cell. The schematic of the hybrid drivetrain for
this bus is shown in Figure 1.3. The fuel cell system is controlled by a programmable
logic controller (PLC), which coordinates the functions of all parts of the fuel cell system
according to the amount of power requested by the vehicle control computer.
The bus is equipped with a data acquisition system installed in a laptop computer,
currently running custom software within LabVIEW. It monitors the vehicle control
computer, the fuel cell system's PLC, and the traction inverter. In addition, it has a GPS
receiver. Real-time data are collected from a variety of other on-board sensors monitoring
fluid temperatures, flowrates, and humidity levels within the fuel cell system.
The overall design of our bus features a battery-heavy hybrid which uses the fuel
cell as a range extender. The basic control strategy is to run the bus in battery-only mode
until the state-of-charge (SOC) reaches a threshold value. This value can be
reprogrammed, but defaults to 65%. Once the SOC reaches 65%, the fuel cell turns on
with a power request governed by
( )d c onehr moving avgFC power request SOC SOC Pα= − +
23
where dSOC is the threshold value, cSOC is the current calculated SOC, α is the
proportionality constant and onehr moving avgP is the moving average power use of the bus
over the last hour.
The vehicle runs solely on the battery at the start of the drive cycle resulting in a
steady drop in SOC. As soon as the SOC reaches the threshold value of 0.65 the fuel cell
stack is turned on and after ramping up at a desired rate, it delivers an average power to
sustain the battery SOC at 0.6. The stack is turned off after the completion of the route
and the bus returns to its garage on batteries alone to deplete it further. This mode of
operation, known as charge depletion, not only reduces hydrogen consumption but also
allows the NiCd batteries to be cycled over a large fraction of their capacity, which helps
to avoid the effect known as "voltage depression" and thus maintain usable capacity.
The electrification of the propulsion system has made the hybrid design process
increasingly complex. An optimal hybrid system can be achieved through appropriate
sizing of the battery and/or ultracapacitor, and fuel cell or ICE, and intelligent energy
management between them. Such decisions depend on many factors such as the vehicle
size, performance targets, type of application, fuel economy, component lifetime, and
cost. There is no simple or direct way to select and harness the advantages of different
components and satisfy the desired targets. The design process is iterative and requires
advanced powertrain modeling and simulation efforts in order to facilitate the analysis
and optimization of the new generation of vehicle power train. The overall goal of this
thesis is to develop and refine a power train modeling and simulation effort and
demonstrate its utility on our fuel cell hybrid bus.
24
1.6 Organization of the Thesis
Chapter 2 presents the development of the fuel cell hybrid powertrain simulator
called LFM. The weaknesses of the previous versions are discussed along with the
modifications and improvements. The description of the simulator is followed by a
validation study with compares the results of the simulation with operational data from
the vehicle. Subsequent chapters detail the use of the simulator in a number of studies
aimed at improving the design, performance, efficiency, and lifetime of our fuel cell bus.
Chapter 3 presents a prediction-based power management strategy for fuel
cell/battery plug-in hybrid vehicles with the goal of improving overall system operating
efficiency. The results obtained from the implementation of the proposed strategy in
LFM are presented along with a sensitivity of parameters important for the study. The
effectiveness of the strategy is evaluated by comparing it with a conventional control
strategy within the simulation environment. Finally, the validation of the simulated
results is demonstrated by implementation of the strategy in the fuel cell bus and by
conducting real time testing.
Chapter 4 explores the load reduction effects on the battery in blended energy
storage (batterty+ultrapacitor) hybrids. This is done by comparing a battery only hybrid
and blended energy storage hybrid within the simulation environment. The chapter first
introduces the factors leading to battery degradation and briefly describes the role that an
ultracapacitor can play. Next, the hybrid topology, energy storage and power
management scheme of the two hybrid systems are presented, followed by simulation
results.
25
Chapter 5 attempts to understand the thermal behavior of Altairnano LiTi cells
using two approaches: experiments, and modeling and simulations. First, the features of
the new Altairnano LiTi cells are introduced. The experiments performed on the cells and
analyses of results are described next. A 2D mathematical model for the evaluation of
heat generation is presented, followed by a transient thermal simulation of the cell
undergoing charge/discharge cycles.
Finally, Chapter 6 summarizes the contributions made in this thesis and presents
conclusions. Also, the possible future work for each topic presented here is discussed.
26
2 LFM Simulator
2.1 Introduction
The hybrid vehicle simulator used in this thesis to perform hybrid-powertrain
performance and optimization studies is called LFM (Light, Fast and Modifiable). LFM
was originally developed by the Electric Power Research Institute (EPRI) in Palo Alto,
CA, and was subsequently modified at the UD Center for Fuel Cell Research. The
accuracy of a simulator is critical for reliable predictions. The LFM simulator has
undergone substantial improvement at UD and has proved its capabilities through
rigorous validation exercises. This chapter describes the LFM simulator, its development,
and the various enhancements that have been incorporated into it. We begin with a
general introduction of the simulator and provide a detailed description of its subsystems.
Next, we discuss the shortcomings of the earlier versions, followed by the modifications
and improvements made during the present work. Finally, a validation exercise is
presented to show the agreement between the modified LFM results and test data
collected from the Phase 1 UD Fuel Cell Bus.
Figure 2.1 Different subsystems of the LFM Simulink model
[J ]
[D]
[G ]
[F]
[E ]
[A ]
[C]
[H]
[B ]
[I ]
[J]
[D]
[H]
[G ]
[F]
[E ]
[A]
[B ]
[I]
[C]
Battery
Fuel Cell
Power Combiner
Accessory
Load Combiner
Motor Transmission Chassis
Controller
E
I
F
G
H
J
A
B
C
D
G F
H
J
E I
D
A
B C
27
2.2 LFM
LFM is a hybrid vehicle simulator, designed to simulate the performance of a fuel cell
and battery powered electric driveline under given driving conditions. The first version of
LFM was developed by Marcus Alexander of EPRI. It consisted of a series of subsystems
built and connected in Simulink using electrical, mechanical, and control signal links
(figure 2.1). The basic structure of the simulator is drive-cycle based, which implies that,
the simulation is driven by an input drive cycle. At each time step during the drive cycle,
the simulator compares the current speed with the desired speed and calculates the
appropriate power command required to propel the vehicle along its virtual trajectory.
The command then propagates through various subsystems of the drivetrain to the power
sources. The LFM solver is a variable time-step ode45 (Dormand-Prince) solver. It is part
of the Runge-Kutta family of ordinary differential equation solvers, and is well suited for
a variable time-step simulation. Besides the drive cycle, LFM requires a number of
quantitative inputs to accurately describe the drivetrain components such as the vehicle,
transmission, drive motor, accessory load, fuel cell system, and battery.
In the first version of the simulator, these data were stored in an Excel spreadsheet
and read using a special class of objects into Simulink in real time. The Simulink results
were written in MATLAB workspace and were available for analysis after the simulation
had ended. Darren Brown (M.S. 2008, University of Delaware) modified the input and
output system of the simulator to facilitate rapid iterations [4, 5] and modified various
subsystems in order to simulate the Phase 1 bus [4].
Despite this first round of modifications, LFM contained additional shortcomings
which are addressed in the present work. These include the lack of a reliable and
28
complete fuel cell system (fuel cell + balance-of-plant) model, an over-simplified
balance-of-plant (BOP) model incorrectly lumped with the accessory load, an over-
simplified battery model, an inaccurate depiction of the hybrid controller, and the need
for full-scale validation. Modifications to the LFM simulator were performed to address
these drawbacks, and are discussed in the following sections while introducing the LFM
subsystems.
2.3 LFM Subsystems
2.3.1 Fuel Cell
The fuel cell subsystem receives a power request which is then converted to a DC current
request. The current is used to calculate hydrogen consumption and update the voltage
output of the stack using a lookup table created in the fuel cell data spreadsheet.
Modifications to the fuel cell model from the current work are described next.
Additions
The two crucial aspects while setting up a vehicle simulation model for validation
are – (i) a reliable physical model, and (ii) the use of technical specifications that
accurately reflect the actual performance of the component. Therefore, the fuel cell data
and model were revised for better prediction of fuel cell output parameters such as
voltage, current, net power, gross power, BOP load, hydrogen consumption and system
efficiency. The polarization curve was modified based upon fuel cell data acquired in real
time during a test run for our Phase 1 bus (figure 2.2).
29
0 50 100 150 200 250 30060
65
70
75
80
85
90
95
100
105
110
Current (A)
Vol
tage
(V
)
Figure 2.2 Voltage vs current corresponding to the Ballard Mark 9 SSL 110-cell stack
employed in our Phase 1 bus
The hydrogen flow rate is calculated using standard equations and incorporated a purge
rate for better accuracy.
22
2fc H st
H purge
n M Im m
F
• •= + (2.1)
where fcn is the number of cells in the stack (provided by the manufacturer- 110 for the
single stack and 220 for the dual stack), 2HM is the molar mass of hydrogen, stI is the
stack current, F is the Faraday number, and purgem•
is the purge rate of hydrogen whose
average value over 2.5 hours of fuel cell testing was recorded to be close to 0.01 kg/hr for
a single stack.
A significant component of fuel cell model is the BOP load which is required to
operate the stack. The power required to run the BOP is provided by the stack itself. The
earlier version of LFM used an over-simplified BOP model. It also lumped the BOP load
30
with the accessory load which introduces inaccuracies, both in fuel cell gross and net
power, and therefore affects all the fuel cell output parameters. In order to address this
weakness, a more sophisticated and accurate BOP model was created in the present work
and the BOP load has been modeled separately from the accessory load.
Amongst the fuel cell BOP components, the air compressor consumes the largest
portion of power. The power consumption of the BOP has been modeled as the sum of a
variable air-compressor load, and a constant load of 1 kW which accounts for the
hydrogen pump and the radiator [6]. The compressor power consumption,cpP , depends on
the stack ambient pressure ratio and the air flow rate, and is given by
( )1 /
1p amb smcpcp
m cp amb
C T pP m
p
γ γ
η η
−•
= −
(2.2)
where smp is the pressure in the supply manifold, ambp is the ambient pressure, γ is the
ratio of pC , specific heat capacity of air at constant pressure, and vC the specific heat
capacity of air at constant volume, ambT (20 °C) is the ambient temperature, mη (80%)
and cpη (70%) are the efficiencies of the compressor motor and compressor respectively,
and cpm•
is the mass flow rate of air through the compressor. To predict compressor power
consumption, cpP , as a function of the stack current, a knowledge of the quantities in
equation 2.2 as a function of stack current is needed. For a given stack current, the
stoichiometric inlet oxygen mass flow rate to the cathode is given by
22
4fc O st
O
n M Im
F
•= (2.3)
31
where2OM is the molar mass of oxygen. The mass flow rate of air to the cathode is given
by
2 2 2 2
2 2
,, ,
4o o rct o fc O st
a ca in
o o
m n M Im
y y F
λ λ•
•= = (2.4)
where 2Oλ is the oxygen excess ratio which is assumed to be maintained at a constant
value of 1.6, and the molar fraction of oxygen in air 2oy is 0.21. Thus the total air flow
rate through the compressor is given by
( ), , ,1cp a cp v cp a cpambm m m mψ• • • •
= + = + (2.5)
where ambψ is the humidity ratio of the atmospheric air, and subscripts cp ,a , andv
denote compressor, air, and water vapor respectively. Also, the mass flow rate of dry air
at the cathode inlet and compressor outlet can be assumed to be the same under steady
state conditions. Therefore,
2 2
2
, ,, , 0
, ,
1 14
o fc O stv amb sat amb v amb sat ambcp a ca in st
a a amb a a amb o
n M IM p M pm m if I I
M p M p y F
λφ φ• • = + = + >
2 2
2
0,
,
14
o fc Ov amb sat ambcp
a a amb o
n M IM pm otherwise
M p y F
λφ• = +
(2.6)
where aM and vM are the dry air and water vapor molar masses respectively, ambφ is the
relative humidity of the ambient air (assumed to be 0.7), ,sat ambp is the vapor saturation
pressure at ambient temperature, and ,a ambp is the pressure of the dry atmospheric air. In
the vehicle smp varies from 13.5 psig to 17 psig. The mass flow rate is a constant below a
32
certain threshold current 0I (150 A). Based on the above calculations the gross stack
power, ,fc grossP , and net power, ,fc netP , can be calculated in the following way:
, ,fc net fc gross BOPP P P= − (2.7)
BOP const cpP P P= + (2.8)
Figure 2.3 depicts the gross power, net power, and BOP load, BOPP , as a function of the
stack current for the single stack. Note that compressor power for the single stack can
vary from 1 KW at low current draw to 2 KW at high current draw.
50 100 150 200 250-2
0
2
4
6
8
10
12
14
16
18
Stack Current (A)
Po
we
r (kW
)
Gross PowerNet PowerCompressor PowerConstant BOP Power
Figure 2.3 Variation of gross stack power, net stack power, compressor power, and rest of
BOP load with stack current.
33
The variation of parasitic losses in the fuel cell system with current naturally leads to an
evaluation of system efficiency over its operating range. The fuel cell system efficiency is
defined as the ratio of the net power delivered by the fuel cell to the lower heating value
(LHV) of the fuel, which is hydrogen in our case.
2 2
,,
fc netfc sys
H H
P
m LHVη •= (2.9)
where ,fc netP is the net power delivered by the fuel cell (i.e. gross fuel cell power minus
the power consumed by the BOP) and 2Hm•
is the corresponding fuel consumption rate.
The system efficiency is plotted against net power for the single stack in figure 2.4, and
offers us guidance into the desired power range of the fuel cell in order to achieve high
efficiencies.
34
0 2 4 6 8 9 10 12 140
5
10
15
20
25
30
35
40
45
50
Fuel Cell Net Power (kW)
Effi
cien
cy (
%)
Test Data
Model
Figure 2.4 Variation of fuel cell system efficiency with net fuel cell power as recorded by
the Phase 1 bus and as predicted by the model.
