ETAS Webinar - ASCMO Calibration. DOE & Statistical Modeling Injection Consumption Ignition Torque AFR HC EGR P-rail NOx Inlet-cam Outlet-cam 1 1 Soot T-exhaust Roughness
What is Design of Experiments? DOE is a formal mathematical method for systematically planning and conducting scientific studies that change experimental variables together in order to determine their effect of a given response DOE makes controlled changes to input variables in order to gain maximum amounts of information on cause and effect relationships with a minimum sample size. DOE is more efficient that a standard approach of changing one variable at a time in order to observe the variable s impact on a given response. 2 DOE generates information on the effect various factors have on a response variable and in some cases may be able to determine optimal settings for those factors. Source: DOE by R.C. Baker
Col_2 Factor_B Analysis Contours of Estimated Response Surface N1 N2 N3 N4 N5 N6 N7 Box-and-Whisker Plot 6 55 5 45 4 35 3 Factor_C=6. 1 14 18 22 26 3 Factor_A Var_3 9. 9. 9. 9. 11. 1. 1. 1. 1. 111. 11. 11. 11. 11.5 11.7 11.9 12.1 12.3 Col_1 3
DOE Plan -> Model -> Optimize Approach Model training using measured data from DoE-Plan Mathematical Model DoE-Plan Systeminputs Systemoutputs Optimising outputs based on the model 4 4
Number of ECU-Labels Motivation: Challenges of todays ECU-calibration* Complexity Time to market 4. 3. 2. 1. Shorter development times Pressure through competition Setting of market trends 5 21 215 New technologies: Hybrid, FlexFuel,... Differences in legislation Worldwide development cooperation between OEMs, Suppliers, Tier1s and Engg. Service Providers More individualization New markets Variance *Source: Bosch Distributed development 5
Number of ECU-Labels Motivation: Challenges of todays ECU-calibration* Complexity Time to market 4. 3. 2. 1. Shorter development times Pressure through competition Setting of market trends 5 21 215 New technologies: Hybrid, FlexFuel,... Differences in legislation Worldwide development cooperation between OEMs, suppliers and Service Providers More individualization New markets Variance *Source: Bosch Distributed development 6
Main Complexity Drivers: Increasing no. of engine parameter Sharpening targets for emission and fuel consumption (CO 2 ) Example: Modern Gasoline Engine Operating Range: Speed Load Engine Parameter: Injection Timing Ignition Timing Fuel Pressure Exhaust Gas Recirculation Exhaust Camshaft Intake Camshaft Swirl Valve Complex Interactions Conflicting Targets Targets: Consumption/CO 2 Emissions: Soot / Particle NO x HC Stability (CoV) Noise Exhaust-Temperature... Every new engine parameter leads to a multiplication of ECU-Labels Calibration effort increases exponentially if classical methods are applied 7
speed [rpm] 9 6 6 mode speed [rpm] 9 6 mode speed [rpm] 9 mode 3 relative internal torque [%] 3 relative internal torque [%] 3 relative internal torque [%] speed [rpm] 9 6 6 mode speed [rpm] 9 6 mode speed [rpm] 9 mode 3 relative internal torque [%] 3 relative internal torque [%] 3 relative internal torque [%] Calibration Process: Classical Methods Calibration- Phase Base Emission & Fuel Optimization Engine Calibration Base ECU Models Start & Fuel Compensation Vehicle Calibration Mixture- Control Emission & Driveablity OBD Monitoring Test & Validation Challenge High No. of Engine Parameter Transient behaviour of Engine, Catalyst, Vehicle Dependencies & Interactions of whole ECU with Vehicle Classical Method Time, costs and availability of prototypes! Extensive Measurement at the Engine Test Bench Numerous Vehicle Test at the Roller Bench and Test Track Numerous Vehicle Trips on Proving Ground & Road 8
