Ph.D. received on: 8/7/1997

E-mail: carlo.siviero@bologna.marelli.it

Tutor: Prof. Riccardo Scattolini, Università degli Studi di Pavia
 
 

 ___________________________________________________________________________________________________________

Identification and Control of Internal Combustion Engine   ___________________________________________________________________________________________________________

Advisor:

Prof. Riccardo Scattolini, Università degli Studi di Pavia

Summary:

The levels of pollutant emissions in internal combustion engines, recently imposed by the European Community to car makers, have commanded a new and scientific approach to the title (air/fuel relative ratio) closed loop control and to the idle speed control.  The goal of the first one is the control of the air/fuel ratio in the mixture entering the cylinders: the catalytic converter, the most used component in the automotive industry for emissions limitation, can guarantee an acceptable pollutant reduction only if the air/fuel ratio is kept on a small range around a given value related to the hydrocarbons properties, the so called “stoichiometric ratio” (approximately one part of fuel mass for 15 parts of air mass). The idle control main task is to keep constant at the desired value the engine crank shaft speed, also when a disturbance torque is requested by the user for additional load (as the air-conditioner power) and in different atmospheric pressure or air temperature conditions. Looking for a scientific and modern approach to these problems, the so called mean Value Models (MVM) have been analysed and implemented, between the different techniques presented by the literature devoted to the engine control. The MVM are able to predict the mean value in the engine cycle of the variables related to the engine control task (mainly the air flow through the throttle valve, the inlet manifold air pressure, the relative air/fuel ratio and the engine crank shaft speed). They are formed by a restrict number of non-linear continuous time differential equations which describe the more relevant engine dynamics (as the inlet manifold pressure dynamics, the fuel dynamics at the inlet wall, the engine speed dynamics and combustion). They are build from balance and energy conservation equations, and the unknown parameters (which take into account the thermal, friction and air pumping dynamics) are fitted by using a “grey box” approach or by using a “black box” identification technique of non-linear static models. Between them, the polynomial models, the additive models with cubic smoothing splines and the neural net models have been tested on real engine data, and their properties have been stressed, mainly in term of model predictive performance and algorithm computational cost.
The greatest limitations of the MVM are their complexity in the combustion and engine speed sub-model representation, and the fact that expensive tests are needed for their identification. A different approach is given by the NARX (Non-linear AutoRegressive with eXogenous inputs Models) where the model is fitted starting from sequences of the input-output engine variables, with for instance the “step-wise” regression technique. The two different approaches (MVM models and NARX models) have also been used on data related to three different engines. The thesis documents the good performances of both the models, stressing their relative advantages and disadvantages. The obtained models have also been used in an engine speed control problem at idle, for the tuning of a classical Linear Quadratic Integral (LQI) controller, and in the air fuel/ratio control and injection components diagnosis problem, by means of non-linear “sliding mode” observers with unknown inputs. These two problems give different examples of the possible use of such techniques in the automotive industry, and this justifies the growing interest of the car makers in the model-based control design and tuning.
 

 _______________________________________