Suchergebnisse

International audience ; Nowadays, automotive manufacturers are submitted to strong constraints in engine calibration: lowest fuel consumption, emission-control legislation and driver requests for driving comfort and performances. These constraints lead to an increasingly complexity of the engines and thus an increasingly number of parameters to be tuned, making the empirical engine calibration by a scan of parameter values impossible at engine test-bench. New methodologies in automated engine calibration based on statistics and optimization have emerged in order to limit the number of experimental tests to be run. The optimization problem of engine calibration consists in the determination of engine tuning parameter maps that minimize the cumulated fuel consumption and pollutant emissions, under combustion noise constraints, on a driving cycle. The usual way to get this result is to select specific operating points representing this cycle in the engine working range and to define upper bounds applied on the different engine responses (allocations) for each of them, in order to obtain a weighted sum of these local responses respecting the global targets. The underlying problem is a multi-objective optimization problem: different compromises between fuel consumption, noise and pollutant emissions on each operating point are possible. We propose an adapted optimization method based on the MO-CMA-ES method (Multi-objective Covariance Adaptation Evolution Strategy) which takes into account the non trivial limits of the engine parameter variation domains and some robustness constraints. An other point addressed in this paper is the map optimization which consists in a global optimization of engine responses cumulated on the driving cycle. This method avoids the cumbersome choice of allocations for each considered operating point and includes directly the map regularity constraints in map parameterizations. Finally, application on real dataset obtained at automated test-bench for a diesel engine are presented .

International audience ; Nowadays, automotive manufacturers are submitted to strong constraints in engine calibration: lowest fuel consumption, emission-control legislation and driver requests for driving comfort and performances. These constraints lead to an increasingly complexity of the engines and thus an increasingly number of parameters to be tuned, making the empirical engine calibration by a scan of parameter values impossible at engine test-bench. New methodologies in automated engine calibration based on statistics and optimization have emerged in order to limit the number of experimental tests to be run. The optimization problem of engine calibration consists in the determination of engine tuning parameter maps that minimize the cumulated fuel consumption and pollutant emissions, under combustion noise constraints, on a driving cycle. The usual way to get this result is to select specific operating points representing this cycle in the engine working range and to define upper bounds applied on the different engine responses (allocations) for each of them, in order to obtain a weighted sum of these local responses respecting the global targets. The underlying problem is a multi-objective optimization problem: different compromises between fuel consumption, noise and pollutant emissions on each operating point are possible. We propose an adapted optimization method based on the MO-CMA-ES method (Multi-objective Covariance Adaptation Evolution Strategy) which takes into account the non trivial limits of the engine parameter variation domains and some robustness constraints. An other point addressed in this paper is the map optimization which consists in a global optimization of engine responses cumulated on the driving cycle. This method avoids the cumbersome choice of allocations for each considered operating point and includes directly the map regularity constraints in map parameterizations. Finally, application on real dataset obtained at automated test-bench for a diesel engine are presented .

International audience ; Nowadays, automotive manufacturers are submitted to strong constraints in engine calibration such as: low fuel consumption, emission-control legislation and driver requests for driving comfort and performances. These constraints lead to an increasing complexity of the engines and thus an increasing number of parameters to be tuned, making the empirical engine calibration by a scan of parameter values impossible at engine test-bench. New methodologies in automated engine calibration based on statistics and optimization have emerged in order to limit the number of experimental tests to be run. The optimization problem of engine calibration consists in the determination of engine tuning parameters that minimize the cumulated fuel consumption and pollutant emissions on a driving cycle generally associated with legislation norms. This cycle is decomposed in a set of stationary operating points of the engine characterized by its speed and its torque (the transient behaviors of the engine are not taken into account in the stabilized calibration). Then, the optimal tuning parameters of the engine should be defined for each operating points, the functions defining these parameters on the whole engine operating domain are called the engine maps. These two-dimensional optimal engine maps are then integrated in the engine control unit in the vehicle. We illustrate the difficulties associated with this application and propose adapted optimization methodologies: LoLiMoT models for engine map parameterization in order to handle intrinsic constraints on the map regularity, multi-objective optimization method based on CMA-ES approach. Finally , application on real dataset obtained at IFP automated test-bench for a diesel engine are presented. 2. Keywords: Engine calibration, LoLiMoT, Multi-objective optimization, Evolutionary algorithm 3. Introduction Engine calibration consists in fulfilling the engine tuning maps that are used in engine controls of the vehicle, i.e. in defining the optimal tuning of ...

