Multi-objective Optimization of the Front-End Structure of an Automobile Body Based on Active-Semi-Supervised Learning

The application of machine learning techniques in the field of optimal design of the front-end structure of an automobile body is facing a major challenge: the performance of traditional machine learning methods is largely limited by the number and quality of data samples. Due to the high complexity and expensive computational cost of body crash simulation models, the limitation of sample size may negatively affect the accuracy of machine learning models and the reliability of optimization results. To address this issue, a novel algorithm based on active-semi-supervised learning is proposed in this study. The unique feature of this algorithm is that it utilizes a modified TS-SSL semi-supervised learning method to generate virtual samples, thereby expanding the sample set and improving the prediction accuracy of the surrogate model. Besides, this algorithm applies an active learning strategy during the optimization process to locate the optimal design at a finer level of detail. The advantages of this algorithm are demonstrated in the optimal design of the front-end structure of an automobile body. Firstly, the algorithm significantly reduces the reliance on complex and time-consuming numerical simulation models, thus greatly improving the efficiency of the optimization process. Secondly, by using an active learning strategy, it can more accurately find a front-end structure design that meets the multi-objective optimization requirements. In a practical application, we performed a multi-objective optimization of the crash performance of the front-end structure of an automobile body. Through parametric modeling, a parametric model was automatically generated and the associated response results were derived. Compared to the base model, we found that the energy absorption ratio remained stable despite a 2% mass reduction of the front-end structure of an automobile body, while the peak acceleration was reduced by 6.5% and the maximum compression was reduced by 14.4%. These results validate the significant superiority of our proposed algorithm in improving the safety performance and optimizing the design efficiency of the front-end structure of an automobile body.