The close match between the efficiency values derived from vehicle data and the
enhanced model as evident in figure 2.4 validates the fuel cell system model used in the
present work. It also justifies adopting the same approach for modeling the dual stack
employed in our Phase 2 bus. The significance of figure 2.4 has been discussed in
Chapter 3 as part of an optimization study. For the dual stack, constP is proportionately
increased to 2 kW. The compressor power,cpP , is calculated using equation 2.2. The dual-
stack compressors are assumed to operate at lower pressures varying from 4 psig to 10
psig, assuming that an improved fuel cell stack will contain larger humidifiers.
35
0 50 100 150 200 250 300-5
0
5
10
15
20
25
30
35
40
Current (A)
Po
wer
(kW
)
Gross PowerNet PowerCompressor PowerConstant BOP Power
Figure 2.5 Variation of gross stack power, net stack power, compressor power, and rest of
BOP load with stack current corresponding to our Phase 2 bus employing the dual stack
as predicted by LFM
36
0 5 10 15 17 20 25 30 350
5
10
15
20
25
30
35
40
45
50
Net Fuel Cell Power (KW)
Fue
l Ce
ll S
yste
m E
ffici
enc
y (%
)
Figure 2.6 Variation of fuel cell system efficiency with net fuel cell power corresponding
to the Phase 2 dual-stack bus as predicted by LFM
2.3.2 Battery
The battery is modeled as a voltage source in series with a resistance, both of which vary
with the state-of-charge (SOC).
int, if 0oc disV V IR I= − >
int, if 0oc chV V IR I= − < (2.10)
where ( )ocV f SOC= is the open circuit voltage of the battery pack, int, ( )disR f SOC= and
int, ( )chR f SOC= are discharge and charge internal resistances, respectively, for both the
battery strings combined.
37
00( )
t
batt
I dtSOC t SOC
Cη= − ∫
(2.11)
where C is the charge capacity (Ah) of both the strings combined, and
1 if 0batt Iη = >
0.85 if 0batt Iη = <
The battery subsystem takes current request as the input to update the state-of -
charge (equation 2.11) and calculates the voltage output of the stack (equation 2.10). The
charging reaction in NiCd chemistry is accompanied by a side reaction (electrolysis) due
to which not all of the charging current goes towards converting active material. The
conversion factor battη is higher at low SOCs (less electrolysis), and lower at high SOCs
(more electrolysis). Based on data provided by the vehicle manufacturer we have used an
average conversion factor of 0.85 over the entire SOC range during charging. No side
reactions are present during discharging and so the conversion factor is set to 1.0.
Modifications to the battery system from the current work are described next.
Additions
An extra subroutine has been added to the battery subsystem which calculates the
discharge and charge power limits of the battery pack at every time step. Battery
discharge power limit (BDPL) is the maximum power that the battery can supply at a
given instant. Similarly, battery charge power limit (BCPL) is the maximum power that
the battery can accept at a given instant. The purpose of this calculation is to limit the
traction motor power by accounting for the power available from the battery and the fuel
cell, and subtracting the power needed by the accessory load at a given time step
(equation 2.4).
38
,max ,max( ) ( ) ( ) ( )Traction Battery FC AccessoryP t P t P t P t= + − (2.12)
The limits are important because they restrict the power request in the simulation to only
the range of values that the on-board power sources are capable of providing. For
example, at 100% state-of-charge the battery cannot accept power from regenerative
braking. The BCPL ensures that the battery does not absorb any power under such a
condition.
The limits are calculated as follows:
If min,max
int,
OCBatt
dis
V VI
R
− ≥
, ( ),max int, ,maxOC Batt dis BattBDPL V I R I= − × (2.13)
Else, minmin
int,
OC
dis
V VBDPL V
R
−= ×
Similarly,
If max,min
int,
OCBatt
ch
V VI
R
− ≥
, ( ),min int, ,minOC Batt ch BattBCPL V I R I= + × (2.15)
Else, maxmax
int,
OC
ch
V VBCPL V
R
−= ×
The voltage and current limits (max min max min, , ,V V I I ) are constants and are generally
provided by the manufacturer. For our batteries the values used are 370 V, 240 V, 300 A,
and -300 A, respectively.
It has been observed that the charging reaction in Nickel Cadmium batteries at
high current and high SOC is accompanied by a rise in its internal impedance, and the
initiation of a side reaction resulting in the evolution of oxygen. This occurs due to mass
transportation limitations inside the cell that limits the charge acceptance rate of the
battery. In such a situation, the energy recovered from regenerative braking is also
39
limited. The phenomenon is generally observed at SOCs higher than 0.75. Since the
model does not capture this phenomenon, all the simulations have been carried out with
an initial SOC of 0.75. However, the complete SOC range can be employed while
simulating other battery chemistries such as lithium–ion due to the absence of the side
reaction phenomenon. The model used for simulating the other battery chemistries
remains the same, except that the battery specifications would differ (OCV, resistance,
voltage, and current limitations).
2.3.3 Ultracapacitor
An ultracapacitor subsystem has been added to the Simulink model in the current work in
order to expand the simulation capability, and simulate hybrid systems consisting of a
battery, ultracapacitor, or a combination of both. Chapter 4 presents a study of battery
stress reduction in a fuel cell/battery/ultracapacitor series hybrid vehicle.
The ultracapacitor is also modeled as a voltage source in series with a resistance.
ocV V IR= − (2.12)
oc
QV
C= (2.13)
00( )
tI dt
SOC t SOCQ
= − ∫ (2.14)
where ocV is the open circuit voltage, R is a constant internal resistance, Q is the charge
and C is the capacitance of the ultracapacitor pack. Similar to the battery subsystem, the
ultracapacitor subsystem incorporates a subroutine to calculate the charge and discharge
power limits. This is particularly useful for the study in Chapter 4 where small
ultracapacitors are used which frequently reach their power limits during a typical drive
40
cycle. The ultracapacitor data are stored in Excel spreadsheets and loaded with the rest of
the data before running the simulation.
2.3.4 Hybrid Controller
The most important role of the hybrid controller is to decide the traction power request.
The hybrid controller houses a Driver or Cockpit subsystem where the force required to
propel the vehicle at the target drive-cycle speed is calculated.
( ) ( ) ( )( )2 11sin tan
2Total D rrF ma C AV mgC V mg gradeρ − = + + +
(2.16)
where m is the vehicle mass, a is the acceleration, DC is the drag coefficient, A is the
vehicle frontal area, V is the velocity, and rrC is the rolling resistance coefficient. The
terms constituting the required force are acceleration, air drag, rolling resistance, and
inclination, respectively. The calculated force is propagated through different subsystems
to the power sources as the electrical power requirement. The hybrid controller also
determines the amount of mechanical braking required if the available regeneration
power exceeds the acceptance limit of the energy storage (battery, ultracapacitor, or
both). Another important function of the hybrid controller is to calculate the fuel cell
power request according to the power management strategy used in the simulation. For
simulating our Phase 1 bus, for example, the controller takes SOC and vehicle power
requirement inputs to compute the one-hour average load and the SOC correction term.
Modifications to the hybrid controller from the current work are described next.
41
Additions
The force calculation at the cockpit is the starting point of the simulation. Correct
calculation of the force and its propagation through the hybrid power train is crucial for
accuracy in power requests at the power sources. Accordingly, the mass of the vehicle
was measured at a weighing station and corrected in the simulation. For simulating a
route traced by the bus, the drive cycle (velocity) data are obtained from an onboard GPS
device. However, the GPS data do not include elevation, and hence cannot be used to
calculate the inclination of the route. Therefore, an accelerometer was installed within the
vehicle, and the accelerometer data were used to calculate the inclination of the route as
follows:
1 1sin accel
dVa
g dtθ − = −
(2.17)
The inclination data derived from the accelerometer data were used during validation on
a more recent vehicle test drive and is presented in section 2.4. Also, the inclusion of an
energy storage charge limit is expected to improve the calculation of the amount of
braking energy available for regeneration because regenerative energy is occasionally
limited by the charge acceptance capacity of the battery.
In order to validate the simulator performance with the vehicle test data, the fuel
cell power request had to undergo a few modifications in order to replicate the actual
vehicle. Specifically, the time period for the average power calculation and the
proportionality constant in the SOC correction term were changed to the values used by
the bus #1. The earlier version used a 10-minute average power instead of one-hour
average, and a value of 80000 for the proportionality constant which was corrected to
30000. The difference between the two values is significant enough to cause major errors
42
in the predicted energy contributions of the fuel cell and battery. Consequently, the fuel
cell parameters have shown close agreement with the vehicle test data and will be
discussed in section 2.4 Also, new power management strategies have been added to the
embedded MATLAB functions within this subsystem which have been used for
simulation studies detailed in Chapters 3 and 4.
2.3.5 Power Combiner
This subsystem is responsible for distributing the vehicle (traction + accessory) power
demand to the fuel cell stack and the battery. In the presence of multiple energy storage
devices such as a battery and ultracapacitor, the subsystem is modified to distribute the
power request according to the governing strategy to all the respective power sources.
2.3.6 Accessory Load
In the previous version of LFM, an oversimplified model of the BOP was lumped with
the accessory load. The BOP component has been deleted in the new version and the rest
of accessory load has been retained as before.
The accessories in the original model comprised of the vehicle air compressor,
hydraulic pump, battery chiller pump, battery chiller compressor, 12V accessories, and
air conditioning, and have been modeled as a constant average load. Air conditioning and
the battery chiller compressor constitute the bulk of the accessory load. Air conditioning
is optional and can be turned on or off before running the simulation. The battery chiller
compressor is operated intermittently to cool the NiCd battery pack. The compressor is
turned on and off when corresponding predefined threshold temperatures are reached.
43
The chiller compressor power requirement is assumed as a constant value and is based on
its duty cycle.
2.3.7 Motor
The traction motor subsystem receives a positive or negative torque request based on the
drive cycle requirement at a particular instant. The torque request is bounded within the
specified limits and then used to calculate the power requirement (torque multiplied by
angular speed) and the expected losses based on the supplied value of motor efficiency.
2.3.8 Transmission
The transmission in the vehicle employs a single gear ratio, and therefore this subsystem
is responsible for performing two very simple calculations. It scales the torque request
and angular velocity at the motor and wheel end based on the gear ratio. It also calculates
the torque loss due to transmission losses using the supplied efficiency value.
2.3.9 Wheels/Chassis
This is the final subsystem in the model which receives the torque input at the wheels
from the traction motor and transmission, and calculates the acceleration of the vehicle as
well as updates the vehicle velocity and wheel angular speed. The vehicle acceleration is
given as:
( ) ( )( )2 11sin tan
2input
D rrwheel
Torquema C AV mgC V mg grade
rρ − = − − −
(2.18)
44
2.4 Validation
As mentioned earlier, the simulator can be used as a useful tool for various design and
optimization studies. However, it is crucial to ensure the validity of the simulator before
making decisions based on its predictions. Therefore, the simulation development is
followed by a validation exercise using the Phase 1 fuel cell hybrid bus described in
section 1.2 as the test vehicle.
The validation procedure begins by driving the test vehicle on a predefined route
to gather drive cycle information and powertrain data using a variety of onboard sensors.
The same drive cycle and initial conditions are then provided as input to the LFM
simulator and the simulation is executed. Key outputs from the simulator are analyzed
and compared with the vehicle data. The validation exercise has been conducted by
comparing LFM results with vehicle data on two distinct drive cycles.
Drive Cycle 1
During the first test run the vehicle was driven on the UD Express Route for a total of 16
miles and 1.5 hours (figures 2.7 and 2.8).
45
Figure 2.7 Aerial view of UD Express Route
46
0 1000 2000 3000 4000 50000
2
4
6
8
10
12
14
16
18
Time (s)
Spe
ed
(m
/se
c)
Figure 2.8 Drive Cycle profile of UD Express Route
Output parameters obtained from the simulation are plotted with the corresponding real
time test data from the vehicle. Figure 2.9 indicates that the battery SOC starts at the
same initial point in the actual vehicle and in the simulation. Since the vehicle runs solely
on the battery at the start of the drive cycle a steady drop in SOC is observed. As soon as
the SOC reaches the threshold value of 0.65 the fuel cell stack is turned on both in the
actual vehicle and in the simulation. From then on the controller calculates the fuel cell
power request and sustains the battery SOC.
47
0 1000 2000 3000 4000 5000
0.65
0.7
0.75
0.8Battery SOC
0 1000 2000 3000 4000 50000
5
10
15
Fuel Cell Gross Power (kW)
Simulation OutputVehicle Test Data
0 1000 2000 3000 4000 5000
0
5
10
Fuel Cell Net Power (kW)
0 1000 2000 3000 4000 50000
0.2
0.4
0.6
0.8Hydrogen Consumption (kg)
Figure 2.9 Comparison of simulation output and vehicle data for Drive Cycle 1
48
Good agreement between the simulated and real SOC throughout the drive cycle reflects
the reasonably accurate simulation of vehicle powertrain and ensures that the fuel cell
stack turned on at about the same time. This in turn ensures correct prediction of
contributions of the stack and battery towards meeting vehicle’s energy requirement at
the end of the drive cycle. Also, a close match between reality and simulation with regard
to fuel cell net power and hydrogen consumption validates the modifications in fuel cell
system model and specifications used. The corrected fuel cell power request as discussed
in section 2.2.4 also contributes to the reasonably accurate results. The comparison can
also be observed in Table 2.1 which quantifies the errors with respect to important
vehicle output parameters. A hybrid vehicle is a fairly complicated system and there are
many sources of errors involved within a vehicle simulator. Therefore error values within
10 % are quite acceptable. The good agreement between traction and regenerative energy
is due to the changes made within the hybrid controller.
Table 2.1 Quantitative Comparison corresponding to Drive Cycle 1
Output Parameters Vehicle
Data
Simulation
Output Error (%)
Battery Energy (Wh) 5885 5728 2.7
Fuel Cell Net Energy (Wh) 10050 9191 8.5
Fuel Cell Gross Energy (Wh) 12250 11713 4.4
Traction Energy Input (Wh) 18423 17267 6.3
Energy Recovered (Regenerative Braking) (Wh)
6207 6857 10.5
Battery State of Charge Drop 0.126 0.13 2.9
Hydrogen Consumed (Kg) 0.7084 0.6786 4.2
49
Drive Cycle 2
During the second test run the vehicle made six trips of a selected route and drove a total
of 24 miles for 100 minutes (Figure 2.10 & 2.11).