International audience ; Nowadays, automotive manufacturers are submitted to strong constraints in engine calibration such as: low fuel consumption, emission-control legislation and driver requests for driving comfort and performances. These constraints lead to an increasing complexity of the engines and thus an increasing number of parameters to be tuned, making the empirical engine calibration by a scan of parameter values impossible at engine test-bench. New methodologies in automated engine calibration based on statistics and optimization have emerged in order to limit the number of experimental tests to be run. The optimization problem of engine calibration consists in the determination of engine tuning parameters that minimize the cumulated fuel consumption and pollutant emissions on a driving cycle generally associated with legislation norms. This cycle is decomposed in a set of stationary operating points of the engine characterized by its speed and its torque (the transient behaviors of the engine are not taken into account in the stabilized calibration). Then, the optimal tuning parameters of the engine should be defined for each operating points, the functions defining these parameters on the whole engine operating domain are called the engine maps. These two-dimensional optimal engine maps are then integrated in the engine control unit in the vehicle. We illustrate the difficulties associated with this application and propose adapted optimization methodologies: LoLiMoT models for engine map parameterization in order to handle intrinsic constraints on the map regularity, multi-objective optimization method based on CMA-ES approach. Finally , application on real dataset obtained at IFP automated test-bench for a diesel engine are presented. 2. Keywords: Engine calibration, LoLiMoT, Multi-objective optimization, Evolutionary algorithm 3. Introduction Engine calibration consists in fulfilling the engine tuning maps that are used in engine controls of the vehicle, i.e. in defining the optimal tuning of parameters used by engine control strategies. Due to the highly increased number of these parameters (especially for diesel engines but spark ignition engines are following the same trend) and the reduction of the development schedule available for the calibration process, manual tuning of engine parameters is now replaced by mathematically assisted calibration process. Such a process is based on the design of experiments with associated modeling methods, in order to reduce the number of tests used to build engine response models depending on engine control parameters, and optimization techniques to determine the optimal settings within the model definition domain. In order to perform the tests in a more productive way, these mathematical techniques are generally associated with test automation, requiring well controlled measurement methods and reliable test equipments. This paper describes the optimization methods developed for this application and illustrates their effectiveness on a real case of a common rail diesel Engine. The first section introduces the classical steps of the calibration process and discusses the associated difficulties. In the second section, we propose the Multi-Objective Covariance-Adaptation Evolutionary Strategy method for solving the optimization problem associated with a given engine operating point defined by the engine speed and the engine load. In the third part, an integrated approach is proposed in order to directly optimize the engine maps on the whole driving cycle (associated with legislation norms) instead of the individual optimization of each engine operating point. 4. Engine calibration 4.1. Sketch of the engine calibration process The emission calibration workflow is classically divided into four steps: 1. a preliminary phase consisting in choosing a sample of operating points (referred to as OP in the 1

International audience ; Nowadays, automotive manufacturers have to deal with strong constraints: lowest fuel consumption, emission-control legislation and driver requests for driveability and performance. For this purpose, the classical engine has evolved towards a very complex system combining many hi-tech components with advanced control strategies. It leads to an increasingly complexity of the underlying optimization issue. The classical method to perform this optimization is to use statistical response surface models obtained from optimally designed experiments at test bench in steady-state conditions. In this approach, the transient effects observed during the driving cycle are not taken into account. A promising idea to include these transient effects in the optimization process is to couple physical modelling of the engine (to get the conditions in the combustion chamber before each combustion event) with statistical models (to obtain engine-out emissions after this combustion event). Powertrain system simulation is shown to provide a helpful tool to reach this objective thanks to a good compromise between accuracy and low CPU time. A dedicated parameterization of the engine maps and a Derivative Free Optimization method is then used to minimize the cumulative fuel consumption and pollutant emissions, under combustion noise constraints, on a driving cycle. An application on a real dataset obtained at automated test-bench for a diesel engine is presented.

International audience ; Nowadays, automotive manufacturers have to deal with strong constraints: lowest fuel consumption, emission-control legislation and driver requests for driveability and performance. For this purpose, the classical engine has evolved towards a very complex system combining many hi-tech components with advanced control strategies. It leads to an increasingly complexity of the underlying optimization issue. The classical method to perform this optimization is to use statistical response surface models obtained from optimally designed experiments at test bench in steady-state conditions. In this approach, the transient effects observed during the driving cycle are not taken into account. A promising idea to include these transient effects in the optimization process is to couple physical modelling of the engine (to get the conditions in the combustion chamber before each combustion event) with statistical models (to obtain engine-out emissions after this combustion event). Powertrain system simulation is shown to provide a helpful tool to reach this objective thanks to a good compromise between accuracy and low CPU time. A dedicated parameterization of the engine maps and a Derivative Free Optimization method is then used to minimize the cumulative fuel consumption and pollutant emissions, under combustion noise constraints, on a driving cycle. An application on a real dataset obtained at automated test-bench for a diesel engine is presented.