Research And Application Of Workpiece Surface Defects Image Adaptive Recognition Methods

During the production and using of the workpiece,various types of defects always occur on the workpiece surface.The presence of the defects on the surface of the workpiece can affect the performance of the mechanical product and even cause the failure of the mechanical product.Therefore,studying the rapid identification of defects on the surface of the workpiece can not only monitor the quality of the machining process of the mechanical product in real time,but also improve the performance of the product.It can also analyze the causes of surface defects on-line so that the occurrence of the defects can be suppressed.Taking into account the morphological features of the workpiece surface and their positional correlations,this thesis proposes an adaptive image defect recognition method for the surface of workpieces,which mainly focuses on the following aspects:1.Research on Feature Defects of Workpiece Surface Image and Its Experimental Acquisition.The defect features of the workpiece surface were studied to determine which features were used to describe the defects.At the same time,the experimental program for determining the milling process and the image acquisition program were determined.The milling experiment platform was set up based on the 5-axis linkage machining center,and the milling experiment was performed on the No.15 steel using a vertical end mill.After the milling experiment,an image acquisition platform was built and the workpiece surface images under different light conditions were collected.2.Research on Adaptive Hybrid Manifold Clustering Recognition of Workpiece Surface Defects.For the collected surface image of the workpiece,the image was preprocessed by using HSI color space transformation,image enhancement,image sharpening,image denoising and threshold segmentation to extract the surface defect binary image of the workpiece.The similarity weights in the hybrid manifold clustering method are improved based on the morphological characteristics of the surface defects of the workpiece,so that the mixed probability principal component analyzer trained in the hybrid manifold cluster can better learn the defect manifold on the workpiece surface.At the same time,the idea of adaptive clustering was introduced into the hybrid manifold clustering to automatically determine the clustering center point in the process of hybrid manifold clustering,so as to realize the self-adaptive recognition of image defects on the workpiece surface with mutual interference.3.Research on Surface Defects Identification with Convolutional Neural Networks.Image segmentation,image added noise,and image color intensity transformation based on principal component analysis are used to expand the surface image data set of the workpiece.After analyzing the requirements for image defect recognition on the workpiece surface,the output structure of the convolutional neural network was determined.Based on the collected surface images of the workpiece and the image defect recognition results obtained by adaptive hybrid manifold clustering,an image defect recognition convolutional neural network model was established,and we trained the convolutional neural network model to quickly identify the various types of defect information in the surface image.The theoretical analysis and experimental results show that the adaptive defect recognition methods proposed by this paper can solve the traditional problem that defect recognition rely on manual detection.The results of each defect separation are basically consistent with the manual detection results.The trained convolutional neural network of image defect recognition has the advantages of efficiently,accurately and has the ability to identify the detailed information of the defect image,stabilizing the production quality of mechanical processing products and reducing the labor cost.