Workshop Model based calibration methodologies Begrüßung, Einleitung

Transcription

1 Workshop Model based calibration methodologies Begrüßung, Einleitung Stefan Jakubek, Technische Universität Wien, Institut für Mechanik und Mechatronik m Thomas Winsel, AVL List GmbH, Graz Powertrain Calibration Technologies Workshop Model based calibration methodologies Begrüßung, Einleitung 1/29

2 Inhalt 1 Introduction 2 Team 3 Reference Projects 4 Research Topic A 5 Research Topic B 6 Summary Workshop Model based calibration methodologies Begrüßung, Einleitung 2/29

3 General aspects The new Christian Doppler Laboratory pursues the development of new and integrated methodologies for model based calibration of automotive systems: Combustion engines, Powertrain systems, Hybrid components Different input parameters VVT-Timing Intake Exhaust Injection timing Spark advance A/F ratio Driver Environment EGR VTG load speed Fuel Efficiency Conflicting goals Driveability Performance Emissions Workshop Model based calibration methodologies Begrüßung, Einleitung 3/29

4 m Institute of Mechanics and Mechatronics Introduction Team Reference Projects Research Topic A Research Topic B Summary Faculty for Mechanical Engineering and Business Science Five departments Vehicle dynamics and biomechanics (multibody simulation in vehicle dynamics) Applied mechanics (nonlinear stability theory) Machine dynamics Measurement and actautor technology Control and process automation Dept. of Control and Process Automation: 3 research areas Active vibration damping, Process control, Automotive control Methodology Nonlinear system identification State observation/kalman Filters Linear/nonlinear predictive control H 2- and H -control (robust control) Workshop Model based calibration methodologies Begrüßung, Einleitung 4/29

5 Goals: Track geometry estimation with regular vehicles Estimation of track geometry by measurements on the train Linear and nonlinear model identification Process and measurement noise estimation Nonlinear Kalman filter with constraints Sensor placement Virtual test bed (Multi-Body System) Valuation of safety limits Evaluation of track-vehicle interactions v y 4 u 4 y 3 u 3 y 2 u 2 y 1 u 1 Workshop Model based calibration methodologies Begrüßung, Einleitung 5/29

6 MEMBAT Modelling, Emulation and Management of high complexity traction batteries Goals: Precise and highly dynamic emulation of energy storage systems by battery simulators for hybrid powertrain testbeds Innovative concepts to control the output voltage despite the influence of uncertain loads Impedance emulation of real battery or supercap systems Complex battery models suitable for real-time simulation Workshop Model based calibration methodologies Begrüßung, Einleitung 6/29

7 BioNetControl Predictive Control and optimization of district heating networks with biomass power plants Goals Modelling of relevant system components Higher-level optimal control of the total system Lower-level optimal control of the plant Modular design and structured configuration Robust MPC with respect to model mismatch Biomass power plant I Consumer Grid Steam turbine District heating net Consumer Gas power plant Consumer Biomass power plant II Workshop Model based calibration methodologies Begrüßung, Einleitung 7/29

8 Team des CD-Labors TU-Wien, Inst. E325 Laborleiter: Prof. Stefan Jakubek PostDoc: Dr. Christoph Hametner PhD 1: DI Christian Mayr PhD 2: DI Markus Stadlbauer AVL-List GmbH DI Horst Pflügl: Entwicklungsleiter Powertrain Calibration Technologies Dr.-Ing. Thomas Winsel: Research Manager Powertrain Calibration Technologies Dr.-Ing. Timo Combe: Fachteamleiter Methodology, Development and Calibration Dr. Mario Schweiger: Application manager Hybrid testing systems and battery management systems DI Manuel Nebel: Calibration of Diesel engines for series production of passenger cars Extern Prof. Stefan Volkwein (Univ. Konstanz): Hull Algorithms Team of Prof. Steindl (TUW): Stability Analysis Workshop Model based calibration methodologies Begrüßung, Einleitung 8/29

9 General aspects The proposed CDL pursues two main targets which are organised in research topics: Research Topics of the CDL: Research Topic A: The basic development of methodologies for an integrated model based calibration workflow. Research Topic A Research Topic B: The systematic employment and enhancement of these methodologies to essential calibration tasks. Research Topic B Workshop Model based calibration methodologies Begrüßung, Einleitung 9/29

