2007 sae commercial vehicle engineering congress – …...• model-based approaches and...

©20

07 T

he M

athW

orks

, Inc

.

® ®

2007 SAE Commercial Vehicle Engineering Congress – MathWorks Hosted Panel on Analytical Calibration

Analytical Calibration PanelAnalytical Calibration PanelIntroductory remarksIntroductory remarks

SAE Commercial Vehicle Conference, Rosemont, IL, October 31st 2007

Giorgio RizzoniDirector

The Ohio State University Center for Automotive Research

CalibrateCalibrate

“To check, adjust, or determine bycomparison with a standard.”

Σ+

−

CalibrateCalibrate


Σ+

−

So So ……Does Control = Calibration?Does Control = Calibration?Do we need Calibrators?Do we need Calibrators?

CalibrateCalibrate


Σ+

−

We get this box …So So ……Does Control = Calibration?Does Control = Calibration?Do we need Calibrators?Do we need Calibrators?

No!No!

CalibrateCalibrate


Σ+

−

… from our knowledgeof this box.


No!No!

CalibrateCalibrate


Σ+

−


No!No!

Yes!Yes!

CalibrateCalibrate


Σ+

−


No!No!

Yes!Yes!Typically, Calibrators arethe plant experts!

Modeling Modeling →→ Control Control →→ CalibrationCalibration

How do we go from How do we go from thisthis……

Modeling Modeling →→ Control Control →→ CalibrationCalibration

To this?To this?

… and do it with a short development cycle, keeping in mind dimensionality, modularity, adaptability, scalability, all the while admitting a rigorous calibration process?

An Obvious Answer: Good ModelingAn Obvious Answer: Good Modeling

All models are lies (some are better than others)(Box)

The “real” answer goes beyond simply “modeling”

““ModelModel--Based ControlBased Control””–– a process, not a techniquea process, not a technique

Define the problem

Understand the plant

Pick a control theory

Control-Oriented model

Calculate a control law

Make it work


Define the problem





Make it work

Manager(with help)

(the “real” problem)


Define the problem





Make it work

“Modeler”

- Physics-Matlab simulation- some data


Define the problem





Make it work

Control TheoristHow many academicians start here ….


Define the problem





Make it workCalibrator andcontrol

practitioner


Define the problem





Make it work

How many academicians start here ….


Define the problem





Make it work


… and finish here?


Define the problem





Make it work


How manypractitioners start

here ….



Define the problem





Make it work



here ….

… and finishhere?



Define the problem





Make it work



here ….

… and finishhere?


Usually, one person cannotplay all these roles …

A Preview of the PanelA Preview of the Panel

• Emissions legislation has significantly increased the complexity of the calibration problem, and extended this problem to heavy-duty and off-highway applications

• Complex combustion and exhaust aftertreatmentbehavior coupled with insufficient sensor information make the task of achieving open loop calibration for new emissions standard very challenging.

• Model-based approaches and computer-aided calibration tools can assist in this process, however the current state of models

• For example, the problem of transient system response is still a very challenging one: models of engine transient behavior, especially vis-à-vis emissions, are still inadequate.

Courtesy: Bosch, FKFS

Growth in complexity Growth in complexity --19901990--20052005

• Processor: 8-bit → 32-bit• Performance (MIPS): <1 → 300• Transistors: < 1M → 25M• Memory: 33 kB → 4,000 kB• Application parameters: 500 → 8,000

– Expected number in 2007: 20,000• Connector pins: 50 → 150• ECU manual ~5,000 pages!!


ECU mapping processECU mapping process


Manual mapping processManual mapping process


Automated mapping processAutomated mapping process

AcknowledgmentsAcknowledgments

• Steve Yurkovich, – Professor of Electrical and Computer Engineering, Center for

Automotive Research, CAR, The Ohio State University.

• Yann Guezennec, – Professor of Mechanical Engineering, Center for Automotive

Research, CAR, The Ohio State University.

• Michael Bargende, – Professor of Mechanical Engineering, Center for Automotive

Research Institute of Automotive Engineering and Vehicle Engines, FKFS, Universität Stuttgart

Analytical Calibration for Diesel EnginesLisa Farrell, PhDCummins Inc

2Cummins Inc.

Evolution of Complexity for HD Diesel Engine Calibration

0

5

10

15

0 0.1 0.25 0.6 1.0

1974 EPA (HC + NOx)

1988

199019911994

1998

Particulate [g/(HP-hr)]

NO

x[g

/(HP -

hr)]

20042010?