Figure 2.10 Aerial view of second test run
50
0 1000 2000 3000 4000 5000 60000
2
4
6
8
10
12
14
16
18
20
Time (secs)
Spe
ed (
m/s
ec)
Test Drive Cycle (100 minutes, 24 miles)
Figure 2.11 Drive Cycle profile of second test run
The second drive cycle involves a different power management strategy. Both the
simulation and real data in figure 2.12 show a decline of battery SOC from an initial state
of 0.6. At a threshold SOC value the fuel cell is turned on and a constant power request of
is sent to the stack. Similarly, the variation of battery SOC and fuel cell parameters in
figure 2.12, and quantitative comparison of key vehicle parameters in table 2.2 validate
the simulator output.
51
0 1000 2000 3000 4000 5000 60000.35
0.4
0.45
0.5
0.55
Time (s)
Battery State of Charge
0 1000 2000 3000 4000 5000 6000
0
2
4
6
8
10
12
Time (s)
Fuel Cell Gross Power (kW)
Simulation ResultVehicle Test Data
0 1000 2000 3000 4000 5000 6000
-2
0
2
4
6
8
10
Time (s)
Fuel Cell Net Power (kW)
0 1000 2000 3000 4000 5000 6000
0
0.2
0.4
0.6
0.8
1
Time (s)
Hydrogen Consumption (kg)
Figure 2.12 Comparison of simulation output and vehicle data for Drive Cycle 2
52
Table 2.2 Quantitative comparison corresponding to Drive Cycle 2
Output Parameters Vehicle
Data
Simulation
Output Error (%)
Battery Energy (Wh) 9493 10116 6.6
Fuel Cell Net Energy (Wh) 13503 13584 0.6
Fuel Cell Gross Energy (Wh) 17186 16952 1.4
Traction Energy Input (Wh) 26356 26264 0.3
Energy Recovered (Regenerative Braking) (Wh)
7877 7408 6
Battery State of Charge Drop 0.198 0.1885 4.8
Hydrogen Consumed (Kg) 0.9063 0.2 5.8
One of the major sources of error in the hybrid simulator is the force calculation
at the cockpit in the hybrid controller subsystem which uses a basic and commonly used
formula (equation 2.16). In order to enhance accuracy, an advanced model for drag and
frictional force calculation is needed along with precise drag and friction coefficient
values. Slight inaccuracies in drive cycle information obtained from the GPS are another
source of error. In the earlier part of the chapter the importance of component
specifications and physics-based modeling was emphasized. Some of the deviations
between simulation results and vehicle data could be reduced with detailed component
maps and advanced models. For example, the simulator uses a constant value of traction
motor efficiency which in reality depends upon the motor torque and speed. A detailed
torque, speed, efficiency map can improve the prediction of traction motor electrical
power demand.
53
There is definitely scope to achieve higher accuracy with the power train
simulator, but the level of accuracy achieved during this exercise for system level
simulations is very reasonable, and the simulator can be used for further optimization
studies in this modified form (Chapters 3 and 4).
2.5 Summary
This chapter presented the LFM simulator and identified avenues of improvement. The
individual subsystem models were discussed and major additions and improvements were
highlighted. In particular, the fuel cell, battery, and hybrid controller were revisited. The
data pertaining to different subsystems were updated according to the observations from
real time data. The simulator was then validated against two test drive cycles. The
comparison between simulator output and vehicle data demonstrated good agreement
between the two. The vehicle energy requirements (traction + accessory) were predicted
with reasonable accuracy. In particular, by virtue of the extensive fuel cell system model,
the simulator showed good results with respect to fuel cell parameters (net power, gross
power, voltage, current, and fuel consumption). The possible sources of errors were
discussed and it was concluded that LFM can be used as a reliable tool for design and
optimization studies.
54
3 Prediction-based optimal power management in a fuel
cell/battery plug-in hybrid vehicle
3.1 Introduction
This chapter presents a prediction-based power management strategy for fuel cell/battery
plug-in hybrid vehicles with the goal of improving overall system operating efficiency.
The main feature of the proposed strategy is that, if the total amount of energy required to
complete a particular drive cycle can be reliably predicted, then the energy stored in the
onboard electrical storage system can be depleted in an optimal manner that permits the
fuel cell to operate in its most efficient regime. The strategy has been implemented in
LFM and its effectiveness was evaluated by comparing it with a conventional control
strategy. A sensitivity analysis has also been conducted to study the effects of inaccurate
predictions of the remaining portion of the drive cycle on hydrogen consumption and the
final battery state-of-charge. Finally, the advantages of the proposed control strategy over
the conventional strategy have been validated through implementation in the University
of Delaware’s fuel cell hybrid bus with operational data acquired from on-board sensors.
Power management strategies have been a subject of study in both fuel cell and IC
engine hybrids. Rodatz et al. [7] proposed a control strategy (equivalent consumption
minimization strategy) to determine the real-time optimal power distribution. Peng et al.
[8] formulated a combined power management/design optimization approach and
proposed a parameterizable and near-optimal controller for power management
optimization using a stochastic dynamic programming algorithm. Paladini et al. [6]
performed an optimization of vehicle configuration and control strategy to minimize
55
hydrogen consumption while sustaining battery state-of-charge. Paladini et al. [9] have
performed control strategy optimization for charge-sustaining operation of batteries and
have reported good fuel economy and final battery state of charge (SOC) for a fuel
cell/battery hybrid system. These proposed control strategies are aimed towards fuel
savings for a charge-sustaining operation in Hybrid Electric Vehicles (HEVs).
This chapter is aimed at optimizing the control strategy for a charge-depletion
operation while maintaining safe power requests to the fuel cell. As stated earlier, the
objective of the charge-depletion operation is to exploit the energy stored in the onboard
electrical storage system in an optimal manner such that the fuel cell is able to operate in
its most efficient regime. In addition to efficiency, any power management strategy must
also maintain operating conditions that prolong the life of the fuel cell system.
Specifically, it is well known that the transient nature of the power load can influence
fuel cell durability and its long-term performance. For example, Kusoglu et al. [10] have
shown that the proton exchange membrane can undergo compressive, plastic deformation
due to hygrothermal loading, resulting in residual tensile stresses after unloading. These
residual in-plane stresses in the membrane may explain the occurrence of cracks and
pinholes in the membrane under cyclic loading. Pei et al. [11] have studied the effects of
four different kinds of operating conditions on the fuel cell and have concluded that 56%
of deterioration is due to load-change cycling and 33% due to start-stop cycling.
Furthermore, frequent exposure of the cells to high voltages typical of open circuit
conditions can accelerate membrane and catalyst degradation [12]. It is therefore
desirable that the hybrid controller sends a stable power request to the fuel cell stack and
avoids frequent load changes and multiple starts and stops of the stack.
56
The primary factors that affect the life of a battery pack are storage conditions,
charge and discharge control, and depth-of-discharge. Fast charge and discharge are
inevitable when the batteries operate within an automotive drivetrain. The permissible
depth-of-discharge and hence the available energy density is an important factor that
decides the suitability of batteries in HEVs and Plug-in Hybrid Electric Vehicles
(PHEVs). Some batteries, such as NiMH, are suitable for powering HEVs in which the
energy from the fuel is used to keep the batteries charged up. In such applications the
battery cycle life is conserved by cycling to shallow depths-of-discharge. This mode of
operation is termed as charge sustaining. For application in plug-in hybrid vehicles,
batteries must be deep-discharge, long cycle-life batteries [13]. Recent advancements in
Li-ion technology have led to the development of Lithium-titanate batteries which have
higher energy density, more than 12000 cycles (at 100% depth-of-discharge) and life
expectancy of 20 calendar years [14] and thus are quite suitable for use in plug-in
hybrids. The Nickel Cadmium (NiCad) battery, if cycled to a certain shallow depth-of-
discharge for a large number of cycles may not yield a storage capacity as large as that
corresponding to normal discharge-charge cycles [15,16]. A phenomenon known as
“memory effect” occurs due to a sudden depression of voltage as a result of highly
repetitive patterns of use [16]. While the effect is completely reversible, it requires a
dedicated and lengthy maintenance schedule [17]. It has therefore been found that it is
best to discharge the NiCads as deeply as possible at the end of the drive cycle, followed
by slow recharge to 100% state-of-charge thus reducing the need for maintenance cycles.
Therefore, despite a limited cycle life (1200 cycles) this renders the NiCads suitable for
use in PHEVs.
57
The following sections describes the analysis, implementation and validation of a
prediction-based power management strategy that reduces fuel consumption while
managing power flow in a manner that promotes fuel cell stack life and performance,
while depleting the battery to a desired state-of-charge at the end of the drive cycle. The
main feature of the proposed strategy is that, if the total amount of energy required to
complete a particular drive cycle can be reliably predicted, then the energy stored in the
battery pack can be depleted in an optimal manner that permits the fuel cell to operate in
its most efficient regime.
The following sections describe the methodology and algorithm of the proposed
strategy, LFM simulation results including a sensitivity analysis, and validation of the
simulation results by an actual implementation of the proposed strategy in our first fuel
cell bus.
3.2 Power Management Strategy
Power flow from onboard energy sources has to be managed in order to maintain the
battery SOC at a desired level. It is assumed that the battery is charged to a state of 0.75
at the start of a drive cycle in our LFM simulations. It has been observed that at SOCs
higher than 0.75, the charging reaction in the NiCad battery is accompanied by the
initiation of a side reaction and a limited ability to recover energy due to regenerative
braking. The LFM simulator does not model this phenomenon and hence, the initial SOC
is set to 0.75. Ordinarily, a charge depletion operating mode can be achieved by driving
all electric until the battery is depleted to the desired SOC, followed by turning on the
fuel cell system to sustain the battery at the desired SOC. This power management
58
strategy, denoted as the baseline strategy, is depicted in Figure 3.1. The following
relations hold for this mode,
Fuel cell turn-on condition: ( ) dSOC t SOC≤
Fuel cell power request: ( ) ( ( ))avg dP t P SOC SOC tα= + − (3.1)
where avgP is the power consumption of the traction motor and accessory load combined,
averaged over a moving time frame (one hour in this case), dSOC is the SOC to which
the battery is desired to be depleted, and α is a constant in the correction term which
alters the power request based on the deviation of the real time SOC from the desired
value. The value of α used in the current simulations is 600,000 W. Hence, if the SOC
differential is 1% for example, then the fuel cell power request is incremented by 6 kW
over avgP . The overall performance of this strategy is relatively insensitive to the value of
α. For instance, the only effect resulting from a smaller α would be somewhat larger
fluctuations in the subsequent time trace of SOC because the fuel cell would take longer
to restore the SOC to the desired value.
59
0 1000 2000 3000 4000 5000 6000 7000 80000.2
0.3
0.4
0.5
0.6
0.7
Time (secs)
Ba
ttery
Sta
te o
f Ch
arg
e
0 1000 2000 3000 4000 5000 6000 7000 8000
0
10
20
30
Time (secs)
Fu
el C
ell
Ne
t Po
we
r (k
W)
Baseline strategyPrediciton based strategy
Figure 3.1 Battery SOC drop and fuel cell net power corresponding to the baseline and
the predictive control strategy for SC03 (~2 hours, 46 miles)
Such a power management strategy suffers from a lack of control over the
operating point of the fuel cell stack. For example, referring to Figure 3.1, it is possible
that when the fuel cell needs to be turned on, the fuel cell power request is higher
thanmax
Pη , the value at which the fuel cell efficiency is maximized. This is because the
power request to the stack is essentially governed by the average power demand of the
drive cycle and the deviation of the battery SOC from the desired level. Consequently,
this baseline power management strategy does not yield the highest possible fuel
efficiency as the fuel cell will be operating at lower efficiency. We will use this baseline
60
strategy as a benchmark to compare the results from the prediction-based strategy which
can deliver higher efficiencies as proposed below.
Transit buses have been the most widely chosen platforms for fuel cell technology
demonstration for a number of reasons as outlined in [18]. The proposed prediction-based
power management strategy uses a priori knowledge of the driving route that would be
typically available in transit applications and hence is particularly well suited for transit
buses. This information can be exploited to manage power flow from onboard energy
sources and achieve the following objectives:
• Operate the fuel cell stack in an efficient zone.
• Reduce fuel consumption.
• Send a smooth power request to the stacks and operate them without multiple
starts and stops or frequent load changes.
• Discharge the battery to a desired state-of-charge at the end of the drive cycle.
3.3 Methodology and Algorithm
The key to meeting the objectives stated above, is the knowledge of the expected net
energy, ,fc netE , required from the fuel cell stack, which will also be referred to as the
predictive parameter in this chapter. This can be achieved either with the help of
simulation software and a priori knowledge of the drive cycle or from data acquired in
real time during an excursion of the drive cycle. Now, the ideal way to meet this energy
demand is to draw net power from the fuel cell system such that the stack functions at
peak efficiency. This logic is implemented in the prediction-based strategy by
determining the stack turn-on time and net power request as outlined in the following
61
algorithm. It should be noted that battery also contributes to the energy requirement of
the vehicle. However, only the fuel cell energy is considered in the equations because the
goal is to maximize operating efficiency of the fuel cell system.
