Hardware-in-the-Loop Validation of a Power Management Strategy forHybrid Powertrains

Youngki Kima, Ashwin Salvia, Jason B. Siegela, Zoran S. Filipib, Anna G. Stefanopouloua, Tulga Ersala,∗

aMechanical Engineering, University of Michigan, Ann Arbor, 48109bAutomotive Engineering, Clemson University, Greenville, South Carolina, 29607

Abstract

Previously, a hybrid powertrain management strategy was developed that controls the power sources basedon frequency content, mitigating aggressive engine transients. This article presents a hardware-in-the-loopvalidation of this strategy, with a real engine and battery integrated into a diesel hybrid electric vehiclesimulation, thereby allowing for a realistic evaluation of fuel economy, soot emissions, and battery life.Considering an aggressive drive cycle and a state-of-charge regulation strategy as a benchmark, the frequency-based strategy yields 5.9% increase in fuel economy, 62.7% decrease in soot emissions, and 23% reduction ineffective Amp-hours processed, which should yield an increase in battery life.

Keywords: hardware-in-the-loop simulation, power management, hybrid electric vehicles, optimization,validation

1. Introduction

Vehicle powertrain hybridization is one of thepromising pathways for improved fuel economy andreduced tailpipe emissions, where energy storage de-vices, such as hydraulic or pneumatic accumulators orbatteries, are used in conjunction with internal com-bustion engines. Various topologies for hybridiza-tion have been explored; e.g., series (Jalil et al.,1997; Filipi and Kim, 2010), parallel (Liu et al.,2008; Yang et al., 2012), and power split (or series-parallel) (Liu and Peng, 2008; Li and Kar, 2011).They all demonstrated improvements in fuel economyand some showed reduction in emissions.

∗Corresponding author. 1231 Beal Ave., Ann Arbor, MI48109. (+1 734) 763-7388. tersal@umich.edu

Email addresses: youngki@umich.edu (Youngki Kim),asalvi@umich.edu (Ashwin Salvi), siegeljb@umich.edu(Jason B. Siegel), zfilipi@clemson.edu (Zoran S. Filipi),annastef@umich.edu (Anna G. Stefanopoulou),tersal@umich.edu (Tulga Ersal)

Hybrid powertrain technology has already beensuccessfully deployed on some passenger vehicles(Lave and MacLean, 2002). Heavy-duty military ve-hicles could benefit from this technology, as well.Even though they have significantly different perfor-mance requirements and driving patterns than thoseof the passenger vehicles, the goals of reducing fuelconsumption and emissions are still the same. Min-imizing soot emissions is extremely desirable withinthe military context to reduce the vehicles visual sig-nature and increase survivability. Further require-ments such as silent watch, increased mobility, en-hanced functionality for on-board power, and im-proved export-power capabilities make hybrid electricconfigurations more attractive than other hybrid ar-chitectures. Among various hybrid electric configura-tions, the series configuration has drawn interest dueto greater flexibility in vehicle design when it comesto considerations such as the V-shaped hull design tomaximize the survivability of the crew during blastevents (Ramasamy et al., 2009). Therefore, with the

This is an Author’s Accepted Manuscript of an article published in Control Engineering Practice 29 (2014) 277-286

specific military application in mind, the focus of thisarticle is on the series hybrid electric architecture.

The performance of a hybrid powertrain in termsof reducing both fuel consumption and emissions crit-ically depends on the power management strategy;that is, the supervisory control algorithm that deter-mines how the total power demanded by the driverwill be shared between the engine and, for exam-ple, the battery. Many power management strate-gies for series hybrid electric vehicles have been pro-posed to fully exploit their potential for minimizingfuel consumption, emissions, and/or battery health(Jalil et al., 1997; Caratozzolo et al., 2003; Pisu andRizzoni, 2005; Konev et al., 2006; Kim and Filipi,2007; Di Cairano et al., 2012; Li and Feng, 2012; Kimet al., 2012b; Serrao et al., 2011; Michel et al., 2013;Serrao et al., 2013).

Among many strategies proposed, Konev et al.and Di Cairano et al. highlight the importance of asmooth engine operation, where smoothness is char-acterized by the rate of change in power. Dependingon the engine specifications, different rate of changethresholds can be used to define the smooth opera-tion threshold. Smooth operation is important fortwo reasons: (1) it allows the engine to operate closeto the steady-state conditions where the operation isoptimal in terms of pointwise powertrain efficiency;and (2) reducing aggressive transients also reducessoot emissions. To achieve such smooth operation,Konev et al. and Di Cairano et al. propose meth-ods to smoothen the power demand that is requiredfrom the engine. In their work, Konev et al. and DiCairano et al. focus on the benefits of this strategyfrom the engine perspective only and within the con-text of passenger vehicles with spark-ignition engines.Serrao et al. developed a power management strat-egy accounting for tailpipe emissions such as NOxin (Serrao et al., 2013); however, soot emissions – asignificant factor in military vehicles – were not con-sidered. Therefore, the impact of this strategy withinthe context of military vehicles with diesel engines isstill an open-research question. Furthermore, the im-pact of engine power smoothing strategy on the bat-tery operation and battery health has yet to be stud-ied. Michel et al. accounted for the battery thermalbehavior in the equivalent consumption minimization

strategy in (Michel et al., 2013) where battery healthwas indirectly considered based on knowledge thatcapacity fade can be accelerated at high tempera-tures; however, the battery life estimation was notconducted. Thus, this article is aimed to investigatethe effects of a power-smoothing strategy in a serieshybrid electric military vehicle with a diesel engine.The effects are addressed from the perspective of boththe engine and the battery.

Towards this end, a frequency-domain power dis-tribution (FDPD) strategy is considered that hasbeen proposed by Kim et al. (Kim et al., 2012b).The FDPD strategy manages power flow by split-ting power demand into low and high frequency com-ponents through low-pass filtering incorporated withload-leveling. Model-based simulations have shownthe method to be capable of achieving: 1) reducedbattery electric loads, 2) smooth engine transients;and 3) less fuel consumption. However, a methodto tune the FDPD has not yet been proposed. Fur-thermore, the simulation-based validations were per-formed with static maps to represent the engine andits emissions. The true transient performance ofFDPD with real hardware has not yet been studied.

