EVS28

KINTEX, Korea, May 3-6, 2015

Design Space Exploration and Hybridization of the
Kiira-EV SMACK

R.Madanda1 , P.I.Musasizi2 , P.Korukundo3 , A.T.Asiimwe4 , J.Africa5 , and S.S.Tickodri-Togboa6

2
KIIRA Motors Corporation, [email protected],[email protected]

Abstract

Hybridization and electrification of vehicles has seen aggressive research in recent years due to the ever-increasing
fuel economy and efficiency demands. Several hybrid powertrain configurations have been developed in commercial
vehicles with Toyotas and Fords power split configurations giving some of the best fuel efficiency values. Kiira is
a vehicle brand of Kiira Motors Corporation (KMC), the first vehicle Original Equipment Manufacturer (OEM) in
Uganda. KMC is building a flagship hybrid vehicle, the KIIRA-EV SMACK, with the intent of mass production
by 2018. Engineering a functional EV or hybrid powertrain is less cumbersome of recent because of the availability
of numerous off-the-shelf products for quick integration. It is important to also note that to achieve a properly
sized powertrain requires a meticulous search from a myriad of alternatives available. In this paper the techniques
adopted in efficiently searching the design space for the KIIRA-EV SMACK hybrid vehicle are presented. To tune
parameters for the KIIRA-EV plug-in hybrid vehicle, several combinations of traction motors, batteries, generators
and control strategies are considered. Other design factors considered include vehicle architecture, cost, component
volume and drive regimes. The design space given by such choices is huge. Design Space Exploration (DSE) and
reduction using the Genetic global optimization algorithm is used to quickly search and reduce the design space.
The powertrain problem is represented as a multi-objective design problem. The optimization criteria used follows
energy consumption and vehicle performance aspects including range, top speed and gradeability. Optimization
yields a minimal set of pareto-optimal design solutions which are used in Autonomie for final vehicle level verification
and validation. The final design solutions for the different architectural components are used in the selection of
off-the-shelf KIIRA-EV SMACK powertrain components.

Keywords: Design Space Exploration(DSE), Hybridization, Powertrain, Global Optimization

1
1 Introduction not be possible. The common DSE methods applica-
ble in common engineering problems include;
Design Space Exploration (DSE) refers to the activity
1. Exhaustive search: This involves trying out
of exploring design alternatives prior to implementa-
all possible scenarios e.g branch and bound and
tion [1]. DSE is a powerful tool for rapid prototyp-
depth first search. This is only feasible for small
ing and system integration. In hybrid electric vehicle
search spaces.
design, several design constraints should be satisfied
2. Random search: Where irregular and unpre-
simultaneously as illustrated by Shaiket al [2]. There-
dictable search spaces exists, this is a method of
fore, the powertrain design problem is a multi-objective
choice. This is suitable for large irregular search
constrained non-linear optimization problem which can
spaces.
to be solved by DSE technique as illustrated by Markus
et al [3]. 3. Guided Search: This may be used where there
is sufficient knowledge of the design space and
The industry standard for powertrain design involves
the trend of choices made in the future. Such
use of forward (driver driven) or backward looking (ve-
design spaces may be traversed using heuristic
hicle driven) models [4]. In a backward-looking model,
algorithms e.g the greedy algorithm.
the desired vehicle speed is instructed from the vehi-
4. Simulation-based techniques: This is a tech-
cle model back to the engine to finally find out how
nique for evaluating performance of single points
each component should be used to follow the speed
in design space using an executable model.
cycle. These models capture a variety of design con-
straints during design but a lot of other factors like cost 5. Analytical methods. Here by reasoning or by

and volume are not generally considered. The design using a suitable algorithm, different solutions are

process with such vehicle models also involves a lot of examined and pruned. Analytical techniques and

iterations of different vehicle specifications and drive simulation based calibration can be combined to

cycles to arrive at an optimal solution. This process is arrive at a solution.

in most cases time consuming. The approaches presented above are generic in nature

The iterative nature of manually searching for an op- and are used in several branches of science for design

timum solution using the vehicle models necessitates space exploration. In the next section, a review of com-

application of faster design space reduction techniques. mon design space traversal and optimization methods