The fuel cell stack is turned on and continues to operate the moment the following
condition is met:
max
,fc netcycle corr
Et T
Pη
δ
≥ − +
(3.2)
The power request is given by
,fc net
cycle turn on corr
EPower Request
T t δ=
− − (3.3)
If the battery SOC reaches dSOC at any point during the drive cycle, the battery is
operated in charge-sustaining mode for the rest of the drive cycle as has been discussed
while introducing the baseline approach. The net power request and implementation
condition is given by
( ( )) If ( )avg d dPower Request P SOC SOC t SOC t SOCα= + − ≤ (3.4)
where t is the current time
turn ont denotes the time when the stack is turned on
max
Pη is the net fuel cell power corresponding to maximum system efficiency
,fc netE is the energy requirement from the fuel cell for the duration of the drive
cycle
corrδ is a correction time to start the stack earlier so as to account for the deficit in
power supply during ramp up and is equal to half of the ramp up time
62
cycleT is the total duration of the drive cycle
The term max
,fc netcorr
E
Pη
δ
+
denotes the time for which the stack should be operated with a
net power supply of max
Pη to meet the energy requirement,fc netE . The conditions stated in
Equations (3.2) and (3.3) can be understood by considering three cases that arise. They
are
Case 1: max
,fc netcycle corr
ET
Pη
δ
> +
implies that the duration for which the stack needs to
operate is less than the total duration of the drive cycle. As the drive cycle progresses,
time t increases from 0 (at the start) until it reaches the value
max
,fc netturn on cycle corr
Et T
Pη
δ
= − +
which is when the stack turns on and continues to operate
till the end of the drive cycle. Substituting for turn ont in Equation (3.3) we
obtainmax
Power Request Pη= . This is exactly the desired objective.
Case 2: max
,fc netcycle corr
ET
Pη
δ
= +
implies that the duration for which the stack needs to
operate is equal to the duration of the drive cycle. Therefore,
max
, 0fc netturn on cycle corr
Et T
Pη
δ
= − + =
and max
Power Request Pη= .
63
Case 3: max
,fc netcycle corr
ET
Pη
δ
< +
implies that the duration for which the stack needs to
operate is greater than the duration of the drive cycle. The earliest the stack can start is at
the beginning of the drive cycle, 0t = . This condition is enforced by the inequality of
Equation (3.2). An obvious deduction is that the energy requirement ,fc netE is met by
drawing net fuel cell power which is higher thanmax
Pη and is given by Equation (3.3) with
turn ont = 0.
It should be noted that the implementation of charge-sustaining operation
(equation 18) ensures that the stack is operating at required power the moment the battery
state of charge drops down to dSOC thus safeguarding against the danger of draining the
battery completely due to a delayed turn on time, obtained from the condition specified in
Equation (3.2). Such a miscalculation in stack turn on time can result from inaccurate
prediction of ,fc netE and will be discussed in the following sections.
3.4 Simulation Results
The proposed power management strategy has been implemented in the LFM simulation
software and compared with the baseline approach for drive cycles of different lengths
which have been created by simply repeating the standard cycle multiple times as shown
in Figure 3.2. Figure 3.1 demonstrates the difference between the predictive strategy and
the baseline approach for the dual stack bus. Based on the prior information of net energy
requirement from the fuel cell, it can be seen that the fuel cell stack was turned on at an
earlier time within the drive cycle such that the power requirement corresponds to the
64
maximum efficiency point of the fuel cell system. The earlier start time of the stack
results in a slower rate of SOC drop from the moment the stack begins to operate.
0 1000 2000 3000 4000 5000 6000 7000 80000
5
10
15
20
25
Time (secs)
Spe
ed
(m/s
ec)
SC03 (~2hours, 46 miles)
0 1000 2000 3000 4000 5000 6000 7000 80000
5
10
15
20
25
30
Time (secs)
Spe
ed (m
/sec
)
UDDS (~2hours, 45 miles)
Figure 3.2 Longer drive cycles formed by repeating standard cycles
Fuel consumption, average fuel cell operating efficiency, and final battery state-
of-charge are reported in Tables 1 and 2. A comparison of average fuel cell operating
efficiency between the two control strategies indicates that prediction-based power
management allows the stack to operate in a more efficient regime thereby reducing fuel
consumption. The final battery SOC is within 3% of the desired value (0.3). The extent of
fuel savings is evidently dependent upon the average operating efficiency of the fuel cell
system with the baseline control strategy. For example, the average efficiency
corresponding to the SC03 (supplemental drive cycle for number 3 for federal test
65
procedure) driving schedule is 41.5% as opposed to 44.5% for UDDS (Urban
Dynamometer Driving Schedule). This explains the relatively higher fuel savings when
the new power management strategy is applied to SC03.
Table 3.1 Comparison of prediction-based and baseline strategy for SC03 as shown in
Figure 3.2
Drive
Cycle
Length
Output Parameters
Prediction
Based
Strategy
Baseline
Strategy
Fuel
Savings (%)
Hydrogen Consumption (kg) 1.5855 1.8362 13.65
Average FC System Efficiency (%) 47.62 40.23 ~2 Hours
46 miles Final Battery SOC 0.3003 0.2931
Hydrogen Consumption (kg) 3.1866 3.6466 12.32
Average FC System Efficiency (%) 47.71 41.29 ~3 Hours
68 miles Final Battery SOC 0.298 0.2935
Hydrogen Consumption (kg) 6.4971 7.2621 10.53
Average FC System Efficiency (%) 47.03 41.85 ~5 Hours
111 miles Final Battery SOC 0.295 0.294
Hydrogen Consumption (kg) 9.9537 10.8776 8.49
Average FC System Efficiency (%) 46.13 42.03 ~7 Hours
154 miles Final Battery SOC 0.2938 0.2935
It should be noted that the preceding results have been generated using the same
drive cycle which was also used for obtaining the parameter ,fc netE . Therefore, the
predicted value of net fuel cell energy is identical to the actual value. However, in reality
two realizations of the same route could lead to different drive cycles (velocity vs. time
profile) due to factors that cannot be completely predicted such as instantaneous traffic
66
conditions and ridership. Consequently, the net energy delivered by the fuel cell stack
during one excursion on a chosen route may differ from the value obtained during a
different excursion on the same route. These variations can lead to an inaccuracy in the
predicted parameter and its effect has been studied by means of a sensitivity analysis in
the following section.
Table 3.2 Comparison of prediction-based and baseline strategy for UDDS as shown in
Figure 3.2
Drive
Cycle
Length
Output Parameters
Prediction
Based
Strategy
Baseline
Strategy
Fuel
Savings
(%)
Hydrogen Consumption (kg) 1.3383 1.4309 6.47
Average FC System Efficiency (%) 47.69 44.49 ~2 Hours
45 miles Final Battery SOC 0.3033 0.3007
Hydrogen Consumption (kg) 2.3946 2.5724 6.91
Average FC System Efficiency (%) 47.8 44.48 ~3 Hours
60 miles Final Battery SOC 0.3037 0.3009
Hydrogen Consumption (kg) 5.5909 5.9937 6.72
Average FC System Efficiency (%) 47.86 44.49 ~5 Hours
104 miles Final Battery SOC 0.3099 0.3009
Hydrogen Consumption (kg) 8.292 8.8448 6.25
Average FC System Efficiency (%) 47.62 44.49 ~7 Hours
142 miles Final Battery SOC 0.3097 0.3009
Sensitivity Analysis:
The inconsistency in ,fc netE can be modeled by varying the prediction parameter
corresponding to a given drive cycle and then using the modified value in the predictive
control strategy for the same drive cycle. The effect of such an inaccuracy has been
67
studied by varying the parameter by the following percentages (-15%, -10%, -5%, 5%,
10%, 15%) to reflect different degrees of inaccuracy. The modified value is then inserted
into by the prediction-based strategy in order to calculate fuel cell turn-on time and
determine the power request. Modifying,fc netE by -x% implies that we are intentionally
under predicting the parameter value such that it is smaller than the correct value by x%.
Modifying ,fc netE by -15% results in under prediction of fuel cell net energy
required to execute the chosen drive cycle. Consequently, the fuel cell turns on later than
it should and the battery depletes to the desired SOC before reaching the destination as
shown in Figures 3.3 and 3.4.
0 1000 2000 3000 4000 5000 6000 7000 80000.2
0.3
0.4
0.5
0.6
0.7
Time (secs)
Ba
ttery
Sta
te o
f Ch
arg
e
Under Prediction -15%Over Prediction 15%Accurate PredictionBaseline Control
0 1000 2000 3000 4000 5000 6000 7000 8000
0
1
2
3
x 104
Time (secs)
Fu
el C
ell
Ne
t Pow
er (
W)
Figure 3.3 Deviation in battery SOC drop and fuel cell net power corresponding to
inaccuracy in prediction for the SC03 (~2 hours, 46 miles)
68
On reaching the desired SOC the strategy switches to charge-sustaining mode in
accordance with the control algorithm such that the net energy supplied by the fuel cell is
still equal to the original, unscaled, ,fc netE value. However, the average operating
efficiency decreases because, late in the cycle, the fuel cell is required to produce power
at a higher rate at which its efficiency is lower than the maximum possible efficiency.
Similarly, scaling ,fc netE by 15% results in an over prediction of fuel cell net energy. But,
unlike under prediction, in case of an over prediction, the net energy supplied by the
stack is greater than required. Consequently the terminal battery SOC stays higher than
0 1000 2000 3000 4000 5000 6000 7000 80000.2
0.3
0.4
0.5
0.6
0.7
Time (secs)
Ba
ttery
Sta
te o
f Ch
arg
e
Under Prediction -15%Over Prediction 15%Accurate PredictionBaseline Control
0 1000 2000 3000 4000 5000 6000 7000 8000
0
1
2
3
x 104
Time (secs)
Fu
el C
ell
Ne
t Pow
er (
W)
Figure 3.4 Deviation in battery SOC drop and fuel cell net power corresponding to
inaccuracy in prediction for the UDDS (~2 hours, 45 miles)
69
the desired SOC and fuel savings decline (Figures 3.3 and 3.4). In both cases of
inaccurate drive cycle predictions, it is of interest to analyze fuel savings with respect to
the baseline control strategy which is shown in Figures 3.5 and 3.6.
A decrease in the magnitude of fuel savings is observed for increasing degree of
under prediction. For a 74 km (46-mile) SC03 drive cycle, for example, the savings are
reduced to 11.39 % for an under prediction of -15 % as opposed to 13.65 % for accurate
prediction (Figure 3.5). The reason, as has been stated earlier, is attributed to a decrease
in average operating efficiency of the fuel cell system. A similar trend is observed for
~2hours, 46 miles ~3hours, 68 miles ~5hours, 111 miles ~7hours, 154 miles-10
-5
0
5
10
15
Fu
el S
avin
gs
(%)
~2hours, 46 miles ~3hours, 68 miles ~5hours, 111 miles ~7hours, 154 miles0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Fin
al B
atte
ry S
OC
-15% UP -10% UP -5% UP AP 5% OP 10% OP 15% OP
Figure 3.5 Fuel savings and final battery SOC for varying degree of inaccurate predictions and for
variable drive lengths for the SC03 driving schedule
UP - Under Prediction, AP – Accurate Prediction, OP – Over Prediction
70
drive cycles of increasing lengths. However, the key inference from this part of study is
that, the fuel savings are still positive; i.e. there is still an overall reduction in hydrogen
consumption as compared to the baseline strategy while maintaining the battery SOC
close to the desired level (within 3%). As expected, the magnitude of improvement
diminishes with increasing amounts of under predictions.
Increasing the degree of over prediction also results in a decline in fuel savings.
However, in this case, the decline occurs because the fuel cell provides more energy than
what is required with the result that the battery is not discharged to the desired level. For
~2hours, 45 miles ~3hours, 60 miles ~5hours, 104 miles ~7hours, 142 miles-10
-5
0
5
10
Fu
el S
avin
gs
(%)
~2hours, 45 miles ~3hours, 60 miles ~5hours, 104 miles ~7hours, 142 miles0
0.2
0.4
0.6
0.8
Ba
ttery
Fin
al S
OC
-15% UP -10% UP -5% UP AP 5% OP 10% OP 15% OP
Figure 3.6 Fuel savings and final battery SOC for varying degree of inaccurate predictions and for
variable drive lengths for the UDDS driving schedule
UP - Under Prediction, AP – Accurate Prediction, OP – Over Prediction
71
a 74 km (46-mile) drive cycle of SC03, the terminal battery SOC is 0.35 for a 15% over
prediction compared to an SOC of 0.3 for an accurate prediction (Figure 3.5). For drive
cycles of greater lengths the terminal battery SOC increases. This not only leads to a
decrease in fuel savings, but may also result in higher fuel consumption compared to the
baseline approach. Hydrogen consumption can be expected to be higher in comparison to
the baseline strategy in the case of over prediction and the probability increases with the
degree of over prediction and the drive cycle length.
It should be noted that for each drive cycle considered in the present work, the
average power required to sustain battery SOC, avgP is greater than max
Pη . This is expected
for cost-effective power source configurations where the fuel cell is down-sized
compared to the battery pack and is just enough to meet the average power requirement
of urban transit drive cycles [18]. If however, avgP is less thanmax
Pη , the situation always
degenerates to the baseline control strategy as depicted in Figure 3.7.
72
Figure 3.7 Possible SOC profiles corresponding to the condition maxavgP Pη<
Trajectory ADC shows the variation of SOC with time for maxavgP Pη< if a prediction-
based strategy is followed without enforcing the charge-sustaining mode at dSOC .
Evidently, the SOC reaches the desired level at B before the turn-on time at D as
calculated by Equation 16. Since it is not desirable to let the SOC fall below dSOC , the
charge-sustaining mode comes into effect at B which implies no fuel savings as the fuel
cell power is below the level at which efficiency is maximized. An alternative approach,
depicted by trajectory ABEC is to turn on the stack when the desired SOC is reached (at
B) and draw ,fc netE amount of energy at max
Pη before shutting it down (at E). In this
manner, the stack can be operated at peak efficiency with additional savings in fuel.
B
A
D C
E
iSOC
dSOC
ABC – Baseline Strategy
73
3.5 Validation
Both the prediction-based and baseline power management strategies were evaluated by
implementing them on the University of Delaware’s fuel cell/battery hybrid bus. The
vehicle selected for this test was UD’s first fuel cell bus; as described earlier it is
equipped with a single stack rated at 19.4 kW and 60 kWh of NiCad batteries. The test
was conducted by driving the bus on a defined route (Figure 3.8) on two separate days,
first with the baseline control strategy, and next with the prediction-based strategy.