With this motivation in mind, this article makestwo original contributions to this body of litera-ture. First, it provides a design methodology to tunethe frequency-based supervisory controller. Controlparameters are systematically optimized through amodel-based two-stage optimization process. Sec-ond, the FDPD strategy is evaluated experimentallythrough a networked simulation setup with a real en-gine and battery in the loop. This hardware-in-the-loop simulation allows not only for a more realisticevaluation of the fuel economy benefits of the con-troller compared to the purely model based evalua-tions in the literature, but also for an assessment ofits soot emissions benefits for the first time. In ad-dition, the battery life is estimated using a weightedAmp-hour (Ah) processed model (Serrao et al., 2009;Onori et al., 2012) to account for thermal effects. Athermostatic control strategy is also considered as abaseline power management strategy, and the FDPDis compared to the baseline strategy in terms of per-formance. This article is based upon the preliminarywork reported in (Kim et al., 2012a) and extends it

2

by putting a real battery in the loop in addition tothe engine and by providing emissions measurements,as well. A more detailed estimation of the battery lifeis also included.

The rest of this article is organized as follows. Sec-tion 2 gives an overview of the power managementstrategies considered in this article and also proposesa method to tune the FDPD strategy. Section 3presents the vehicle system considered as a case studyand optimized control parameters. Hardware-in-the-loop setup and experimental results are presentedand discussed in Section 4, and conclusions are drawnin Section 5.

2. Overview of Power Management Strategies

The primary task of a power management strategy(PMS) is determining the power flow between the ve-hicle, engine and battery to minimize a cost functionsuch as fuel consumption and emissions. Specifically,a series hybrid configuration can take advantage ofthe decoupling of the engine from the wheels to op-erate the engine at the optimal conditions. However,the decrease in total system efficiency due to inherentmultiple energy conversions, and other constraintssuch as battery voltage and current limitations makethe power management problem a challenging task.Therefore, the design of power management strategyis important to improve fuel economy while reducingengine emissions and to ensure safe battery opera-tions.

2.1. Thermostatic battery state-of-charge (SOC)control

Thermostatic SOC control, a heuristic controltechnique, has been widely employed for series hy-brid electric vehicles (Caratozzolo et al., 2003; Leeet al., 2011; Li and Feng, 2012). This strategy isadvantageous because of its ease of implementation,the effectiveness of SOC regulation, and improvedfuel economy. Thus, the thermostatic SOC controlis considered as a baseline strategy in this study.

Figure 1 summarizes the principle of the thermo-static SOC control. As long as current SOC is higherthan the target SOC, the engine provides zero power.

The engine starts charging the battery with the pre-determined power level when SOC drops to the tar-get SOC. A dead band is implemented to preventfrequent engine on/off’s. When the power demandfor vehicle propulsion is higher than the battery dis-charging power limit, the engine operates in powerassisting mode.

1.0

SOC

Engine Power

Max. Power

ThresholdPower

0TargetSOC

Dead band

Max PowerSOC

Figure 1: The schematic of thermostatic SOC control

However, the thermostatic SOC control has severaldrawbacks. Since the engine is commanded to pro-vide power demand above threshold level, the engineoperation changes suddenly and aggressively fromzero power demand. This behavior considerably de-teriorates tailpipe emissions (Hagena et al., 2011) andalso prevents the engine from following the optimaloperation line (Di Cairano et al., 2012). Moreover,the heavy-duty diesel engine cannot follow the aggres-sive command because of its large inertia and turbo-charger lag. This is a problem because the batteryhas to provide the remainder of the power demand,resulting in more electrochemical-mechanical stressesin the battery (Lee et al., 2011). Finally, in termsof improving the fuel economy, this strategy cannotavoid multiple power conversions since it prefers us-ing the battery power to the engine/generator power.

2.2. Frequency Domain Power Distribution Strategy

The separation of power demand in frequency do-main provides control inputs to each power sourcetailored according to speed of response. Unlike theturbo-charged diesel engine, the battery can absorb

3

and provide high frequency power demands withoutdelay in response. Therefore, by splitting the totalpower demand into low frequency and high frequencycomponents, each power source can be utilized moreeffectively. The low frequency components capturethe smooth trajectory of the power demand, whereasthe high frequency components cover the small am-plitude but aggressive and transient power demand.The smooth power demand transitions can help re-duce engine emissions, whereas the reduced ampli-tude of the electric load is beneficial to mitigate elec-trical stress on the battery which will be discussed inSection 4.

Low frequency

Pdmd,0

SOCref SOC I

∆SOC

∆ Pdmd

+ -

+ -

Coulomb Counting

Battery

Pdmd,1

Engine/ Generator

High frequency

Low-pass Filtering

Pdmd,2 Mode Decision

Pdmd,3 Power Calculation

SOC Regulator

FDPD

Figure 2: The schematic diagram of the FDPD strategy

Figure 2 shows the structure of the proposed strat-egy consisting of: 1) FDPD module; 2) SOC regu-lation module; and 3) mode decision module. TheFDPD module for the hybrid electric vehicle (HEV)mode determines the engine/generator power de-mand by splitting the total power demand into lowand high frequency components. The power demandPdmd,1 is determined as follows:

if Pth,1 ≤ Pdmd,0 ≤ Pth,2 thenPdmd,1 = Pdmd,0 + ∆Pdmd

else if Pdmd,0 > Pth,2 thenPdmd,1 = Pth,2 + ∆Pdmd

elsePdmd,1 = ∆Pdmd

end

where Pdmd,0 and Pdmd,1 are power demand for vehi-cle propulsion and total power demand respectively.Parameters Pth1 and Pth2 are threshold power levelsfor HEV mode incorporated with load-leveling, and

τLF is the time constant of a low-pass filter. Then,the power demand Pdmd,2 is filtered using a first orderlow-pass filter:

τLFdPdmd,2

dt+ Pdmd,2 = Pdmd,1. (1)

A first order filter is used in this study since a first-order filter outperforms higher order filters in termsof fuel economy and engine smoothness as suggestedby Kim et al. (Kim et al., 2012b).

