Fault Diagnosis Method Based On Support Vector Data Description

The substance of fault diagnosis is a problem of pattern recognition with small-sample. It can not expect an ideal result with usual diagnose method such as ANN for there are not enough fault data. Therefore, the lack of fault data is always a problem that limits the development of intelligent fault diagnosis.Support vector data description is a one-class classification method that developed from the statistic learning theory and support vector machine. The basic idea of support vector data description method is to find a super-sphere in feature space and limit the volume of the sphere to be the smallest, in the same time include as possible as more target data and exclude as possible as more non-target sample data. Thus the data can be classified and the point included in the super-sphere was thought as target data and the point excluded the super-sphere was thought as non-target data. This method has the advantage of rapid calculation, have well robust, and can deal with few fault data. Applied to machine fault diagnosis and condition monitoring, this method can monitor machine condition and diagnose its faults simply by using normal condition signals. For some key equipment, fault is not allowed or very low fault rate is desired. Support vector data description method is very useful for the online condition monitoring and evaluation of these equipments with small fault data or even no fault data. The application of the method on fault diagnosis field can expected to solve the problem of the of the fault data.The support vector data description method was applied to fault diagnosis and these aspects of the method were researched based on a brief introduction of statistic learning theory. (1) Support vector data description can classify with only normal state data. But when fault data is available, how to use the fault data to improve the veracity of fault diagnosis?; (2) The preprocess of the data in support vector data description was researched, mainly the feature extraction technology which is important in fault diagnosis. The wavelet packet decomposing technology was used to feature extraction and good classification effect was obtained in experiments. (3) Kernel technology in support vector data description was researched and contrast of different functions was done through experiments to show the different effect of them in support vector data description method in this paper. (4) The mixture of support vector data description method was researched in this paper and the validity of the method was validated thr ough experiment on rolling bearing. (5) The incremental support vector data description method was researched to deal with the data varying with time.