Sequential Deep Learning Model Development for Battery Remaining Useful Life Forecasting and Anomaly Detection

Lithium-ion batteries have become indispensable in various energy storage applications, powering a wide array of devices. Anomaly detection and remaining useful life forecasting are critical tasks in battery management for predictive maintenance and reliability testing. An integrated approach that combines both remaining useful life forecasting and outlier detection is proposed, by monitoring the deviation between prediction and ground truth. This approach is validated using real-world CALCE data and augmented datasets generated from it. First, the capacity degradation of the battery is predicted, then an anomaly is detected if the error crosses a predefined threshold. The models employed achieve high accuracy, forecasting errors limited to 1%, with minimal false positives. This establishes their reliability for practical deployment and makes them comparable to state-of-the-art approaches.