Application of multi-task regression in multi-objective optimization for automotive crashworthiness and lightweight

The aim of this article is to discuss the application of multi-task regression (MTR) in multi-objective optimization of automobile crashworthiness and lightweight. The applicability of MTR in such tasks is assessed by comparing the fitting accuracy of single-task regression (STR) and MTR. Two typical cases were selected for the study: an automotive front-end structure (frontal rigid wall crash) and a battery pack system (side rigid column crush). A surrogate is constructed using a multi-layer perceptron (MLP), and comprehensive correlation coefficient is used for feature screening and Bayesian optimization is used for hyper-parameter optimization. This study finds that MTR has higher prediction accuracy for objectives with poor STR performance when the STR surrogate’s performance is disparate between objectives. In the three-objective case, combining with two better-performing objectives brings more significant improvement than combining with a single objective. Ultimately, multi-task regression-based multi-objective optimization is solved using the U-NSGA-III algorithm and a compromise solution is selected using the Entropy-Gray method. It is shown that MTR has the potential to improve the efficiency and effectiveness of optimization in crashworthiness and lightweight multi-objective optimization tasks.