Multi-Objective Optimization of Single-Cylinder Diesel Engine Gas Exchange via the Surrogate Neural Network Model

Optimization of the working process parameters is one of the main methods for improving the performance of piston engines, ensuring the growth of the engineering level and competitiveness. For high-speed engines, the gas exchange has a serious impact on the effective power and fuel consumption. So, its improvement can significantly improve engine performance. In this work, numerical simulation of gas exchange of the TMZ-450D engine was performed using a software based on the one-dimensional linearized Godunov method. The simulation results are generalized by creating a surrogate model of the “multilayer Perceptron” type, which describes the volumetric efficiency and pump losses. A multiobjective optimization of the engine gas exchange parameters (including the dimensions of the pipes and valve timing angles) was performed for increasing the volumetric efficiency and reducing fuel consumption. Three alternative configurations of parameters are proposed that provide (according to an estimate based on a one-dimensional model) a significant improvement in engine performance.