DSE techniques can be applied in tandem with the ve- applied to electric and hybrid vehicles are discussed.

hicle models to arrive at a high quality design solution The rest of the paper is organised as follows. Section.2
in a shorter time. It should be noted that the choice reviews the current optimization approaches applied
of the DSE method to be applied highly depends on to hybrid vehicles, section 3 and 4 gives the high level
the nature of the alternatives available and their im- vehicle definition and constraints of the KIIRA EV
pact on results. If a components impact on the output SMACK, section 5 discusses in detail the application
is predictable, then application of a search strategy is of the DSE technique and finally section 6 the results
possible. If a components impact on performance is from the DSE approach and their use in the vehicle
unpredictable then an optimum search strategy may model are presented.

2
EVS28 International Electric Vehicle Symposium and Exhibition
2 Related Work Murgovski et al [11] uses convex optimization to search
the design space. In convex optimization, a multivari-
Sharer et al [5] presents an iterative component sizing
able function is optimized subject to constraints which
approach for hybrid vehicle design space search. The
have a convex constraint function.
iterative component sizing algorithm simply searches
Xiaolan et al [12], use a parallel chaos optimization
for a vehicle mass which solves the constraint Equa-
algorithm(PCOA) where a multiobjective function is
tion 1;
defined so as to minimize the drivetrain cost consider-

Pm (Mveh ) Peng (Mveh ing the drive performance requirements as constraints.

+ + Ncell (1.3Pbatt ...
m eng (1)
2Ebatt )XMcell (1.3Pbatt , .2Ebatt ) + Mglider = Mveh
3 Vehicle Definition
where Pm is the motor peak power, Peng is the engine
peak power, Pbatt is the battery peak power, Ebatt is
Table 1 lists the initial KIIRA EV SMACK vehicle
the battery total energy capacity, Mveh is the vehicle
level goals. The rest of the vehicle definition parame-
total mass, Mglider is the mass of the vehicles glider,
ters are outline in the Table 2
Ecell is the mass of a battery cell, m is the specific
Table 1: Vehicle Definition
power of the motor, eng is the specific power for the
Requirement Specification
engine,and Ncell is the number of cells connected in
Top Speed 180 km
series to form the battery pack
AER 50 kms
Jainet al [6], Shahirinia et al [7], Morteza et al
Gradeability 17.63 at 30 km/hr
[8], Jianping et al [9], Kumaret al [10] use a ge-
netic algorithm (GA) for design space search. A Table 2: Vehicle Specifications
genetic algorithm is a stochastic global search tech-
GVW 1500 kg
nique which mimics the process of natural biological
Lenght 4840
evaluation(survival of fitness) used for solving both
Track width 71657 mm
constrained and unconstrained optimization problems
Height 1386 mm
based on a natural selection process that mimics bio-
Frontal Area 2.3m2
logical evolution. The algorithm repeatedly modifies a
Wheels 215/65R18
population of individual solutions at each iteration. At
Drag Coefficient 0.28
each step, the GA randomly selects individuals from
capacity and type passenger Hybrid
the current population and uses them as parents to
produce the children for the next generation. Over suc-
3.1 Vehicle Architecture
cessive generations, the population evolves toward
an optimal solution. It is important to note that the The KIIRA EV SMACK is a series hybrid electric vehi-
GA is derivative free and good at arriving at a global cle, refer to Fig.3.1. The drive train architecture com-
optima rather than local minima as a final solution. prises of two energy sources; a Lithium ion battery
This motivates the choice of its application in this pa- bank of 211 V and 40 AH lithium ion and a genera-
per. tor/engine combination of 100 kW. The traction motor

3
EVS28 International Electric Vehicle Symposium and Exhibition
of peak power 80 kW is coupled to a single speed trans- the generator and the batteries are operated simulta-
mission with a fixed ratio of 8:1. neously.

