Abstract
Keywords: Dynamic Modeling, Design of Experiments, Volterra Series, Mutual Information, LARS, Variable Selection, Pollutants Emissions
Due to the increasing number of engine control parameters, emissions optimization has increasingly presented to be a challenge for automotive engineers. The object of this optimization process is to minimize fuel consumption while fulfilling exhaust emissions regulations. In order to address this problem, black box models based on classical Design of Experiments (DoE) are frequently used. This steady state approach is time consuming in bench tests, particularly when dealing with a large number of control parameters acting upon DoE dimensions. This paper reviews this dilemma in an effort to provide a description of the global process for engine emissions modeling at PSA Peugeot Citroen. We present here a local dynamic modeling approach that is based on dynamic Design of Experiments. We, then, used measured data to identify nonlinear models. Variable selection methods are considered in order to enhance the generalization performance of our models. Next; we propose a mixture of the local models to generate a global model that applies in a large area of the (torque, rpm) plane. The mixture method is based on Support Vector Machines. Lastly, we conclude this paper with a brief perspective concerning the use of our global model to optimize consumption while complying with exhaust emissions regulation.