Battery Safety: Detecting and Predicting Anomalous Spikes in Li-ion Batteries

The safety and reliability of Lithium-ion (Li-ion) batteries remain critical concerns for Electric Vehicles (EV). Thermal stress can cause abnormal spikes in battery behaviour, and employing detection methods is crucial to ensure safe operation. We propose a method that combines Isolation Forest anomaly detection with Median Absolute Deviation (MAD) based spike categorisation to distinguish between normal and abnormal deviations. Using an Arrhenius-based model, we simulate how a Li-ion battery's State of Health (SoH) and voltage decline accelerate under increasing temperatures. We perform experiments on NASA battery datasets and detect 13 abnormal spikes out of 168 cycles in the B0006 battery. This corresponds to 7.7% of the cycles. Temperature and SoH have a Spearman correlation coefficient of -0.8011 and p-value < 0.05. We use Machine Learning (ML) to predict the SoH and voltage changes when the temperature varies. Random Forest Regression (RFR) and Gradient Boosting (GB) achieve SoH Mean Squared Error (MSE) of 0.000602 and 0.000676 respectively. This work provides a proactive solution for battery health monitoring, enabling predictive maintenance, reducing the risk of unexpected failures from thermal stress and transitioning to sustainable electric mobility.