10 Research Topic A Main Focus: Development of new methodologies and the related basic research for all steps in model based calibration. There are three research tasks: A1: Experiment design A2: Nonlinear system identification A3: Controller stability analysis and design Workshop Model based calibration methodologies Begrüßung, Einleitung 10/29

11 Research Topic B Main Focus: Integration of Topic A results into application workflows for essential calibration tasks: B1: Battery Managment System Calibration B2: Gasoline & Diesel engine calibration workflow B3: Cold start calibration Workshop Model based calibration methodologies Begrüßung, Einleitung 11/29

12 Interconnection of research topics A & B Research Topic A Research Task A1 Experiment design Research Task A2 Nonlinear system identification Research Task A3 Controller stability analysis & design Research Task B1 Battery Managment System Calibration Research Topic B Research Task B2 Gasoline & Diesel engine calibration workflow Research Task B3 Cold start calibration Workshop Model based calibration methodologies Begrüßung, Einleitung 12/29

13 Research Task A1: Experiment design State of the art: Experiment design for static models - Analytic representation of the area of operation (direct & indirect methods) - Mixture criterion for candidate selection (A,S,D-optimality) λ [ ] load [Nm] 5000 speed [rpm] Workshop Model based calibration methodologies Begrüßung, Einleitung 13/29

14 Research Task A1: Experiment design Research programme: Experiment design for dynamic models - Determination of the dynamic area of operation (test based, model based) - Analytic representation of the dynamic area of operation (local driveable areas) - Creation of a dynamic experiment design (time space, distribution) 1 u1 u time [samples] time [samples] u u time [samples] Workshop Model based calibration methodologies Begrüßung, Einleitung 14/29

15 Model Based DoE Optimal Design of Experiments based on a process model offline online existing model initial model based DoE train/adapt model online DoE Motivation DoE subsequent to component changes DoE for changed environmental conditions Use a model of a comparable process (e.g. a similar engine) Use a model from online training to optimize future process inputs (online DoE) Adherence to process constraints (inputs & outputs) Workshop Model based calibration methodologies Begrüßung, Einleitung 15/29

16 Model Based DoE Optimality Criteria The Fisher Information Matrix I is used as a measure: 1 1 I( ψ) = (k n max) σ ψ T ψ 2 Example for model structure: Dynamic multilayer perceptron network: u u(t 1) q 1 u(t 2) q 2 : y(t 1) q 1 y(t 2) q 2 1 ω11 ωjk ωnhnϕ ω10 ω12 ωj0 f1 : : fj : W11 W1j 1 W1n h W10 ŷ(t) fn h Workshop Model based calibration methodologies Begrüßung, Einleitung 16/29

17 Research Task A2: Nonlinear System Identification State of the art: Nonlinear dynamic system identification - Neural networks (black box), physical models (white box) - Grey-Box: Incorporation of qualitative physical knowledge - Advanced parameter estimation methods (GTLS) 30 injection mass [mg] speed [rpm] Workshop Model based calibration methodologies Begrüßung, Einleitung 17/29

18 Research Task A2: Nonlinear System Identification Research programme: Nonlinear dynamic system identification - Workflow of system identification for non experts - User-defined performance criteria (relative error, data tansformation) - Operating regime based models (user-defined pre-partitioning) - Constraints (= incorporation of quantitative knowledge) - Model adaptation and extension (new operational conditions) Inputs N, Q Main, r EGR, r VNT P Rail, phi MI dynamic local dynamic local dynamic dynamic dynamic static static static static static Outputs NO x,soot, HC m fuel, m Air p 2, p 3, T 2, T 3 Md, n Turbo Data Model NOx [ppm] time [samples] Workshop Model based calibration methodologies Begrüßung, Einleitung 18/29

19 Operating Regime Model Modelling strategy local nonlinear model κ j Local nonlinear models Embedded in the operating regime model User-defined pre-partitioning Split into segments representing the main physical effects (using load and speed) load speed Workshop Model based calibration methodologies Begrüßung, Einleitung 19/29

20 Operating Regime Model Training of the local nonlinear models One local nonlinear model in each operating regime Mean squared error (MSE) at the training data for each local model (exemplarily for P max) Stem plot illustrates locally different complexity: Highly nonlinear behaviour of the engine MSE varies in a wide range which reflects locally different noise levels and input sensitivities x Q 10 MSE N Workshop Model based calibration methodologies Begrüßung, Einleitung 20/29

