Surface Defect Detection Of Automotive Air Conditioning Swash Plate Based On Deep Learning

The air conditioning compressor swash plate is a kind of circular metal plate with a special surface coating,which is an important part of the automotive swash plate air conditioning compressor.Its surface quality directly affects the working state of the compressor.In actual industrial production,there are defects such as scratches and bumps on the surface of the swash plate.Currently,the method for detecting the surface defects of the swash plate in enterprises is mainly manual inspection,which judges whether the swash plate has defects based on worker experience.Manual testing methods are inefficient,wasteful,and limited by worker experience.Therefore,this thesis uses deep learning technology to intelligentize the detection of swash plate surface defects.The specific work is as follows:(1)Design an automatic swash plate defect detection device,and design an automatic swash plate detection process.Build a data set of swash plate surface defects.Build a defect collection platform with a CMOS industrial camera as the main body and a lens,light source,and bracket;Analyze the characteristics of the collected defects,divide them into five types: scratches,oil stains,bubbles,color differences,and bumps,and label their characteristics to create a dataset label.(2)Establish a surface defect detection model for Faster RCNN,SSD,and Yolo v5 swash plates.Analyze the structure and key modules of the three algorithm models;Three models were trained using a swash plate surface defect dataset,and the model training results were analyzed.An experimental comparison of the three models was conducted.The results showed that the Yolo v5 detection model had a 76% m AP accuracy and a 125 Fps detection speed,with the best defect recognition effect.(3)In order to solve the problems of false detection,missing detection,and inaccurate prediction frame during the detection of Yolo v5 model,this thesis proposes Yolo v5 CC model and makes three improvements.Increase the number of network layers,introduce CA attention mechanism,and improve the sensitivity of the model to defect areas;Improve the upsampling operation and replace it with CARAFE upsampling to improve feature extraction capabilities and further enhance model accuracy;Improved Io U loss function using α-CIo U ensures prediction frame regression accuracy and accelerates network training speed.The experimental results show that the average detection accuracy of Yolo v5 CC reaches 91%,and the detection speed is 125 Fps.Compared with the original Yolo v5 model,significant progress has been made in terms of detection accuracy and speed,meeting the requirements for detecting surface defects on the swash plate.(4)Design visual software for detecting surface defects of swash plate.Use Py Qt5 to design a visual interface for Yolo v5 CC,and use Auto py to exe to package the model and interface.The software implements operations such as detection parameter adjustment,defect detection,and defect statistics.After testing,it meets the actual detection operation requirements.