Matthias Schüler

Fast neural networks for diesel engine control design

Abstract Advanced engine control systems require accurate dynamic models of the combustion process, which are substantially nonlinear. This contribution presents the application of fast neural net models for engine control design purposes. After briefly introducing a special local linear radial basis function network (LOLIMOT) the process of building adequate dynamic engine models is discussed in detail. These neuro-models are then integrated into an upper-level emission optimization tool which calculates a cost function for exhaust versus consumption/torque and determines optimal engine settings. A DSP-based process computer system allows a fast application of the optimization tool at the engine test stand.

Einsatz schneller neuronaler Netze zur modellbasierten Optimierung von Verbrennungsmotoren

Aufgrund der steigenden Anzahl von Stellgrosen und zu optimierenden Ausgangsgrosen an modernen Verbrennungsmotoren werden modellbasierte Verfahren zur Optimierung des Motorverhaltens immer wichtiger. Der erste Teil dieses zweiteiligen Beitrages enthalt eine Einfuhrung in spezielle neuronale Netze,geht auf geeignete Prufstandsmessstrategien ein und stellt eine Modellbildung des Motor- und Abgasverhaltens mit Hilfe schneller neuronaler Netze vor. Mit Hilfe der Modelle kann dann offline eine Applikation der Motorsteuergeratefunktionen durchgefuhrt werden,ohne Prufstandszeit zu beanspruchen. Die Untersuchungen wurden im Rahmen des DFG-Sonderforschungsbereichs „Integrierte mechanisch-elektronische Systeme fur den Maschinenbau“ am Institut fur Automatisierungstechnik der TU Darmstadt durchgefuhrt.

Fast Neural Networks for Diesel Engine Control Design

Abstract Advanced engine control systems require accurate process models. This paper presents neural net models for combustion engines. After briefly introducing a special local linear RBF network (LOLIMOT) two applications are described. Different methods for developing exhaust gas models are compared and a dynamic model for the charging pressure dynamics of a turbocharger is presented. Finally, an exhaust vs. consumption optimization is presented for optimizing the injection angle dependent on given weighting factors for specific emissions, the fuel consumtion and the current driving situation.

Dynamical identification and control of combustion engine exhaust

This paper presents a new approach for model based control of combustion engine exhaust. Fast neural networks of the LOLIMOT-type are used to dynamically simulate different emissions from diesel engines. Neuro-models for the exhaust gases and the fuel consumption are integrated into an upper-level optimization tool. The tool calculates the cost function for exhaust vs. consumption and determines an optimal injection angle dependent on the engines exhaust performance, its fuel consumption and the current driving situation.

A FUZZY-SYSTEM FOR A CLASSIFICATION OF THE DRIVER BEHAVIOR AND THE DRIVING SITUATION

Abstract A fuzzy system is presented which recognizes and classifies the driver behavior between economic and sportive. The aim of this approach is to supply an engine management system with information about the driver demands. Based on measured driving signal data, features are calculated and then evaluated with a fuzzy classificator initialized with a priori heuristic knowledge. This first fuzzy system could be improved by nonlinear optimization of the position and shape of the membership functions of the premises and the position of the conclusion-singletons. A further examination of the information contents of each signal lead to a reduced, simplified structure with fewer input signals. Finally the first result of an initial fuzzy-system for the recognition of the driving situation (city, highway and road) are presented.

Model-Based and Signal-Based Fault Detection of Diesel Engines with Turbocharger

Abstract Rising demands in automotive development and strict emission standards enforce the application of modern conrrol and supervision strategies to combustion engines. This contribution shows the shaping and adaption of model based fault detection and direct signal analysis methods when applied to a turbocharged diesel engine. First a real lime supervision of fuel mass and injection angle based on dynamic cylinder pressure measurement is described. This is followed by a method for engine misfire detection using only a low resolution crankshaft speed signal. Then fault detection for a diesel engine turbocharger with nonlinear neural networks is proposed. Finally the results of a diagnosis of multiple faults with a neural network are presented. All methods have been implemented and tested experimentally on a dynamical engine test stand at the Technical University of Darmstadt.