21 Research Task A3: Controller Stability Analysis and Design State of the art: - Gain scheduled PID control - Gain scheduled observation/estimation (e.g. SoC) - Manual tuning with simulation models LM 1 LM 1 LC 1 LM 2 LM M Linearised model Controller design Operating point dependent LM 2 LM M LC 2 LC M Controller design Operating point dependent a) Standard b) Controller network Workshop Model based calibration methodologies Begrüßung, Einleitung 21/29

22 Research Task A3: Controller Stability Analysis and Design Research programme: Controller stability analysis and design - Open loop stability analysis of dynamic model networks: Lyapunov stability theory - Closed loop stability conditions (local controller network) - Support for design of controllers/estimators - Effect of model parameter uncertainties on controller design (robustness) 2 LM1 LM2 LM3 LM4 LM5 ẏ(t) y(t) Workshop Model based calibration methodologies Begrüßung, Einleitung 22/29

23 Global Stability Motivational Example Stable and unstable combinations of two stable local models Two stable State-Space models: x(k + 1) = A i x(k) i = 1, 2 Global stability of the LMN depends on the partitioning! Partitioning along x 1 direction: Partitioning along x 2 direction: x2 0.5 x x x1 Workshop Model based calibration methodologies Begrüßung, Einleitung 23/29

24 Direct Lyapunov Methods Three different approaches Common Quadratic Lyapunov Function Piecewise Quadratic Lyapunov Function Fuzzy Lyapunov Function Workshop Model based calibration methodologies Begrüßung, Einleitung 24/29

25 Physical Engine Models Reduction by Galerkin Projection State of the art and challenges: - Various physical engine models are available - High complexity often prevents analytic insight - Model based controller design requires reduced order models - Realtime execution can be achieved with surrogate models Research Programme: - Flat Galerkin projection methods - Sub-Manifold is obtained from snapshot approach - Nonlinear Galerkin methods - Applicable to both ODEs and PDEs Quelle: Mercedes-Benz engineering center steyr Workshop Model based calibration methodologies Begrüßung, Einleitung 25/29

26 Research projects within Topic B Research task B1: Battery Management System Calibration - Model architecture for battery/cell models - Experiment design for battery test system - Related SoC estimation techniques - Battery models used in battery simulator or hybrid vehicle simulation Main Challenges - Hysteresis - Cell ageing - Proper DoE - Observer Robustness Workshop Model based calibration methodologies Begrüßung, Einleitung 26/29

27 Research projects within Topic B Research task B2: Gasoline & Diesel engine calibration workflow - Integrated workflow: DoE, identification, optimisation/calibration - Dynamic models for torque, fuel consumption, NO x, HC, CO, boost pressure, air mass flow - Split of the overall system behaviour into static and dynamic submodels - Integration of physical submodels Inputs N, Q Main, r EGR, r VNT P Rail, phi MI dynamic local dynamic local dynamic dynamic dynamic static static static static static Outputs NO x,soot, HC m fuel, m Air p 2, p 3, T 2, T 3 Md, n Turbo Workshop Model based calibration methodologies Begrüßung, Einleitung 27/29

28 Research projects within Topic B Research task B3: Cold start calibration Iterative procedure comprising - Cold start experiment: A single experiment with a fixed parameter setting - Model adaptation: Adjustments of the process model based on experiment data - ECU parameter adjustments: Optimisation of engine start reliability, engine speed overshoot, HC emissions and catalyst temperature Throttle Injection Air Path Fuel Path Cylinder Ignition Air-Fuel-Mixture & Combustion Emissions Indicated Torque Motored Torque Speed Motion Coolant Load Temperatures Workshop Model based calibration methodologies Begrüßung, Einleitung 28/29

29 Summary Christian Doppler Labor TU-Wien & AVL-List GmbH: Model based calibration methodologies Research Topic A: Basic research Experiment design Nonlinear system identification Controller analysis and design Topic A Research Topic B: Application oriented research Battery Managment System Calibration Controller calibration workflow Cold start and warm up Topic B Return to TOC WS Programme Workshop Model based calibration methodologies Begrüßung, Einleitung 29/29