Machine learning prediction study on collision failure risk of lithium-ion battery packs considering cyclic aging

Lithium-ion batteries in new energy vehicles are prone to safety accidents during collisions. However, the performance uncertainty of battery packs caused by cyclic aging makes it difficult to accurately and rapidly predict collision failure risks. This study focuses on the cyclic aging state of batteries to conduct rapid prediction research on failure risks: quasi-static compression tests of battery cells with different cycle counts are performed to analyze the effect of aging on failure modes; a neural network rapid prediction model for failure risks is established based on test and RVE simulation results, with its reliability and robustness verified; finally, the failure risks of battery cells in battery packs during collisions are predicted and analyzed using the neural network. The results show that cyclic aging advances the failure of lithium-ion batteries and significantly reduces failure displacement; the machine learning model built by integrating test and RVE simulation data achieves R 2 > 0 . 99 ; the feasibility of the method is validated through risk predictions under X / Y -direction squeeze plate collision and underbody conditions of battery packs.