A dynamic memory model for mechanical fault diagnosis using one-class support vector machine

Due to the mechanical failure data is cumulatively acquired and has uncertain features, the memory model for fault diagnosis is required to adapt with the information updating. In this paper, a dynamic memory model using one-class support vector (OCSVM) is proposed to extract and keep diagnostic information. The feature of each failure type is respectively processed by incremental learning algorithm of OCSVM to construct the optimal distribution region in high-dimensional feature space. Moreover, the minimum decision function, which indicates the distance between failure data and the distribution space, is used to recognize the failure state. The memory model can facilely generate new failure type and update the distribution of existing failure. Evaluation results of simulated and experiential data showed that the memory model satisfies the demands of fault diagnosis effectively.