Granular computing with application to fault diagnosis

To get accurate diagnosis of the faults in mechanical equipments, especially the early stage weak and the compound faults, a new approach to intelligent fault diagnosis of the machinery based on granular computing (GrC) is proposed. A complicate problem can be divided into several small ones which are easily understood and solved according to the idea of GrC. The tolerance granularity space mode is constructed by means of the inner-class distance defined in the attributes space. The fault information can be decomposed into different granularity levels, and be clearly analyzed at each level. To improve the diagnosis accuracy, a reduction method of the fault features based on GrC is also proposed, which directly gets the minimal reduction to construct the tolerance granularity space mode for the best classification accuracy. The proposed approach is applied to the fault diagnosis of locomotive bearing. The results show this mode is endowed with better classification performance than RBF neural network, and the attribute reduction method provide a good way to improve the diagnosis efficiency.