Online Fault Tolerant RUL Prediction Strategy for Lithium-Ion Batteries Using Machine Learning

The deterioration of lithium-ion batteries can lead to electrical system failures and potentially catastrophic consequences. Consequently, predicting the remaining useful life (RUL) of batteries is essential to prevent such failures and related issues. Reliable, accurate, and straightforward RUL prediction is crucial for effective power management in electric vehicles and to mitigate the risk of battery failure. This study introduces a highly available fault-tolerant prediction framework designed to forecast the RUL of lithium-ion batteries in challenging scenarios where key features such as voltage, current, and temperature are unavailable. The framework utilizes an Improved Convolutional Long Short-Term Memory Deep Network (Imp-CLD), a hybrid machine learning algorithm integrating Deep Neural Networks, Convolutional Neural Networks, and Long Short-Term Memory networks. Performance metrics including Mean Absolute Error, Absolute Error, Relative Error, and Root Mean Square Error are used to assess the accuracy of RUL predictions. The model’s ability to maintain accuracy is influenced by the point at which predictions begin, as earlier predictions introduce higher levels of uncertainty due to accumulating errors over time. The proposed method is experimentally validated using datasets from the Massachusetts Institute of Technology and the Center for Advanced Life Cycle Engineering. The results indicate that this approach demonstrates greater resilience in predicting RUL, thereby enhancing lifetime control strategies and the safety monitoring function of the battery. This framework maintains prediction accuracy even when all features are missing at the prediction time, potentially preventing short-term catastrophes.