Up until 1990's, performancewas fixed by design – no realdegrees of freedom

In 1991, variable injection timing was a degree of freedom

In the late 1990's, electronic fuel systems added many degrees of freedomIn 2004,

VG, EGR added

3Cummins Inc.

Evolution of Complexity for HD Diesel Engine CalibrationIn 2007, robust particulate filter after-treatment was introduced for HD US on-highway enginesWhat is coming for 2010?

May introduce new degrees of freedomMay introduce new after-treatment challenges

Increasing complexity of Diesel engine systems requires the application of analytical calibration methods

4Cummins Inc.

Analytical Calibration Benefits – The Cummins Inc PerspectiveDesign of experiments has led to reduced data collection timesKey benefit is constant development cycles with increasing system complexityA means to optimize engine systems with many degrees of freedom outside the test cell environmentRigorous problem definition

Constraints – mechanical/emissionsObjective function – emissions/fuel consumption

5Cummins Inc.

400

1000

1400

1800

2230

80

770

020406080

100120

Turbo Speed (krpm)

Speed (rpm)Torque (ft-lb)

Global ModelTurbo Speed = f(Speed, Load,Timing, VG, EGR, Rail Pressure,Pilot, Post, etc.)

Engine Family Torque Curves

0

200

400

600

800

1000

1200

1400

1600

1800

2000

600 1000 1400 1800 2200

Speed (RPM)

Torq

ue (f

t-lb)

CalibrationPlanning

DOEData Collection

ModelDevelopment

Optimizationand Validation

Itera

tion

Engine Speed

Torq

ue

Variation inSpeed, Load, TimingVG, EGR, Rail PressurePilot, Post, etc.

Analytical Calibration DevelopmentSteady-state Performance

6Cummins Inc.

Benefits of Analytical CalibrationFuel Consumption is optimizedover a cycle while constraining cycle bsNOx and surface smoke targets.

Optimal set of control surfaces are determined.

Excellent benefit and capabilityfor steady-state performance.

7Cummins Inc.

Transient Calibration Development

The traditional approach for transient tuning has been centered on transient testingCurrent analytical calibration methods for transient tuning are limited, if available at allApproach for Cummins Inc is a quasi-steady cycle optimization

Captures transient NOx trends reasonablyDoes not capture transient PM trends

8Cummins Inc.

Transient Results

Comparison of measureddilution tunnel NOxemissionsvs. quasi-steady predictedNOx emissions

9Cummins Inc.

What is Needed for Dynamic Calibration of Diesel Engines?Can dynamic models for key performance parameters be incorporated into analytical techniques?What statistical or physical models are appropriate for transient response?What is the optimization method for transient response tuning parameters?

10Cummins Inc.

What is the Future of Diesel Engines?

OxidationCatalyst

DieselParticulate

Filter

TT

SCR Catalyst

TT

NH3 Slip Cat

Temperature Sensors

Engine + After-treatment (TBD)

+

11Cummins Inc.

How Can Dynamic After-Treatment Performance Be Included?

0

50

100

150

200

250

300

350

400

450

0 200 400 600 800 1000 1200

Time [second]

Exha

ust

Tem

pera

ture

[C]

100 150 200 250 300 350 400 450 5000

20

40

60

80

100

Temperature [C]

New

Cat

alys

t Ef

ficie

ncy

[%]

ActualDesired

EngineControls

ExhaustConditioning

400°C = 752°F

After-treatment performance is linked to engine exhaust conditions

12Cummins Inc.

Summary

Analytical calibration methods have been very successful when applied to steady-state performanceDynamic tuning capability needs further development of analytical methodsInclusion of after-treatment modeling techniques would enhance analytical calibration methods for diesel engines in the future

®

1

Analytical Calibrationat

International Truck and Engine Corporation

2007 SAE Commercial Vehicle CongressEduardo Nigro

®

2

DoE Automation

CALGEN

Map Plots

Physical Model Fitting

Job Split

Progress Report

Database

Calibration Compare

Combustion Development

Aftertreatment Calibration

Steady State Calibration Transient Calibration

Vehicle Calibration

Desktop Calibration

Diagnostic Calibration

AUTOMATION &

OPTIMIZATION TOOLS

CALIBRATION PROCESS

CALIBRATION MANAGEMENT

Calibration Process Overview

®

3

DoE Automation

CALGEN

Map Plots

Physical Model Fitting

Job Split

Progress Report

Database

Calibration Compare

Combustion Development

Aftertreatment Calibration

Steady State Calibration Transient Calibration

Vehicle Calibration

Desktop Calibration

Diagnostic Calibration

AUTOMATION &

OPTIMIZATION TOOLS

CALIBRATION PROCESS

CALIBRATION MANAGEMENT

Calibration Process Overview

®

4

Steady State Optimization Process

• Procedure built on CAMEO (AVL) and CALGEN (in-house) provides state-of-the-art tool for steady-state calibration optimization

