Steady-State Engine Modeling for Calibration: A Productivity and Quality Study MathWorks Automotive Conference 2007 Hyatt Regency, Dearborn, MI Ulrike Schoop John Reeves Satoru Watanabe Ken Butts IAV GmbH A&D Technology Toyota Motor Corporation Toyota Motor Engineering and Manufacturing, NA
Presentation Outline 1. Motivation and Introduction 2. Advanced Calibration Process Considerations 3. Engine Test-bench Infrastructure 4. Engine Modeling: Productivity and Quality Assessment 5. Future Work 2
Motivation New powertrain technologies benefit society * 1. EFl : Electronic Fuel lnjection *2. VVT-i : Variable Valve Timing-intelligent *3. D-4 : Direct lnjection 4 stroke gasoline engine *4. D-4S : Direct lnjection 4 stroke gasoline engine superior version (Source: Toyota Motor Corporation - http://www.toyota.co.jp/en/tech/environment/) but their development carries planning implications Complexity Uncertainty Engine development is already on Toyota s critical development path! 3
Consider Engine Mapping Engine Map: Steady-state engine response to control inputs Establishes operating limits Used to set control input bias versus operating point (steady-state calibration) Typically an empirical model due to accuracy requirements Traditional Methods use Full-Factorial Experiment Design Development Planning Implication: 1,000,000 100,000 Engine technology vs. Measurement requirements 238,140 26,460 476,280 # of Measurements. 10,000 1,000 100 2,940 We need a different way! 10 1 Base Engine 12 x 7 x 7 x 5 + Intake VVT 9 levels + Exhaust VVT 9 levels + Manifold Tuning 2 levels 4
Model-based Calibration Our Process Model : Model-based Calibration Toolbox Definition of factors and responses Experimental design Atlas ATLAS (A&D) Measurements on the test bench Model-based Calibration MBC Toolbox Toolbox (TMW) CAGE (TMW) and scripts (IAV) Modeling Optimization & evaluation of DoE- models X Filling tables and fitting models of ECU X X not considered in this presentation 5
Purpose of our study 1. Investigate (and Develop where necessary) Model-based Calibration processes, methods, and tools infrastructure for engine mapping to increase productivity. 2. Quantify productivity benefit and verify quality 1. Relative to known metrics from previous production development 1. I-4 engine system with Intake VVT 2. Super-Ultra-Low-Emission-Vehicle (SULEV) emission target 6
Advanced Calibration Process Considerations 1. Model Structure 2. Experiment Design 3. Model Generation 7
Model Structure based on IAV experience 1. One-Stage Model 2. Engine Map is composed of five over-lapping regional models with interpolation WOT Lambda < 1 (catalyst temperature) Load / % Region 2 Region 4 Region 5 Region 1 Region 3 600 1000 2000 3000 4000 5000 limited CAM timing 6000 Speed / rpm 8
Response Models Engine Model Output Model Inputs Purpose torque speed, load, VVT, λ, spark timing optimization fuel consumption speed, load, VVT, λ, spark timing optimization CA50 (crank angle @ 50% burn) speed, load, VVT, λ, spark timing optimization MBT spark timing speed, load, VVT, λ constraint knocking limit spark timing speed, load, VVT, λ constraint exhaust gas temperature speed, load, VVT, λ, spark timing constraint catalyst temperature speed, load, VVT, λ, spark timing constraint HC, NOx speed, load, VVT, λ, spark timing constraint engine roughness (COV of IMEP) speed, load, VVT, λ, spark timing constraint ECU calculated engine load speed, load, VVT scaling 9
Method to coordinate the regional experiment designs 1. Analyze the initial preparation measurements and engineering knowledge (i.e. full load, zero torque, VVT, spark timing, and limits) to determine the experiment constraints for each region. Capture these constraints in five corresponding test-plans in a single Model-Based Calibration Toolbox Project. 2. Use the Model-Based Calibration Toolbox to sequentially design an experiment for each test-plan while enforcing common measurement-points in overlapping areas. 3. Export the regional experiment designs and combine them into a single design. Sort the measurement points by speed and then load in ascending order. 4. Include repeatability measurement points at regular intervals to allow the test-automation procedures, test engineers, and modeling engineers to assess data quality. 10
