储能科学与技术 ›› 2022, Vol. 11 ›› Issue (8): 2537-2545.doi: 10.19799/j.cnki.2095-4239.2022.0226

• 电化学储能安全专刊 • 上一篇    下一篇

基于WOA-BPNN的锂电池极片涂布缺陷检测识别

钟健平1,2(), 费韬1   

  1. 1.华南理工大学自动化科学与工程学院
    2.精密电子制造装备教育部工程 研究中心，广东省高端芯片智能封测装备工程实验室，广东 广州 510641
  • 收稿日期:2022-04-26 出版日期:2022-08-05 发布日期:2022-08-03
  • 通讯作者: 钟健平 E-mail:897045115@qq.com
  • 作者简介:钟健平（1995—），男，硕士研究生，主要研究方向为计算机视觉与锂电池元器件缺陷检测，E-mail：897045115@qq.com。
  • 基金资助:
    国家自然科学基金资助项目(61573146)

Defects detection and recognition of lithium battery electrode plate coating based on WOA-BPNN

Jianping ZHONG1,2(), Tao FEI1   

  1. 1.School of Automation Science and Engineering, South China University of Technology
    2.Engineering Research Center for Precision Electronic Manufacturing Equipment, Ministry of Education & Guangdong Provincial Engineering Laboratory for Advanced Chip Intelligent Packaging Equipment, Guangzhou 510641, Guangdong, China
  • Received:2022-04-26 Online:2022-08-05 Published:2022-08-03
  • Contact: Jianping ZHONG E-mail:897045115@qq.com

RichHTML

40

PDF

392

摘要：

锂电池的正、负极片是锂电池的重要组成部分，极片涂布的质量很大程度上影响着电池的性能和使用寿命，而有缺陷的极片往往是电池安全隐患的根源。为了进一步提高锂电池极片涂布缺陷检测与识别的自动化性能水平，本工作提出了一套基于WOA-BPNN的锂电池极片涂布缺陷检测识别算法。首先，对采集到的锂电池极片涂布图像进行图像预处理操作；接着，将图像中的缺陷目标区域分割出来后，提取其形态、灰度、纹理特征；然后，搭建误差反向传播网络(back propagation neural network，BPNN)，并将串行融合后的融合特征向量作为网络的输入；最后，在训练神经网络分类模型的过程中，使用鲸鱼优化算法(whale optimization algorithm，WOA)用于辅助调参，以进一步提高模型的识别准确率。本工作算法可精确实现对划痕、漏金属、孔洞、裂纹、异污、脱碳等8种常见的锂电池极片涂布缺陷的检测与识别，实验结果证明，当检测的锂电池极片宽度为200 mm，检测精度为0.05 mm，检测速度为60 m/min时，本工作算法的平均漏检率为1.68%，平均误检率为0%，平均分类识别准确率为97.08%。本工作算法能够有效应用于高速高精度的锂电池极片涂布缺陷检测场合，在锂电池智能制造领域具有一定的实用价值。

关键词: 机器视觉, 缺陷检测, 图像分类, 锂电池极片

Abstract:

The positive and negative electrodes of a lithium battery are crucial parts of the battery. The electrode coating quality is significantly related to the battery's performance and service life, and a defective electrode is often the cause of the battery's potential safety hazard. This study presents a set of algorithms based on a whale optimization algorithm-back propagation neural network (WOA-BPNN) to further improve the automation level of defect detection and recognition for lithium-battery electrode plate coating. First, image preprocessing is performed on the collected lithium-battery electrode plate coating image. Then, the shape, gray, and texture features of the defect target area are extracted after being segmented from the image. Next, the BPNN is built, and the fused feature vector after serial fusion is used as the network's input. Finally, the WOA is used to assist in parameter adjustment while training the neural network classification model to further improve the model's recognition accuracy. This algorithm can accurately detect and recognize eight types of common lithium-battery electrode plate coating defects, including scratch, metal exposure, hole, crack, abnormal pollution, and carbon spalling. The experimental results show that when the width of the lithium-battery electrode plate is 200 mm, the detection accuracy is 0.05 mm and the detection speed is 60 m/min, and the average missed detection rate of this algorithm is 1.68%, the average false detection rate is 0%, and the average classification and recognition accuracy is 97.08%. This algorithm can be used to effectively detect defects in lithium-battery electrode plate coating at high speeds and with high precision, and it has practical applications in the field of lithium battery intelligent manufacturing.

Key words: machine vision, defect detection, image classification, lithium battery electrode plate

中图分类号: