A Machine Learning Approach to Capacity Estimation of Lithium-Ion Batteries using Electrochemical Impedance Spectroscopy

Accurate battery health prediction is vital for ensuring the reliability and longevity of electric vehicles and energy storage systems, leading to optimal performance and dependability. This study explores the potential of Electrochemical Impedance Spectroscopy (EIS) data in predicting battery health through the application of machine learning techniques. By uncovering the intricate relationship between EIS signatures and capacity, this work endeavors to enable monitoring and proactive maintenance strategies for batteries. Through the application of a Bayesian Neural Networks (BNNs) based probabilistic framework, this work aims to provide reliable battery health estimates. The validation has been carried out over diverse charging/discharging and temperature conditions. With a confidence interval of 95%, the predictive model is found to predict capacities with RMSE below 0.98%.