The Device Structure Defects In Non-destructive Automatic Recognition

Nondestructive testing is a new subject used all-arround application technology. It helps manufacture to control the products'quality, and also assures the equipment's safety which is still running. So, nondestructive testing is widely used in varied industry, such as mechanism, metallurgy, electronic, chemical plant, railway, shipping, nuclear energy, aviation, spaceflight, electric power, and etc.In civil aviation, every time the plane takes off and lands, its surface must endure the pressure and decompression process, this make great damage to the material around rivets which fix the airplane surface, and so, tiny radial cracks form. When crack's length extends to the critical length, cracks will extends so rapidly that it will cause the surface structure break, this kind of structure damage is one of the most important reasons of aerial accident. The aerial organizations all over the world pay more and more attention to detection technology research for aged aircrafts. In order to detect the aircraft surface defects through Magneto Optic image, this thesis develops the recognition algorithm of structural defect detection robot. A new method based on fuzzy Support Vector Machine (FSVM) is presented. The original magneto optic image contains many kinds of disturbs, such as snake-like lines and the variation of the light distribution. To be more convenient for recognition, series of pretreatment are done, in turn, subtraction from a mask, denoising, erosion/dilatation process, binarization and boundaries detection.Considering the irregular cycle character of rivet's magneto optic image, threshold method is used to get the approximate center of rivet region, star radial vector emitted from this center is taken as the recognition basic feature, and FSVM classifies the direction of cracks around rivets. Thereinto, radial basis kernel function is selected in the SVM, cross-validation and fuzzy membership function (FMF) are used for higher recognition ability. SVM mode is optimized by utilizing cross-validation method and the wrong and refusal recognition problems are solved through FMF in multi-classifier. Experiment on test samples proves high recognition ability.In addition, this research can be used for reference in detection of electric power equipment which is still in service.