A method of anomaly detection and fault diagnosis with online adaptive learning under small training samples

Several methods of the modern intelligent anomaly detection and fault diagnosis have been developed to provide more efficient solutions. However, lacking of fault samples, the training stage and testing stage being mutually independent, and not recognizing new fault type restrict their application in some cases. This paper presents a method of anomaly detection and fault diagnosis with online adaptive learning under small training samples. This approach has classification function and clustering function at the same time. The samples of known fault type are categorized and the samples of unknown fault type are clustered with this approach. To determine the performance and possible advantages of the approaches, the experiments on ball bearing fault data and Iris data were performed. Results show that our proposed approach outperforms the other methods, when the training samples are inadequate to cover all of the fault types. The less the known fault types are, the more advantages it has. To a certain extent, this approach could make up for the disadvantages of other methods of anomaly detection and fault diagnosis.