Data-Driven Safety Risk Prediction of Lithium-Ion Battery

Inevitable safety issues have pushed battery engineers to become more conservative in battery system design; however, battery-involved accidents still frequently are reported in headlines. Identifying, understanding, and predicting safety risks have become priorities to further accelerate technology and industry development. However, diverse loading scenarios, significantly varied stress-induced short circuit mechanisms, and highly coupled mechanical-electrochemical safety behaviors have remained grand challenges. Herein, the safety risk is termed as the probability of the mechanical triggering of an internal short circuit, to reflect the safety related behaviors of lithium-ion batteries. Based on a mechanical model and experimental results, a sufficient dataset is generated consisting of strain states and their corresponding safety risks, covering both cylindrical and pouch cells, various states of charges, and loading conditions. Machine-learning tools combined with the established finite element mechanical model are applied to predict the safety risks of the cells. The results achieve a high level of accuracy on the test data (the relative error of the average short circuit prediction deviation is less than 6.2%.). This work underpins the safety risk concept and highlights the promise of physics combined with data-driven modeling methodology to predict the safety behaviors of energy storage systems.

Automated summary

In response to inevitable safety issues, battery engineers have become more conservative in design, yet accidents remain common. Identifying, understanding, and predicting safety risks are top priorities for technology and industry advancement, but challenges persist in diverse loading scenarios and highly coupled mechanical-electrochemical safety behaviors.