Rechargeable Battery State Estimation Based on Adaptive-Rate Processing and Machine Learning

The generalization of the use of electronic systems and their integration in industrial systems and different aspects of modern life (internet of things, electric vehicles, robotics, smart grids), give rise to new challenges related to the storage and optimized management of energy. Lithium-on batteries perfectly meet this objective due to their good qualities such as high energy density, small installation size, low self-discharge and high supply capacity. However, their wide application requires further research on battery failure prediction and health management. Intelligent “battery management systems” (BMSs) employ the real-time estimation and control algorithms to improve the battery safety while enhancing its performance. Nevertheless, BMS are complex and require increased processing power which could lead to more power consumption. In this context, the present article provides a new approach for efficient prediction of the “Lithium-ion” (Li-ion) battery cells capacities by analysing and exploiting the battery parameters, acquired by an event-driven module. It acquires the intended cells voltages, currents and temperature values during the charge-discharge cycles. The solution is based on the machine learning algorithms and event-based segmentation. The "National Aeronautics and Space Administration" (NASA) has provided a high-power Li-Ion cells dataset for the purpose of research and innovation. This dataset is used to test and evaluate the suggested approach. The evaluation of the overall performance of the chain has shown encouraging results of the proposed approach.