Dynamic early recognition of abnormal lithium-ion batteries before capacity drops using self-adaptive quantum clustering

Abstract

Lithium-ion batteries may exhibit an abnormal degradation due to causes such as lithium plating, characterized by a rapid capacity drop after a period of normal capacity degradation, posing a major threat to the system reliability and safety. To ensure continuously safe use of the system, this paper proposes a dynamic early recognition framework to distinguish the abnormal batteries from normally degrading batteries before their capacity drops. An unsupervised machine learning method called quantum clustering is introduced to identify normal and abnormal batteries, and it is further improved by using a form of Weibull wave function, which is more sensitive to the abnormal battery features. To eliminate the subjective effects on clustering performance, a self-adaptive parameter estimation method for the quantum clustering is also developed. Through applying to two types of lithium-ion batteries, the proposed dynamic early recognition framework is proven to be highly effective, where all abnormal batteries are recognized before capacity drops, and quantitatively, the average recognition points are 45.78% and 34.75% earlier than the average of knee-points where capacity begins to drop, showing great advantages compared with existing methods.