Research On Surface Defect Detection Of Parts Based On Deep Learning

As the core component of automobile,engine is composed of tens of thousands of precise parts.The surface processing quality of these parts directly affects the performance and life of the engine.Once the parts with surface quality defects are used,serious consequences will be caused.For this reason,parts manufacturing enterprises must carry out strict inspection on the parts produced.But at present,most enterprises still use manual inspection or traditional visual defect detection methods,and therefore the low accuracy and efficiency of detection can’t meet the needs of the national intelligent manufacturing strategy.In this paper,the surface defect detection algorithm of parts is taken as the research object.In view of the low accuracy and efficiency of the existing defect detection methods,a deep learning target detection algorithm is proposed to detect the surface defect of parts.Taking the spherical joint parts as the detection object,the surface defect image dataset is established,and the deep learning target detection algorithm is studied and improved,so as to finally achieve the surface defect detection of parts with high accuracy and high efficiency,which is of great significance to enhance the level of Intelligent Manufacturing in China.The main research contents of this paper are as follows:(1)Engine spherical joint parts with higher detection accuracy and difficulty level in practical applications are used as detection object.And the three kinds of defects on the parts surface,such as bump injury,turning tool scratch,and impurity,are used as the detection targets.And the surface defect image dataset of parts are constructed.The image capture scheme of the dataset is designed for the detection target.Capturing defect images by building an image capture platform.At the same time,data augmentation and preprocessing are carried out for the captured images,and the image dataset are converted into the PASCAL VOC dataset format.(2)To cope with the features of form subtle and complex on surface defects of engine parts,the scheme of Faster R-CNN algorithm and Res Net101 feature extraction network as the basic framework of the detection algorithm is determined by analysizing and comparing the target detection algorithm based on candidate regions with the target detection algorithm based on regression method in deep learning target detection algorithm.(3)To cope with the features of form subtle and complex on surface defects of engine parts the detection object in this paper,using Faster R-CNN as the algorithm framework,an improved Faster R-CNN surface defect detection algorithm of parts is proposed.Firstly,the clustering algorithm and the guided anchoring algorithm are introduced to generate the Anchor scheme,which solves the problem that the Anchor scheme in the Faster R-CNN original algorithm is not targeted to the surface defect size of the spherical joint,and improves the quality of the region proposal bounding box;Secondly,the multi-level ROI pooling layer structure is introduced to reduce the bias generated by the rounding operation during the ROI pooling process and further improve the accuracy of the detection algorithm.(4)Through the defect detection experiment,it is verified that the proposed algorithm can effectively improve the accuracy and efficiency of defect detection;at the same time,the proposed algorithm is compared with the YOLOv3 algorithm to verify the advancedness of the algorithm in this paper for defect detection tasks.Experimental results show that the proposed algorithm can improve the mean average precision of the detection to 98.9%,and achieve a detection speed of 4.1 fps,which shows the highly accurate and efficient surface defect detection of parts is achieved.In addition to this,a graphical user interface software was developed based on QT to perfect the surface defect detection system of parts.