Rapid failure mode classification and quantification in batteries: A deep learning modeling framework

Unique, rapid identification and quantification of the dominant aging modes in lithium-ion batteries (LiBs) with early and non-specialized test data is a significant scientific challenge. Leveraging synthetic-data, deep-learning (DL) techniques have great potential to enable fast and robust classification and quantification of battery aging modes that produce different patterns of cell aging. This study presents, for the first time, a synthetic-data-based DL modeling framework for rapid and automatic classification and quantification of battery-aging modes and resultant aging, with experimental validation of the technique. Availing synthetic dQ.dV(-1) curves for roughly 26,000 initial conditions and aging modes, the framework classified the dominant aging modes for cells undergoing fast charge in less than 100 cycles. Upon classification, the framework quantified the evolution of the aging modes, which were often not uniform with cycling, for 22 Gr/NMC532 pouch cells tested up to 600 cycles at different charging rates (1C-9C).

Automated summary

This study presents a synthetic-data-based deep learning modeling framework for rapid classification and quantification of battery aging modes, with experimental validation of the technique.