Online Classification And Optimization Of Gear Surface Defects Based On Support Vector Machine

Surface defect inspection is an essential part before gear products go out of factories.The purpose of this paper is to conduct the online real-time classification of defects by collecting local images of gear surfaces on the production line.The collected defect images were analyzed and three common surface defects were identified.The main cause of cracks is the appearance of problems with blanks during processing.Scratches are mainly generated on machine tool process lines,and the occurrence of pit-shaped indentations are mainly due to the falling off of impurities in the blank during processing.Thus,it is possible to solve the problems according to their types and causes,thereby improving the production efficiency.Firstly,the surface defect image obtained is preprocessed.Due to the influence of the collection environment,there are uneven illumination and certain noise in the collected pictures;then the homomorphic filter is used to remove illumination,and the median filter is used for denoising.The defect image is segmented using the Otsu threshold method so that defects can be separated from the background.In order to extract the Fourier descriptors,the edge detection of the defect images is conducted using the Canny operator and the edges of the defects are obtained.Secondly,feature extraction and selection are performed on the images that have been operated.For three common surface defects,Hu moment invariant,aspect ratio and circularity,mean and variance,Fourier descriptors are extracted.Then,the extracted features are selected,and the first four orders were determined.A seven-dimensional vector consisting of the first four invariant moments and its mean,aspect ratio,and circularity is used as the input vector of the support vector machine for defect classification.Finally,the support vector parameters are optimized using common optimization algorithms.The experimental results show that the immune algorithm has greatly improved the accuracy,but the recognition time cannot meet the real-time requirements,so the classification time is optimized to make it meet the requirements of onlineclassification in terms of accuracy and speed,and solves the problem of on-line classification of gear surface defects.