Application of artificial neural networks in engine modelling

Abstract

Abstract

A study was conducted to develop an accurate simulation tool with a small computer resource footprint for engine design. The modelling approach uses artificial neural networks (ANNs) based on multilayer perceptrons (MLPs). The ANN is used to represent engine in-cylinder processes by training the ANN to approximate computational fluid dynamics (CFD) simulation results of the engine. The ANN approach was applied to model a turbo-charged direct injection (DI) diesel engine over a wide range of operating conditions. Seven primary diesel engine control parameters were varied over their possible ranges: engine speed, engine load, start of injection, injection pressure, mass in the first injection pulse of a split injection, boost pressure and exhaust gas recirculation (EGR). The model output includes five quantities: cylinder pressure, cylinder temperature, cylinder wall heat transfer, NO_X emission and soot emission (the elemental carbon fraction of particulate matters). The cylinder pressure, temperature and heat transfer are crank angle resolved, while the emissions are cycle resolved. In total 71 engine steady state operating conditions were simulated with CFD, and five MLPs were trained individually to approximate the five engine output parameters as a function of the seven engine control parameters. The testing results showed that the five trained MLPs achieved satisfactory capabilities of predicting engine responses and representing the characteristics of the engine over a wide range of operating conditions. Additionally, the ANN modelling accuracy was improved by incorporating prior knowledge into the ANN design and using a committee of networks instead of the best single network to make predictions.