Diagnosis and prognosis of battery degradation through re-evaluation and Gaussian process regression of electrochemical model parameters

Abstract

Lithium-ion battery degradation is complex, and many mechanisms occur concurrently. In-depth degradation is traditionally investigated by post-mortem characterization in lab-settings. If mechanisms could instead be identified in-operando, utilization could be adjusted, and battery lifetime extended. We investigate changes in electrochemical model parameters during battery testing and their correlation with degradation observed in a traditional post-mortem characterization. Commercial batteries are cycle-aged using different stationary storage service cycles and a novel reference performance test is applied intermittently. This test is based on current profiles optimally designed with respect to maximized sensitivity for individual electrochemical parameters and embedded within a charging procedure. Usage dependency of parameter trajectories over the course of ageing is demonstrated and coupled to observed micro-structural changes. Subsequently, the parameter trajectories are extrapolated using Gaussian Process Regression for physics-based state-of-health estimation and remaining-useful-life prediction. We demonstrate and validate estimation of full cell performance under constant load at a later state in life.