Application of machine learning to optimize the combustion characteristics of RCCI engine over wide load range

Abstract

The objective of this study is to investigate, optimize, and compare the combustion features and intake variables affecting the kinetics of gasoline/diesel RCCI engine by applying the genetic algorithm. Six decision variables include the fuel flow rate (FFR), premixed fuel rate (PFR), exhaust gas recirculation (EGR) rate, first injection mass ratio (IMR1), and the first and the second start of injection (SOI1 and SOI2). The objective of the optimization was to simultaneously maximize the engine gross indicated efficiency (GIE), the second law efficiency (SLE), and minimize exhaust emissions and the maximum pressure rise rate. The optimal space filling design of experiment method was applied to evaluate the effect of the decision variables on the objective functions with a minimum number of numerical experiments. The relationships between the decision variables and objective functions were approximated by the genetic aggregation response surface model. Finally, the non-dominated sorting genetic algorithm II was employed to find the optimum values. In the optimal case GIE, SLE, and Fmerit increased by 1.56%, 1.25%, and 15.14% while NOx concentration and engine noise reduced by 16.67% and 33.45%, respectively. The best range of SOI2 for GIE is between –22 and −14 degrees ATDC, while for improving SLE, SOI2 should be advance. Sensitivity analysis shows that PFR and FFR are the most influential variables to moderate the combustion kinetics and increasing IMR1 has the proper effect on combustion by creating more opportunities for a homogeneous charge and reducing hot spots.