Engineering Testing Protocols for Machine Learning-Based SoH Estimation in Lithium Metal Batteries

Lithium metal batteries (LMB) represent one of the most promising battery technologies due to their high energy density. However, the commercial adoption of LMBs has been hindered by their limited lifespan and safety concerns resulting from lithium metal's high electrochemical reactivity. Effective health estimation could enable optimized LMB operation to improve utilization efficiency and reduce failure risks, as well as accelerate LMB development iteration. In this work, we first propose novel diagnostic tests and aging cycling protocols that enable the tracking of LMB degradation from voltage-current measurements through extracting internal resistance, relaxation voltage statistics, and charging impedance descriptors. We present an integrated data-driven health estimation framework for LMBs, using Bagged Gaussian Process Regression (Bagged-GPR). Validation over test cells demonstrates accurate health tracking performance throughout LMB lifetime using one cycle of voltage-current data. The proposed model provides degradation insights for LMBs without the need for additional internal sensors.