Data-Driven Thermal Anomaly Detection in Large Battery Packs

The early detection and tracing of anomalous operations in battery packs are critical to improving performance ensuring safety. This paper presents a data-driven approach for online anomaly that uses real-time voltage temperature data from multiple Li-ion cells. Mean-based residuals generated cell groups evaluated using Principal Component Analysis. then thresholded cumulative sum control chart detect anomalies. mild external short circuits associated with balancing detected the signals necessitate retraining after balancing. Temperature prove be critical, enabling module events within 14 min unobservable residuals. Statistical testing proposed is performed on experimental electric locomotive injected model-based has low false-positive rate accurately detects traces synthetic compared direct thresholding mean-based shows 56% faster time, 42% fewer false negatives, 60% missed anomalies while maintaining comparable rate.