Experiment Regional Design Recipe Measure the engine response at: 1. the spark timing that yields either the Maximum Brake Torque (MBT) or the engine knock limit and 2. some spark timing delta from the MBT / knock limit spark timing as scheduled by the experiment design. Specify a 4th order polynomial with 3rd order interaction model then: 1. generate a D-optimal design with 20% more points than is minimally required. The D-optimal design gives good test coverage at the borders of the model. 2. add 20% more points with a V-optimal design. The V-optimal design adds coverage of the interior of the model. 3. add 20% more validation points with a V-optimal design. These measurements are not used for the fitting of the model. 11
Model Generation Recipe Semi-automated scripts that use the Model-Based Calibration Toolbox Command Line Interface: 1. automatically generate alternative polynomial and Radial Basis Function models that are initially evaluated based on validation Root-Mean-Square-Error. 2. manually tune candidate polynomial models based on condition number and the effect of transformation. 3. manually inspect candidate models using single-influence plots to ensure that the responses match physical intuition. 12
Engine Test-bench Infrastructure 1. Combustion analysis system to measure knock, misfire, and Coefficient of Variation of Indicated Mean Effective Pressure (COV of IMEP). 2. Test-bench automation 3. Engine water and oil temperature conditioning Prototype ECU Engine Dyno. Emissions Analyzer CAL CAS DAC ASAP3 CAL ASAP3 CRAMAS & Rtype ECU Prototyping Tool CAS Combustion Analaysis ADAPT Data Acquisition and Control ATLAS Calibration Tool 13
Test-bench Automation using Atlas by A & D Technology 1. Atlas communicates with test-bench data-acquisition and control and engine ECU calibrations systems via ASAM standard protocols. Our set-up allowed ECU calibration command and response rates at up to 10HZ. 2. Atlas readily imports and executes experiment designs from the Model-Based Calibration Toolbox. 3. Atlas can dynamically access MATLAB for on-line data processing. We use this feature to fit a spark sweep from as few as three measurements. 4. Atlas can execute parallel threads of execution. We separate our engine safe operation monitor and our test-execution processes into parallel, communicating execution threads. 14
Test-bench Automation Example (! " #$ #$ %&' ( #$ 15
Engine Modeling: Productivity Run # Experiment Design Test Automation Test Bench # of points & Test Time (compared to baseline) 1 5 independent regions; No overlap matching Baseline; Sequential program flow. Heat exchanger only # points: X 0.24 Test time: X 0.47 2 5 dependent regions; Overlap Commonality General Improvements: Parallel program. As above. # points: X 0.17 Test time: X 0.36 3 As above. Improved COV of IMEP With heater # points: X 0.17 Test time: X 0.28 16
Engine Modeling: Quality Assessment 10.000 8.000 Run2 Ave Repeatability (Std/Mean) Run2 Model Error (RMS/Range) Run3 Ave Repeatability (std/mean) Run3 Model Error (RMSE/Range) Percent 6.000 4.000 2.000 0.000 Torque Fuel Flow CA50 Cat Temp In Cat Temp Rear HC NOx COV of IMEP ECU KL 17
Given Successful Demonstration: Future Work 1. Installation of the test-bench infrastructure in Toyota facilities for detailed evaluation. 2. Confirmation of the methods on a second engine application in close cooperation with Toyota mass-production engineers. 3. Porting the test-bench automation logic to a new Simulink / Stateflow based test-automation environment from A&D. 4. Extending the test-bench automation methods in a general way to address more advanced engine technologies. 5. Connecting the test-automation to Toyota standard ECU communication tools. 18
Acknowledgements 1. Frank Biens, Tony Gullitti, Mirko Knaak, and Karsten Roepke, (IAV). 2. Andy Hall, Brian Moore, and Ray Skinner (A&D Technology). 3. Allen Lock (Denso International of America). 4. Kotaro Tanaka, Harunaga Uozumi (Fujitsu-Ten) 5. Harufumi Muto and Masato Ehara (Toyota Motor Corporation.) 19