Utilizing Bayesian Optimization to Facilitate Differential Voltage Analysis of Lithium Ion Batteries

Abstract

Differential voltage analysis (DVA) is a non-destructive, in situ method for analyzing degradation in lithium ion (Li-ion) batteries, and can be used to identify major modes of performance degradation, such as loss of lithium inventory (LLI) and loss of active material (LAM). Using the cycling voltage derivative with respect to capacity of half cells and full cells, DVA can be used to estimate which mode is the main contributor to degradation. Analysis of the signature peaks in the half cell to the full cell can reveal the degradation mode of full cell. Shifts in the peak can indicate LLI, while intensification and narrowing of peaks can indicate LAM. Analyzing the peak changes after cycling allows quantitative determination of how much cell aging has occurred, and what mode is involved in the cell aging process. However, the process of analyzing peak changes manually is inefficient and time consuming. In this work, we utilize a Bayesian optimization approach to facilitate the process of fitting full cell DVA data with pristine half-cell data. The Bayesian optimization algorithm finds the best deformation of pristine half-cell data to fit full-cell data automatically and efficiently for multiple chemistries. The only manual procedures for each fitting are reading data files into the program and determining a coarse bound. The rest is automated and takes less than ten minutes for each fitting. The fitting accuracy is comparable to manual fitting and significantly less time consuming. This method facilitates battery aging analysis and can be generalized to different kinds of data sets as well.