Cold Engine Emissions Optimization Using Model Based Calibration

19th – 20th June 2007

International Automotive Conference

Dr. Clive TindleGeneral Motors Holden Ltd.

Overview

• Cold engine emissions.• What is model-based calibration?• Design of experiments.• Statistical modelling.• Response model trade-offs and optimisation.• Implementation.

Cold Engine Emissions

• “Cold Engine” refers to operation after engine start up.• 20˚C emissions tests.• Catalytic converter has not achieved light-off.• Engine operation is referred to as “Catalyst Heating.”

Challenges before light-off

• Minimise exhaust pollutants.• Achieve acceptable combustion stability.• Maximise fuel economy.• Minimise piece costs, e.g. catalytic converter, engine

hardware.

Engine Control Options

• Ignition timing, and fuelling.• Continuously, variable valve timing for both intake and

exhaust.• Split injection, semi-stratified charge combustion.

Increased complexity means more degrees of freedom !

PFI - 3-4 variables.DI - 7-8 variables.

Our brains struggle to visualise 2 or 3 variables at a time !

Model-Based Calibration Toolbox

• An add-on toolbox for MATLAB®

• Specifically developed for engine calibration• Has been commercially available for approximately 5 yearsModel-Based Calibration Toolbox provides design tools for calibrating powertrain systems. The toolbox is built on the high-performance technical computing environment of MATLAB® and the modeling capabilities of Simulink®. Model-Based Calibration Toolbox enables the development of optimized calibrations for complex high-degree-of-freedom engines that are difficult to calibrate using traditional methods. Using the toolbox, you can develop a process for systematically generating calibrations that find an optimal balance of engine performance, emissions, and fuel economy.

Model Based Calibration Concept

Data ModelingExperiment Design

Calibration Implementation

Data Collection

Design of Experiments

• Design of experiments is a technique used to select the most statistical useful data.

• Essential for high degree of freedom systems.• # of points influenced by degrees of freedom and model.

Benefits• Significant reduction in test points compared with ‘one

parameter at a time’ or factorial test methods.• Data collected randomly minimising influence of ‘noise’

parameters.

Examples of DOE

Factorial designsLV = # of pointsL = Levels.V = Variables.3 levels and 3 variables = 27 points.

Central Composite Design (CCD)A classical DOE. Minimum of 15 points.Almost 50 % reduction in test points.

Optimal designs and augmented space fills

• Computer based design, flexible and useful for constrained design spaces.

Model Based Calibration

Data ModelingExperiment Design

Calibration Implementation

Data Collection

Statistical Modelling

• The choice of variables (factors and responses) should be selected based on physical knowledge.

• Models capture the shape of the response and confidence intervals.

• Modelling is split into two parts;• Local model.• Global model.

Typical Models are;• Polynomials.• Radial Basis Functions (RBF).

Local Model

• Local model refers to spark sweep data. Spark is the variable onthe x-axis, all other variables are held constant.

• Fitting local models help identify bad data and ‘expected’ trends.

- 3 0 - 2 0 - 1 0 0 1 0 2 0 3 0

0

0 . 0 0 5

0 . 0 1

0 . 0 1 5

0 . 0 2

0 . 0 2 5

0 . 0 3

0 . 0 3 5

0 . 0 4

Spark Retard

HC e

mission

s

Local Response Model for HC

Outlier

Towards MBT

33

2210 SPARKbSPARKbSPARKbbHC +++=

Global Model

• Coefficients from the local model are fed into the global model.

-30 -20 -10 0 10 20 300

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

Spark

HC

[g/s

]

4 Cubic Fits to Spark Sweeps

• A model now represents the local coefficients, enabling reproduction of the spark sweep at any combination of global conditions.

Response Surface

• The response surface can be calculated from the local and global model coefficients.

