Research On Rough Sets And Neural Network In Fault Diagnosis Expert System

Fault diagnosis data, in other words, the input data of neural network have some disadvantage, for example, data is widely redundancy,incompatibility, which cause neural network training slowly and convergence diffcult, and it affect neural network's further application in expert system. In this paper rough sets is used to eliminate redundancy of network input(fault diagnosis data) and network structure,improve the application capbility of expert system.Firstly,it is discussed that redundancy objects affect rough set attribution in decision table, on this condition, a algorithm of computing the simplest decision table is put forward. With the concept of object frequency, algorithm deletes the object which is same or inconsistent and low frequency, construct the simplest decision table. This algorithm is integrated with distinguish object group algorithm-it is named SDTRA algorithm, it could be applied not only consistent decision table, but also inconsistent decision table.Secondly, optimizal theory of BP network structure is thoroughly lucubrated,the advantage and disadvantage of weight elimination, sensitivity pruning algorithm and correlation pruning algorithm is analyzed . For the character of BP learn and weight adjustment and the excellence of redundancy information can be reduced by rough sets,it is supposed that decision-making table would be established when the network is trainning,after the table would be reduced, redundancy weight and node can be deleted, at last a best optimizal network would be acquired. It is a new method of designing neural network-rough sets method.Finally, a new model for fault diagnosis expert system based on rough sets is put forward, which is based on reduction algorithm of rough sets and structure design method of neural network. It is combined tightly between Rough sets and neural network, rough sets is used to not only reducting and optimizing the fault diagnosis data, but also pruning and optimizing the structure of neural network. On this condition, a valve clearance diagnosis expert system of diesel engine is implemented. The results show that the expert system has better efficiency and diagnosis accuracy than the other system, so it is estimated that the expert system would be applied in fault diagnosis.