Research And Development Of Flaw Detection In Metallics Suface

Metal workpiece is widely used, it is an indispensable component of many instruments.Along with the development of productivity, the users require higher quality, and the surfacequality is more restrictive. The metal workpiece surface defects affect not only the appearanceof the product, but also reduce the performance of the product, so those workpiecescontaining defects must be removed befor leaving the factory. Currently, most of themanufacturers rely on visual detection, the speed is low, the results are linked with theworkers, lacking of scientific guidance. So many companies demand the advanced surfacedefects detection technology and equipment. Foreign equipments and technology are not onlyexpensive, but also we do not have independent property rights. All of thoes force us todevelop automatic defect detection equipment and technology.In this paper, considering the owing metal workpiece as research objects, combined withmachine vision, image processing and pattern recognition to complete the research.(1) Researched the algorithm and methods of surface defect detection, such as imagepreprocessing, image segmentation and feature extraction, and then experimented some ofthem.(2) Considering the bearings as research object, tried to find out how to defect thesurface defects on bearing dust cover, and proposed a method. Using the blue coaxial lightovercomed the metal reflection; Using the least squares method fit the bearing outer circle;Using Otsu’s method and Roberts edge extraction processed the shield image, compared withthe template data, obtained the phase angle, then separateed the character region andno-character region; there was no interference when the two parts defect detecting.(3)Considering the iron oxide-magnetic tile as the research object, using texture analysismethods to achieve feature extraction. Constructed Gabor filters in different scales anddifferent directions.Improved the Gabor filters for the magnetic tile surface defects. In orderto remove the data redundancy, using PCA and ICA.(4) Studied the basic principle of BP neural network and SVM, using additionalmomentum and variable learning rate learning improved the shortcomings of the BP neuralnetwork; Contrapose the difficult selecting of nuclear parameter c and support vectormachines used the grid method and K-CV method to realise the c and g. At last, compared andanalysised the result of thoes two classifier with the magnetic tile surface defects data.The experiment results that the proposed method of bearing dust cover, captured thebearing image clearly, and the correct rate is more than than96%, it can complete the defectsdetecting automatically in real time. Using improved Gabor filters, the magnetic tile surfacedefects feature can be extracted obviously. Using the PCA and ICA analysis, can reduce thefeature dimension. Using this paper’s classification algorithm, the overall accuracy rate canreach93%or more, this paper provides a new way of defects detecingt and classification.