Modelling and identification of an internal combustion engine to provide simulator for control purposes

Field of ?System Identification? has become an important discipline. Identification is basically the process of developing a mathematical representation of a physical system using experimental data. The identification of nonlinear dynamical systems is a substantial part of the control science and therefore appropriate models should be developed to control nonlinear dynamic systems. The main idea that lies under the procedure is to obtain a regular and mathematically tractable model of the system of interest. Automotive internal combustion engine (ICE) control is one of the most complex control problems for control system engineers and researchers. Among all the engine control variables, the engine torque is one of the most important performance variables of an ICE and, for this reason, a technique based on optimizing the engine torque control can improve substantially the performance of the overall vehicle. There are two objective of this work. First one is to develop a steady-state model of a gasoline engine torque and brake specific fuel consumption by using neural network and genetic programming and second one is to develop an accurate and robust model of a spark ignition (SI) engine torque by using the most common nonlinear black-box parametric models namely Hammerstein model, nonlinear auto-regressive with exogenous inputs (NARX) model and neural network model that is including multilayer feedforward neural network (FFNN) model, radial basis function (RBF) neural network model and Elman type recurrent neural network model. These developed methods are implemented to an existing 1400 cc, four cylinder Fiat SI engine. The artificial neural network (ANN) is a newly developed technique among the other identification methods. There are various methods used for training of ANN. Two of them are included in this study. These are, namely, the bacpropagation method and the Levenberg-Marquardt algorithm. The different nonlinear identification approaches are used in this thesis. The neural network based model has captured the dynamics very well and the method has been found suitable for modeling the SI engine torque. However all the nonlinear identification methods identified the SI engine torque dynamics at acceptable levels of accuracy.