A Numerical Methodology for the Multi-Objective Optimization of an Automotive DI Diesel Engine

Nowadays, an automotive DI Diesel engine is demanded to provide an adequate power output together with limit-complying NOx and soot emissions so that the development of a specific combustion concept is the result of a trade-off between conflicting objectives. In other words, the development of a low-emission DI diesel combustion concept could be mathematically represented as a multi-objective optimization problem. In recent years, genetic algorithm and CFD simulations were successfully applied to this kind of problem. However, combining GA optimization with actual CFD-3D combustion simulations can be too onerous since a large number of simulations is usually required, resulting in a high computational cost and, thus, limiting the suitability of this method for industrial processes. In order to make the optimization process less time-consuming, CFD simulations can be more conveniently used to build a training set for the learning process of an artificial neural network which, once correctly trained, can be used to forecast the engine outputs as a function of the design parameters during a GA optimization performing a so-called virtual optimization. In this paper, a numerical methodology for the multi-objective virtual optimization of the combustion inside an automotive DI Diesel engine, based on artificial neural networks combined with genetic algorithms, is presented. Copyright © 2013 SAE International.