Multiobjective model-based optimization of diesel injection rate profile by machine learning methods

The contribution of this article is to present a model-based machine learning methodology for automatic and simultaneous optimization of the power output and exhaust emissions of diesel internal combustion (IC) engines. We carry out parametric optimization of the rate profile at which fuel is injected into the cylinder for producing minimal nitrogen oxide (NOx) emissions and maximal cylinder power (nIMEP) output, on a computational simulation model of an Agco Power 44 AWI engine calibrated by measurements. Our results display the tradeoffs in reaching these two contradictory optimization objectives on the Pareto frontiers. We show that the so-called boot injection profile, which is commonly used in practice, also emerges through mathematical optimization as a reasonable compromise of the objectives.