Design Of Material Surface Defect Detection System Based On Machine Vision

Material surface quality is an important parameter in evaluating the product’s durability and commercial value.It has a strong impact on products’ service life and business value,especially in aerospace and other areas where material defects can have major safety implications.Aluminum as one of the metals used in industrial fields is widely applicable in many fields due to its good mechanical and physical properties.In reality production,due to various factors such as complexities caused by multiple factors,aluminum extrusions may present some defect issues.Therefore,it is necessary to screen out aluminum extrusion with defects for improving the quality of alloys obtained from them.The inspection of plating involves the use of image recognition technology combined with target detection algorithms to design a machine vision based automatic flaw inspection system that replaces manual inspection.This system also applies to other materials surfaces which needs to be screened and recommended for widespread adoption.The aluminum material surface defect detection system researched and designed in this thesis is composed of a main control unit,an LCD display unit and an image acquisition unit.The main control unit adopts RK3399 core processor,and the function development is based on the Linux system.At the same time,it is also used as an image processing unit,calling the transplanted YOLOv5 model to detect the input image,which greatly reduces the development cost compared to using a PC,and uses the Python language and the Py Qt5 image interface framework to complete the development of the system software;the original image and defect detection result images are displayed in real time through the LCD screen.The image acquisition unit is composed of camera,lens and light source.By selecting the ring light source and vertical lighting method to obtain higher quality images and reduce the difficulty of defect detection.In this thesis,based on YOLOv5 network,adaptive learning rate algorithm is introduced to make the convergence of the model better.The detection method of machine vision is used,based on domestic and international research,combined with feature point-based image alignment fusion algorithm to process the original image and obtain a new image after processing to make the defect features more prominent.Then the YOLOv5 network with the introduction of adaptive learning rate is used to optimize the clustering of the new image and generate a defect detection model,and finally the input image is detected according to the model to complete the output of object category and confidence level to achieve accurate classification and localization of defects on the surface of aluminum profile material.Through the test of the system function,the results show that the overall detection accuracy of the defect detection method based on machine vision proposed in this thesis is 93.25%,and the average inspection time per piece is 2.71 seconds,and the overall detection accuracy is significantly higher than the single deep learning algorithm.It can realize the identification and location of surface defects of aluminum profile materials,and can meet the design requirements of the system and carry out long-term stable work.Compared with manual detection,it not only improves the detection accuracy,but also improves the reliability and speed of detection results.It has the advantages of high efficiency and real-time,and has high application value and economic benefits.