A Comparison of Online Model-Based Anomaly Detection Methods for a Lithium-Ion Battery Cell

Deployed Lithium-ion batteries are often equipped with battery management systems (BMSs) that monitor sensor readings, estimate states, and ensure safe operation. Unfortunately, additive bias anomalies could corrupt sensor readings used in state estimation, leading to degraded performance or hazardous conditions. Existing protection features of BMSs can be augmented with model-based anomaly detection by combining battery modeling, estimation, and detection algorithms. In this paper, an equivalent circuit model and charge reservoir model were used to model a battery cell, an unscented Kalman filter was used for estimation, and four online model-based anomaly detection methods (chi-squared test, cumulative sum (CUSUM) algorithm, summation detector, and Shewhart control chart) were compared. When evaluated in terms of false positive rate and detection capability, and the CUSUM algorithm was the superior model-based anomaly detector due to its false positive rate of 0%, its ability to detect large and small magnitude anomalies, and its ability to classify anomalies as positively or negatively biased.