Support vector data description with model selection for condition monitoring

Condition monitoring is very important in machinery engineering study. In most conditions, normal signals are acquired easily but fault samples are difficult to be gained. Because of lacking enough fault samples, the machine diagnosis meets difficulties. Support vector data description (SVDD) is a single classifier and it can distinguish the normal and fault condition just using normal samples. In this paper we first describe the basic algorithm of SVDD. 5-cross validation is used as model selection to optimize the parameter of SVDD. Extracting two-dimension spectrum entropy of signals as the input of the SVDD classifier, we got high classification. Compared neural network (ANN) with SVDD, the experiment result represents that in the environment which lacks of enough fault samples condition monitoring, SVDD has better classification than ANN.