CREATE DOEs

BUILD LOCAL MODELS

DEFINE KEY POINTS / KEY

REGIONS

RUN DoEs IN TEST

CELL

BUILD GLOBAL MODELS

LOCAL OPTIMIZATION

GENERATE CALIBRATION

MAPS

EXPORT TABLES TO

ECU

GLOBAL OPTIMIZATION

KEY POINT

KEY REGION

SOI FUP

EGR

P

KEY POINT FACTORS

OPTIMIZATION POINTS

SURFACE FIT MODEL BSFC

BSNOx

BSS

OO

T

OPTIMAL MAP OPTIMAL FACTORS

AUTOMATIC DoE ADAPTATION

CAMEO

CALGEN

®

5

+ +

Analytical Tools

Analytical tools must be end-user-oriented

CALGEN is used by engine calibrators

+LOW COST FASTPROCESS

ORIENTEDEASY TO

USE

®

6

Vision for Tool Development

Steady-State Modeling

Static Modeling of Transient Parameters

Transient Modeling with Dynamic Systems Identification

Transient Modeling with Dynamic Systems Models (Model-Based)

TOOL DEVELOPMENT TIMELINE

CALGEN

Under Development

®

7

Analytical Models

Analytical models strengthen the calibration process

Benefits in process flow, data quality and development time

Systematic Approach

Better Knowledge Transfer

Easy-to-do Offline Calibration Modifications

Clear Hardware Limit Determination

Optimal Calibration Generation

ANALYTICAL MODELS

12007 SAE Commercial Vehicle Panel on Analytical CalibrationY. He10/31/07

MathMath--based Control Development based Control Development and Analytical Calibration and Analytical Calibration

Yongsheng HeGeneral Motors Research and Development

October 31, 2007


OutlineOutlineIntroductionIntroduction

Math-based control developmentAnalytical calibration

MathMath--based control developmentbased control developmentDetailed 1D engine modelMean value engine model

Results and DiscussionResults and DiscussionStep transientsFTP cycle

SummarySummaryCurrent capabilitiesFuture outlook


IntroductionIntroductionMathMath--based control development in automotive industrybased control development in automotive industry

Much of control design and development process could be done off-line using computer simulationsDramatically reduce development time and risk

Integrated engine and control system model valuable Integrated engine and control system model valuable Accurately evaluate control algorithmsExplore different control strategies & study parameter sensitivity

Before experiments conductedBefore hardware selected and built

Analytic calibration critical to develop modern embedded Analytic calibration critical to develop modern embedded powertrainpowertrain controllers (complexity, speedcontrollers (complexity, speed--toto--market, etc.)market, etc.)

Physical dyno and/or vehicle testing to be minimizedComputer simulations also to be reduced


Model Accuracy Model Accuracy vsvs Model SpeedModel SpeedFastFast--running engine model with sufficient accuracyrunning engine model with sufficient accuracy

Efficient evaluation of control algorithms and control strategiesExploration of the classical trade-off in the modeling process

Detailed 1D engine modelDetailed 1D engine modelPredict gas dynamics and engine performance within 3-5%Run speed on the order of 100~1000 times slower than real time

Mean value engine modelMean value engine modelCapture dynamics over one or more engine cyclesRun speed close to or faster than real time

Model AccuracyModel Accuracy Model SpeedModel Speed


Integrated Engine & Control System SimulationIntegrated Engine & Control System Simulation

fuel temp (C)

coolant temp (C)

mph

Intake_oxy_frac_out

Exh_tmp_K_out

Pexh_bar_out

MAP_eng_out

MAF_eng_out

fqc_q_desired_out

pos_itv_target_out

power_eng_out

BMEP_eng_out

trq_eng_out

turbo_speed_out

Exh_oxy_frac_out

pos_vnt_target_out

pos_egr_target_out

PPS (%)

