Research On Bearing Ring Surface Defeat Detection Method Based On Depth Learning

Bearing industry is an indispensable part of China’s manufacturing industry.For mechanical equipment,bearing is one of the most important basic components to ensure its normal work.For bearings,the quality of the bearing ring directly affects the stability of their normal operation.In view of the traditional defect detection method for bearing ring surface defect detection,which cannot meet the real-time production requirements,this paper applies machine vision deep learning and other related knowledge.The main contents include the following:Aiming at the flat defect on the bearing ring,an improved YOLOv4 network is proposed to evaluate the bearing ring defect data based on the current mainstream deep learning target detection algorithm.First,for the original YOLOv4 network,the default prior box is not suitable for the bearing collar defect data collected in this paper,so using K-means algorithm to cluster the bearing collar defect data set to obtain new anchor boxes,then insert CBM module in the backbone network residual module,add a small size of output to improve the detection of smaller defects,borrow the idea of asymmetric convolution replace 33 convolution layer in the PAnet module,improve the detection accuracy without increasing the number of parameters.Also chose YOLOv4,Faster R-CNN,SSD,YOLOv5 this several target detection commonly used algorithm in the field of comparison test,the experimental results show that the improved algorithm for bearing ring defect data detection has obvious improvement,accuracy reached 91.5%,a 5.9% higher than the original algorithm,compared to other algorithms also have or big or small.The detection speed can also achieve good practicability,which provides a new detection method for the defect detection of industrial bearing sleeve ring.