Defect Detection of Bearing Side Face Based on Sample Data Augmentation and Convolutional Neural Network

Bearing surface quality has significant impact on the working performance and durability of the mechanical transmission equipment. The traditional visual detection methods for bearing surface defects face the problems of weak versatility, low efficiency and poor reliability. In this paper, a deep learning detection method for bearing side face based on data augmentation and convolutional neural network is proposed. Firstly, image expansion based on circle detection and polar coordinate transformation is utilized to facilitate the labeling process and improve the significance of bearing defect area. Secondly, a bearing sample data augmentation method is designed to construct the defect data set. Semi-supervised data enhancement based on local defect features, improved RA strategy, and Mosaic algorithm are used to augment the initial bearing sample data set. Thirdly, an improved Faster R-CNN framework for bearing defect detection is established. The ROI align pooling is used to improve the continuity of output features. The Resnet101 network and Leaky Relu activation function are used to avoid the tiny defect feature loss and function dead zone. Furthermore, the FPN is integrated into Resnet101 to improve the detection precision for multi-scale bearing defects. Experimental results show that the proposed method can effectively achieve accurate and rapid defect detection of bearing surface, with a mAP of 98.18%. The proposed data augmentation strategy and defect detection framework show great application potential in the automatic surface detection of mechanical components.