In sustainable mobility, electric powertrains are crucial, with batteries holding utmost importance. Manufacturers compete based on time-to-market for innovative technologies. Battery analytics demand advanced modeling for aging models and a scalable environment for processing data from large vehicle fleets in real-time. This paper focuses on methodologies for enhancing State of Health (SoH) and automatically detecting abnormal behavior in battery components. Combining a SoH observer and data-driven module enables a more accurate health estimation. While machine learning predicts degradation trajectories and system anomalies thus addressing limitations of onboard systems and physics-based approaches..

Existing diagnostics rely on onboard system IO from physics-based approaches, facing computational challenges and potential deviation from real operating conditions. Our solution introduces a cloud-based machine learning SoH prediction model trained with fleet data, using service shop aging data as a truth source. Interpretable features enhance accuracy, validated across diverse vehicles and battery types. A confidence metric quantifies uncertainty, aiding experts in decision-making, and outperforming existing model-based methods in both error performance and efficiency.

In addition to age monitoring, severe battery issues can arise during vehicle product life. Data analytics presents a methodology for monitoring anomalies and predicting failures. A supervised machine learning model, trained on fleet data, predicts failures, and assigns risk scores, allowing initiative-taking measures and reducing warranty costs for OEMs. This AVL methodology enhances battery system safety by identifying critical conditions prior to a thermal incident, while also allowing for selective and preventive measures to be taken, resulting in significant cost reduction.