Intake tmp (K)

rpm

EGR cooler tmp (K)

EGR lift (%open)

VNT pos (%close)

fuel (mm^3/st)

eng spd (rpm)

baro (bar)

IAT (K)

EGR cooler out temp (K)

MAF (kg/s)

MAP (bar)

Pexh (bar)

Exh tmp (K)

Intake O2 frac

Exh O2 frac

Turbo spd (rpm)

Trq (Nm)

BMEP (bar)

Power (kW)

Diesel Engine

MAF (kg/s)

MAP (bar)

PPS (%)

rpm_eng

mph_veh

Baro (bar)

fuel_temp (C)

tmp_air_intake (K)

coolant_temp (C)

egr_target_lift (%)

vnt_position (%)

itv_position (%)

fuel (mm^3/st)

Controller (ECM)

Baro (bar)

(SAE Paper 2006-01-0439)


FTP Cycle: FTP Cycle: Simulation Simulation ResultsResults

BlowBlow--up of the up of the FTP results to FTP results to compare compare simulations and simulations and experiments experiments (200(200--300 s)

200 210 220 230 240 250 260 270 280 290 3000

0.05

0.1

MA

F (k

g/s)

200 210 220 230 240 250 260 270 280 290 300

1

1.2

1.4

MA

P(b

ar)

200 210 220 230 240 250 260 270 280 290 300

1

1.5

2

2.5

Pex

h (bar

)

200 210 220 230 240 250 260 270 280 290 300400

600

800

1000

T exh (K

)

200 210 220 230 240 250 260 270 280 290 3000

0.1

0.2

0.3In

t O2

200 210 220 230 240 250 260 270 280 290 3000

0.1

0.2

0.3

Exh

O2

200 210 220 230 240 250 260 270 280 290 3000

2

4

6

8x 104

Turb

o S

peed

(rpm

)

Time (sec)

(a)

(b)

(c)

(d)

(e)

(f)

(g)

SimulationExperiment

Simulated Thermocouple Temp

300 s)ExperimentSimulation

(SAE Paper 2006-01-0439)


Detailed 1D Engine ModelDetailed 1D Engine Model

Inlet

Outlet

Turbine CylindersExhaust Manifold

Compressor IntercoolerIntakeThrottle

IntakeManifold

EGRValve

EGRCooler

(SAE Paper 2007-01-1304)


Input Variables and DOEInput Variables and DOETurbocharged V6 Turbocharged V6 diesel engine with diesel engine with external EGRexternal EGR

Focus on the Focus on the control of fueling, control of fueling, EGR, and VNT EGR, and VNT

DOE: Constrained DOE: Constrained Latin Hypercube Latin Hypercube

Consider the physical constraints of engine operations

500 1000 1500 2000 2500 30000

20

40

60

Fuel

(mg/

cycl

e)

500 1000 1500 2000 2500 30000

0.2

0.4

0.6

0.8

1

Engine Speed (rpm)

EG

R L

ift F

ract

ion

500 1000 1500 2000 2500 30001

1.1

1.2

1.3

1.4

Boo

st P

ress

ure

(bar

)

500 1000 1500 2000 2500 30001

1.5

2

2.5

Engine Speed (rpm)

Bac

k P

ress

ure

(bar

)

Engine Speed (rpm) [530 3000]Total Fueling (mg/cycle) [0 55]EGR Valve Lift Fraction [0 1]

Boost Pressure (bar) [1 1.4]Back Pressure (bar) [1 2.4]

(SAE Paper 2007-01-1304)


Mean Value Engine Modeling Mean Value Engine Modeling –– Final ModelFinal Model

Hybrid RBF (ModelHybrid RBF (Model--Based Based Calibration Toolbox, MATLAB) Calibration Toolbox, MATLAB) to approximate cylinder to approximate cylinder quantities for better accuracyquantities for better accuracy

Hybrid RBFHybrid RBFVolumetric EfficiencyVolumetric Efficiency 0.9990.999

Indicated EfficiencyIndicated Efficiency 0.9670.967

Exhaust Energy FractionExhaust Energy Fraction 0.9790.979

2R

(SAE Paper 2007-01-1304)