4 Design Constraints

All vehicle design constraints are typically derived from

the high level requirements. In [5], [10] and [13], for
the PHEV, the main constraints considered are per-
formance constraints. In this paper, cost and weight
related constraints which use the estimation of unit
Figure 1: KIIRA EV SMACK
cost and weight of the different components are added.
These requirements should be satisfied on a typical
Kampala road cycle. A typical UDDS cycle imple-
mented in Autonomie and a flat grade road with a
constant speed of 50 km/hr are also added. These con-
straints form the basis for the mathematical constraint
and objective functions used in the GA discussed later.
In this paper fuel efficiency and carbon emissions are
not considered because they are of less importance to
the design goals of the project.

Maximum speed on level road (120 km/hr)

Figure 2: Vehicle architecture Acceleration 0 100 km/hr in 10s

Vehicle weight 1500 kg

3.2 Control Strategy AER Range 50 km

Grade ability 17 % at 30km/hr

The initially envisaged control strategy governing the
Cost estimate between USD. 20, 000 and USD.
operation of the drive train was the engine turn on and
30, 000
off . Vehicle start up is supported by the battery bank
for speeds up to 50 km/hr. The engine is run below 50 SOC limit 20 %

km/hr only when the battery State of Charge(SoC) is

below 20 (in this case the generator charges the batter-
5 DSE Approach
ies) or when the requested power is much greater than
the battery can sustain for a long period like during hill The approach employed follows a Y-chart design pro-
climbing. Above 50 km/hr, the vehicle is run on the cess shown in Fig. 5. By benchmarking vehicles in
generator and batteries occasionally switching off the the category of the KIIRA sedan i.e Honda Civic GX,
generator when the battery charge is regained above a Honda Civic Hybrid, Chevrolet Volt, Fisker Karma
certain point of SoC. To maintain the top speed, both and Toyota Prius hybrid, lower and upper specification

4
EVS28 International Electric Vehicle Symposium and Exhibition
bounds are drawn for the different components i.e mo- ampere hour rating AHA, cost of components, final
tors, batteries and generators. This is simply achieved gear ratio GR , DC bus voltage VDC and motor torque
by linear extrapolation using weight and performance. Tm . Equation.2 is typical in defining the pertinent ob-

The models based on the road-load equations are de- jective functions for most of these constraints.

veloped for both the objective function, e.g the Ampere

V
hour (AH) rating and the constraint function, e.g the Pout = (Mv g(fr + cos) + ...
t m
(2)
consumption rate constraint. 1 dV
a C D Af V 2 + M v )
2 dt
The objective and constraint functions are optimized where Pout is the output power from the motor, V is
using the GA, which produces candidate solutions from the velocity of the vehicle, t and m are the transmis-
a population of solutions. Since GA is stochastic, each sion and motor efficiencies, Mv is the vehicle weight,
run produces slightly different results which are all V is the vehicle speed, g is acceleration due to gravity,
pareto-optimal. It is applied 20 times to increase the Af is the vehicle frontal surface area, CD is the coeffi-
population of solutions. The optimal solutions are used cient of drag, and a is the air density. Sections 5.1.1
in a suitable model based simulation (model in the loop to 5.1.3 elaborate further the model representation of
or out of the loop) to arrive at one final candidate. the different components.

5.1.1 Battery Pack Representation

Simplifications are made by assuming that the vehi-

cle uses only series connected batteries in its drive. A
separation of concerns is applied where the vehicle is
treated first as purely electric for a speed range 0-50
km. This criteria is employed so as to independently
design the electric version of the vehicle and also the
conventional version without paying much attention to
the vehicle level control strategy in the initial stages.
Equations.3, 4 , 5 and 6 are used as the objective func-
tions for the battery pack design.

CR AER
AHA = (3)
Figure 3: Y chart design space search VDC
where AHA is the overall ampere hour Rating of the
battery system, CR the consumption rate , AER is
5.1 Problem Representation
the all electric range. The quantity CR AER is the
Using the GA, each candidate solution is looked at a energy content of the energy storage system (ESS). It
chromosome with the genes which are design param- is assumed that from the control requirements, elec-
eters. The genes of the powertrain were identified as tric only system is able to support the vehicle at top
peak generator power PG , maximum motor power PM , constant speed of 50 km/hr for 50 km on a level road.

