Detecting Anomalous Cell Behavior in Electrochemical Battery Testing Using Machine Learning

Abstract

The growing availability of structured battery databases opens new opportunities for data‐driven quality control in battery manufacturing. Efficient identification of anomalous cells is critical for ensuring safety, performance, and longevity. In this work, we present a machine‐learning‐based framework for detecting anomalous cell behavior using rate‐test data and electrochemical impedance spectroscopy (EIS) measurements. A curated battery database serves as a reference and filter, enabling systematic comparison across 29 distinct coin cell configurations, each comprising 4–14 nominally identical cells. Rate‐test anomaly prediction using a support vector classifier (SVC) with a radial basis function (RBF) kernel yields a test accuracy of 90%, whereas anomaly detection based on EIS data with a convolutional variational autoencoder (CNN‐VAE) achieves an accuracy of 83.3%. Overall, these findings demonstrate the promise of combining structured battery datasets with advanced machine‐learning techniques to enable automated, accurate, and scalable diagnostic pipelines for next‐generation energy storage systems.