Integrated Engine & Controller Model Integrated Engine & Controller Model –– Updated Updated with Mean Value Modelwith Mean Value Model

fuel temp (C)

coolant temp (C)

mph

Intake_oxy_frac_out

Exh_tmp_K_out

Pexh_bar_out

MAP_eng_out

MAF_eng_out

fqc_q_desired_out

pos_itv_target_out

power_eng_out

BMEP_eng_out

trq_eng_out

turbo_speed_out

Exh_oxy_frac_out

pos_vnt_target_out

pos_egr_target_out

PPS (%)

Intake tmp (K)

rpm

EGR cooler tmp (K)

EGR lift (%open)

VNT pos (%close)

fuel (mm^3/st)

eng spd (rpm)

baro (bar)

IAT (K)

EGR cooler out temp (K)

MAF (kg/s)

MAP (bar)

Pexh (bar)

Exh tmp (K)

Intake O2 frac

Exh O2 frac

Turbo spd (rpm)

Trq (Nm)

BMEP (bar)

Power (kW)

Diesel Engine

MAF (kg/s)

MAP (bar)

PPS (%)

rpm_eng

mph_veh

Baro (bar)

fuel_temp (C)

tmp_air_intake (K)

coolant_temp (C)

egr_target_lift (%)

vnt_position (%)

itv_position (%)

fuel (mm^3/st)

Controller (ECM)

Baro (bar)


(SAE Paper 2007-01-1304)


Model Validation: Vehicle TestingModel Validation: Vehicle TestingSeries of different cruising and acceleration conditionsSeries of different cruising and acceleration conditions

Selected for validation: 3 step transients (ST)

0 100 200 300 400 500 600 700 800 900 1000 11000

20

40

60

80

Ped

al (%

)

0 100 200 300 400 500 600 700 800 900 1000 1100500

1000

1500

2000

2500

3000

Eng

ine

spee

d (rp

m)

0 100 200 300 400 500 600 700 800 900 1000 11000

20

40

60

80

Veh

icle

spe

ed (m

ph)

Time (sec)

(a)

(b)

(c)

PPS

RPM

MPH

ST3 ST1 ST2

(SAE Paper 2007-01-1304)

Step Transient: Simulation Results (1/3)Step Transient: Simulation Results (1/3)

0 5 10 15 201500

2000

2500

3000

Eng

ine

Spe

ed (r

pm) (a)

0 5 10 15 200

20

40

60

80

Ped

al (%

)

(b)

0 5 10 15 200

20

40

60

80

100

120

EG

R L

ift (%

)

Time (sec)

(c)

0 5 10 15 200

20

40

60

80

Fuel

(mm

3 /st)

(d)

0 5 10 15 2020

40

60

80

100

120

VN

T P

ositi

on (%

) (e)

0 5 10 15 200

0.05

0.1

0.15

MA

F (k

g/s)

Time (sec)

(f)

Mean Value (RBF)Detailed ModelST2

(SAE Paper 2007-01-1304)




0 5 10 15 200.8

1

1.2

1.4

Inta

ke P

ress

ure

(bar

)

(a)

0 5 10 15 20

1

1.5

2

Exh

aust

Pre

ssur

e (b

ar)

(b)

0 5 10 15 20

200

400

600

800

1000

Exh

aust

Tem

pera

ture

(K)

Time (sec)

(c)

0 5 10 15 200.05

0.1

0.15

0.2

0.25

0.3

Inta

ke O

xyge

n Fr

actio

n

(d)

0 5 10 15 20-0.1

0

0.1

0.2

0.3

Exh

aust

Oxy

gen

Frac

tion

(e)

0 5 10 15 200

2

4

6x 104

Turb

ine

Spe

ed (r

pm)

Time (sec)

(f)

Mean Value (RBF)Detailed Model

ST2

(SAE Paper 2007-01-1304)


0 5 10 15 20-50

0

50

100

150

Eng

ine

Pow

er (k

W)

(a)

0 5 10 15 20

-100

0

100

200

300

400

500

Eng

ine

Torq

ue (N

m)

(b)

0 5 10 15 200

20

40

60

80

100

EG

R R

ate

(%)

Time (sec)

(c)

0 5 10 15 200

20

40

60

80

100

120

Vol

umet

ric E

ffici

ency

(%)

(d)

0 5 10 15 200

20

40

60

80

100

Indi

cate

d E

ffici

ency

(%)

(e)

0 5 10 15 200

20

40

60

80

100

Exh

aust

Fra

ctio

n (%

)

Time (sec)

(f)


ST2

(SAE Paper 2007-01-1304)



Model Validation: FTP CycleModel Validation: FTP Cycle

0 200 400 600 800 1000 1200 14000

20

40

60

80

Ped

al (%

)

