# Machine Learning‐Assisted Prediction of State of Health in Lithium Metal Batteries with Electrochemical Impedance Spectroscopy

## TL;DR

This study uses machine learning to predict the health of lithium metal batteries using EIS data, achieving over 95% accuracy and highlighting the importance of phase shift.

## Contribution

The study introduces a novel AI-driven method for EIS-based SoH prediction, emphasizing phase shift as a key feature.

## Key findings

- Four machine learning algorithms achieved over 95% accuracy in predicting battery SoH.
- Phase shift was identified as a critical feature in SoH prediction, often overlooked in traditional methods.
- EIS features were linked to specific electrochemical phenomena, clarifying the model's physical basis.

## Abstract

Electrochemical impedance spectroscopy (EIS) offers a nondestructive means of diagnosis for the battery's state of health (SoH). However, traditional equivalent circuit‐based approaches—relying on extensive modeling and fitting of complex EIS data such as real and imaginary impedance components, phase shift, and frequency—are time‐consuming and heavily dependent on expert interpretation, which can compromise reliability. In this context, artificial intelligence‐based models present a faster and more reliable alternative for interpreting EIS data. These models can uncover hidden patterns and parameters that may be overlooked by human experts, thereby enabling more accurate prediction of the battery's SoH. In this study, four machine learning algorithms are employed to predict the SoH of lithium metal batteries based on EIS data, achieving predictive accuracies exceeding 95%. Feature importance analysis indicated that phase shift—an often underutilized parameter in conventional EIS interpretation—plays a critical role in the SoH prediction process. Furthermore, the analysis enabled the attribution of specific EIS features to their corresponding electrochemical phenomena, thereby elucidating the physical basis of the model predictions. The resulting models exhibit high precision in forecasting battery discharge capacity and diagnosing degradation mechanisms, demonstrating their potential as powerful tools for advancing battery diagnostics and performance optimization.

This study investigates the artificial intelligence‐driven method for predicting the state of health of lithium metal batteries based on electrochemical impedance spectroscopy (EIS) data. Four machine learning algorithms achieve accuracy exceeding 95%, identifying phase shift as a critical feature for prediction and establishing a connection between EIS‐derived parameters and underlying electrochemical degradation mechanisms.© 2025 WILEY‐VCH GmbH