Chemomechanical damage prediction from phase-field simulation video sequences using a deep-learning-based methodology

Understanding the failure mechanisms of lithium-ion batteries is essential for their greater adoption in diverse formats. Operando X-ray and electron microscopy enable the evaluation of concentration, phase, and stress heterogeneities in electrode architectures. Phase-field models are commonly used to capture multi-physics coupling including the interplay between electrochemistry and mechanics. However, very little has been explored regarding developing predictive models that would forecast imminent failure. This study explores the application of convolutional long short-term memory networks for damage prediction in cathode materials using video sequence from phase-field simulations as a proxy for video microscopy. Two models are examined making use of, respectively, the damage video only and the damage and hydrostatic stress videos combined. We use customized quantitative metrics to compare the performance of the models. Our work demonstrates the outstanding capability of deep learning models using limited data to predict fracture behavior of battery materials, including crack propagation angle and length.