0 200 400 600 800 1000 1200 14000

1000

2000

3000

Eng

ine

spee

d (rp

m)

0 200 400 600 800 1000 1200 14000

20

40

60

Veh

icle

spe

ed (m

ph)

Time (sec)

(a)

(b)

(c)

PPS

RPM

MPH

(SAE Paper 2007-01-1304)


FTP Cycle: Simulation Results (1/2)FTP Cycle: Simulation Results (1/2)

0 200 400 600 800 1000 1200 14000

20

40

60

80

Fuel

(mm

3 /st) (a)

0 200 400 600 800 1000 1200 1400

60

80

100

VN

T P

ositi

on (%

)

(b)

0 200 400 600 800 1000 1200 14000

50

100

EG

R L

ift (%

)

Time (sec)

(c)


(SAE Paper 2007-01-1304)


0 200 400 600 800 1000 1200 14000

0.1

0.2

MA

F (k

g/s) (a)

0 200 400 600 800 1000 1200 14001

1.5

MA

P (b

ar) (b)

0 200 400 600 800 1000 1200 14001

1.5

2

Pex

h (b

ar)

(c)

0 200 400 600 800 1000 1200 14000

500

1000

Texh

(K)

(d)

0 200 400 600 800 1000 1200 14000.1

0.2

0.3In

t O2

(e)

0 200 400 600 800 1000 1200 1400

0

0.1

0.2

Exh

O2

(f)

0 200 400 600 800 1000 1200 14000

5

x 104

Turb

ine

Spe

ed (r

pm)

Time (sec)

(g)


FTP Cycle: FTP Cycle: Simulation Simulation Results (2/2)Results (2/2)

(SAE Paper 2007-01-1304)


FTP Cycle: Simulation Results BlowFTP Cycle: Simulation Results Blow--up (1/2)up (1/2)

200 210 220 230 240 250 260 270 280 290 3000

20

40

60

80

Fuel

(mm

3 /st)

200 210 220 230 240 250 260 270 280 290 300

60

80

100

VN

T P

ositi

on (%

)

200 210 220 230 240 250 260 270 280 290 3000

50

100

EG

R L

ift (%

)

Time (sec)


DetailedMean Value

BlowBlow--up of the FTP results for comparison (200up of the FTP results for comparison (200--300 s)300 s)

(SAE Paper 2007-01-1304)


200 210 220 230 240 250 260 270 280 290 3000

0.1

0.2

MAF

(kg/

s)

200 210 220 230 240 250 260 270 280 290 3001

1.5

MAP (b

ar)

200 210 220 230 240 250 260 270 280 290 3001

1.5

2

Pex

h (b

ar)

200 210 220 230 240 250 260 270 280 290 3000

500

1000

Texh

(K)

200 210 220 230 240 250 260 270 280 290 3000.1

0.2

0.3In

t O2

200 210 220 230 240 250 260 270 280 290 300

0

0.1

0.2

Exh

O2

200 210 220 230 240 250 260 270 280 290 3000

5

x 104

Turb

ine

Spe

ed (r

pm)

Time (sec)

Mean Value (RBF)Detailed ModelFTP Cycle: FTP Cycle:

Simulation Simulation Results Results BlowBlow--up (2/2)up (2/2)

BlowBlow--up of the up of the FTP results for FTP results for comparison comparison (200(200--300 s)300 s)

DetailedMean Value

(SAE Paper 2007-01-1304)


Model Accuracy Model Accuracy vsvs Model Speed (Summary)Model Speed (Summary)Mean value engine model developed in this studyMean value engine model developed in this study

Accuracy slightly compromised (cylinder quantities)About 40 times faster than the detailed model

0

20

40

60

80

100

Model Run Time (x Real Time)

Detailed 1D Model [15-16] Mean Value Model (RBF)

Model Error (%)(SAE Paper 2007-01-1304)


SummarySummaryA fastA fast--running mean value engine model with sufficient running mean value engine model with sufficient accuracy developed for control applications accuracy developed for control applications

Reduced from a detailed engine model in GT-PowerConstrained Latin Hypercube to consider physical constraints Hybrid RBF to approximate cylinder quantities for better accuracyCompletely simplified (cylinders, intake & exhaust system)

Model development time & model throughput minimized

The developed mean value model integrated with a The developed mean value model integrated with a comprehensive controller model for control analysiscomprehensive controller model for control analysis

