Multi-objective design optimization of a CO2 gascooler for electric vehicle applications under all-climate conditions via the machine learning method

초록

The adoption of CO2 as a refrigerant is rapidly increasing due to stricter environmental regulations, which has led to a growing interest in CO2-based heat pump systems for electric vehicles. Improving the efficiency of these systems requires optimizing the design of heat exchangers, particularly the gascooler, which plays a crucial role in heat rejection and absorption. In this study, a machine learning-based optimization framework was developed to improve the geometric design of a CO2 gascooler while reducing its pressure drop, weight, and volume. Experimental data and correlation-based calculations were used to construct a surrogate model that predicts the thermal and hydraulic performance of the gascooler. The surrogate model was trained using a neural network and coupled with a multi-objective genetic algorithm to optimize heat transfer rate, pressure drop, and mass simultaneously. A SHAP-based sensitivity analysis was also performed to interpret the contribution of each geometric parameter to the predicted performance and to identify the dominant design variables. Unlike previous studies that focused on a single operating mode, this work considered three representative operating conditions corresponding to summer, winter, and transitional climates. The optimized gascooler achieved up to 150 % increase in heat transfer performance, 32.7 % reduction in pressure drop, 59.0 % reduction in weight, and 45 % reduction in volume compared with baseline. System-level evaluation further confirmed that the optimized designs improved performance under both heating and cooling operations, demonstrating that all-climate optimization effectively enhances the overall performance of CO2 heat pump systems for electric vehicles.

키워드