AI-enabled thermal monitoring of commercial (PHEV) Li-ion pouch cells with Feature-Adapted Unsupervised Anomaly Detection

Distributed temperature profiling of lithium-ion batteries provides valuable insights, aiding thermal management and minimising risk of battery failures. Highlighted by Batteries Europe as crucial for battery safety, advances in thermal monitoring are imperative to continuous safe adoption of battery technology. Deep Learning techniques have recently emerged as powerful tools for anomaly detection (AD) in many thermal mapping applications. These data-driven methods can handle common challenges like data unavailability or environment variations. Our study devises a methodology to leverage Deep Learning with thermal data from commercially available pouch cells and an infrared camera. We explain the building blocks of FAUAD (Feature-Adapted Unsupervised Anomaly Detection), which models the normality of the input data and synthesizes anomalies in its feature space. The resulting model is benchmarked against some of the latest state-of-the-art methods and achieves high anomaly detection capability; Area Under the ROC Curve (AUROC) score of 0.971 on simulated data, 0.990 on contaminated real data, and a perfect score of 1.0 on real clean data. While maintaining a compact size of 15 MB. FAUAD offers a notable advancement in unsupervised anomaly detection for battery thermal monitoring. The proposed method is cell chemistry agnostic and open to usage scenarios beyond this works’ scope. • Unsupervised thermal anomaly detection. • Li-ion cell thermal profiling. • Deep learning using images as training data.