Typical Workflows with the Tool ETAS ASCMO [R->L->M] Cycle-Optimizer DoE Test planning Modeling & Optimization Cycle-Prognosis & Map Optimization Experimental Design Engine on Test Bench Modeling Visualization Optimization ECU-calibration ECU-Data Virtual Testbench Generating Virtual Data from Model ETAS ASCMO-Toolchain Benefits: Efficiency: Reduction of measurement effort up to 8 % Quality: optimizer systematically finds best calibration Cost: Real prototypes can be replaced by models Integrated development: Reuse of models from engine calibration for other tasks Model Export Export to Matlab, C, Simulink, Implementation in HiL-System 9
Comfort Calibration Process: Support with Model Based Methods Calibration- Phase Base Emission & Fuel Optimization Engine Calibration Base ECU Models Start & Fuel Compensation Vehicle Calibration Mixture- Control Emission & Driveablity OBD Monitoring Test & Validation Challenge High No. of Engine Parameter Transient behaviour of Engine, Catalyst, Vehicle Dependencies & Interactions of whole ECU with vehicle Appropriate Model based Method Models can not fully replace real prototypes but significantly reduce their number Dynamics Data driven Engine Models (Global DoE ) Models + single ECU-Functions Criteria based DoE-Methods Combination of Models with real ECU (HiL-System) 1
Comfort Calibration Process: DoE Methodology for Engine Calibration Calibration- Phase Base Emission & Fuel Optimization Engine Calibration Base ECU Models Start & Fuel Compensation Vehicle Calibration Mixture- Control Emission & Driveablity OBD Monitoring Test & Validation Challenge High No. of Engine Parameter Transient Behaviour of Engine, Catalyst, Vehicle Dependencies & Interactions of whole ECU with Vehicle Appropriate Model based Method Dynamics Data driven Engine Models (Global DoE ) Models + single ECU-Functions Criteria based DoE-Methods Combination of Models with real ECU (HiL-System) 11
Data-driven Modelling and DoE (Design of Experiment) Basic Principle Description of the engine with a model that is based on real data Mathematical approximations are used for modelling, no physical description necessary DoE minimises the number of required measurements by an statistically optimal distribution Model training using measured data from DoE-Plan DoE-Plan Mathematical Model CO2 Ordnung 3, Minimum bei (17Nm,3kW) 165 16 155 Systeminputs 15 145 1 8 6 4 35 25 15 2 Systemoutputs Optimising outputs based on the model 12 12
Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination Charge Determination But often, full factorial engine mapping is still the standard calibration process Example: Increase of measurement effort for MAF & Torque calibration with variable camshafts: Engine with 1 variable Camshaft: ~2. MP Engine with 2 variable Camshafts: ~25, MP ~1h ~ 1h ~h 13 13
Requirements to the Modelling Algorithm for broad Use in Calibration: High accuracy and easy parameterization even for strongly nonlinear behavior Polynomial- or neural net models do not fully meet these requirements Development of new modelling algorithms and integration in the tool ASCMO* Model training using measured data from DoE-Plan DoE-Plan Mathematical Model Systeminputs Systemoutputs Optimising outputs based on the model *ASCMO = Advanced Simulation for Calibration, Modeling and Optimization, joint project between Bosch, BEG and ETAS 14 14
Polynomials or Neuronal Nets Principle: Search in a given class of functions (polynomial, neuronal net,...) Fit the model parameter by experts and validation measurements Disadvantages: Limited flexibility & danger of over-fitting High expertise and assumptions necessary Statistical machine learning methods Principle: Search in a complete function space: Automatic determination of the most likely function Advantages High flexibility without assumptions or expertise Gives local confidence interval (model variance) Robust against outliers Training Data Training Data & Model Prediction Model Prediction Modelvariance & Validity Modeling a complex 1-D signal with classical DoE-Models ( Advanced Polynomials ) Modeling a complex 1-D signal with statistical machine learning methods 15 15