5
EVS28 International Electric Vehicle Symposium and Exhibition
 Equations 7, 8 and 9 are set as the objective functions
Pout for the genetic multi objective optimization. The lower
CR = (4)
V and upper bounds are informed by benchmarking pro-
Where V is the vehicle velocity. cess of commercial vehicles. Acceleration timeta ), final
speed Vf and gear ratio GR are taken as the variable
V AHA unitCost parameters for optimization. The bounds on these are
Cb = (5)
3.2
set to values in Table.4
Where Cb is the cost of the entire battery pack and
unit cost is the prevailing cost per AH.  
1 Mv 2 1
PM = + Mv gfr Vf + a CD Af Vf3 (7)
t m 2ta 3 5
Wb = AHA (W/AHA) (6) where Vf is the final vehicle speed, Vb is the vehicle
base speed, ta is the time required to accelerate from
Where Wb is the weight of the total battery pack and
the base speed to the final vehicle speed.
W/AHA is the mean weight per AHA.

The consumption rate at 50 km/hr is at least equal to

or greater than Pout/V and the range is not more than P out rd
Tm = torque) (8)
GRmin
50 km at constant speed of 50 km/hr. These restric-
tions form the constraints for the battery pack design. Where Tm is the required motor torque, rd the radius

The lower and upper bounds for CR, AER and VDC of the vehicle wheels and GRmin is the lowest applica-

in the GA are all set to values in Table.3. These are ble gear ratio.

obtained by benchmarking of commercial vehicles.

Nmmax rd
Table 3: Lower and upper battery variable bounds GRmin = (9)
30Vmax
CR AER VDC
Where Vmax is the vehicle maximum speed, Nm is the
LB 0 50 100
motor torque.
UB 200 60 400
Table 4: Lower and upper bounds of motor variables
5.1.2 Motor Specifications ta Vf GR
LB 0 0 0
The most important motor parameters are the peak
UB 20 100 12
power, torque and motor speed. The peak power of
the motor is determined by the acceleration power re-
5.1.3 Peak Generator Power PG
quired to attain a top speed of 96 km(60 mph) within a
given acceleration time. The power required is defined The peak generator rating is derived from the required
in Equation.7. The torque supplied by the motor de- power to maintain the vehicle at a particular speed
pends on the wheel radius and the final effective gear or grade. According to the control strategy, it is re-
ratio used as shown in Equation.8. The base speed or quired that the generator is started at speeds above 50
the required motor speed(rpm) is computed according km/hr. In a series hybrid, the engine is considered as
to Equation. 9. another power source which delivers power to the DC

6
EVS28 International Electric Vehicle Symposium and Exhibition
bus. Therefore its optimization is similar to the opti-
mization of the battery. Other parameters like weight, Table 5: Optimization method
cost can also be optimized. It is assumed that above
50 km/hr, the battery delivers all the available power Optimization Process
with the generator assistance. This leads to a global Step 1:Set Vehicle constraints ie Range and Top Speed
optimization function given according to Equation.10. Step 2:Compute and optimize the consumption Rate
Optimization for the generator is subject to the same Step 3:Apply GA to AH, Weight,
constraints as the motor. cost and volume ,PM , RPM, Tm and Pg
Step 4:Extract the population of points
PG = P out P batt (10)
Step 5: Apply pareto front to the population of points
Where Pgen is the required engine rating at a given de- Step 6: Extract pareto point and run a vehicle model
sign top speed, Pout is the required power at the driven in Autonomie
wheel and Pbatt is the maximum continuous power from Step7: Compare SoC and Top speed results
the batteries. and percentage of road cycle missed

6 Results and Discussion

P G = e PE (11)

