Synthesized fault diagnosis method reasoned from rough set-neural network and evidence theory

When traditional machinery fault diagnosis methods are used to handle diagnostic problems, the problems such as low diagnosis accuracy and bad real-time capability will arise if there are lots of data and various complex faults. An integrated fault diagnosis reasoning strategy based on fusing rough sets, neural network, and evidence theory is presented using the principles of data fusion and meta-synthesis. Firstly, use the the parallel neural network structure to improve diagnosis ability of the local diagnosis networks; preprocess the data with rough set theory to simplify the complex neural networks; and eliminate redundant properties in order to determine the topological structure of network. By this way, the shortcomings of network, such as large scale and slow classification, can be overcome. Secondly, a new objectified method of basic probability assignment is given. Besides, the accuracy and efficiency of the fault diagnosis can be improved obviously according to the various redundant and complementary fault information by using the combination rule of the evidence theory to synthesize and make decisions on the evidence. The example of rotating machinery diagnostic given in the paper proves the method to be feasible and available.