Global Cycle Optimization based on the Engine Model and Vehicle Data: Minimize fuel-consumption for given cycle-emission constraints and other targets Vehicle data as mass,transmission Drive cycle (speed-profile) Speed/load trajectory (weighted O.P. s) Prognosis ASCMO Model Calibration Maps Consumption Emission, FC: NO x : 5,3 l/1km,25 g/km : Optimization Cycle Values, Local Constraints, Smooth Maps 16
Global Optimization of the ECU Base Maps for a given Cycle with ASCMO : Minimize fuel-consumption, keep constraints for emissions, stability and smooth maps 17
Comfort Calibration Process: Criteria based DoE Methodology for Vehicle Calibration Calibration- Phase Base Emission & Fuel Optimization Engine Calibration Base ECU Models Start & Fuel Compensation Vehicle Calibration Mixture- Control Emission & Driveablity OBD Monitoring Test & Validation Challenge High No. of Engine Parameter Transient Behaviour of Engine, Catalyst, Vehicle Dependencies & Interactions of whole ECU with Vehicle Appropriate Model based Method Dynamics Data driven Engine Models (Global DoE ) Models + single ECU-Functions Criteria based DoE-Methods Combination of Models with real ECU (HiL-System) 18
Comfort Calibration Process: Support with Model Based Methods Calibration- Phase Base Emission & Fuel Optimization Engine Calibration Base ECU Models Start & Fuel Compensation Vehicle Calibration Mixture- Control Emission & Driveablity OBD Monitoring Test & Validation Challenge High No. of Engine Parameter Transient Behaviour of Engine, Catalyst, Vehicle Dependencies & Interactions of whole ECU with Vehicle Appropriate Model based Method Dynamics Data driven Engine Models (Global DoE ) Models + single ECU-Functions Criteria based DoE-Methods Combination of Models with real ECU (HiL-System) 19
Comfort Calibration Process: Integration of data driven models in a HiL Calibration- Phase Base Emission & Fuel Optimization Engine Calibration Base ECU Models Start & Fuel Compensation Vehicle Calibration Mixture- Control Emission & Driveablity OBD Monitoring Test & Validation Challenge High No. of Engine Parameter Transient Behaviour of Engine, Catalyst, Vehicle Dependencies & Interactions of whole ECU with Vehicle Appropriate Model based Method Dynamics Data driven Engine Models (Global DoE ) Models + single ECU-Functions Criteria based DoE-Methods Combination of Models with real ECU (HiL-System) 2
Enabling a HiL-System for Calibration by Integrating a Global Engine Model ETAS ASCMO Engine Model: Emissions, Fuel, Torque,... + Project Specific Parameter: Inj.-Flow, Manifold- Vol. Integration in Standard LabCar Model-Structure = LABCAR as a Virtual Vehicle for Calibration Optional + Comp. Models: Catalyst, Powertrain, Sensors, 21
Prediction of Raw-Emissions from a HiL (LABCAR) for a ECE-Cycle, Gasoline Engine: Error in cumulated emissions: HC & NO x < 2%, CO < 1% The HiL system can now be used for vehicle calibration tasks like OBD or emission validation 22
ASCMO : In Conclusion New data driven modelling methods can be used in different phases of the calibration process, reducing time and the demand of prototypes. By the integration in the easy to use tool environment ASCMO, model based calibration methods are no longer restricted modelling experts The shown methodology is today broadly used at Bosch, BEG by >35 engineers and other customers in gasoline & diesel calibration and other development areas. ETAS ASCMO 4.3 is available for use. positive feedback from customer evaluations and benchmarks* *Klar, H.; Klages, B.; Gundel, D.: Automation of Model-Based Calibration in Engine Development. 6th Conference on DoE in Engine Development. Berlin: Expert Verlag 211 23 23
ASK for ASCMO! Thank You for Your Attention! Questions? 24 24