Research On Defect Detection Of Lithium Battery Electrode Sheet Based On Machine Vision

Lithium battery is the most widely used dynamical power all over the world. As one of its main components, the electrode sheet is a principal constituent which effects the battery quality. The battery electrode sheet defects result in poor battery quality, and even lead to potential safety hazard. Compared with the traditional manual defect detection of electrode sheet, applying machine vision is beneficial to working efficiency and cost. Thus, the increase of product qualified rate and improvement of battery safety are achieved. Therefore, research on the defect detection of lithium battery electrode sheet has important theoretical significance application value.This paper introduces the research on defect detection of lithium battery electrode sheet based on machine vision. The electrode sheet defect includes tabs, edge and polar net. The type of polar net defect comprises scratch, foreign material and bubble. The scheme of detection system is designed, and detection methods of important defects are discussed in detail. The main contents are shown as follow.Firstly, two defect detection methods based on image processing are proposed. One method is aiming at tab defects and the other is against edge and polar net defects. After image pre-processing, median filter and Sobel operator are utilized to detect damage, wrinkle and deletion on tabs. For defects on edge and polar net, the image is processed by Gaussian filter and median filter respectively to get two intermediate images and the defect information can be extracted from the subtraction between them. The experiments show that proposed methods accomplish defect detection effectively and accurately.Secondly, a polar net bubble detection approach based on the depth information of electrode sheet acquired by three-dimensional reconstruction is presented. According to laser triangulation measurement, the height information of an object can be acquired by laser irradiation. Then the bubble and its depth can be extracted by the arithmetic operator. In experiments, three-dimensional model of cube samples and polar net bubble are reconstructed by the proposed approach. At the same time, the reconstruction outcomes and measuring error are analyzed. The experimental results show that the depth information of polar net bubble can be obtained effectively and satisfy the precision requirement.Finally, the defect classification of polar net bubble is studied. The training set and testing set are both composed of images that demonstrate eligible polar nets and with bubble defect ones. After studying training samples by Support Tucker Machines, the proposed algorithm disposes testing samples and gets classification result based on Support Tucker Machines. Comparing the classification accuracy and ROC curve, the Support Tucker Machines has better performance on accuracy and ROC curve property than other classification methods, for instance, Support Vector Machines.