Research On Defect Recognition Method Of Stamping Parts Based On Machine Vision

Metal stampings are essential products in the fields of domestic OEMs,machinery manufacturing industry,domestic military and national defense.Due to the degree of matching between different stamping equipment and different molds,as well as multiple factors such as relatively complicated pressure parameters,plate stress characteristics,plate oil coating,etc.,different types of defects appear randomly on the surface of the plate after the stamping is formed,such as Cracks,necking,indentation and other types of defects,the above defects will cause the quality of the stamped parts to be damaged,and the dimensions of the stamped parts will also deviate.Therefore,it is necessary to improve the level of surface defect recognition for stamping parts after forming,which is conducive to improving the detection level of products,ensuring product quality and improving detection efficiency,and can provide higher market competition for stamping parts and manufactured finished products for factories and manufacturing industries force.Aiming at the problem of surface defect identification of stamping parts,this paper makes an in-depth study on image processing methods,the combination of different kinds of algorithms,and defect classification and defect identification.Based on the process of collecting the surface images of stamping parts,an original image based on the area array CCD camera for scanning the stamping parts on the assembly line is proposed,and the method of pixel calculation and processing of the image is used to filter and remove the collected defect images.Noise interference.The image is sharpened according to the equalization principle of the image gray histogram.After the edge detection algorithm based on the Canny operator and the adaptive threshold determination method proposed,an appropriate threshold is determined and then the image is binarized.The original image of the required defect target area can be segmented to extract the defect features of the image and lock in a more accurate range.If the number of selected features is too large,the efficiency of the classifier will be low,and even training will be overfitting.In order to determine the main defect features and reduce the number and dimensions of defect features,the PCA dimension reduction method was used,based on the combination of different types of features,using the cumulative contribution rate as the criterion,and finally selected five defect features,and these features as Effective features for classification and identification of surface defects.Using the SVM classification algorithm to analyze the characteristics of shape features,moment features,and statistical texture features for defect classification description,and according to the test results,an effective feature combination is selected,which improves the accuracy of the system to identify defect types "one to one" with SVM Classifier accuracy.Determine the influence of different parameters in the SVM classifier on the prediction results,select the selection and setting of different parameters of the SVM classification algorithm suitable for the current classification needs.The experiment first performed defect detection on the acquired pictures,and used 20 sets of defect feature parameters as input samples according to the detection results.The SVM classifier accurately achieved the classification of two types of defects,cracks and scratches.The effectiveness of the algorithm and classifier,the defect classification accuracy rate reached more than 95%.