Research of mechanical system fault diagnosis based on support vector data description

In order to solve the problem of insufficient fault samples in intelligent monitoring and diagnosis for machinery, a new method of one-class classification of mechanical faults-support vector data description is proposed. With this method, the outlier objects can be distinguished from target objects if the information of the target class is available without knowing the outlier class. Applying this method to mechanical condition monitoring and fault diagnosis, machine condition can be monitored only by using normal condition signals. It is unnecessary for this method to preprocess the signals to extract their features. The experimental results show that support vector data description method has stronger classification ability and higher efficiency than conventional classification method of neural network.