The generator and engine specification are related be- 2D and 3D pareto graphs are plotted for the different

cause the engine rating depends on the calculated gen- objective functions. The GA generally gives different

erator requirement where e is the conversion efficiency optimal results during each run because it is stochas-

of the engine tic, therefore, it is applied within 20 iterations to ob-

tain different optimal results which are later used in
The cost associated with the particular generator en-
the simulink model for verification.
gine combination is given as
In Figs. 6,6 and 7 most of the points from the de-
X
CG,E,M = Pi U nitcost (12) sign space population appear in the pareto front. This
i=G,E,M
is because GA is probabilistic and results in different
Where CG,E,M Is the cost of the generator, engine and
points but most of the points are always close to the
motors.
optimal solutions in one or more objectives. The values
at these pareto points are reproduced in Tables 6,7 and
5.2 Matlab Implementation 6.1. In Figs.6, the battery pack Energy capacity can
be designed to values between 41 AH and 121 AH, for
The the optimization algorithm is applied as illustrated
a DC voltage rating between 119 VDC and 328 VDC.
in Table.5.
The final solution is subject to other factors which can-
not be explored extensively by the optimisation. These
factors are investigated in Autonomie.

In Fig.7, the traction motor design values ranges be-

tween 56 kW and 172 kW, for acceleration times rang-

7
EVS28 International Electric Vehicle Symposium and Exhibition
ing between 18s and 3 seconds respectively. The engine
specifications range between 15 kW and 88 kW.

Figure 7: Motor power vs speed vs torque

6.1 Vehicle Level Verification

Figure 4: Battery Energy Vs DC bus Voltage
The genetic optimization does not produce one final set
of specifications which can be used as a blue print for
the powertrain but prunes down the design choices to a
few viable results. To arrive at one set of design speci-
fications, a suitable vehicle model in Autonomie is run
with a combination of several parameter specifications
from the optimization. The specifications obtained
from the pareto graphs are used. The motor, battery
and engine specifications are given in the Tables 6, 7
and 8 respectively. To test for battery performance, an
all electric drive is used with 50 km/hr top speed, 50

Figure 5: Engine power vs vehicle speed km range and the motor specifications obtained from
the pareto graphs. Two versions of the same vehicle
are developed, a purely electric vehicle and a hybrid
vehicle. The most important statistics obtained from
the simulations are the battery SoC, battery voltage
decline, percentage of drive cycle missed, the maxi-
mum attainable speed and the acceleration times for
the different vehicle configurations.

Figure 6: Motor power vs acceleration time

8
EVS28 International Electric Vehicle Symposium and Exhibition
 shelf components used in the final design were informed

Table 6: Motor specification by these specifications.

Future research shall focus on optimization of the over-
Power 92 67 110 120 69 56
all vehicle weight, vehicle volume, prototyping costs
ta 8 12 7 6 12 15
and control strategy.
172 75 74
4 10.6 11

Table 7: Battery specification Acknowledgement

DC 328 197 259 158 240 203 248 The authors would like to thank the government of
AHA 41 68 53 84 58 66 54 0 the Republic of Uganda for supporting the KIIRA EV
119 245 281 research and development program.

121 55 5
References
Table 8: Engine specifications
[1] Eunsuk Kang, Ethan Jackson, and Wolfram
Speed (km) 90 100 110 120 130 140 150 Schulte. An approach for effective design space

Engine(kW) 15 20 25 32 39 47 56 exploration. In Foundations of Computer Soft-

160 170 180 ware. Modeling, Development, and Verification of

65 76 88 Adaptive Systems, pages 3354. Springer, 2011.

[2] Shaik Amjad, S Neelakrishnan, and
7 Conclusion R Rudramoorthy. Review of design considera-
tions and technological challenges for successful
In this paper, the design process for the KIIRA EV
development and deployment of plug-in hybrid
SMACK Hybrid powertrain has been demonstrated.
electric vehicles. Renewable and Sustainable
We have shown that by using Design Space Explo-
Energy Reviews, 14(3):11041110, 2010.
ration techniques with a suitable component optimiza-
[3] M Steinhauer, K Shea, et al. Efficient design space
tion method, the Genetic algorithm , the design choices
exploration and optimization using the example
can be pruned without missing the optimum values.
of plug-in hybrid electric vehicle architectures. In
These specifications can then be used in an simulation
DFX 2010: Proceedings of the 21st Symposium on
environment which gives a more realistic idea about
Design for X, Buchholz/Hamburg, Germany, 23.-
the impact of the choices on other parameters like fuel
24.09. 2010, 2010.
efficiency. [4] ANL. Vehicle modelling approaches.
The powertrain specifications for the desired perfor- [5] Phillip Sharer, Aymeric Rousseau, Sylvain
mance obtained from the optimization and simulation Pagerit, and Paul Nelson. Midsize and suv vehi-
are; motor power of 110 kW, DC bus voltage of 259 cle simulation results for plug-in hev component
VDC and AHA of 53 AH with a generator of 88 kW requirements. Technical report, SAE Technical
for a top Speed of 180 km/hr. Other parameters like Paper, 2007.
gear ratios are not extensively discussed but were con- [6] Manu Jain, Chirag Desai, and Sheldon S
sidered during the entire design process. The off-the- Williamson. Genetic algorithm based optimal