Figure 3.8 Aerial view of the trajectory traced by the fuel cell hybrid bus
74
During each test run the vehicle made six trips on the route and drove a total of 38.6 km
(24 miles) for 100 minutes. The route includes two bus stops and the duration of each
round trip is matched to a typical time-bound transit operation. The drive cycle (Figure
3.9) includes high and low speed segments with an average of 23.3 km/h (14.5 mph).
0 1000 2000 3000 4000 5000 60000
2
4
6
8
10
12
14
16
18
20
Time (secs)
Spe
ed (
m/s
ec)
Test Drive Cycle (100 minutes, 24 miles)
Figure 3.9 Profile of the test drive cycle
The initial and desired SOC were chosen to be 0.6 and 0.4 respectively, which
allowed the control strategies to be tested on a drive cycle of smaller distance and
duration. While operating with the baseline control strategy, the fuel cell was turned on
when the SOC reached 0.41 (Figure 3.10). This allowed for some warm up time so that
the stack could ramp up and provide 13.5 kW of net power in order to sustain the battery
75
at 0.4 SOC. In Figure 3.10 the periodic sharp declines in SOC correspond to high power
demands when the vehicle executes the high speed segment of the drive cycle. On the
other hand, frequent occurrences of SOC rise are attributed to cell charging while the
vehicle is idling at a bus stop or a traffic intersection. The optimal net fuel cell power of
the test vehicle was obtained experimentally as 9 kW with a corresponding fuel cell
system efficiency of 45.9 %.
0 1000 2000 3000 4000 5000 6000
0.35
0.4
0.45
0.5
0.55
0.6Test Drive Cycle
Bat
tery
Sta
te o
f Ch
arg
e
Time (secs)
0 1000 2000 3000 4000 5000 6000-5000
0
5000
10000
15000
Fue
l Ce
ll N
et P
owe
r (W
)
Time (secs)
Baseline StrategyPrediction Based Strategy
Figure 3.10 Battery SOC drop and fuel cell net power corresponding to baseline and
predictive control strategy
The optimal power along with the net energy spent by the fuel cell during the first
run (baseline strategy) was used as an input to determine the stack turn on time for the
second run that employed the prediction-based strategy. Figure 12 shows that for the
76
prediction-based strategy the stack turned on earlier and operated at stable optimal power
for the rest of the drive cycle. A quantitative comparison of the key output parameters
confirms the benefits of using the proposed power management (Table 3). Through
intelligent management of energy flow and with no additional costs, the stack was
operated at higher efficiency resulting in 11.7 % savings in fuel consumption. Moreover,
the stable operation of the fuel cell system also extends the life of the stack. The battery
SOC at the end of the drive cycle was close to the desired lower limit, which is one of the
considerations for plug-in hybrid operation.
Table 3.3 Comparison of prediction-based and baseline strategy for test drive cycle
Output Parameters
Prediction
Based
Strategy
Baseline
Strategy
Fuel Savings
(%)
Hydrogen Consumption (Kg) 0.9063 1.0124 11.7
Average FC System Efficiency (%) 44.7 39.5
Final Battery SOC 0.4115 0.402
3.6 Summary and Conclusions
A new prediction-based power management strategy for fuel cell/battery plug-in hybrids
has been proposed and implemented in the LFM simulation software. Simulation results
for the prediction-based strategy showed significant improvements in fuel cell system
efficiency and reduction in hydrogen consumption compared to a conventional, baseline
strategy of charge sustenance. The importance of a stable power request to the fuel cell
has been stated and realized. A sensitivity analysis was conducted to study the effects of
77
inaccurate predictions. Results indicate that under prediction reduces the magnitude of
fuel savings, and in the borderline case, may show results identical to the baseline
strategy. A large degree of over prediction, on the other hand, may even lead to higher
fuel consumption than the baseline strategy while resulting in a higher terminal battery
SOC than desired. A conservative approach may therefore be adopted by using a
downscaled predicted parameter value, which results in fuel savings that may be less than
the maximum possible but will safeguard against entering into the over predicted zone
and the associated risk of increased fuel consumption. The implementation of the
proposed strategy and its comparison with the baseline control strategy in a fuel cell and
battery powered hybrid bus has confirmed the benefits predicted from simulation studies.
Finally, good agreement between the simulator outputs and data acquired in real time
confirms the validity of the power-train simulator.
78
4 Reduced battery stress through blended energy storage
4.1 Introduction
The estimated lifetime of the battery is an important consideration while designing a
hybrid power train for automotive applications. The objective of the present work is to
use our validated LFM simulation tool to evaluate one approach to reduce battery loads
by adding an ultracapacitor module, and thereby enhance battery lifetime.
Batteries and ultracapacitors are the most commonly used energy storage systems
(ESS) in hybrid electric vehicles. Batteries usually have high energy density but limited
power density, while ultracapacitors (Ucaps) have high power density but low energy
density. Due to these complementary properties, batteries can be combined with Ucaps to
create a lightweight, compact ESS that exhibits a good compromise between energy and
power densities.
Another significant difference between the two systems is their cycle life.
Batteries typically lose their effectiveness after a few thousand charge-discharge cycles.
The best cycle life for commercial battery systems is that of Altairnano’s Lithium-
Titanate cells which have shown up to 12000 cycles (100% depth of discharge) at 2C
charge and discharge, and 25 °C (Table 4.1). Charge and discharge current of a battery is
typically measured in C rates. A current rate of 1C is equal to the current required to fully
charge the battery to its rated capacity in one hour; a current rate of nC is n times the
current at 1C. In contrast, Ucaps are able to maintain performance for about one million
cycles. Table 4.1 compares an advanced technology battery with an Ucap. Storage system
lifetime is therefore another metric which can be enhanced by employing a suitable
combination of the two energy storage systems.
79
Table 4.1 Comparison of advanced technology battery and Ucap
Altairnano (LiTi cell) Maxwell
Ultracapacitor
Peak W/kg* 760 5900
Wh/kg 72 5.96
Cycle Life
>12000 cycles at 100 % DoD** (2C rate and 25 °C)
>4000 cycles at 100% DoD (1C rate and 55 °C)
1 million cycles at 50 % DoD
* The peak powers are calculated based on peak pulse currents of the ESS which may not be allowed by the traction inverter ** DoD is Depth of Discharge
Yang et al. have mentioned that stress factors such as temperature, SOC swing,
current load (C rate), energy throughput, and also calendar time affect the cell
degradation rate [9]. Amongst these factors, the adverse effects of current load (C rate)
and energy throughput can be mitigated by using an Ucap to share the load with batteries.
In the literature, batteries and Ucaps have been considered separately on most occasions
while studying the hybrid powertrain. Gao [20] developed and implemented a fuzzy logic
based energy management on a fuel cell/battery/Ucap hybrid bus. Bauman and Kazerani
[21] performed optimization studies on fuel cell/battery/Ucap vehicle to find the optimal
configuration with respect to acceleration performance, fuel economy, and cost. Blended
energy storage has rarely been studied with the objective of reducing stress on the battery
and improving its lifetime.
The goal of this chapter is to investigate and compare the battery stress for a
battery-only ESS with a blended ESS (battery+Ucap) using our previously validated
LFM simulation tool. The specific objective here is to conduct an analysis to quantify
how a blended ESS relieves the load on the battery and thereby extends its life. We begin
80
by describing the blended ESS topology, energy management scheme, and energy storage
details, followed by the simulation results.
4.2 Blended ESS topology and energy management
The vehicle platform used for this analysis corresponds to the 22-ft UD fuel cell bus
described in chapter 1. The analysis is conducted for a battery-only ESS, followed by a
blended ESS. The hardware and energy management schemes for each are described
next.
4.2.1 Battery-only ESS
The drivetrain topology for a fuel cell/battery series hybrid vehicle is shown in figure 4.1.
As described earlier, this drivetrain corresponds to the UD fuel cell bus. While the ESS
on the bus currently consists of NiCd batteries, the analysis presented in this chapter
employs LiTi batteries. Power from the battery and fuel cell feeds the traction motor and
the accessory load. Note that power flow is bidirectional in the traction motor and battery.
The battery can accept power from either the fuel cell, or the traction motor during
regenerative braking. The fuel cell is rated at 20 kW and the battery pack comprises 144
Altairnano (50Ah Li-Ti) cells (Table 4.2).
81
Figure 4.1 Topology of a fuel cell/battery hybrid
Table 4.2 Battery Description
Altairnano (50 Ah cells)
Number of cells 144
Max/Min Voltage 400/240 V
Max. Current 300 A
Max. Power 120 kW
Available Energy 16.5 kWh
Hybrid Energy Management: The fuel cell net power is given by
, ( )FC net avg d cP P SOC SOCα= + − (4.1)
Traction Motor
Accessory Load
Fuel Cell
Battery
Unidirectional flow
Bidirectional flow
DC/DC Converter
82
where avgP is the combined power consumption of the traction motor and accessory load
averaged over a moving time frame (one hour in this case), dSOC is the SOC to which
the battery is desired to be depleted, and α is a constant in the correction term which
alters the power request based on the deviation of the real time SOC ( cSOC ) from the
desired value.
The battery power is given by
( ),Battery tract acc FC netP P P P= + − (4.2)
where tractP and accP are the power consumption of the traction motor and accessory load,
respectively. Note that Ptract is negative during regenerative breaking.
4.2.2 Blended ESS
The topology of a series hybrid with blended ESS is shown in Figure 4.2. This hybrid
system includes an Ucap module in addition to the battery and the fuel cell. Since the
Ucap operating voltage (50V to 120V) is smaller than the bus voltage (240V to 400V), a
DC/DC converter is added to boost the voltage of the Ucap.
For the present analysis, the system uses the same fuel cell and battery as in the
case of the fuel cell/battery hybrid described in Section 4.2.1. However, an additional
component consisting of a Ucap module is considered here to create a blended ESS. Two
Maxwell Ucap modules consisting of 48 and 36 cells are considered as described in Table
4.3. For a given drive cycle, the size of the Ucap module determines the extent of battery
load reduction. The battery load is expected to reduce with increasing Ucap module size.
The above modules sizes were selected to obtain 25 to 35 kW of average Ucap power
which is expected to demonstrate an appreciable degree of battery load sharing.
83
Figure 4.2 Topology of fuel cell/battery/ultracapacitor series hybrid
Table 4.3 Ultracapacitor Description
Maxwell (BCAP 3000) Ultracapacitor
Number of cells 48 36
Max/Min Voltage 120/60 V 90/45 V
Max. Current 400 A 400 A
Max. Power 48 kW 36 kW
Available Energy 94 Wh 70 Wh
Hybrid Energy Management: The fuel cell net power remains unchanged in the present energy management scheme
and is given by Equation 4.1; it should be noted that the SOC here still refers to the
battery state-of-charge.
Traction Motor
Accessory Load
Fuel Cell
Battery
Unidirectional flow
Bidirectional flow
DC/DC Converter
Ultra Capacitor
DC/DC Converter
84
The ultracapacitor power request, ,Ucap reqP is given by
,Ucap req ESS cutoff ESS cutoffP P P if P P= − ≥
,Ucap req ESS cutoff ESS cutoffP P P if P P= + ≤ −
, 0Ucap reqP otherwise= (4.3)
( ),ESS tract acc FC netP P P P= + − (4.4)
The condition PESS ≤ -Pcutoff arises mostly during regenerative braking when Ptract is
negative. It can also occur for small positive values of Ptract.
Based on the traction power, accessory power, and fuel cell net power,ESSP is the
resulting power requirement from the ESS, which in this case is shared by ultracapacitor
and battery. cutoffP is a threshold value beyond which the Ucap starts contributing.
Therefore, if ESSP is greater thancutoffP , the extra power request is sent to the
ultracapacitor. If this extra power request can be met by the Ucap, then the battery only
needs to provide power up tocutoffP . If, however, the Ucap power supply is limited by its
size, the remaining power request is again met by the battery. The actual power supplied
or accepted by the Ucap is given by UcapP . The rationale for using a threshold power
parameter is to allow the ultracapacitor to contribute only at high power demands and
reduce the peak power demand on the battery.
The battery power request is given by
Battery ESS UcapP P P= − (4.5)
Thus, the remaining energy storage power requirement is met by the battery.
85
4.3 Simulation Results
The energy storage performance was simulated on the UD Express Route for battery-only
as well as blended ESS topologies using LFM. As stated earlier, the blended ESS analysis
was conducted for two Ucap module sizes (36 and 48 cells), and the effect of the
parameter cutoffP was also analyzed.
4.3.1 Simulation results with 48-cell Ucap
Figure 4.3 demonstrates the reduction in the frequency of high C-rate current
draws from the battery for a 48-cell, Ucap-assisted energy storage system. Our
expectation is that the battery in a blended ESS would experience fewer occurrences of
current draws within any given range. For example, figure 4.3 shows that a battery-only
ESS experiences current draws in the 2C to 4C range during 8.99% of the drive cycle. In
contrast, the battery in a blended ESS experiences 2C to 4C currents during only 1.11 to
3.18% of the drive cycle. The frequency of high battery-current draws decreases because
of load sharing by the Ucap. Ucaps have very low charge storage capacity (Ah) as
compared to the batteries. As the cutoff power is raised from 0 kW to 30 kW, there is
further reduction in C rate frequency. Low values of cutoff power can result in situations
when the Ucap runs out of available energy while providing nominal power (low drive-
motor power demands and accessory load) and has nothing left to contribute if a peak
power request is sent by the traction motor. Raising the cutoff power ensures that the
ultracapacitor energy is reserved for situations when the power requirement is high,
thereby reducing possibilities of premature energy drain out. Therefore, as evident in
figure 4.3, higher cutoff powers increase the Ucap’s capability to share peaks loads and
86
thus reduce the occurrence of high battery current. It should be noted that in an extreme
event when the cutoff power is higher than the peak power requirement, the Ucap will be
rendered useless for the entire drive cycle.
