Abstract: Because the number of category of sample is unknown, the selection of parameters of model of unsupervised SVC becomes very difficult. Therefore, the supervised SVC method is put forward and applied to fault diagnosis of machine. The parameters of model are selected on the basis of the number of clustering regions and the number of SVs. The probability density of sample distribution is estimated on the basis of taking SV as the core, and typical samples of categories of different clustering regions are selected on the basis of estimated values of probability density. Different faults are classified by introducing k nearest neighbor. The result of classification shows that the SVC method is effective.

Key words: Fault diagnosis, k nearest neighbor, Support vector clustering