UCLA

This thesis presents a novel method for onboard detection of lithium plating in lithium-ion batteries by combining machine learning (ML) techniques with differential pressure sensing. The study aims to address the challenge of lithium plating, which can lead to capacity fade and cell failure, particularly during fast charging. By measuring the pressure changes during charging and discharging, differential pressure (dP/dQ) can serve as a binary classifier for lithium plating detection. While such detection methods are limited by their invasiveness or specialized equipment requirements, we can overcome these limitations through the proposed approach that utilizes ML models to predict the differential pressure (dP/dQ) signal associated with lithium plating. The best regression model achieved an accuracy of 97.75% on the test set, providing an accurate means of calculating dP/dQ without the need for load cells or external pressure sensors. The study also analyzes the feature importance of the model, revealing key factors influencing the prediction, such as cycling protocol, phase transitions, and battery health. Although further refinement is needed, this research offers a promising avenue for real-time lithium plating detection in LIBs, facilitating safer and more efficient battery management.