Figure 4.3 Simulated battery C-rate frequency distribution for a battery only, as well as
blended ESS (48-cell Ucap) at different threshold levels
Load sharing by the Ucap also reduce the energy throughput of the battery pack.
Raising the cutoff power diminishes the Ucap’s ability to reduce the battery energy
throughput (figure 4.4).
Battery current distribution corresponding tp UD Express Route with 48 cell Ultracapacitor
0.56
3.18
0.13
2.75
0.07
2.60
0.05
1.11
0.01
8.99
012
34567
89
10
2C<x<=4C 4C<x<=6C
C Rate Categories (x=Crate)
Freq
uenc
y of
occ
uren
ce (%
)
Battery only
Battery+Ucap(Pcutoff=0 kW)
Battery+Ucap(Pcutoff=10kW)"
Battery+Ucap(Pcutoff=20kW)
Battery+Ucap(Pcutoff=30kW)
87
Battery energy throughput corresponding to UD Express Route with 48 cell Ultracapacitor
43
18
29
3539
0
5
10
15
20
25
30
35
40
45
50
Ene
rgy
(Wh)
Battery only
Battery+Ucap(Pcutoff=0kW)
Battery+Ucap(Pcutoff=10kW)
Battery+Ucap(Pcutoff=20kW)
Battery+Ucap(Pcutoff=30kW)
Figure 4.4 Simulated energy throughput for a battery only, as well as blended ESS (48-
cell Ucap) at different threshold levels
A plot of ultracapacitor SOC for 0 kW and 30 kW cutoff powers (figure 4.5) indicates the
reason for the trend seen in figure 4.4. At Pcutoff = 0, the Ucaps experiences high SOC
swings indicating a high degree of participation during the drive cycle. The Ucap
provides all the energy it can, regardless of power demand during an acceleration or
cruising event and absorbs the maximum possible energy during a regenerative braking
event. Due to repetitive usage at Pcutoff = 0, the Ucap is able to deliver or absorb much of
the ESS energy flow thereby reducing battery energy throughput by more than 50%. On
the contrary, at Pcutoff = 30 kW, the Ucap only participates in the ESS energy flow when
the ESS power demand exceeds the cutoff value. In such situations, the Ucap contributes
only occasionally, and although it takes care of peak currents, there is not much energy
flow as indicated by the shallow Ucap SOC swings.
88
0 200 400 600 800 1000 1200 1400 1600
0.5
0.6
0.7
0.8
0.9
1
Time (s)
Ultracapacitor SOC (48cell)
Pcutoff=0kWPcutoff=30kW
Figure 4.5 SOC swing of 48-cell Ucap module at 0 kW and 30 kW threshold power
corresponding to UD Express Route
4.3.2 Simulation results with 36-cell Ucap
The energy storage simulations with a 36-cell Ucap module shows similar
behavior (figures 4.6 & 4.7) as discussed above. The only difference is that the
corresponding C rate frequency or energy throughputs are smaller because of the smaller
size of the Ucap.
89
Battery current distribution corresponding UD Express Route with 36 cell Ultracapacitor
0.56
4.26
0.17
3.48
0.10
3.15
0.10
2.19
0.04
8.99
0123456789
10
2C<x<=4C 4C<x<=6C
C Rate Categories (x=Crate)
Freq
uenc
y of
occ
uren
ce (%
) Battery only
Battery+Ucap(Pcutoff=0 kW)
Battery+Ucap(Pcutoff=10kW)"
Battery+Ucap(Pcutoff=20kW)
Battery+Ucap(Pcutoff=30kW)
Figure 4.6 Simulated battery C-rate frequency distribution for a battery only, as well as
blended ESS (36-cell Ucap) at different threshold levels
Battery energy throughput corresponding to UD Express Route with 36 cell Ultracapacitor
43
22
31
3639
0
5
10
15
20
25
30
35
40
45
50
Ene
rgy
(Wh)
Battery only
Battery+Ucap(Pcutoff=0kW)
Battery+Ucap(Pcutoff=10kW)
Battery+Ucap(Pcutoff=20kW)
Battery+Ucap(Pcutoff=30kW)
Figure 4.7 Simulated energy throughput for a battery only, as well as blended ESS (36-
cell Ucap) at different threshold levels
90
As can be seen from the preceding plots for both the 48- and 36-cell Ucap
modules, the variations of C-rate frequency and energy throughput with cutoffP show
opposing trends. Therefore, selecting the optimal Pcutoff value depends on the impact of
these two factors on battery life. Similarly, selecting the appropriate Ucap-module size
depends upon the hybrid performance targets, expected usage (duty cycle), battery
lifetime goals, and cost. While the present analysis has been conducted with Li-Ti
batteries which already boast a high cycle life, similar conclusions would hold for any
battery chemistry, be it Li-ion, Nickel, or lead-acid. Whether or not a battery’s lifetime
can match the vehicle’s lifetime is always a concern and integrating an Ucap in a battery-
dominant hybrid represents a good option to stretch the lifetime limit. A comprehensive
life cycle model can bring more perspective to such studies and help immensely in using
simulations for actual decision making.
4.4 Summary
This chapter has presented the concept of battery-load reduction through blending with
ultracapacitors. The topologies for the two ESS systems were presented and an energy
management scheme for blended energy storage was proposed. Simulation of energy
storage performance on the UD Express Route showed substantial reduction in battery
current-load and energy throughput for a blended system which are two of the
contributing factors towards battery degradation. The sensitivity of the results to the
parameter cutoffP was analyzed and the resulting tradeoff was explained. The importance
of a reliable battery lifetime model for actual decision making was stated.
91
5. Battery Thermal Model
5.1 Introduction
Thermal management of batteries in hybrid vehicles is essential for effective operation in
all climates. The electrochemical performance, charge acceptance, power and energy
capability, cycle life and cost are influenced by the operating temperature of the battery.
The use of Li-ion batteries in particular presents a safety issue and requires a reliable
thermal management system. The goal of a thermal management system is to maintain an
optimum average temperature of the battery pack with acceptable temperature variations
between the cells within the pack. However, the thermal management system has to
satisfy other constraints such as compactness, low weight and volume, low cost,
packaging ease, and compatibility with the onboard location. In addition, it must
consume low parasitic power, be accessible for maintenance and most importantly be
reliable under a wide range of temperatures. The system design therefore entails many
decisions and exploratory trial and error experimentation, a process which can be made
easier and cost effective with the help of reliable thermal model and simulations. Thermal
modeling is also essential to understand the effect of design and operating variables on
the thermal behavior of batteries. Therefore, battery thermal models can prove useful in
the battery design process as well as subsequent thermal management development.
This chapter attempts to understand the thermal behavior of Altairnano LiTi cells
using two approaches - experiments and modeling and simulations to understand the
thermal behavior of batteries under typical operating conditions. Altairnano LiTi cells
incorporates a novel chemistry and hence have different electrical and thermal
performance which has not been discussed in literature. The Phase 3 UD fuel cell hybrid
92
bus will incorporate these novel LiTi cells and hence an understanding of these cells is
particularly important for the UD Fuel Cell Bus Program.
In the past few years battery thermal modeling, experiments and thermal
management have been an important area of research. Hallaj et al. studied the thermal
behavior of Li-ion cells with active and passive cooling systems for non automotive
applications [22-26]. One-dimensional electrochemical thermal models were developed
for Li-ion cells [27, 28]. The thermal implications of extreme conditions such as thermal
abuse and internal short-circuit are also found in literature [29, 30]. Kim et al. presented
thermal model for prismatic lithium cells that could capture temperature variation on the
cell surface [32-33]. The National Renewable Energy Laboratory has conducted
significant research on advanced multi-physics and multi-dimensional electrochemical
thermal model of lithium batteries (cylindrical and large prismatic) and extensive battery
testing [34-37].
The following sections will introduce the LiTi cells, describe the experiments that
were performed on the cells as part of this thesis, and the model used to predict
temperature rise followed by results and conclusions.
5.2 Altairnano Lithium-Titanate Cells
Altairnano LiTi cells are 2.3 V prismatic cells (figure 5.1). The Altairnano cells replace
graphite (commonly used in Li-ion cells) with a proprietary, high surface area lithium-
titanate oxide based anode material. Consequently, the Altairnano cells possess fast-
charge kinetics and hence faster charge and discharge rates. Due to the absence of the
93
solid electrolyte interface (SEI) layer, these cells have better thermal stability when
compared to Li-ion cells and can operate over a wide temperature range (-40C to 50 C)
Figure 5.1 An Altairnano Lithium-Titanate cell (2.3 V, 50 Ah, 25x25x1.2 cm). Electrical
contacts are made using the two tabs at the top of the cell.
5.3 Battery Tests
This section describes the experiments performed on LiTi cells and discusses the results.
The objective of these experiments is to observe the battery temperature change during
charging and discharging, and extract the necessary electrical and thermal properties
which are required by the thermal model. A battery pack of 5 cells was assembled for
high current testing by bolting the battery tabs in series. A power supply (15V DC and
640 A) was used for charging the battery pack. Discharge currents were achieved using a
bank of 24 resistors in parallel, each connected to an automatic switch to provide 24
possible resistor combinations thus allowing different magnitudes of discharge current.
The tests were controlled and monitored using a Data Acquisition (DAQ) device and
LabVIEW program.
94
Figure 5.2 Schematic of the 5 cell stack and thermistor locations
The system was equipped with sensors that recorded the voltage of each cell and
temperature at seven different locations (six at the tabs and one at the base of the 4th cell)
(figure 5.2). The tab is expected to be the hottest part of the cell and hence all tabs were
monitored. One thermistor was located at the bottom of the 4th cell to study spatial
variations in temperature across the height of the cell. The 4th cell was chosen because it
is isolated from the cooling effects of metal interconnects. In addition, the temperature
field on the outer surface of the cell was captured in the form of infrared thermal images
from an IR Camera.
Cell 1
Cell 2
Cell 3
Cell 4
Cell 5
Tab0
Tab1
Tab2
Tab3
Tab4
Tab5
Cell4 bottom
Thermistor Locations
Shunt Resistor
Interconnects
95
Figure 5.3 Snapshot of the battery pack of five cells used for experiments. The fourth cell
is inverted to allow direct imaging of its surface
The battery pack was subjected to different magnitudes of charge and discharge current
and the temperature rise was recorded. One of the cells in the middle of the pack was
inverted to be able to capture a thermal image of the its surface and avoid the effects of
thermal mass due to interconnects at the ends of the pack. Figure 5.4 shows the
temperature distribution of the inverted cell with the passage of time while charging at
400 A.
Inverted cell for thermal IR imaging
Interconnects
Metal bars to clamp battery tabs
96
t=0 min t=1 min
t=2 min t=3 min
t=4 min t=5min
Figure 5.4 Temperature distribution on the surface of the cell recorded by IR Camera at
different time instants during charging at 400 A
After five minutes of charging at 400A, the tabs exhibited a greater temperature rise than
the rest of cell’s surface. Note that the view of the tab on the right hand side of the cell is
Blocked view Tab
97
blocked by another cell in the battery pack as indicated in figure 5.3. The test result
confirms the temperature imbalance in large prismatic cells as reported in the literature
[33-37]. The resulting temperature profiles are attributed to two reasons: 1) Higher heat
generation occurs near the tabs because the tabs are the pathway through which the entire
current either enters or leaves the cell, and 2) higher heat generation results from the
contact resistance between the tabs of two connected cells.
Figure 5.5 Temperature distribution on the surface of the cell recorded by IR Camera at
the end of 15 minutes of charging time with 100 A of current
A similar test conducted with 100 A of charging current shows that the temperature
difference across the cell surface is not appreciable (figure 5.5). This is because the heat
generation rate is much slower at 100 A current which allows sufficient time for heat
dissipation within the cell. The temperature change at lower times was insignificant.
Therefore in the next set of experiments the battery was subjected to longer duration of
charge and discharge cycles.
Figure 5.6 shows the result of multiple charge and discharge at 200 A for a period
of 1.5 hours. The temperatures at tabs0 and 5 (the two ends of the pack) are much lower
98
than the rest because they are clamped to interconnects from the power supply and the
resistor bank, respectively, which absorb some of the heat. Tabs 2, and 3 belong to the 4th
cell in the pack. It was observed that the temperature rises during discharge and falls
slightly when the battery is being charged. The trend can be explained with the help of
the following commonly used equation for heat generation of batteries where heat
generation is given by
2 ocdVq I r IT
dT
•= − 0I > : discharge (5.1)
where, I is the current, r is the cell internal resistance, T is the temperature in Kelvin, and
Voc is the open circuit voltage of the cell. The first term on the right hand side represents
ohmic heating. The second term is the heat generation behavior due to entropy change of
the cell. The term ocdV
dTis negative. Therefore the second term results in heat evolution
during discharge and heat absorption during charging. The differences between the tab
temperatures in figure 5.6 arise from different contact resistances at the tabs.
99
0 1000 2000 3000 4000 500025
30
35
40
45
50
55
60
65
Tem
per
atu
re (
C)
0 1000 2000 3000 4000 5000-200
-100
0
100
200
Ch
arg
ing
Cu
rren
t (A
)
Tab 0Tab 1Tab 2Tab 3Tab 4Tab 5Cell 4 Bottom
Figure 5.6 Temperature readings at different locations of the battery pack during 200 A
charge/discharge cycles
A similar charge/discharge cycle test performed at 100 A shows a greater temperature
drop during the charging phase (figure 5.7). This is because at lower currents the heat
absorptive effect of entropy change becomes comparable to the heating generation effects
due to ohmic losses.
100
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000
28
30
32
34
36
38
40
42
44
46T
emp
erat
ure
(C
)
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000-100
-50
0
50
100
Ch
arg
ing
Cu
rren
t (A
)Tab 0Tab 1Tab 2Tab 3Tab 4Tab 5Cell 4 bottom
Figure 5.7 Temperature readings at different locations of the battery pack during 100 A
charge/discharge cycles
The relationship of Voc and temperature is specific to the cell chemistry and was
determined experimentally in this study for the LiTi cell. For this purpose the cell was
heated slowly by silicone rubber heating pads and the entire assembly was insulated with
low-conductivity foam. The resulting plot of Voc vs temperature is shown in figure 5.8.