9
EVS28 International Electric Vehicle Symposium and Exhibition
 powertrain component sizing and control strat- Authors
egy design for a fuel cell hybrid electric bus. In
Vehicle Power and Propulsion Conference, 2009.
VPPC09. IEEE, pages 980985. IEEE, 2009.
[7] AH Shahirinia, SMM Tafreshi, A Hajizadeh Gas-
taj, and AR Moghaddomjoo. Optimal sizing of
hybrid power system using genetic algorithm. In
Future Power Systems, 2005 International Con-
ference on, pages 6pp. IEEE, 2005.
[8] Morteza Montazeri-Gh and Amir Poursamad. Ap-
plication of genetic algorithm for simultaneous Richard Madanda has worked as a researcher power-
optimisation of hev component sizing and con- train and charging infrastructure at the Center for Re-
trol strategy. International Journal of Alternative search in Transportation Technologies now Kiira Mo-
Propulsion, 1(1):6378, 2006. tors Corporation. He is currently pursuing a Masters
[9] Jian Ping Gao, Zhen Nan Liu, Zhi Jun Guo, and degree at the Eindhoven University of Technology, the
Yue Hui Wei. Optimization of hybrid electric bus Netherlands.
control strategy with hybrid optimization algo-
rithm. Applied Mechanics and Materials, 341:924
930, 2013.
[10] Hillol Kumar Roy, Andrew McGordon, and
Paul A Jennings. Real-world investigation of a
methodology for powertrain component sizing of
hybrid electric vehicles. Paul Isaac Musasizi is an assistant lecturer in the De-
[11] Nikolce Murgovski, Lars Johannesson, Jonas partment of Electrical and Computer Engineering at
Sj
oberg, and Bo Egardt. Component sizing of a Makerere University. He is also the Associate Princi-
plug-in hybrid electric powertrain via convex op- pal Investigator(Engineering) at Kiira Motors Corpo-
timization. Mechatronics, 22(1):106120, 2012. ration.
[12] Xiaolan Wu, Binggang Cao, Xueyan Li, Jun Xu,
and Xiaolong Ren. Component sizing optimiza-
tion of plug-in hybrid electric vehicles. Applied
Energy, 88(3):799804, 2011.
[13] Vincenzo Galdi, Lucio Ippolito, Antonio Piccolo,
and Alfredo Vaccaro. A genetic-based methodol-
ogy for hybrid electric vehicles sizing. Soft Com- Arthur Tumusiime Asiimwe an assistant lecturer in the
puting, 5(6):451457, 2001. department of Electrical and Computer Engineering.
He is the Principal Researcher(Electrical Engineering)
at Kiira Motors Corporation.

10
EVS28 International Electric Vehicle Symposium and Exhibition
Junior Africa currently works as a Researcher Power- Sandy Stevens Tickodri-Togboa is an Engineering Sci-
train and Charging Infrastructure department at Kiira entist and Professor of Electrical and Computer Engi-
Motors Corporation. neering at Makerere University, Uganda. He received
his PhD in Digital Communications in 1985, MSc in
Radio Engineering in 1979 and BSc in Electrical En-
gineering in 1973. He is the Principal Investigator of
the Kiira Motors Corporation.

11
EVS28 International Electric Vehicle Symposium and Exhibition