The feedback power demand ∆Pdmd for the bat-tery SOC regulation is determined through theproportional-integral (PI) controller given by

∆Pdmd = kP ∆SOC + kI

∫∆SOCdt, (2)

where ∆SOC is the difference between the currentand reference SOC values; kP and kI are proportionaland integral control gains respectively.

The mode decision module determines drivingmodes. The modes change between an electric-vehicle (EV) mode, a hybrid electric vehicle (HEV)mode and a performance vehicle (PV) mode accord-ing to the following:

if Pdmd,2 ≤ Pth,1 thenPdmd,3 = 0: EV mode

elseif Pdmd,2 ≥ Pdmd,0 − Pbatt,max then

Pdmd,3 = Pdmd,2: HEV modeelse

Pdmd,3 =min(Peng,max, Pdmd,0 − Pbatt,max): PVmode

end

end

where Peng,max and Pbatt,max are the maximum avail-able engine power and battery discharging power, re-spectively. Consequently, the performance of FDPDstrategy is determined by five control parameters;namely, τLF , Pth1, Pth2, kP , and kI . These five pa-rameters are determined through a model-based two-stage optimization process as described next.

4

2.3. Model-based Control Parameter Optimization

In this section, the formulation of the optimiza-tion of control parameters for the thermostatic andFDPD strategies using a model-based simulation ispresented. A hybrid vehicle is a complicated systemthat includes both energy conversion and energy stor-age among various power/energy sources. Since nu-merical round-off, interpolation inaccuracy, and dis-crete events in the vehicle simulation lead to disconti-nuity and computational noise in the objective func-tion (Assanis et al., 1999; Gao and Porandla, 2005),gradient-based optimization algorithms are not fre-quently used. Thus, a two-stage optimization frame-work was used in this study to take advantage ofboth derivative-free (global) and gradient-based (lo-cal) optimization algorithms. First, a non-gradientbased optimization algorithm searches for the globalminimum over a bounded domain. Then, the set isused as an initial point for a gradient-based algo-rithm with fast convergence. The DIviding RECTan-gles (DIRECT) algorithm is used for the global opti-mization, wherein the feasible region of design vari-ables is divided into n-dimensional hyper-cubes andhyper-rectangles and the objective function is eval-uated at the center of the hyper-cubes and hyper-rectangles. This algorithm has several advantages(Jones et al., 1993): (1) it searches for global andlocal optima; (2) parameter tuning is not required;(3) both equality and inequality constraints can beeasily handled; (4) it is robust for nonlinear prob-lems. For the subsequent local optimization algo-rithm, Sequential Quadratic Programming (SQP) isused. Both DIRECT and SQP are implemented inMATLAB through the gclsolve.m code by Holmstrom(Holmstrom, 1989) and the built-in MATLAB func-tion fmincon, respectively.

The control parameter optimization can be math-ematically formulated as following:

Objective : Maximize fuel economy

Subject to |∆vveh | ≤ ∆vveh,ref within 1 second

SOCL ≤ SOC ≤ SOCU

SOCend,L ≤ SOCend ≤ SOCend,U

P (Peng ≤ α) ≥ β

where the subscripts ref and end represent the refer-ence and the end of driving cycles and the subscriptsL and U denote the lower and upper bounds, respec-tively. The difference between the desired and actualvehicle speeds is represented by ∆vveh and ∆vveh,refis set to 1.6 km/h based on the regulation of USEnvironmental Protection Agency (EPA)1. The SOCshould be bounded during the vehicle operation andthe SOC at the end of the driving cycle should be suf-ficiently close to the target SOC. The variable Peng isthe derivative of engine power demand with respectto time; thus, the last constraint can be interpretedas a minimum probability of β for an engine powercommand rate less than α. The parameters α andβ are set to 40 kW/s and 95% adopted from (Kimet al., 2012b). For the purpose of accounting for theremaining battery SOC, two consecutive driving cy-cles are considered; the fuel economy is calculated by

fuel economy =1

Ns

Ns∑k=1

∫ tk+tcycletk

vveh dt∫ tk+tcycletk

mf dt, (3)

where tcycle is the total time of the given drivingcycle, vveh is the velocity of the vehicle and Ns isthe total number of tk’s that satisfy the condition:tk � SOC (tk) = SOC (tk + tcycle).

The same optimization formulation is used to tuneboth the thermostatic SOC and FDPD control strate-gies.

3. Description of the Hardware-in-the-loopStudy

As a case study, a hybridized Mine Resistant Am-bush Protected All-Terrain Vehicle (M-ATV) is con-sidered to explore the effectiveness of the FDPDstrategy in severe circumstances including frequentand high power demand. The specifications of theM-ATV are summarized in Table 1

A networked hardware-in-the-loop simulation (Er-sal et al., 2011, 2012, 2013) of this vehicle systemis considered to enable a system integration despite

1EPA, Code of Federal Regulations, Title 40 Chapter I,§600.109-08 EPA driving cycles

5

Table 1: Vehicle Specification

Parameter Value UnitWeight 14403 kgPayload 1814 kg

Frontal area 5.72 m2

Diesel engine 260 kWGenerator 236 kWBattery 9.27 kWhMotor 380 kW

the fact that the components resided in different ge-ographic locations. The engine and battery are thehardware components located in two different loca-tions, and the remaining components of the vehi-cle system (i.e., generator, motors, vehicle dynamics,and driver) are mathematically modeled and simu-lated in a third location to enable a future study witha human driver in the loop setup at a location not col-located with the engine or battery similar to (Ersalet al., 2011). One of the important considerationsin such networked simulation is selecting the locationof the coupling point – how to distribute the modelsbetween the sites (Ersal et al., 2013). The couplingpoint significantly affects how the system dynamicsare affected by the network dynamics (e.g., delay).To this end, for the three-site networked setup con-sidered in this work, there are three options for wherethe PMS can be placed: with the driver, battery, orengine. Among these three options, placing the PMSwith the driver is advantageous, because this is theonly option which does not require a third communi-cation channel between the engine and battery loca-tions. Working with only two communication chan-nels (between the driver site and the engine site, andbetween the driver site and the battery site) decreasesthe sensitivity of the simulation to communicationdelays.