2 5

3 0

3 5

4 0

4 5

5 0

5 5

6 0

6 5

7 0

- 2 5

- 2 0

- 1 5

- 1 0

- 5

0

5

1 0

- 2

- 1

0

1

2

3

4

5

6

7

8

x 1 0

- 3

Air Flow

Effect of Global Inputs on HC Emissions

Intake Cam Advance

HC

Em

issi

ons

• Interrogation increases understanding of trends.

Model Based Calibration

Data ModelingExperiment Design

Calibration Implementation

Data Collection

Optimisation and Trade-offs• Optimise controlled variables to meet emissions targets and

satisfy customer expectations.• Optimisation can focus on one response or consist of trade-

offs.• Input variables can be optimised manually considering all

responses.• Optimization Toolbox algorithms can be used for automatic

optimisation.

Single Objective Optimisation

• Appropriate when no trade-off exists or is necessary.• Constraints are used to limit other responses, e.g. combustion

stability.

- 3 0

- 2 0

- 1 0

0

1 0

2 0

3 0

2 5

3 0

3 5

4 0

4 5

5 0

5 5

6 0

6 5

7 0

0

0 . 5

1

1 . 5

2

2 . 5

3

3 . 5

x 1 0

4

Exhaust Energy and HC Emissions fn(mass flow rate, spark advance)

Spark AdvanceMass Flow Rate

Exha

ust

Ene

rgy

- 3 0

- 2 0

- 1 0

0

1 0

2 0

3 0

2 5

3 0

3 5

4 0

4 5

5 0

5 5

6 0

6 5

7 0

- 0 . 0 0 5

0

0 . 0 0 5

0 . 0 1

0 . 0 1 5

0 . 0 2

0 . 0 2 5

0 . 0 3

0 . 0 3 5

Mass Flow Rate

Hyd

roca

rbon

sApproximate constant torque vector

Approximate constant torque vector

Spark Advance

Multi-Objective Optimisation

• Necessary when optimising one response, results in the deterioration of another.

• Generates a Pareto, more time consuming to select optimum calibration.

Pareto: HC Emissions and Exhaust Gas Energy

2 .4

2 .5

2 .6

2 .7

2 .8

2 .9

3 .0

3 .1

3 .2

4 2 4 4 4 6 4 8 5 0 5 2 5 4 5 6

Exhaust Energy

Engi

ne O

ut H

C E

mis

sion

s

Stability <= 100 % Stability <= 120 %

Pareto extended by increasing Stability limit

Re-use of Models

• Collect data once.• Re-use models as many times as required.• Constraints can be changed.• Models are an engine test bed simulator.

Model Based Calibration

Data ModelingExperiment Design

Calibration Implementation

Data Collection

Implementation

• Optimum values must translate into ECU maps.• Take into account strategy deficiencies. • Can be limited by engine hardware and other calibration

areas.• Cycle analysis.

Calibration Verification On FTP Drive Cycle

0

10

20

30

40

50

60

70

80

90

100

0 25 50 75 100 125 150

Time [s]

Nor

mal

ised

NM

HC

[%]

0

10

20

30

40

50

60

70

Vehi

cle

Spee

d [m

ph]

Vehicle Speed

HC Baseline before Optimisation

HC After Optimisation

Modal Data

Conclusions

• Model-based calibration and design of experiments is a key tool in developing high-technology engines with demanding emissions targets and driveability constraints.

• Use of models allows careful scrutiny of multiple responses and increases understanding of engine operation across all degrees of freedom.

• Use of models as engine test bed simulator reduces expensive retesting.

• Experience is a key factor in successful application of this process.

Acknowledgements

Thanks to Kevin Yardley (Manager – Powertrain Calibration) for making it possible for me to present this material at this conference.

The author would like to acknowledge the assistance of the following people at General Motors Holden Ltd. Greg Horn, Martin Jansz, Richard Hurley, Kevin Yardley, Joshua Wood, and Julian Banfield.

Thanks also to David Sampson at The MathWorks for his continuingsupport of Model-Based Calibration Toolbox.

Questions?