Plug‐in Hybrid Powertrain Configuration C i U i Gl b l O i i iComparison Using Global Optimization

2008 ASME Dynamic Systems and Control Conference,

Ann Arbor, 

October 21st 2008October 21st, 2008

Dominik A. Karbowski, Karl‐Felix Freiherr von Pechmann, Sylvain Pagerit Jason KwonPagerit, Jason Kwon

Argonne National Laboratory

Sponsored by Lee Slezak

This presentation does not contain any proprietary, confidential, or otherwise restricted information

Outline

Introduction

Vehicle Sizing

Global Optimization Algorithm

Simulation Results

Conclusion

2

PHEVs Come in Various PowertrainConfigurations: How to Compare Them?Configurations: How to Compare Them? Plug‐in Hybrid Electric Vehicles (PHEVs) combine:

– long range, high power density and easy energy refill

– low tailpipe emissions, high fuel displacement (on short daily trips) 

Various hybrid powertrain configurations are possible: how do they compare in a PHEV context ? 

Prototypes exist but built upon different requirements sizes electric ranges Prototypes exist, but built upon different requirements, sizes, electric ranges, etc.

Chevrolet (GM) Volt(Series)(Series)

Toyota Prius PHEV(Power‐Split)

Mercedes Sprinter PHEV(Parallel Pre‐Transmission)

3

Simulation tools must be used for a fair comparison !

Choosing the Most Suitable Simulation Tool

Argonne’s PSAT– Forward‐looking model

l d

Real‐world Oriented

R lt d d t l– Includes Transients

– Rule‐based control

Results depends on control design and tuning

Global OptimizationTheoretical tool

tSOCtP nm ,

0X

...

– Backward‐looking model

– Steady‐state model

– Control is an output

Theoretical tool

Each Vehicle is operated optimally tJtSOCtP '0,000 ,1',1 tJtSOCtP MMMM ',' ,1,1

tU0 tU M

Control is an output

TX

...

Global optimization allows a “fair” comparison

4

Global optimization allows a  fair  comparison

Study ProcessDefine Vehicle Requirements Industry standards

Size VehiclesImplement + Automated 

Sizing

Optimally Simulate Vehicles

optimization code for various

configurationsDifferent cycles

Sizing

Vehicles Different cycles,Different distance

Analyze data

5

Outline

Introduction

Vehicle Sizing

Global Optimization Algorithm

Simulation Results

Conclusion

6

Requirements and Assumptions

Type Requirement Comments

All‐electric Range 10 mi (16 km) on the UDDS 90% to 30% SOC

Grade  6%@65 mph (105 km/h) @GVWR, w/o using the battery

0‐60 mph (97 km/h) time 9.3s @ 30% SOC

M i d 100 h (160 k /h)Maximum speed >100 mph (160 km/h)  

Parameter Value

Glid 1142 kGlider mass 1142 kg

Drag coefficient 0.28

Frontal area 2.3 m2

Comparable to Toyota Camry,Rolling resistance 0.009

Wheel Radius 0.332 m

Comparable to Toyota Camry, Chevrolet Malibu

7

Automated Sizing Routine

Motor and battery power sized for the UDDSMotor Power for Cycle

Vehicle Assumptions

UDDS

Engine power sized for the grade

Engine (and motor in series config.) size increased if 0 60 mph requirements not met

Battery Power

y

increased if 0‐60 mph requirements not met

Battery energy sized to meet the AER requirements

M i f ti f t

Engine Power

Battery Energy

Mass is function of components characteristics

Based on PSAT simulations

Convergence

Yes

No

8

Sizing Results

Configuration Mass (kg)

Parallel 1782

Power‐Split 1824

Series 1793Series 1793

9

Outline

Introduction

Vehicle Sizing

Global Optimization Algorithm

Simulation Results

Conclusion

10

Global Optimization Uses a Backward Model, Based on Look-up TablesBased on Look-up Tables

ICE ESSICE GEN

ICE ACCESS

Thermal Elec. Mech. 

ACC

MG

ACC ESSMG2MG1

TC

GEARBOX

MG

DRIVELINE

TCTC

PLANETARY

DRIVELINEDRIVELINE

Parallel Pre Transmission

Power‐Split SeriesPre‐Transmission

Engine (ICE), electric machines (MG) : input power (thermal or electric) is function of output torque and speed

11

Battery voltage and resistance is function of SOC

Optimization Problem State: Battery State‐of‐charge (SOC)

Command u: – gear and engine torque (Parallel)g g q ( )

– generator power (Series)

– engine power (Power‐split)

Link between state and command:

Initial and final conditions: initial SOC between 90 and 30% , final SOC=30%, 

Cost function to minimize:

Constraints:

– Vehicle follows drive trace

– Keep ICE and EM speed/torque within limits

– Keep battery current within limits 

S i & lit t d d h th t f i ICE t– Series & power‐split: torque and speed such that for a given power ICE operates optimally

Method: application of Bellman’s principle

12

Algorithm Outputs Results for Different SOCSOC

SOC.9

1 vehicle1 cycle1 distance

calculation.8

.6

1 run .4

.3

0.60 0.1 0.3 0.5

SOC

time

Plug‐to‐Wheel Wh/km

Outline

Introduction

Vehicle Sizing

Global Optimization Algorithm

Simulation Results

Conclusion

14

In Series, Engine Starts “Later” and Is Used More Often To Charge The BatteryOften To Charge The Battery

Power at the Wheels Above Which ICE Is ON 95% of the Time

Power‐split and parallel follows similar

4x 104

paralleli Power split and parallel follows similar 

trends

Series “Engine ON” threshold is higher => ICE is ON less often and at higher 2

3

wer

(W)

seriessplit

loads1

Pow

0.15

Cha

rgin

g

parallelUDDS x1

UDDS x1

0 50 100 1500

Electric energy consumption (Wh/km)

0.1

ergy

use

d fo

r C seriessplit

UDDS x1

Ratio of ICE Energy Used to Charge

0

0.05

atio

of I

CE

Ene Ratio of ICE Energy Used to Charge 

the Battery

Much more Charging for the SeriesLoad‐following

15

0 50 100 1500

Electric energy consumption (Wh/km)

Ra

In Series Config., ICE is Used at Higher Power for Higher Efficiency Higher Efficiency

Average Engine Power (when ON)

Different trends:1.5

2x 104

Different trends:– Series = independent of Wh/km

– Parallel = higher for higher Wh/km

Series: ICE power is much higher 

1

Pow

er (W

)

parallelUDDS x1 p g

0 50 100 1500

0.5

Electric energy consumption (Wh/km)

seriessplit

0 37

0.38 parallel

UDDS x1

Electric energy consumption (Wh/km)

0.35

0.36

0.37

ffici

ency

seriessplit Overall Engine Efficiency

Similar trends to average power

Series = engine operates close to maximal

0.32

0.33

0.34Eff Series  engine operates close to maximal 

efficiency (36%)

Parallel = much lower efficiency in CS mode

Power Split = in betweenUDDS x1

16

0 50 100 1500.32

Electric energy consumption (Wh/km)

p

On The UDDS Power Split Seems to Be the Best CompromiseCompromise

Charge‐Sustaining: Series fi f l d

5

m)

parallel

config. uses most fuel, due to poor electric transmission efficiency

Parallel uses the least fuel3

4

n (L

/100

km

parallelseriessplit

Parallel uses the least fuel, as engine‐to‐wheel path is more efficient

In electric‐only, parallel is 2

3

nsum

ptio

n

penalized by the transmission efficiency

Power‐Split performs well i b th it ti0

1

Fuel

Con

is both situations0 50 100 150

0

Electric energy consumption (Wh/km)

17

Similar Hierarchy When Using Other CyclesI h t i i d th diff b t th i In charge‐sustaining mode, the difference between the series configuration and the two other ones is more important on more energy‐intensive cycles (HWY) or aggressive (LA92)

Th diff b t ll l d th th i l i ifi t

6

The difference between rarallel and the other ones is less significant on LA92 or HWY

LA92Parallel

4

5

(L/1

00km

) HWY

UDDS32 km

ParallelSeriesPower‐Split

120

km/h

2

3

Con

sum

ptio

n LA92

HWY

UDDS16 km

∞ km

(charge‐sustaining)

0

0

0

0

0

0120

60

UDDS LA92

HWY

0

1

Fuel

LA92UDDS

HWYUDDS

12 km24 km

0 200 400 600 800 1000 1200 14000 500 1000 15000 200 400 600 80000HWY

s s s

0 50 100 150 200Electrical Energy Consumption (Wh/km)

18

Outline

Introduction

Vehicle Sizing

Global Optimization Algorithm

Simulation Results

Conclusion

19

Conclusion

Comparing different powertrains requires a common set of requirements

An automated routine using PSAT allows to quickly size vehicles

Global optimization can be used to ensure “fair” comparison, because control is optimal

Series is potentiallymore efficient on the “electric‐only” side, parallel (pre‐tx) is potentiallymore efficient on the “charge‐Sustaining” side, power‐split is a good compromise

Global optimization is a theoretical tool, and in real‐world optimal control is harder to get because:– it requires the knowledge of the cycle beforehand

– real‐world must also take into account drivability, safety, etc.

However it provides the theoretical boundaries, as well as control patterns

Future work will use patterns obtained from global optimization to implement adaptive rule‐based controls.

20