More details about the SHEV model are describedin Appendix A. The overview of the networked sys-tem architecture is illustrated in Fig. 3.

Location 2

Location 3

Location 1

Driver PMS

Engine Generator

Battery

Vehicle Motor

Drive cycle

Vehicle speed

Desired E/G power

Enginepower

Throttle,brake

Voltage, current

Generator power

Batterypower

Motorpower

Desired motor powerBrake

Torque

Speed

Figure 3: The overview of the networked hardware-in-the-loopsimulation setup. Each shaded area represent a different geo-graphic location. Italic typeface denotes actual hardware com-ponents. Dashed lines represent the signals communicated overthe network. PMS stands for the power management strategy.

Figure 4: A photo of the engine-in-the-loop testing facility.The engine is on the left, and the dynamometer is on the right.

3.1. Engine-in-the-Loop Setup

A Navistar 6.4 L V8 diesel engine with 260 kWrated power at 3000 rpm and a rated torque of 880Nm at 2000 rpm is used for this study. It is in-tended for a variety of medium-duty truck applica-tions covering the range between classes IIB and VII,and features technologies such as high pressure com-mon rail fuel injection, twin sequential turbocharg-ers, and exhaust gas recirculation. A high-fidelity,AC electric dynamometer couples the physical enginewith the simulation models in real time and operates

6

Figure 5: The schematic of the measurement principle of theDMS system (Reavell et al., 2002)

in speed control mode. The setup is connected toMatlab/Simulink for integration with mathematicalmodels, allowing for a real-time hardware-in-the-loopsimulation. This connection is achieved through anEMCON 400 flexible test bed with an ISAC 400 ex-tension (Filipi et al., 2006). The photo of the engine-in-the-loop setup is shown in Fig. 4.

Transient soot emissions are measured with a Dif-ferential Mobility Spectrometer (DMS) 500 manufac-tured by Cambustion Ltd. in the form of temporallyresolved particulate concentrations. The DMS 500,whose measurement principle is illustrated in Fig.5, offers measurement of different particle sizes byidentifying the mobility of particles with a samplingfrequency of 10 Hz and a response time of 200 ms.Therefore, the DMS 500 makes it possible to ana-lyze the time evolution of the soot emissions. In thisstudy we consider a military application and thus fo-cus on engine out emissions due to the fact that mili-tary vehicles do not use aftertreatment systems. Sootemission measurements are performed for a time win-dow of 750 seconds due to the limitation of the mea-surement instrument for reliable and repeatable data.After 750 seconds, the measurements start becomingunreliable due to contamination in the instrument.

3.2. Battery-in-the-Loop Setup

A cylindrical 26650 power cell manufactured byA123 systems is used for the battery-in-the-looptest. The cell chemistry is LFP with capacity of2.3 Ah. The specifications of the battery pack andcell are summarized in Table 2. Since a single cell

is tested, the battery current (or demand) requestedby the power management strategy Ipack has to bescaled down to the cell level Icell given by Icell =Ipack/np. This scaled current demand is applied tothe battery cell using a Bitrode Battery Test SystemFTV1-200/50/2-60. The battery SOC is estimatedbased on coulomb counting given by

dSOC

dt= − Icell

3600Cb(4)

where I and Cb are current and battery capacity re-spectively.

Figure 6: A photo of the battery-in-the-loop testing facility

As seen from Fig. 6, the battery cell is placed ina designed flow chamber which emulates forced-aircooling conditions inside the battery pack (or mod-ule). Since the flow chamber is located in a ther-mal chamber, the ambient temperature can be con-trolled around a predetermined temperature at 25◦C.In particular, forced-air cooling is performed by con-trolling the fan speed and temperature inside thethermal chamber. To investigate the performance

Table 2: Specifications of battery pack and cell

Pack Symbol ValueNumber of cells in series ns 130

Number of cells in parallel np 10

Cell Symbol ValueNominal capacity Cb 2.3 AhNominal voltage Vn 3.3 V

7

of different power management strategies in termsof battery life, the temperature of the battery dur-ing operation is also measured using a T-type ther-mocouple. The sensor accuracy is the maximumof 0.5◦C or 0.4% according to technical informationfrom the manufacturer. Similar to the engine-in-the-loop setup, the battery-in-the-loop setup is also inter-faced to Matlab/Simulink for a real-time hardware-in-the-loop simulation.

4. Experimental Results and Discussion

0 200 400 600 800 1000 12000

20

40

60

80

Time (sec)

Vel

ocity

(km

/h)

Figure 7: The speed profile of the Urban Assault Cycle

The performances of the two power managementstrategies are investigated using an aggressive mili-tary driving cycle, Urban Assault Cycle (UAC) (Leeet al., 2011), which features frequent acceleration anddeceleration events. The velocity profile of this driv-ing cycle is displayed in Fig. 7. The parameters of thebaseline thermostatic SOC and the FDPD strategies,both optimized for the UAC, are summarized in Ta-bles 3 and 4. Note that the baseline strategy behaveslike proportional control above the threshold powerof 20 kW; that is, the engine power increases by 34.3kW to compensate for every 0.01 decrease in batterySOC. Figure 8 shows the power spectral analysis ofthe UAC and the cut-off frequency τLF obtained asa result of the optimization.

