Research Of Machine Vison Detection Technology Of Regular Machinery Parts Based On Support Vector Machine

With the advantages of non-contact, real-time and fast-speed, the detecting system of machinery parts based on machine vision has been applied to industrial production quite extensively, which works well in detecting defected parts and improving the quality of products. Based on a comprehensive analysis to the scopes and methods of applying machine vision technology to defect detection of machinery parts, this thesis points out the necessity to detect defects of regular porous machinery parts, and then explores taxonomic approaches for guiding the defect detection, with samples of porous gear wheels. Out of different kinds of preprocessing algorithm of images and detection algorithm of edges, this thesis provides with an intelligent algorithm of defected edges, which can be used for complete detection of defect edges of machinery parts. Via extraction of feature of defect, and the employment of classification algorithm of support vector machine (SVM), kernel functions and parameters are selected accordingly, which in the end contributes to a systematic defect detection program for regular machinery parts.This thesis makes researches on all the technologies about image filtering and image edge detection. Through some comparison experiments and the sample parts, this thesis has chosen to preprocess sample images with median filtering and Canny edge detection, get the complete edge of sample parts which is useful for finding out defect feature. This thesis also makes a summary of this processing technology.How to get the edge of defect correctly and fleetly is a premise of getting defect feature, and it directly affects the final detecting precision and the detecting speed. For the porous feature of sample gear wheels, this thesis propose a intelligent algorithm of detecting defected edges based on edge tracking, which can detect the defected edge integrally and provide with the parameters of defect edge for feature extraction.This thesis makes use of the classification algorithm of support vector machine (SVM). Through finding out the feature data of defect edge and then data scaling, feature data of training and classification will be got. After making comparison, this paper chooses the RBF kernel function as this system's kernel function and gets the optimal parameter of kernel function with the method of cross validation and grid search parameters selection. This thesis uses one-to-one voting strategy to train and test sample parts. At last, this thesis designs the software modules with the platform of Matlab and design a GUI interface of this software through which the classification result will be directly examined and analyzed.Through classification testing of defect sample parts, this system achieved to the high detecting precision of 96%, which is higher than 92% of BP neural network. At last, this thesis makes analysis on the measuring errors.