Key Technologies Study Of NDT Based On Magnetic Particle Indication Image

As an effective,mature and low-cost NDT method,magnetic particle inspection was widely used for detecting iron products in the process of quality assurance.Surface crack detection method based on machine vision has become one of the most commonly NDT technology.This paper combined the image processing and pattern recognition technology with traditional magnetic particle inspection,so that to detect automatically the surface crack of bearing part.This paper firstly introduced the latest research progress of NDT.It expounded the basic principle of magnetic image and analyzed the factor of image.Then,the paper presented an integrated solution of magnetic particle inspection plan based on machine vision.Including a hardware platform,used for acquiring the magnetic image,and an algorithm design,mainly used for extracting and describing feature of magnetic image,as well as detecting crack.Finally,as to the problems about extracting and describing feature,the paper propose some effective method.On the part of magnetic mark extraction,the paper firstly introduced the characteristics of the magnetic image and preprocessing algorithm,then adopted the 2DOtsu,Laws and K-Means algorithm to extracting the magnetic mark,so that overcome the difficulty of extracting.Compared to traditional segmentation algorithm based on color and threshold,the paper's idea can get a better effort.In the aspect of feature description,the paper adopt the method of invariant moment theory,color moment and color coherence vector.Relative to the method of regional shape description,by comparing and analysis the test effort,the paper's methods needn't to segment in advance and can be used widely.In the end,this paper extract the 81 dimension features vector,consist of invariant moment,color moment and color coherence,and utilize SVM to detect surface crack of bearing parts.The experiment result shown that the paper's method can effectively distinguish surface contamination,crack and fine products,and the overall recognition rate has reached up to 85.62%,and the method has a certain engineering application value.