To highlight the performance of the power man-agement strategies, specific time periods are shownin Fig 9. There is no difference in vehicle speed be-tween the FDPD and thermostatic strategies, imply-ing that the vehicle performance is not deteriorating.The engine power demand gradually changes underthe FDPD strategy when the power demand is higher

Table 3: Parameters of the thermostatic SOC strategy opti-mized for the UAC

Parameter Value UnitTarget SOC 0.5 -Deadband 0.02 -

Max. power SOC 0.43 -Max. power 260 kW

Threshold power 20 kW

Table 4: Control parameters of the FDPD strategy optimizedfor the UAC

Parameter Value UnitCut-off frequency, τLF 0.21 Hz

Threshold power 1, Pth,1 16.7 kWThreshold power 2, Pth,2 116.7 kW

Proportional gain, kP 836.2 -Integral gain, kI 4.537 -

0 0.1 0.2 0.3 0.4 0.50

0.5

1

1.5

2

2.5

3x 1013

Frequency (Hz)

Mag

nitu

de (W

2 )

LF

=0.21 Hz

Figure 8: Power spectral density of the power demand of theUAC

than 116.7 kW, the threshold power level 2, Pth,2. Ingeneral, the actual engine power can track the desiredengine power very closely, indicating that the enginecan operate very close to the optimal operation line.In contrast, the baseline thermostatic SOC strategyalways commands engine power demand above thethreshold level of 20 kW with high power rate. More-

8

610 620 630 640 650 660 670 680 690 7000

200

400

600

800

Time (sec)

Soo

t (m

g/m

3 s)

610 620 630 640 650 660 670 680 690 7000

100

200

300

Eng

ine

Pow

er (k

W)

610 620 630 640 650 660 670 680 690 700−200

0

200

Bat

tery

Pow

er (k

W)

610 620 630 640 650 660 670 680 690 700

0.48

0.5

0.52

0.54

Bat

tery

SO

C (−

)

610 620 630 640 650 660 670 680 690 700

20

40

60

80

100

Vel

ocity

(km

/h)

FDPDThermostatic

Pdmd,3 PFDPD

act

ThermostaticPdmd,3 Pact

DesiredFDPDThermostatic

FDPDThermostatic

FDPDThermostatic

Figure 9: Comparison of engine power demand, actual enginepower, and soot emissions under different power managementstrategies. Soot emissions are measured from 0 to 750 secondsduring the drive cycle since the electrometer detectors showdrift when they are exposed to high concentrations for a longtime period.

over, it can be seen that the diesel engine cannotfollow an aggressive command with high power ratedue to its slow dynamics; therefore, the battery hasto provide the remaining propulsion power until theengine power demand is satisfied. The thermostaticSOC strategy requires significantly high power de-mand to charge the battery for SOC regulation. On

the other hand, the FDPD strategy reduces the powerdemands higher than threshold power level 2, lead-ing to a reduced occurrence of aggressive transients.Specifically, these smooth engine transients under theFDPD strategy result in 62.7% reduction in the ac-cumulated soot emissions from 0.822 g/km to 0.306g/km for the first 750 s.

Furthermore, the FDPD strategy does not empha-size battery charging as much as the thermostaticSOC strategy does. The decrease in multiple powerconversions could improve system efficiency. Specifi-cally, the fuel economy is improved by 5.9% from 2.87km/l to 3.04 km/l compared to the thermostatic SOCstrategy over the UAC. The term Ns used in the fueleconomy calculation in Eq. (3) is found to be 166and 49 for the thermostatic and FDPD strategies,respectively.

−15 −10 −5 0 5 10 150

0.1

0.2

0.3

0.4

C−rate

Occ

uren

ce

FDPDThermostatic

0 50 100 150 200 250 3000

0.5

1

Power rate (kW/s)

Occ

uren

ce

0 100 200 300 400 5000

0.2

0.4

0.6

0.8

Soot (mg/m 3)

Occ

uren

ce

Figure 10: Battery C-rate, engine power rate, and soot con-centration histograms

Figure 10 shows the histogram of battery cell oper-

9

0 500 1000 1500 2000 25000.45

0.5

0.55

SO

C (−

)

0 500 1000 1500 2000 250025

30

35

40

Time (sec)

Tem

pera

ture

(C

)o

FDPD Thermostatic

Figure 11: Comparison of battery SOC and temperature underdifferent power management strategies

ation and engine operation with the two power man-agement strategies. The frequency of high batterycurrents and aggressive engine power demands (asquantified by the engine power rate) are significantlyreduced in case of the FDPD strategy, leading to thedecrease in soot emissions. Specifically, the amountof time the battery spends in high C-rate (more than±5 C) is reduced from 25% to 18% with the FDPDstrategy. A C-rate is a measure of the rate at whicha battery is discharged relative to its maximum ca-pacity. A 1C rate means that the discharge currentwill discharge the entire battery in 1 hour. Addi-tionally, the high power rate (more than 50 kW/s)operation time of the engine is reduced from 51% to1%. Since Joule heating dominates the heat gener-ation from the battery as discussed in (Kim et al.,2014), a lower operating temperature of the batteryis expected corresponding to lower average C-rate.Indeed, Fig. 11 shows that the battery temperatureis 3◦C lower with the FDPD than the thermostaticstrategy for the same cooling condition, even thoughboth strategies regulate the battery SOC around thetarget value of 0.5. These results could lead to an ad-ditional benefit of a reduction in power/energy con-sumption for battery cooling.

Battery degradation over the driving cycle is es-timated by using the weighted Ah-processed model

discussed in (Onori et al., 2012). This approach usesthe linear cumulative damage concept to analyze bat-tery degradation. To account for the influence ofoperating conditions such as temperature Tbatt andstate-of-charge (SOC) or depth-of-discharge (DOD)on the degree of degradation, Onori et al. intro-duced the severity factor σ, a nonlinear function ofbattery temperature and SOC. The effective accumu-lated Ah-processed is calculated by using

Aheff =

∫σ(Tbatt,SOC)|I|dt. (5)

Even though the severity factor is highly dependenton battery specifications such as chemistry and elec-trode design, it is suggested that the severity factorhas a typical shape as illustrated in Fig. 12 (Onoriet al., 2012). As seen from Fig.11, the operatingrange of the battery SOC under both strategies isnarrow. Thus, it is reasonable to assume that theseverity factor is only a function of temperature. Byusing the severity factor as a first approximation, theFDPD provides a 23% reduction of Ah-processed overthe UAC compared to the thermostatic SOC strategy.This significant decrease in the Ah-processed can beinterpreted as less electrical stress on the battery andlonger battery life.

