Combined classification of battery state-of-life and charging regimes by recurrent deep neural networks

Accurately estimating both the state-of-life and charging regimes of batteries is essential for optimizing performance, ensuring reliability, and extending the battery's operational lifespan. This dual-task estimation is particularly challenging due to the nonlinear degradation behaviors of different battery chemistry and the varying effects of charge-discharge patterns. This work proposes a deep learning-based approach that simultaneously classifies battery life states and identifies charging regimes using a Long Short-Term Memory network. The model is trained on charge-discharge cycle data from Nickel-Metal Hydride and Lithium-Polymer batteries, where each cycle serves as an input sequence for the network. Unlike conventional methods focusing on state-of-life estimation, our approach also determines the charging current, distinguishing between different charging strategies applied over the battery lifespan. By integrating state-of-life assessment with charging regime identification, this study provides a novel perspective on battery health monitoring, enhancing the potential of deep learning in comprehensive, data-driven battery management systems.