Physical-based surrogate model and its application in 3D-1D fusion optimization of a marine two-stroke engine

Abstract

Marine two-stroke low-speed engines face challenges of complex experimentation and high computational costs, making 1D-3D integrated simulation optimization using surrogate models combined with optimization algorithms a potential solution. However, existing surrogate models suffer from issues such as high data dependency and low interpretability. To address this, this study develops simplified physical-based surrogate models, including a scavenging multi-stage model, a scavenging swirl model, and a swirl mixing controlled combustion (SMCC) surrogate model, calibrated via genetic algorithm (GA). Validation against experimental data shows errors within 5%. These models are coupled to enable full-cycle performance simulation. Subsequent multi-objective optimization using Non-dominated sorting genetic algorithm II (NSGA-II) reduces maximum pressure (P_max) and maximum pressure rise rate (dP_max) by 3.23% and 2.06%, respectively, while improving Brake Thermal Efficiency (BTE) by 1.1%. The optimized surrogate model maintains consistency with CFD results, with errors within 5%, but requires significantly less computational time than traditional CFD-based optimization. Therefore, physical-based simplified surrogate models integrated with optimization algorithms provide an effective approach for system-level simulation and multi-objective optimization of marine two-stroke low-speed engines.