Genetic Algorithm for Dynamic Calibration of Engine's Actuators

Modern diesel engines are equipped with an increasing number of actuators set to improve human comfort and fuel consumptions while respecting the restricted emissions regulations. In spite of the great progress made in the electronic and data-processing domains, the physical-based emissions models remain time consuming and too complicated to be used in a dynamic calibrating process. Therefore, until these days, the calibration of the engine's cartographies is done manually by experimental experts on dynamic test bed, but the results are not often the best compromise in the consumption-emissions formula due to the increasing number of actuators and to the nonlinear and complex relations between the different variables involved in the combustion process. Recently, neural networks are successfully used to model dynamic multiple inputs - multiple outputs processes by learning from examples and without any additional or detailed information about the process itself.

In this paper, we fully describe the construction and applications of a nine inputs dynamic emissions' model based on neural networks. The simulations' results are in good agreement with real engine data measured on test bed. The emissions' model is conceived to be used in an upper-level dynamic optimization process based on genetic algorithm. Our goal is to present, while using the minimum number of experimental tests, a fast and practical optimization procedure capable of finding the optimal calibration values of the seven engine's actuators over the New European Driving Cycle (NEDC). The results are very promising.