10 20 30 40 50 60 70 80 90 1001

2

3

4

5

6

7

8

9

DOD [%]

σ T batt,D

OD

Tbatt=25°C

Tbatt=30°C

Tbatt=35°C

Tbatt=40°CTbatt=45°C

Tbatt=50°C

Tbatt=55°C

Tbatt=60°C

increasing Tbatt

Figure 12: Severity factor as a function of DOD parameterizedwith respect to battery temperature (Onori et al., 2012)

In this work, the power management strategies areoptimized and evaluated for one particular drive cy-cle. Other drive cycles may need different controlgains and lead to different savings in terms of fueleconomy, soot emissions, and battery life. Therefore,

10

the presented control gains should not be interpretedas a single set of gains recommended for all drive cy-cles. The typical approach in the literature to ensurea robust performance is to tune the controller for dif-ferent types of drive cycles separately and using apattern recognition algorithm to switch between thebest gains (Murphey et al., 2012, 2013). However, de-veloping such a gain scheduling approach and evalu-ating the robustness of its performance is beyond thescope of this article and is left as future work.

5. Conclusions

The original contributions of this article can besummarized as follows. A control parameter tuningstrategy has been proposed for the frequency-domainpower distribution (FDPD) strategy. Control pa-rameters are systematically determined through themodel-based two-stage optimization process, wherenon-gradient and gradient based algorithms are se-quentially combined to take advantage of both algo-rithms.

A case study has been conducted to experimen-tally compare the performance of the FDPD to thethermostatic SOC strategy as the baseline. A net-worked hardware-in-the-loop simulation platform hasbeen developed for this purpose and a Mine ResistantAmbush Protected All-Terrain Vehicle (M-ATV) hasbeen considered as the vehicle system.

The results show that the FDPD strategy success-fully reduces aggressive engine power demand andexcessive electric battery loads while improving fueleconomy by 5.9% compared to the baseline strategyin the specific scenario considered. The smooth en-gine power demand results in 62.7% reduction of sootemissions from the engine, as well as a reduction ofhigh current operation of the battery during propul-sion. A decrease in high current operation leads tothe lower temperature of the battery. Specifically, thebattery temperature is 3◦C lower under the FDPDstrategy than the baseline strategy. In addition,battery life is estimated by using the weighted Ah-processed model. The results show that the FDPDstrategy can reduce the Ah-processed by 23% andthereby extend the battery lifespan over typical mil-itary driving conditions.

Future work will compare the FDPD strategy tooptimal control strategies such as Dynamic Program-ming or Model-Predictive Control, which requires thedevelopment of an accurate dynamic soot emissionsmodel, and an efficient way of formulating and solvingthe resulting optimal control problem with increaseddimensionality.

6. Acknowledgements

This work was supported by the Automotive Re-search Center (ARC), a U.S. Army Center of Excel-lence in Modeling and Simulation of Ground Vehicles.

Appendix A. SHEV modeling

This appendix presents the SHEV system model.Figure A.1 shows the engine torque map obtained

from a Navistar 6.4L V8 diesel engine in (Johri et al.,2012). The engine torque map is augmented by a PIfuel controller sub-model generating the engine rackposition (ζ(t) ∈ [0, 1]), given by

ζ(t) = kP ∆τe + kI

∫∆τedt, (A.1)

where ∆τe is the error between the desired and ac-tual engine torque; kP and kI are proportional andintegral gains respectively. To represent the effect ofturbocharger lag on transient response during rapidincreases of engine rack positions, the fuel mass isfiltered by a first order filter. The engine-generatorunit is assumed to be fully warmed up so that theeffects of temperature are ignored. Figure A.2 illus-trates the efficiency of the generator obtained from(Argonne National Laboratory, 2002).

The most efficient operating points of the en-gine/generator combined system are different fromthe best engine-efficient operating points. In a se-ries hybrid configuration, the attached generator pos-sibly shifts the best fuel efficient operating pointsof the combined system to other operating points.The combined system brake specific fuel consump-tion (bsfc) map is obtained by dividing the enginebsfc map by the generator efficiency map. The bsfc

11

00

00

100100100

200200200

300300300

400400

400

500500

500

500

600

600600

600

700

700700

700

800

800

800

900

900 1000

Speed (RPM)

Fu

el

rate

(g

/cy

cle

)

1000 1500 2000 2500 30000

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0

100

200

300

400

500

600

700

800

900

1000

To

rqu

e (N

m)

Fu

elin

g r

ate

(g/c

ycl

e)

Speed (RPM)

Figure A.1: Engine torque map as a function of speed and fuelrate

0.70.70.7

0.8

0.80.8

0.85

0.850.850.85

0.87

0.87

0.870.87

0.87 0.9

0.9

0.9

0.9

0.9

0.91

0.910.91

0.910.91

0.91

0.92

0.92

0.92

0.92

0.92

0.920.92

0.92

0.93

0.93

0.93

Speed (rpm)

Torq

ue (N

m)

0 1000 2000 3000 40000

200

400

600

800

1000

1200 efficiency�

m,peak

0.7

0.75

0.8

0.85

0.9

Effic

ienc

y

Torq

ue(N

m)

Speed (RPM)

Maximum torque

Figure A.2: Generator efficiency map as a function of speedand torque

of the engine/generator unit bsfceng/gen can be cal-culated by using

bsfceng/gen = bsfceng/ηgen. (A.2)

The best fuel-efficient operating line is then deter-mined by searching the minimum fuel consumptionpoint for any given power demand. Figure A.3 showsthe combined bsfceng/gen and optimal operation lineof the engine/generator unit which is used for tuning

both the thermostatic and FDPD strategies.

Figure A.3: bsfc of engine/generator unit obtained by combin-ing engine bsfc and generator efficiency and superimposed byoptimal operation lines of the engine/generator unit and theengine only

A 9.27 kWh (281 Ah) lithium ion battery pack withLithium-Iron-Phosphate (LiFePO4 or LFP) cells byA123 is considered and the battery is modeled usingan OCV-R-RC-RC equivalent circuit approach. Thismodel has been parameterized and validated in (Linet al., 2014). The specifications for the LFP batteryare summarized in Table A.1.

