Capacity-fading Behavior Analysis for Early Detection of Unhealthy Li-ion Batteries

Abstract

Reliability testing on lithium-ion (Li-ion) batteries is critical to designing operational back-end strategies for developing portable electronics. In this article, we develop a capacity-fading behavior analysis for the early detection of unhealthy Li-ion batteries during reliability tests by comparing against the capacity-fading behaviors of healthy batteries from qualification. The developed approach uses a local outlier factor for measuring the anomaly scores of the capacity-fading behaviors of test batteries at a certain cycle, kernel density estimation for normalizing the range of anomaly scores over cycles, and a hidden Markov model for estimating the probability that the test batteries are at a certain state (i.e., healthy or unhealthy). Experimental results on Li-ion batteries used for portable consumer electronics confirm that the developed method outperforms previous approaches, reducing the required number of reliability tests for unhealthy batteries to 100 cycles, less than a month in practice.