101
The value of the term OCdV
dTis equal to the slope of the line in Figure 5.8 and was found
to be approximately equal to -8.0e-5.
25 30 35 40 45 50 55 601.9415
1.942
1.9425
1.943
1.9435
1.944
Temperature (C)
Vo
c
Figure 5.8 Experimentally measured variation of open circuit voltage (Voc) with
temperature for LiTi cell
We can use the experimentally derived value of OCdV
dT to approximately calculate the
value of heat generation. For a cell with 0.6 mΩ resistance and current draw of 200A,
heat generation due to ohmic losses is 24 W. Now, heat evolved or absorbed due to the
entropy term at 300K amounts to 4.8 W which is five times smaller than the ohmic
contribution. At a lower current of 100 A, the corresponding contributions are 6W
(ohmic) versus 2.4 W (entropy) due to the entropy term. Thus, the entropy term is less
102
than two times smaller at 100A. Hence it is quite evident that relative magnitude of
entropy term increases as current is decreased.
The specific heat capacity of the LiTi battery was determined from a simple
calorimetric test. The cell was placed in a container and hot water (~ 75 C) was poured
into it until the cell was completely submerged. The temperatures on the cell surface and
of the water (away from the cell) were read through thermistors in LabVIEW. Figure 5.9
shows that the cell acquires heat from the hot water and its temperature rises until the
system reaches thermal equilibrium at about 1000 s. During this process the system loses
heat to the surroundings which is why the system temperature continues to fall gradually.
500 1000 1500 2000 2500 3000 3500 4000 450020
30
40
50
60
70
80
Time (sec)
Tem
per
atu
re (
C)
Cell SurfaceWater
Figure 5.9 Time trace of water and cell surface temperature during calorimetric test for
measuring specific heat capacity of LiTi cell
103
The heat lost to the ambient from the start to the thermal equilibrium point was estimated
through another experiment. In this experiment an equal weight of hot water was poured
into the container (without the cell inside) and the water temperature was recorded as it
lost heat to the environment. The rate of temperature drop dT/dt was plotted against the
temperature of the hot water (figure 5.10). The rate of temperature drop was then used to
estimate the amount of heat lost by the cell to the environment during the calorimetric
experiment.
45 50 55 60 65 70 75
-8
-7
-6
-5
-4
-3
-2
-1x 10
-3
Temperature (C)
dT
/dt
y = - 1.01e-007*x3 + 1.44e-005*x2 - 0.000826*x + 0.0152
Experimental DataFitted Curve
Figure 5.10 Rate temperature drop due to heat loss to the environment as a function of
water temperature
The heat balance is given by the following equation.
104
0
( ) ( ) ( )eqt
w w w eq B B eq B w w
dTm Cp T T m Cp T T m Cp f T dt
dt − = − + = ∫ (5.2)
The first term is the total heat lost by the water, the second term is the heat gained by the
cell and the third term is the heat lost by water to the ambient from start of the experiment
to the point when equilibrium is reached. Equation 5.2 was used to obtain the value of the
specific heat capacity of the battery as 1110 J/(Kg.K).
The next section presents a thermal model and simulation whose goal is to capture
the experimentally observed thermal behavior of the LiTi battery.
5.4 Thermal Model & Simulation
5.4.1 Mathematical Model
The mathematical model presented by Kim et al. [31-33] has been used for evaluating the
heat generation rate within a cell. A cell consists of repeating units of positive electrode,
electrolyte, separator and negative electrolyte, which are packed together to increase the
total Ah (Ampere hours). As the unit is repetitive it is sufficient to develop the
mathematical model for just one such unit, i.e. a positive electrode assembly positive and
negative electrode with electrolyte and separator. The model should then be applicable
for the entire cell. Figure 5.11 shows a schematic of the two electrodes and current flow
between them. Both the positive and negative electrodes consist of active materials on a
metal current collector. The active materials are separated by the electrolyte and
separator. The rectangular tabs are extensions of the current collector and therefore are
made of metal. The current enters the negative electrode current collector tab and leaves
through the positive electrode current collector tab.
105
Figure 5.11 Schematic diagram of current flow in parallel electrodes of a cell
The repeating units are connected in parallel. It is assumed that the distance between the
two electrodes is very small and that the current flow between them is perpendicular.
. 0p pi J in→
∇ − = Ω (5.3)
. 0n ni J in→
∇ + = Ω (5.4)
where pi→
and ni→
are the current per unit thickness vectors (A/m) for the positive and
negative electrode respectively, and J is the current density (A/m^2) that is transferred
through the separator from the negative to the positive electrode. pΩ and nΩ denote the
domains of positive and negative electrodes respectively and i jx y
∧ ∧∂ ∂∇ = +∂ ∂
. Equations
(5.3) and (5.4) merely state that the sum of currents entering an element of the electrode
is equal to the sum of currents leaving the element. According to Ohm’s law pi→
and ni→
can be written in the following way
Current
pi
Current
ni y
x
Positive Electrode Negative Electrode
106
1p p p
p
i V inr
→= − ∇ Ω (5.5)
1n n n
n
i V inr
→= − ∇ Ω (5.6)
where pr and nr are resistances and pV and nV are potentials of the positive and negative
electrodes respectively. By substituting equations (5.5) and (5.6) in (5.3) and (5.4)
respectively the following equations are obtained for pV and nV .
2p p pV r J in∇ = − Ω (5.7)
2n n nV r J in∇ = Ω (5.8)
The relevant boundary conditions for pV are
01 p
p
V I
r n L
∂− =
∂ at the positive tab (5.9)
0pV
n
∂=
∂elsewhere in pΓ (5.10)
where n
∂∂
denotes the gradient in the direction of the outward normal to the boundary, 0I
is the total current entering through the electrodes and L is the length of the tab. The first
boundary condition implies that the total current flow through the tab is equal to0I . The
second boundary condition imposes the restriction that there is no current flow through
the boundary of the electrode other than the tab.
01 n
n
V I
r n L
∂ =∂
at the negative tab (5.11)
0nV
n
∂ =∂
elsewhere in nΓ (5.12)
0nV = at the midpoint of the tab (5.13)
107
Equation (5.11) imposes the condition that the total current inflow through the tab at the
negative electrode is equal to0I . Equation (5.12) implies that there is no current flow
through the boundary of the negative electrode other than the tab. Equation (5.13) assigns
a reference zero value at the midpoint of the edge shared by the electrode and the tab. The
reference values enable assigning values to pV and nV . The electrode resistance
( )p nr r or r is calculated as the equivalent network of parallely- connected resistors of
electrode material and the corresponding current collector.
1
c c e e
rh S h S
=+
(5.14)
where eh and ch are the thickness of the electrode and current collector respectively and
eS and cS are the electrical conductivities of the electrode and current collector
respectively.
The current density is a function of the potential difference between the positive
and negative electrode and is given by the equation (5.15)
( ( ))p nJ Y U V V= − − (5.15)
where U is the open circuit potential and 1Y is the resistance of the electrolyte and
separator between the electrodes. The heat generation source term within the cell is given
by
2 2( ( ))( , ) p n p p n n
dUJTJ U V V i r i r dTq x y
h h
• − − + += + (5.16)
108
where h is the thickness of the entire assembly. The first term on the right hand side
denotes the heat generation due to resistances of the cell components. The second term is
due to change in the entropy.
5.4.2 Results
The governing equations were solved numerically in MATLAB. The steady-state results
are shown for a charging current of 400 A. Figure 5.12 shows an increasing gradient of
electrode potential near the positive tab. This is because all the current that flows into the
positive electrode exits through the positive tab and therefore results in a greater potential
drop.
109
Width (m)
Hei
gh
t (m
)
0 0.05 0.1 0.15 0.20
0.05
0.1
0.15
0.2
2.5585
2.5586
2.5587
2.5588
2.5589
2.559
2.5591
2.5592
2.5593
2.5594
Figure 5.12 Distribution of Vp: Contour plot (above); 3D surface plot (below)
110
Similarly, the negative potential gradient is higher towards the location of the negative
tab (figure 5.13). Again the reason is that the entire current enters the cell through the
negative tab and then gets distributed throughout the negative electrode. Note that
according to the boundary condition given by equation 5.13, the reference value of
negative potential is chosen to be zero at the midpoint of the tab. Therefore the values of
Vn throughout the negative electrode are close to zero.
111
Width (m)
Hei
gh
t (m
)
0 0.05 0.1 0.15 0.20
0.05
0.1
0.15
0.2
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
x 10-3
Figure 5.13 Distribution of Vn: Contour plot (above); 3D surface plot (below)
112
Figure 5.14 shows the plot of potential difference in 2D space. The gradient is higher
towards the negative tab because the resistance of the negative electrode is assumed to be
higher than the positive electrode resulting in a greater potential drop near the negative
tab.
113
Width (m)
Hei
gh
t (m
)
0 0.05 0.1 0.15 0.20
0.05
0.1
0.15
0.2
2.56
2.561
2.562
2.563
2.564
2.565
2.566
2.567
2.568
2.569
Figure 5.14 Distribution of the potential difference (Vp-Vn): Contour plot (above); 3D
surface plot (below)
114
The distribution of current density can be understood from the fact that current follows
the path of least resistance. Since the area near the negative electrode presents higher
resistance, current entering the negative electrode tries to leave the electrode and cross
over to the positive electrode and reach for the positive tab. Therefore, current density is
higher near the location of negative tab and lower towards the positive tab. The current
density decreases as we proceed towards the base of the cell. This is to be expected
because the path length increases with the distance from the tabs and presents greater
resistance to current flow.
115
Width (m)
Hei
gh
t (m
)
0 0.05 0.1 0.15 0.20
0.05
0.1
0.15
0.2
95
95.5
96
96.5
97
97.5
98
98.5
Figure 5.15 Distribution of current density J: Contour plot (above); 3D surface plot
(below)
116
The 2D heat generation was reduced to 1D by averaging the values along the width of the
cell. Figure 5.16 shows the variation of resulting heat generation with the height of the
cell. Due to higher current flow near the tabs, there is greater heat generation in the upper
part of the cell.
0 0.05 0.1 0.15 0.21.4
1.45
1.5
1.55
1.6
1.65
1.7
1.75
1.8
1.85x 10
5
Height (m)
Hea
t G
ener
atio
n (
W/m
3)
y = 8.97e+008*x5 - 4.23e+008*x4 + 7.16e+007*x3 - 4.72e+006*x2 + 1.29e+005*x + 1.43e+005
Mathematical Model Fitted Equation
Figure 5.16 Variation of heat generation rate with the height of the cell
5.4.3 Thermal Simulation
A 3D model of the cell with the tabs and metal bars (used to clamp the tabs in series) was
created and meshed in GAMBIT. The problem is symmetric about the XZ plane passing
through the middle of the cell. Therefore only one half of the cell has been modeled
(figure 5.17). A natural convection boundary condition was applied on the outer surface
117
of the aluminum jacket indicated as (1), and the exposed surfaces of the metal clamps (2)
and (3). A zero heat flux boundary condition is applied on the remaining surfaces.
Figure 5.17 3D model of half LiTi cell created in Gambit: view from the outside (left)
view from the midplane of the cell (right)
The mesh was exported in FLUENT and a transient 3D heat transfer simulation was
performed. The heat-generation equation shown figure 5.16 was used as the source term
for the cell. In addition the source terms were assigned to the tabs which were calculated
from the measurement of contact resistance and the value of current used (400A). The
transient problem was simulated with the same initial conditions as the IR experiment on
Aluminum Jacket
Aluminum Metal clamps
Electrode, collector, separator assembly
Aluminum tabs
X Y
Z
(1)
(2)
(3)
118
an inverted cell subjected to 400A of current, and for the same length of time. Figure
5.18 shows the resulting temperature distribution of the outer surface of the cell.
Figure 5.18 Temperature distribution on the cell surface after 5 minutes of charge at
400A obtained from FLUENT simulation (above) IR imaging (below)
119
The simulated temperature distribution on the cell surface is similar to the temperature
field obtained from thermal IR imaging. Figure 5.18 indicates the model’s ability to
predict the non uniform rise in temperature across the cell surface. However, these are
just preliminary simulation results and the accuracy of the model has to be checked by
simulating temperature rise under a variety of different conditions and compared with
experimental data.
A literature survey during the course of this work has revealed areas where the
model can be further improved. In the current mathematical model, the effect of
temperature on voltage and current distribution has not been modeled. NREL
presentations [37] have revealed that, temperature affects charge transfer kinetics and
therefore the temperature imbalance across the cell is expected to have an impact on other
parameters. Another missing component in the model is a governing equation for the
diffusion of reactant material at the electrodes. The rate at which the active material
diffuses from the inner bulk regions to the surface is the predominant limiting mechanism
during high rate discharge and will affect the current distribution throughout the cell. In
addition, 3D models of candidate cooling systems can be added to the battery model to
simulate the thermal performance of the entire system. A complete, validated battery
thermal model can be used to reliably predict the temperature distribution within the
entire system for a variety of operating conditions over an extended period of time.
5.5 Summary
This chapter has presented an investigation of the thermal behavior of the Altairnano LiTi
battery. The experiments performed on the battery and the results of the tests were
120
described. An uneven temperature distribution across the surface of the cell was
experimentally recorded using an IR camera during charging at high current. A
mathematical model was developed and used to calculate the heat generation within the
cell. A 3D model of the cell was created and meshed in GAMBIT. The transient 3D
problem was solved in FLUENT to simulate temperature rise during charging at high
current. Comparison of the simulation results and the IR image showed a similar
temperature distribution. However, it was concluded that the model should be tested
under a variety of different conditions. Also, avenues for further improvement of the
thermal model were identified.