Terminal voltage Vt of the battery is calculated byusing

Vt = Voc − V1 − V2 − IRs, (A.3)

where V1 and V2 are voltages across the capacitorsC1 and C2, respectively, and calculated based on thefollowing dynamic equations:

dVidt

=1

Ci

(I − Vi

Ri

), i = 1, 2. (A.4)

12

Table A.1: Specification of the battery

Parameter Value UnitNominal voltage 3.3 VMinimum voltage 2.0 VMaximum voltage 3.3 VNominal capacity 2.3 Ah

Number of cells in series 130 -Number of cells in parallel 10 -

0.8

0.82

0.84

0.86

0.88

0.88

0.9

0.9

0.92

0.92

0.92 0.92

0.940.94

0.94

0.94

0.94 0.94

0.960.96

0.96 0.96

Speed (rpm)

Torq

ue (N

m)

0 1000 2000 3000 4000

500

1000

1500

2000

2500

0.75

0.8

0.85

0.9

0.95Ef

ficie

ncy

Torq

ue(N

m)

Speed (RPM)

Peak torque

Continuous torque

Figure A.4: Motor efficiency map superimposed by peak andcontinuous torques

The sign convention is such that positive current de-notes battery discharging.

Figure A.4 shows the efficiency of the motor ηm isexpressed as a function of motor torque τm and motorspeed ωm. Maximum output torque of the motorτm,max is governed between the continuous torqueτm,cont and the peak torque τm,peak accounting forthe heat index γ as follows:

τm,max = τm,cont + (1− γ)τm,peak,

dγ

dt= − 0.3

180

(τm

τm,cont− 1

), γ(0) = 0.3, (A.5)

where τm,cont and τm,peak are a function of the motorspeed ωm as seen from Fig. A.4. The heat index γemulates the change in the torque limit based on op-

erating temperature as introduced in Powertrain Sys-tems Analysis Toolkit (PSAT) developed by ArgonneNational Laboratory (Argonne National Laboratory,2002).

A point-mass representation is used for the vehicle.The longitudinal dynamics of the vehicle is calculatedthrough the equation

Mvehdvveh

dt= Fprop − Fbrk − FRR − FWR, (A.6)

where Mveh is the mass of the vehicle, respectively,Fprop is the propulsion force, Fbrk is the braking force,and FRR is the rolling resistance force expressed by

FRR = frMvehag, (A.7)

where fr is the rolling resistance, ag is the gravita-tional acceleration. The wind resistance force FWR iscalculated by using

FWR =1

2ρairCdAvehv

2veh, (A.8)

where ρair is the air density, Cd is the drag coefficient,and Aveh is the frontal area of the vehicle. The roadgrade is not considered in the driving cycles in thisstudy.

The driver model, which takes the desired and ac-tual vehicle velocities as inputs and provides propul-sion or braking power demands, is adopted from (Er-sal et al., 2011) and is a PI controller with saturationand anti-windup.

References

Argonne National Laboratory, . Powertrain systemsanalysis toolkit. http://www.transportation.

anl.gov/modeling_simulation/PSAT/; 2002.[Online; accessed 18-April-2014].

Assanis, D., Delagrammatikas, G., Fellini, R., Fil-ipi, Z., Liedtke, J., Michelena, N., Papalam-bros, P., Reyes, D., Rosenbaum, D., Sales, A.,Sasena, M.. Optimization approach to hybrid elec-tric propulsion system design. Mechanics of Struc-tures and Machines 1999;27(4):393–421.

13

Caratozzolo, P., Serra, M., Riera, J.. Energy man-agement strategies for hybrid electric vehicles. In:Proceedings of the IEEE International Electric Ma-chines and Drives Conference. volume 1; 2003. p.241–248.

Di Cairano, S., Liang, W., Kolmanovsky, I.V.,Kuang, M.L., Phillips, A.M.. Power smoothingenergy management and its application to a serieshybrid powertrain. Control Systems Technology,IEEE Transactions on 2012;21(6):2091–2103.

Ersal, T., Brudnak, M., Salvi, A., Stein, J.L.,Filipi, Z., Fathy, H.K.. Development andmodel-based transparency analysis of an internet-distributed hardware-in-the-loop simulation plat-form. Mechatronics 2011;21(1):22–29.

Ersal, T., Brudnak, M., Stein, J.L., Fathy,H.K.. Statistical transparency analysis ininternet-distributed hardware-in-the-loop simula-tion. Mechatronics, IEEE/ASME Transactions on2012;17(2):228–238.

Ersal, T., Gillespie, R.B., Brudnak, M., Stein, J.L.,Fathy, H.. Effect of coupling point selection ondistortion in internet-distributed hardware-in-the-loop simulation. International Journal of VehicleDesign 2013;61:67–85.

Filipi, Z., Fathy, H., Hagena, J., Knafl, A.,Ahlawat, R., Liu, J., Jung, D., Assanis, D.,Peng, H., Stein, J.. Engine-in-the-loop testingfor evaluating hybrid propulsion concepts and tran-sient emissions - HMMWV case study. SAE Tech-nical Paper 2006;(2006-01-0443).

Filipi, Z., Kim, Y.J.. Hydraulic hybrid propulsionfor heavy vehicles: Combining the simulation andengine-in-the-loop techniques to maximize the fueleconomy and emission benefits. Oil & Gas Scienceand Technology - Revue d’IFP Energies nouvelles2010;65(1):155–178.

Gao, W., Porandla, S.. Design optimization of a par-allel hybrid electric powertrain. In: Proceedings ofthe IEEE Vehicle Power and Propulsion Confer-ence. 2005. p. 530–535.

Hagena, J.R., Assanis, D.N., Filipi, Z.S.. Cycle-resolved measurements of in-cylinder constituentsduring diesel engine transients and insight intotheir impact on emissions. Proceedings of the In-stitution of Mechanical Engineers, Part D: Journalof Automobile Engineering 2011;225(9):1103–1117.

Holmstrom, K.. gclsolve.m: A Standalone Versionof Direct, Software Documentation, Revision 2.00,HKH MatrisAnalys AB, Sweden. 1989.

Jalil, N., Kheir, N., Salman, M.. A rule-based en-ergy management strategy for a series hybrid ve-hicle. In: Proceedings of the American ControlConference. volume 1; 1997. p. 689–693.

