Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (24): 176-184.DOI: 10.3778/j.issn.1002-8331.2206-0509

• Graphics and Image Processing • Previous Articles     Next Articles

Bearing Full Surface Defect Detection by Dimension Segmentation Method

YANG Dongyi, HUANG Danping, XU Jiale, LIAO Shipeng, YU Shaodong   

  1. 1.School of Mechanical Engineering, Sichuan University of Science and Engineering, Yibin, Sichuan 644000, China
    2.Chengdu Information Technology of Chinese Academy of Sciences, Chengdu 610041, China
  • Online:2023-12-15 Published:2023-12-15

维度分割法轴承全表面缺陷检测

杨冬毅,黄丹平,徐佳乐,廖世鹏,于少东   

  1. 1.四川轻化工大学 机械工程学院,四川 宜宾 644000
    2.中国科学院 成都计算机应用研究所,成都 610041

Abstract: In order to improve the detection accuracy and speed of bearing full-surface defects, a bearing full surface defect detection method based on dimensional segmentation method is proposed. In this paper, by building a bearing defect detection platform independently, The dimensional segmentation method is used to dimensionally divide the bearing visual informationand initially extract the suspected defect areas in each dimension to enhance the integrity of bearing outer surface detection. The VGG16 network model is trained using the small region dataset of each dimension of the bearing to obtain the bearing feature vector and apply the improved Euclidean distance formula to replace the VGG16 fully connected layer to judge the suspicious region. Finally, parallel processing mode is used to run the defect detection algorithm in each dimension. The experimental results show that the method can effectively improve the accuracy and speed of full-surface defect detection of bearings, and the leakage rate is low, while reducing the size and  cost of the equipment, which can meet the needs of online detection of full-surface defects in the bearing industry.

Key words: bearing, dimension segmentation, neural network, multithreading

摘要: 为提高轴承全表面缺陷的检测精度与速度,提出基于维度分割法的轴承全表面缺陷检测方法。通过自主搭建轴承缺陷检测平台,采用维度分割法对轴承视觉信息进行维度划分,并初步对各维度疑似缺陷区域提取,增强轴承外表面检测的完整性;利用轴承各维度小区域数据集训练VGG16网络模型,获取轴承特征向量,应用改进欧式距离公式替换VGG16全连接层对可疑区域进行判断;采用并行处理模式运行各维度缺陷检测算法。实验结果表明,该方法能有效地提高轴承全表面缺陷检测精度与速度,且漏检率较低,同时降低了设备的体积与成本,能够满足轴承工业全表面缺陷在线检测的需求。

关键词: 轴承, 维度分割, 神经网络, 多线程