A Partitioning Decision Way for Fault Diagnosis

In this paper, we propose a fault diagnosis method with the combination of rough set (RS) theory and neural networks (NN). A strategy of establishing the partitioning decision tables is adopted. When there are many condition attributes in the fault diagnosis system, the decision tables can be established by dividing condition attributes. Then the partitioning decision tables are reduced according to the reduction theory of rough sets, and we get the core of decision tables and the minimum condition attributes sets. The learning sample sets corresponding to least condition attributes sets are used as testing sample sets of neural network. BP neural network is designed to implement fault diagnosis. The partitioning decision tables can make the reduction simplify and shorten working time. With the aid of experiments using fault data of roll bearing, the paper analyzes the model of rough set and neural network. The results show that the proposed method is feasible to solve fault diagnosis.