Johri, R., Salvi, A., Filipi, Z.. Real-time transientsoot and NOx virtual sensors for diesel engine us-ing neuro-fuzzy model tree and orthogonal leastsquares. Journal of Engineering for Gas Turbinesand Power 2012;134(9):092806.

Jones, D., Perttunen, C., Stuckman, B.. Lip-schitzian optimization without the lipschitz con-stant. Journal of Optimization Theory and Appli-cations 1993;79(1):157–181.

Kim, Y., Ersal, T., Salvi, A., Stefanopoulou, A.G.,Filipi, Z.. Engine-in-the-loop validation of a fre-quency domain power distribution strategy for se-ries hybrid powertrains. In: Proceedings of theIFAC Workshop on Engine and Powertrain Con-trol Simulation and Modeling. 2012a. p. 432–439.

Kim, Y., Filipi, Z.. Simulation study of a series hy-draulic hybrid propulsion system for a light truck.SAE Technical Paper 2007;(2007-01-4151).

Kim, Y., Lee, T.K., Filipi, Z.. Frequency domainpower distribution strategy for series hybrid elec-tric vehicles. SAE International Journal of Alter-native Powertrains 2012b;1(1):208–218.

Kim, Y., Mohan, S., Siegel, J.B., Ste-fanopoulou, A.G., Ding, Y.. The estima-tion of temperature distribution in cylindricalbattery cells under unknown cooling conditions.IEEE Transactions on Control Systems Technol-ogy, 101109/TCST20142309492 2014;.

14

Konev, A., Lezhnev, L., Kolmanovsky, I.. Controlstrategy optimization for a series hybrid vehicle.SAE Technical Paper 2006;(2006-01-0663).

Lave, L.B., MacLean, H.L.. An environmental-economic evaluation of hybrid electric vehicles:Toyota’s prius vs. its conventional internal combus-tion engine corolla. Transportation Research PartD: Transport and Environment 2002;7(2):155–162.

Lee, T.K., Kim, Y., Stefanopoulou, A., Filipi,Z.. Hybrid electric vehicle supervisory control de-sign reflecting estimated lithium-ion battery elec-trochemical dynamics. In: Proceedings of theAmerican Control Conference. 2011. p. 388–395.

Li, D., Feng, D.. Thermostatic control for series hy-draulic hybrid vehicle (shhv) energy management.Advanced Materials Research 2012;512-515:2676–2681.

Li, Y., Kar, N.. Advanced design approach of powersplit device of plug-in hybrid electric vehicles us-ing dynamic programming. In: Proceedings of theIEEE Vehicle Power and Propulsion Conference.2011. p. 1–6.

Lin, X., Perez, H.E., Mohan, S., Siegel, J.B.,Stefanopoulou, A., Ding, Y., Castanier, M.P..A lumped-parameter electro-thermal model forcylindrical batteries. Journal of Power Sources2014;257:1–11.

Liu, J., Hagena, J., Peng, H., Filipi, Z.S.. Engine-in-the-loop study of the stochastic dynamic pro-gramming optimal control design for a hybrid elec-tric HMMWV. International Journal of Heavy Ve-hicle Systems 2008;15(2-4):309–326.

Liu, J., Peng, H.. Modeling and control of a power-split hybrid vehicle. Control Systems Technology,IEEE Transactions on 2008;16(6):1242–51.

Michel, P., Charlet, A., Colin, G., Chamaillard,Y., Bloch, G., Nouillant, C., et al. Pollution con-strained optimal energy management of a gasoline-hev. SAE Technical Paper 2013;(2013-24-0083).

Murphey, Y., Park, J., Chen, Z., Kuang, M., Mas-rur, M., Phillips, A.. Intelligent hybrid vehiclepower control - Part I: Machine learning of opti-mal vehicle power. Vehicular Technology, IEEETransactions on 2012;61(8):3519–3530.

Murphey, Y., Park, J., Kiliaris, L., Kuang, M.,Masrur, M., Phillips, A., Wang, Q.. Intelligenthybrid vehicle power control - Part II: Online intel-ligent energy management. Vehicular Technology,IEEE Transactions on 2013;62(1):69–79.

Onori, S., Spagnol, P., Marano, V., Guezennec,Y., Rizzoni, G.. A new life estimation method forlithium-ion batteries in plug-in hybrid electric ve-hicles applications. International Journal of PowerElectronics 2012;4(3):302–319.

Pisu, P., Rizzoni, G.. A supervisory control strategyfor series hybrid electric vehicles with two energystorage systems. In: Proceedings of the IEEE Ve-hicle Power and Propulsion Conference. 2005. p.65–72.

Ramasamy, A., Hill, A.M., Hepper, A.E., Bull,A.M.J., Clasper, J.C.. Blast mines: Physics, in-jury mechanisms and vehicle protection. JR ArmyMed Corps 2009;155(4):258–264.

Reavell, K., Hands, T., Collings, N.. A fast responseparticulate spectrometer for combustion aerosols.SAE Technical Paper 2002;(2002-01-2714).

Serrao, L., Onori, S., Rizzoni, G., Guezennec, Y..A novel model-based algorithm for battery prog-nosis. In: Proceeding of the IFAC Symposium onFault Detection, Supervision and Safety of Techni-cal Processes. 2009. .

Serrao, L., Onori, S., Sciarretta, A., Guezennec,Y., Rizzoni, G.. Optimal energy management ofhybrid electric vehicles including battery aging. In:American Control Conference. 2011. p. 2125–2130.

Serrao, L., Sciarretta, A., Grondin, O., Chasse,A., Creff, Y., Di Domenico, D., Pognant-Gros,P., Querel, C., Thibault, L.. Open issues insupervisory control of hybrid electric vehicles: Aunified approach using optimal control methods.

15

Oil Gas Sci Technol – Rev IFP Energies nouvelles2013;68(1):23–33.

Yang, S., Qi, C., Guo, D., Wang, Y., Wei, Z..Topology optimization of a parallel hybrid electricvehicle body in white. Applied Mechanics and Ma-terials 2012;148-149:668–671.

16