Optimized Fault Diagnosis-Driven RUL Prediction for Lithium-Ion Batteries using Multivariate LSTM

Accurately predicting the remaining useful life (RUL) of lithium-ion batteries is crucial for optimizing battery management practices. However, many existing methods overlook the impact of fault occurrences on RUL prediction accuracy. To address this gap, we propose a novel approach that integrates fault diagnosis as a pivotal component of the Long Short-Term Memory (LSTM) model. Leveraging the capabilities of LSTM networks, known for their effectiveness in capturing temporal dependencies and patterns in sequential data, our holistic perspective considers both normal and faulty states of lithium-ion batteries. Validation using NASA's battery dataset, a widely recognized benchmark, demonstrates the superior performance of our approach with the lowest percentage of error based on mean absolute error (MAE) and root-mean-squared error (RMSE) below 3%. Additionally, our method exhibits enhanced stability and reliability in both the normal and abnormal states of lithium-ion batteries, promising advancements in battery management practices and ensuring optimal performance while minimizing maintenance costs.