The integrated engine and control system model extensively validated with satisfactory accuracy achieved

1 Step change, 3 Step transients, 1 FTP cycleControl strategies development & preliminary calibrations beforehardware availability and testing


SummarySummaryCurrent Capabilities:Current Capabilities:

Provide fast-running models for control developmentExplore control strategies and study control parameter sensitivities Generate preliminary calibrations before hardware availability and testing Use for air-EGR system calibrationsAllow easy adaptation to hardware changes …

Future Outlook:Future Outlook:Analytic calibration critical and integral part of modern embedded powertrain controllers development process, but more important in the early development phasePhysical dyno and/or vehicle testing still needed, but to be minimized Computer simulations more accurate, powerful and standardized, but model development time, model throughput, and model runs to be reduced

John Deere Power SystemsAnalytical Engine Calibration at John Deere

John Deere Power SystemsAnalytical Engine Calibration at John Deere

Jason SchneiderEngine Engineering

MotivationMotivation• Performance Optimization

• Off-Road Market– Number of applications – Over 1000 internal and external

• Application Variation– Different Usage Profiles– Different Optimization Objective

• Complexity– HPCR– Cooled EGR – VTG and EGR Valve

Worldwide Engine CustomersWorldwide Engine CustomersENGINE BUSINESS OVERVIEW

External ApplicationsExternal Applications

≅ 50% Engine VolumeIndustrial, Power Generation

and Marine Applications

Internal ApplicationsInternal Applications≅ 50% Engine Volume

Ag, C&F, CC&E Division Applications

Usage Profile ExamplesUsage Profile Examples

Application 1 Application 2

750

900

1150

1350

1500

1600

1700

1800

1900

2000

2100

2200

2300

2412

.5

2475

0

10

20

30

40

50

60

70

80

90

100

110

120

Speed [rpm]

Pow

er [%

of R

ated

]

Islands of Operation

750

900

1150

1350

1500

1600

1700

1800

1900

2000

2100

2200

2300

2400

2450

0

10

20

30

40

50

60

70

80

90

100

110

120

Speed [rpm]

Pow

er [%

of R

ated

]

Islands of Operation

Analytical Calibration ObjectiveAnalytical Calibration Objective

• Generate calibration tables off line from test bed– Comply with emission legislation– Minimize BSFC, subject to application, base engine and

calibration constraints

• Deere developed empirical engine models are used– DOE – Matlab MBC Toolbox– Matlab (MBC Toolbox), Statistica, Table Curve 3D

• Deere Optimized Table Generator (DOTG) interface is used to enter calibration optimizer settings

– Excel driven Matlab optimization

• Final results are calibration set point tables

DOTG Calibration Process Flow

Implementation

Data ModelingDOE Design Data Collection

Deere Optimization GUI

Table ExportTable Import

CalibratorSettings

Calibrations not released without test bed confirmation

Virtual Calibration Lab

TimingBSFCTorque

Fuel Pressure

Input and Output Scheme

Fuel/Air Ratio

Turbo. SpeedPeak PressureVTG Position

ConstraintsAnd

Optimal Tables

Objective

Optimal Tables

EGR

NOx+HCTC AFR

Constraints

PMSmokeEGR Cooler TComp. Out TComp. Out PEOI TimingCoolant HRIntercooler HRPress. Ratio

RPM

Fuel Mass/Inj

Table Set PointComp. Mass

DieselEngine Models

BenefitsBenefits• Performance optimization given application constraints

– 1-3% improvement in application specific fuel consumption compared to conventional techniques

• Reduction of needed testing and associated expense– Measured in hundreds of thousands of dollars compared to

conventional techniques

• Control of constraint usage to minimize errors– Example – peak firing pressure or exhaust temperature– Consistent reliability performance across applications

• Calibration methodology is controlled – NTE compliance– Similar performance output of engine across applications

Optimization Potential –Optimization Potential –8530 Ag Tractor

1 2 3 5 6 7

Emission Mode

BSF

C, g

/kW

h

Fully Optimized Traditional Calibration

0.7%

1.1%

6.6%

1.4%

1.1%1.0%

Industry Record, Nebraska Test 2005:Most Fuel Efficient Row-Crop TractorIndustry Record, Nebraska Test 2005:Most Fuel Efficient Row-Crop Tractor• 8430 Series Tractor / 9.0L PowerTech Plus

