Base On SVDD Method In Roller Bearing Fault Diagnosis Research

Machine fault diagnosis is a problem of pattern recognition with a small sampleessentially. In the engineering applications，the lack of fault data is always aimportant factor that restricts the development of intelligent fault diagnosis andmakes usual diagnose method such as ANN can not expect an ideal result.Support vector data description is a one-class classification method thatdeveloped from the statistic learning theory and support vector machine. It caneffectively solve the problem of lack of fault samples data. The basic idea of SVDDmethod is to find a super-sphere in feature space and limit the volume of the sphere tobe the smallest which includes as possible as more target data, in the same time,non-target sample data as far as possible fall on the hyper sphere in vitro, thus the aimof separating sample of the target and non-target samples can be accomplished. SVDDmethod requires only one kind of sample data to classify; the method has theadvantage of rapid calculation, dealing with few fault samples, well robust. Throughthis method, we can monitor machine condition and distinguish the machine is fault ornot by using normal condition signals. Therefore, the SVDD method has very highpractical value in engineering applications and can expected to solve the problem thatlacking of fault sample data can not distinguish the fault accurately.The main work is as follows:(1) Kernel function is very important for SVDD method. Researching on thekernel function shows that the SVDD method which uses the Gaussian kernel functionhas the highest detection accuracy.(2) A comparative study on the LMD method and EMD method. The result shows,LMD method is better than the EMD method in reducing the number of iterations anddealing with end effect problem.(3)A fault diagnosis approach for roller bearings based on SVDD method andsingular value decomposition technique of LMD is proposed. In practical applications,the mehod can be applied to the roller bearings fault diagnosis effectively.(4) A fault diagnosis approach for roller bearings based on SVDD method andenvelope spectrum of LMD is proposed.Using of this mehod can monitor theoperational status of the roller bearings effectively and identify the fault of the rollerbearings accurately.