Defect Classification And Identification For Workpiece Surface With Milling Textures

In order to solve the problem that the surface defects of workpieces are identified difficultly with the interference of milling texture background,this dissertation studies a machine vision-based method for the classification and identification of surface defects with the interference of milling background texture.It includes the experiments of milling and acquisition of workpiece surface defect,Workpiece surface milling texture background extraction,image processing and defect target segmentation of workpiece surface defects under the interference of milling background texture,the description of feature defects on the surface of the workpiece,and construction of defect feature vectors,defect classification and recognition of workpiece surface,etc.For these contents,the dissertation carried out the following researches:(1)The Milling experiments and the collection of workpiece surface defects image.High-speed three-axis machining center(HNC-180xp/M3)is used as the experimental platform.According to the cutting manual,three groups of commonly used processing parameters were selected to establish a three-factor three-level orthogonal experiment table,and then a series of milling experiments were carried out for No.15 steel,No.45 steel and Titanium alloy TC4,respectively,to obtain workpiece surfaces with different background textures.Based on experience knowledge,the characteristics of common defects on the surface of the workpiece are described and then the surface image of the workpiece is collected.(2)The extraction of milling background textures.The wiener filtering method is used to denoise the workpiece surface images.For the problem of high computational complexity due to the use of the same quantization interval for traditional gray level co-occurrence matrix algorithms,a method for quantizing different regions with different frequency of image texture by using different gray levels is proposed.Based on the improved gray level co-occurrence matrix algorithm,the background texture of the workpiece surface is extracted and the background texture map of the workpiece surface is obtained.(3)The defective object segmentation and the suppression of milling background textures.The background texture image is divided into a number of pixel blocks of the same size,and a number of pixel blocks are randomly sampled,these pexel blocks are arranged in a matrix to represent the background texture of the non-defect area.Then non-negative matrix factorization is used to reduce the size of these pexel blocks,and then the Euclidean distance between these pixel blocks and the same size pixel blocks in the background texture image are calculated.The distance average value are calculated and assign it to the center pixel of the corresponding pixel square in the background texture image to weaken the background texture to highlight the defect target.Then,the K-means clustering algorithm is used to binarize the background texture weakening map on the surface of the workpiece to realize the target image segmentation of the workpiece surface with the interference of the milling background texture.(4)The description of common defects on the surface of the workpiece.The defect features are mathematically described based on the geometric features of the defect target to build a set of defect feature vectors for classification recognition.(5)Establishment and Verification of the Classification and Recognition Model of Workpiece Surface Defects.Based on the set of defect feature vectors,a binary tree support vector machine classifier is designed.A binary tree structure is constructed by using top-down splitting to classify and identify common defects on the surface of the workpiece.The theoretical analysis and experimental results show that the defect recognition and classification methods established in this paper can effectively separate the defect targets from the images with complex background textures and classify them.This method is insensitive to background texture interference,and can avoid the shortcomings of missing detection and low efficiency caused by traditional manual detection.It provides a new idea for non-destructive testing of workpiece surface defects.