Graded Plating Diagnosis of Lithium-Ion Battery With Multifeatures Extraction and Label Weight Guided Clustering

Irreversible lithium plating serves as an essential failure mechanism leading to the degradation of life and safety performance of lithium-ion batteries (LIBs). Motivated by this, herein we propose an improved lithium plating diagnostic method combining multifeatures extraction and label weight guided clustering. For the first time, we find features correlated highly to the lithium plating, including the high-order electrical features, equivalent circuit, and electrochemical impedance spectra parameters. By the efficient fusion of these highly correlated parameters, we further propose a label weight guided density peak clustering method (W-DPC) for graded lithium plating diagnosis. The proposed W-DPC method exploits the prior knowledge of lithium plating to guide the density calculation of data, with the expectation of better classification performance than widely used clustering method. The proposed method is validated with capacity degradation and postmortem analysis. Results show that the proposed method promises a superior diagnostic performance, with diagnosis accuracy of 95.54% and detection ratio of 87.50% on 202 independent batteries. Moreover, the mild and severe plating can be distinguished accurately with the proposed method, allowing for a graded lithium plating diagnosis.