Research On Detection And Classification Of Surface Defects Of Metal Parts

In the power system,metal parts play an indispensable role in switching appliances.In order to achieve a good switching effect,the metal parts in switching appliances should have good electrical properties.During the production process of metal parts in the workshop,due to the influence of force majeure factors such as production tools,parts wear and artificial environment,there are scratches,discoloration,dirt,foaming,impurities and other defects on the surface of the parts.Defects on the surface of metal parts will increase the resistance of the contact surface with the contact element,which will cause the contact part to heat up and cause the metal parts to stick to the contact element.In severe cases,the switch may fail,affecting the quality and service life of high-voltage electrical appliances.This paper uses machine vision and machine learning methods to detect and classify surface defects of certain metal parts in switching appliances.According to the characteristics of a certain metal parts product under study,the hardware platform is built,the structure of the hardware platform is designed,the hardware products are selected,and the image acquisition and lighting system is constructed according to the actual size and surface characteristics of the metal parts.Collect panoramic images of metal parts,and write software to process the collected metal parts pictures.The software can control the hardware and display the processing information and results in real time.It can complete the cracks,scratches and dirt of certain metal parts.,Defect detection and classification of silver drop.Shape-based template matching and cross-correlation-based template matching were performed on metal parts images,and the part location and detection method based on cross-correlation template matching was determined,and the detection effect was improved by optimizing the template matching method.This paper introduces the basic methods of image segmentation,and denoising the images collected by metal parts,and compares the effects of several denoising methods.The method of image enhancement is introduced,and the images of metal parts are enhanced and the results are compared.Aiming at several defects in metal parts,a method of dividing the area is proposed,and different methods are selected for segmentation according to the characteristics of different defects.The feature of the image is introduced and the feature extraction of the defect area of??the metal part is carried out.The feature selection algorithm of the extracted high-dimensional feature is designed to optimize the feature,and the selected feature is formed into a feature vector for classification.Finally,a metal parts surface defect classifier based on random forest algorithm and support vector machine algorithm is constructed.Through comparative analysis,the average classification accuracy based on random forest algorithm is 97.07%,and the average classification accuracy based on SVM algorithm is 85.31%.Considering the operation time,the random forest algorithm is used to classify the surface defects of the parts.