– 8.8% more fuel efficient with 40% less emissions

– Engine optimization

– Vehicle efficiency improvement

JOHN DEERE ENGINES

AccuracyAccuracy

Example Results – OEM RatingExample Results – OEM RatingNOxHC

0 1 2 3 4 5 6 7 8

Modal Point

NO

xHC

, g/k

Wh

-10

-8

-6

-4

-2

0

2

4

6

8

10

% E

rror

Prediction Actual % Error

Example Results – OEM RatingExample Results – OEM RatingSmoke

0 1 2 3 4 5 6 7 8

Modal Point

Smok

e, R

B

-10

-8

-6

-4

-2

0

2

4

6

8

10

% E

rror


Example Results – OEM RatingExample Results – OEM RatingBSFC

0 1 2 3 4 5 6 7 8

Modal Point

BSF

C, g

/kW

h

-5

-4

-3

-2

-1

0

1

2

3

4

5

% E

rror


Example Results – OEM RatingExample Results – OEM RatingVTG Position

0 1 2 3 4 5 6 7 8

Modal Point

VTG

Pos

ition

-5

-4

-3

-2

-1

0

1

2

3

4

5

% E

rror


Transient Certification TestsTransient Certification Tests

• NRTC in effect for Interim Tier 4

– New problem statement for Deere

• Methods heavily depended on emissions technology chosen

– Number of independent variables• EGR vs Non-EGR• After-treatment and interaction

Transient Certification TestsTransient Certification Tests• Perturbation of calibration tables for sequence of transient test

runs

– Select value for calibration tables that minimizes emission tradeoff point by point

– Fuel Pressure, Injection Timing, EGR rate, etc.

• Steady state points weighted to for correlation to transient test

– Calibration process as outlined can be used to reach targets

– System control tuning for refinement of NOx and PM tradeoff for cooled EGR engine

• Transiently accurate emission models

– Most elegant

FutureFutureFirst Steps

• Elimination of confirmation runs for steady state

– Allows further release of expensive calibration resources to earlier stages of product development

• Accurate transient emission models for transient optimization

– Fits well with need for embedded models for systems with AT to predict state of system

FutureFuture

Needs

• Better integration of calibration process with ECU software development process

– Collective mind set

– Comprehensive tool set• Controller Models Engine Models Calibration

• Predictive emission models that are accurate to drive process further upstream

– Engine cycle simulation environment

Thank You

2

Analytical Calibration Process

3

Current Benefits of Analytical Calibration

Enables calibration process prototyping before hardware availability

Provides diagnostic data for later hardware testing

Provides fast-running statistical engine model for control development

Can be used for calibrations related to engine-breathing (e.g., EGR, VE)

Provides a non-hardware training environment for new calibrations

Acts as an executable specification of company calibration processes

Provides a means of determining minimum DoE testing requirements

4

Analytical Calibration Workflow Example

5

Identify Future Physical Test Setup

VGT Pos.

EGR Pos. Brake TorqueEngine Out NO

RPM Fuel Mass/Inj

SOI

Fuel Press.

Exhaust AFRTurbo. SpeedPeak PressureEGR Fraction

Closed-LoopEGR Fraction ControllerEGR Fraction Cmd.

6

Define Optimization Model Setup

VGT Pos.

EGR Pos.

Brake TorqueEngine Out NO

RPM Fuel Mass/Inj

SOI

Fuel Press.

Minimize mode-weighted brake specific fuel consumption,subject to multiple mode-based output constraints


BSFC

Used toBuildExtraTable

Objective

Used To BuildOptimal Tables

7

Design Experiment

8

VGT Pos.

EGR Pos. Brake TorqueEngine Out NO

RPM Fuel Mass/Inj

SOI

Fuel Press.


Closed-LoopEGR Fraction ControllerEGR Fraction Cmd.

High FidelityEngine Model

Execute Virtual Testingwith Distributed Computing

9

Statistically Model Engine Responses

10

Set Up Optimizations With Constraints

11

SOI Table Fuel Mass Table

Fuel Pressure Table VGT Rack Position Table

EGR Mass Fraction Table

Generate Optimal Calibration Tables

12

Future Benefits of Analytical Calibration

Inexpensive calibration adaptation to late program hardware changes

Tighter feedback between engine hardware design and control design using model sharing

Improvement of predictive quality of CAE engine models resultingfrom calibrator feedback

2007 sae commercial vehicle engineering congress – …...• model-based approaches and...

Documents