A methodology for engine design using multi-dimensional modelling and genetic algorithms with validation through experiments

Abstract

Abstract

A methodology for internal combustion engine design has been formulated which incorporates multidimensional modelling and experiments to optimize and simulate direct injection diesel engine combustion and emissions formation. The computer code KIVA-GA performs full-cycle engine simulations within the framework of a genetic algorithm (GA) global optimization code. The methodology is applied to optimize a heavy-duty diesel truck engine. The study simultaneously investigated the effects of six engine input parameters on emissions and performance for a high-speed medium-load operating point. The start of injection (SOI), injection pressure, amount of exhaust gas recirculation (EGR), boost pressure and split injection rate shape were optimized. The convergence of the GA optimization process is demonstrated and the results were compared to those of the experimental optimization study employing a response surface method (RSM), which uses statistically designed experiments to determine an optimum design. In addition, the parameters of the computationally predicted optimum were run experimentally and good agreement was obtained.

The potential for ultra-low emissions levels was assessed through additional computational GA runs that included higher maximum EGR levels (up to 50 per cent). The predicted optimum results in significantly lower soot and NO_x emissions together with improved fuel consumption compared to the baseline design. The present results indicate that an efficient design methodology has been developed for optimization of internal combustion engines, one that allows simultaneous optimization of a large number of parameters.