121
6 Summary and Future Work
6.1 Summary
The focus of this thesis has been modeling, simulation, and optimization of hybrid
powertrain systems. The majority of the present work has centered on improvement of
the LFM hybrid powertrain simulator and using it to conduct powertrain optimization
studies. Chapter 2 introduced the LFM model and highlighted the drawbacks of the
earlier versions. The functionality of every subsystem was described and all
enhancements made were detailed. Finally, the improved LFM simulator was validated
with test data acquired from onboard sensors in the UD Phase 1 fuel cell bus. The
comparison between the simulator’s outputs and vehicle data demonstrated good
agreement. The possible sources of errors were discussed and it was concluded that LFM
can be used as a reliable tool for design and optimization studies.
The improved LFM simulator was used to explore optimization of fuel
cell/battery hybrid power management. A new prediction-based power management
strategy was proposed in Chapter 3 and implemented in the LFM simulation software.
Simulation results for the prediction-based strategy showed significant improvements in
fuel cell system efficiency and reduction in hydrogen consumption compared to a
conventional, baseline strategy of charge sustenance. The importance of a stable power
request to the fuel cell was stated and realized with the help of this novel strategy. A
sensitivity analysis was conducted to study the effects of inaccurate predictions and the
impact on vehicle performance was discussed. Finally, the benefits predicted from
simulation studies were confirmed through implementation of the proposed strategy and
its comparison with the baseline control strategy in the Phase 1 fuel cell/battery hybrid
122
bus. The conclusion was drawn that the prediction-based strategy is beneficial for transit
applications.
Chapter 4 shifts the focus from fuel cell lifetime and fuel economy to the
enhancement of battery lifetime. The validated LFM tool was used to evaluate one
approach to reducing battery loads by adding an ultracapacitor module, and thereby
enhancing battery lifetime. A hybrid power management scheme for blended energy
storage was designed and implemented in the simulation. Simulation of energy storage
performance showed a substantial reduction in battery current-load and energy
throughput for the blended storage system, which are two of the contributing factors
towards battery degradation. The sensitivity of the results to the hybrid control parameter
was analyzed and the resulting tradeoff was explained. The results from this chapter have
opened up a new direction where powertrain simulations can help in further evaluation of
blended energy storage systems and their feasibility and usefulness in electric-drive
automobiles. The importance of a reliable battery lifetime model for such assessments
has been emphasized in the study.
Chapter 5 specifically addresses battery thermal behavior. Several thermal
experiments were conducted on the LiTi battery, and a mathematical model was used to
setup a 3D transient heat transfer simulation in FLUENT and predict temperature rise.
The simulation results show that the model captures the temperature imbalance in
prismatic cells under a high-current regime. A comparison of simulation results with IR
temperature data showed good agreement for the temperature distribution. However, the
model needs to be tested under a variety of different conditions. Also, a literature review
has revealed avenues of further improvement in the thermal model.
123
6.2 Future Work
The following topics have been identified to extend the current effort into the future.
6.2.1 Power train model and simulation
It is extremely important to verify the accuracy of a powertrain simulator before
placing trust in simulation results. At present, the LFM simulator is capable of predicting
hybrid powertrain performance with an accuracy of under 10% which is considered
adequate for both the studies in chapter 3 and 4. However, depending on the nature of the
study, greater accuracy may be desired from the simulator. For example, if certain
parameters values obtained from simulations are very sensitive to the inaccuracies of the
simulator, then the utility of those parameters is compromised. Such situations require
very accurate input data and models. To achieve greater fidelity, the various subsystems
within the simulator should be modeled accurately. Therefore, first, the vehicle resistance
model and data should be improved. The friction, drag coefficient, and inclination data
should be reasonably accurate. An advanced vehicle resistance model, if available, should
be adopted, for example, incorporation of wheel slip. Similarly, transmission losses and
traction motor losses play a significant role. Currently, LFM uses a constant traction-
motor efficiency value although it actually depends on the motor torque and speed. A
reliable map or model to calculate motor losses can help in ensuring high accuracy in
predicting the motor power request. After achieving a dependable motor power request,
the next step is to improve the fuel cell, balance-of-plant, and battery model. It should be
understood that all of these practical constraints impose certain limitations on the
accuracy of powertrain simulator results, and therefore the idea is to plan the modeling
effort so that the return on investment is justified.
124
Besides, depending on the nature of the simulation study, the effort can be
focused upon a particular component as well. For instance, simulations dealing
exclusively with the battery subsystem can employ advanced battery models (advanced
equivalent circuits) so as to capture battery dynamics with better accuracy.
6.2.2 Prediction-based power management strategy
The prediction-based power management strategy developed as part of this thesis
is based on the assumption of a priori knowledge of the vehicle route and the vehicle’s
energy requirement. This assumption is adequate for transit applications but cannot
succeed in other regimes of automobile use. However, GPS devices can be used to
provide information about the route in a manner that can be used by vehicle supervisory
controller to exercise near optimal controls. Therefore, the current study can be extended
to formulate a hybrid power management scheme that uses GPS route information to
improve vehicle performance. An approach to solving such a problem can be to define
and store optimal controls/control parameters for different driving conditions, and then
use the route information to determine the expected nature of the drive cycle and apply
the appropriate controls.
6.2.3 Blended energy storage
The present work on battery-stress reduction through blended ESS presents
opportunities to further extend the analysis. In chapter 4 the battery size was fixed and
two different sizes of Ucap modules were used to demonstrate the effect of Ucap size and
control parameter on battery stress. However, in order to optimize the system, both the
125
battery and Ucap size should be varied so that the best possible combination is achieved.
In addition to ESS size, the variation of control parameters is also crucial. A very
important aspect of this concept is to evaluate the lifetime benefits of blended ESS
against cost. Therefore battery lifetime models and costs, which have not been considered
in the present study, are additional elements for future work. Promising candidates from
this analysis should be validated by demonstrating the concept first through laboratory
testing, followed by implementation in our hybrid bus.
6.2.4 Battery thermal modeling and simulation
Battery thermal modeling, simulation and management is a vast area with
tremendous research opportunities. The present work represents an initial attempt to
understand battery thermal behavior which can be expanded significantly on both
modeling and experimental aspects. The battery model can be improved by adopting
multi-physics, multi-dimensional models of the battery. Besides predicting the
temperature rise of a given cell, such models could give an insight into the factors
affecting the thermal behavior and performance, and prove very useful for battery design.
Similarly, much work can be done on experimental and testing fronts on the new LiTi
cell to acquire better understanding of its characteristics, performance, and optimal
operating conditions. Such tests can involve evaluating the effect of operating
temperature on internal resistance, pulse-power capability, and efficiency (heat
generation rate vs. power production rate). Finally, the design of an effective, efficient,
and reliable thermal management system is another component of research which will be
immensely useful in ensuring safe operation of lithium cells in the fuel cell hybrid bus.
126
6.2.5 Intelligent driving
Frequent acceleration and deceleration over short distances in urban driving
conditions can lead to increased fuel consumption. There is a potential to reduce energy
consumption by adopting an intelligent way of driving that avoids unnecessary speeding
and braking, also commonly known as ‘hypermiling’. Such intelligent driving requires
inputs such as current traffic flow information and computer control to send appropriate
speed recommendations to the driver. The LFM simulator can be employed to evaluate
the practicability and potential fuel economy gains from such intelligent driving in urban
conditions. It will involve challenges such as identifying the precise components of
traffic information flow and how they can be incorporated into LFM to optimize the
shape the velocity profile of the vehicle.
127
REFERENCES
[1] http://altfuels.org/backgrnd/altdrive.html [2]http://www.neoam.cc.ok.us/~rjones/Pages/Online%200123/Chapter%201%20images/fuelcell.gif [3] www.wikipedia.org [4] D. Brown, Simulation and optimization of fuel cell hybrid vehicle (M.S. Thesis, Spring 2008) [5] D. Brown, M. Alexander, D. Brunner, S. G. Advani, A. K. Prasad, Drive-train simulator for a fuel cell hybrid vehicle, Journal of Power Sources 183 (2008) 275-281 [6] V. Paladini, T. Danteo, A. Risi, D. Laforgia, Super-capacitors fuel-cell hybrid electric vehicle optimization and control strategy development, Energy Conversion and Management 48 (2007) 3001-3008 [7] P. Rodatz, G. Paganelli, A. Sciarretta, L. Guzzella, Optimal power management of an experimental fuel cell/supercapacitor-powered hybrid vehicle, Control Engineering Practice 13 (2005) 41-53 [8] M. J. Kim, H. Peng, Power management and design optimization of fuel cell/battery hybrid vehicles, Journal of Power Sources 165 (2007) 819-832 [9] V. Paladini, T. Danteo, A. Risi, D. Laforgia, Control Strategy Optimization of a Fuel-Cell Electric Vehicle, Journal of Fuel Cell Science and Technology May 2008, Vol. 5 [10] A. Kusoglu, A. M. Karlsson, M. H. Santare, S. Cleghorn, W. B. Johnson, Mechanical response of fuel cell membranes subjected to a hygro-thermal cycle, Journal of Power Sources 161 (2006) 987–996 [11] P. Pei, Q. Chang, T. Tang, A quick evaluating method for automotive fuel cell life time, International Journal of Hydrogen Energy, 33 ( 2008 ) 3829 – 3836 [12] A. Ohma, S. Yamamoto, K. Shinohara, Membrane degradation mechanism during open-circuit voltage hold test [13] A. F. Burke, Batteries and Ultracapacitors for Electric, Hybrid, and Fuel Cell Vehicles, Proceedings of the IEEE, Vol. 95, No. 4, April 2007 [14] Altair Nano data sheet. www.altairnano.com
128
[15] Y. Sato, K. Ito, T. Arakawa, K. Kobayakawa, Possible Cause of the Memory effect Observed in Nickel-Cadmium Secondary Batteries, Journal of the Electrochemical Society, Vol. 153, No. 10, October 1996 [16] Y. Sato, S. Takeuchi, K. Kobayakawa, Cause of the memory effect observed in alkaline secondary batteries using nickel electrode, Journal of Power Sources 93 (2001) 20-24 [17] D. Linden, Handbook of Batteries, Second Edition, McGraw Hill Inc. [18] P. Bubna, D. Brunner, J.J. Gangloff Jr., S.G. Advani, A.K. Prasad, Analysis, operation and maintenance of a fuel cell/battery series-hybrid bus for urban transit applications, Journal of Power Sources, 195 (2010) 3939-3949 [19] X.G. Yang, B. Taenaka, T. Miller, K. Snyder, Modeling validation of key life test for hybrid electric vehicle batteries, International Journal of Energy Research, 2010, 34: 171-181 [20] D. Gao, Z. Gin, Q. Lu, Energy management strategy based on fuzzy logic for a fuel cell hybrid bus, Journal of Power Sources, 185 (2008) 311-317 [21] J. Bauman, M. Kazerani, A comparative study of Fuel-Cell-Battery, Fuel-Cell-Ultracapacitor, and Fuel-Cell-Battery-Ultracapacitor Vehicles, IEEE Transactions on Vehicular Technology, Vol. 57, No. 2, March 2008 [22] S. Al-Hallaj, H. Maleki, J.S. Hong, J.R. Selman, Thermal modeling and design considerations of lithium-ion batteries, Journal of Power Sources 83 (1999) 1-8 [23] A. Mills, A. Al-Hallaj, Simulation of passive thermal management system for lithium-ion battery packs, Journal of Power Sources 141 (2005) 307-315 [24] S. A. Khateeb, S. Amiruddin, M. Farid, J.R. Selman, S. Al-Hallaj, Thermal management of Li-ion battery with phase change material for electric scooters: experimental validation, Journal of Power Sources 142 (2005) 345-353 [25] R. Sabbah, R. Kizilel, J.R. Selman, S. Al-Hallaj, Active (air-cooled) vs. passive (phase change material) thermal management of high power lithium-ion packs: Limitation of temperature rise and uniformity of temperature distribution, Journal of Power Sources 182 (2008) 630-638 [26] R. Kizilel, A. Lateef, R. Sabbah, M. M. Farid, J.R. Selman, S. Al-Hallaj, Passive control of temperature excursion and uniformity in high-energy Li-ion battery packs at high current and ambient temperature, Journal of Power Sources 183 (2008) 370-375 [27] V. Srinivasan, C.Y. Wang, Analysis of electrochemical and Thermal behavior of Li-ion cells, Journal of the Electrochemical Society, 150 (1) A98-A106 (2003)
129
[28] K. Smith, C.Y. Wang, Power and thermal characterization of a lithium-ion battery pack for hybrid-electric vehicles, Journal of Power Sources, 160 (2006) 662-673 [29] G.H. Kim, A. Pesaran, R. Spotnitz, A three-dimensional thermal abuse model for lithium-ion cells, Journal of Power Sources 170 (2007) 476-489 [30] H. Maleki, J.N. Howard, Internal short circuit in Li-ion cells, Journal of Power Sources 191 (2009) 568-574 [31] K.H. Kim, C.B. Shin, T.H. Kang, C.S. Kim, A two-dimensional modeling of a lithium-polymer battery, Journal of Power Sources 163 (2006) 151-157 [32] U.S. Kim, C.B. Shin, C.S. Kim, Effect of electrode configuration on the thermal behavior of a lithium-polymer battery, Journal of Power Sources 180 (2008) 909-916 [33] U.S. Kim, C.B. Shin, C.S. Kim, Modeling for the scale-up of a lithium-ion polymer battery, Journal of Power Sources 189 (2009) 841-846 [34] D. Bharathan, A. Pesaran, and G. Kim, Electro-Thermal modeling to improve battery design, National Renewable Energy Laboratory, 2005 [35] M. A. Keyeser, A Pesaran, M. Mihalic, J.S. Yu, S.R. Kim, M. Alamgir, D. Rivers, Thermal characterization of advanced lithium-ion polymer cells, Third Advanced Automotive Battery Conference, 2003 [36] G.H. Kim, K. Smith, Multi-Scale Multi-Dimensional Model for Better Cell Design and Management, (NREL) 1st International Conference on Advanced Lithium Batteries for Automotive Applications, 2008 [37] G.H. Kim, K. Smith, Evaluation Study for Large Prismatic Lithium-Ion Cell Designs Using Multi-Scale Multi-Dimensional Battery Model, (NREL) 215th ECS meeting, 2009
