Unlocking the Potential of Second-Life Batteries: A Nonlinear Frequency and Machine Learning Approach to Health Prediction

Abstract

The growing demand for sustainable energy solutions highlights the need to extend the use of lithium-ion batteries (LIBs) in first-life applications (e.g., electric vehicles) or repurpose them for second-life uses like energy storage. However, most existing research primarily focuses on first-life applications, with limited attention to the unique challenges of second-life batteries, where accurate estimation of the State of Health (SOH) at low levels (<80%) becomes increasingly difficult due to nonlinear degradation mechanisms. This study addresses this gap by introducing a machine learning approach leveraging Nonlinear Frequency Response Analysis (NFRA) data for SOH estimation in second-life applications down to 60%. NFRA outperformed traditional Electrochemical Impedance Spectroscopy (EIS), achieving >98% prediction accuracy for second-life batteries, even when trained on first-life data alone, and >96.7% using published datasets. NFRA captures nonlinear responses, such as energy losses linked to Li+ transport and solid-electrolyte interface dynamics, which EIS fails to detect. Two predictive models, Long Short-Term Memory (LSTM) networks and Nonlinear Autoregressive with External Input (NARX), were tested, with LSTM reducing root mean square error (RMSE) by up to 30% compared to NARX. NFRA consistently reduced RMSE by over 39% relative to EIS in second-life phases. These findings establish NFRA as a reliable tool for enhancing SOH predictions, enabling safer, more efficient battery repurposing and extended lifetimes.

Keywords: Lithium-ion battery, nonlinear frequency response analysis, electrochemical impedance spectroscopy, second life application, machine learning, LSTM, NARX.