Interpretable battery health prognostics coupling physics with machine learning

Non-destructively battery health prognostics play significant roles in ensuring safe operation, supporting optimal management and maintenance, as well as cutting expenditures and customer anxiety. This project aims at coupling the physics of battery aging mechanisms with machine learning to achieve more accurate, reliable, and interpretable health prognostics. It involves high-fidelity battery digital twin modeling based on electrochemical models coupling complex aging mechanisms, which supports physical feature extraction that reflects internal aging reactions. Machine learning will be designed to achieve aging mode prognosis via the onboard measurements, where the domain discrepancies are minimized to ensure effective prognostics under variable working conditions. Coupling the physical features with the internal states of the machine learning model, the final model is physically interpretable and domain-adaptable for more accurate and reliable battery health prognostics.