Abstract:

Lithium plating is one of the main causes of internal short-circuit failures in energy storage lithium-ion batteries, and online monitoring of the lithium plating process is a key research direction for ensuring battery safety. However, in engineering applications, existing methods are limited to monitoring reversible lithium plating, and machine learning-based online monitoring algorithms face challenges in model training due to the lack of lithium plating data from actual batteries. To address these issues, this work proposes a monitoring method for irreversible lithium plating based on unsupervised domain adaptive transfer learning. The source and target domain data for the irreversible lithium plating monitoring model are generated through simulation using an electrochemical-thermal-aging model and low-temperature lithium plating aging experiments. Features related to lithium plating are extracted from discharge curves, and an irreversible lithium plating monitoring model based on multilayer perceptron is developed within the framework of unsupervised domain adaptive transfer learning. This allows for the transfer of the irreversible lithium plating monitoring model from the source domain data to the target domain data. The results show that for simulated data, the algorithm achieves an accuracy rate exceeding 99% in determining the occurrence of irreversible lithium plating. For experimental data, the qualitative judgements of the algorithm on the irreversible lithium plating are consistent with the actual conditions. This demonstrates that the proposed method can effectively monitor irreversible lithium plating in batteries and provides a new approach for the online monitoring of irreversible lithium plating in energy storage lithium-ion batteries.

Key words: Lithium-ion batteries, Lithium plating monitoring, Pseudo two-dimensional model, Multilayer perceptron, Transfer learning