Comparison of Pattern Recognition Approaches for Identification of Failure-prone Battery Cells

This paper compares three numerical classification algorithms in terms of their accuracy in performing classification for pre-screening battery cells as part of a pre-acceptance qualification program: support vector machine (SVM), linear discriminant analysis (LDA), and principal component analysis (PCA) pre-processing followed by LDA (PCALDA). The paper augments previous research which examined only one classifier, the simple generalized classifier (SGC). Key findings of the paper are that the SVM, LDA, and PCALDA all outperformed the SGC in terms of overall classifier accuracy in a numerical study. The PCALDA is shown to yield the greatest overall classifier accuracy (97%), while LDA is shown to give the greatest accuracy (98%) in identifying failure-prone cells. Based on these observations, the SVM, LDA and PCALDA methods are potentially promising candidates to perform battery cell pre-screening classification.