Research On Defect Detection Algorithm Of Stamping Parts Contour Based On Machine Vision

As an important sheet metal forming processing method,stamping has the advantages of high production efficiency,good interchangeability of parts,and low cost.However,due to the influence of various factors such as molds,equipment,and the environment during the stamping process,the surface contour of the workpiece is prone to defects.The traditional quality inspection of stamping parts mainly adopts methods such as manual visual inspection and gauge measurement,which have high error detection rate,low efficiency and poor flexibility.Based on machine vision theory,this paper applies image processing and deep learning technology to carry out algorithm research on contour defect detection of stamping parts.Firstly,according to the forming process of stamping parts in actual production,the principle of stamping forming process is analyzed,and the reasons for the formation of contour defects are discussed,and then the defects are classified reasonably.Then,according to the requirements of detection indicators,the selection of cameras,lenses and light sources are completed,and a visual inspection system for the surface quality of stamping parts is designed.Finally,a stamping part contour defect dataset is constructed,and the images in the dataset are annotated.Defect detection algorithm research based on image processing technology,mainly through image segmentation,feature extraction,defect identification and other steps to detect defects.(1)Aiming at the problem that the image segmentation algorithm based on markov random field is prone to fall into local optimality,an improved image segmentation algorithm based on markov random field is proposed by combining pixel features,regional features and edge features,which improves the accuracy of image segmentation;(2)Using the Canny edge detection operator,a defect area localization method is designed,which can effectively determine the accurate range of defect feature extraction;(3)A defect classifier based on support vector machine is designed,geometric features and texture features of defect are used as effective features for classification and identification,and grid search method is used to determine its optimal parameters.Using the designed algorithm for defect detection experiments,the results show that the average accuracy of the algorithm reaches 91.00%,but the detection accuracy of some defect types is low and the detection speed is slow.Research on defect detection algorithm based on deep learning theory,the YOLOv3 target detection algorithm is optimized,and an improved YOLOv3 algorithm is proposed.(1)Based on the attention mechanism,the convolution block attention module is embedded in the network structure to improve the network’s ability to obtain key information in the image;(2)The fourth scale prediction is added to further improve the network’s detection of small target defects;(3)Designing a loss function that adds generalized intersection over union loss and integrates Focal Loss,which solves the problem of using L2 loss and class imbalance in samples.After experimental verification,the average accuracy of the improved YOLOv3 algorithm reaches 94.50%,and the average detection speed reaches 37ms/sheet,indicating that the algorithm has good real-time detection efficiency and high detection accuracy for stamping contour defects.