Development Of Intelligent Hybrid Fault Diagnosis System Based On Rough Sets And Neural Networks

At present, intelligent hybrid fault diagnosis system has been a research hotspot in the field of fault diagnosis. If the training data and diagnostic data used in neural networks diagnosis have redundant information, it will affect the diagnostic efficiency; whereas, rough sets can delete irrelevant or unimportant information under the condition of keeping the classified ability of decision table; so, combining them will make the fault diagnosis system more efficient and practical.The paper's purpose is to establish an intelligent hybrid fault diagnosis system based on rough sets and neural networks and realize its program. The paper pushes the correlation works in two aspects.Firstly, on the basis of predecessor, optimizes the realization method of fault diagnosis furtherly, which mainly contains following content: (1) Adopt SOM neural networks to realize the discretization of training data. This neural networks use abundant training data to adjust the networks' power value by self-organizing, and the process only needs to be given the clustering number, the result of discretization can reflect data's distribution objectively.(2) Adopt heuristic attribute reduction algorithm based on attribute frequency function to realize reduction of decision table formed by discretized data. This algorithm takes core as the start of reduction, introduces attribute significance as heuristic information; Commonly, it ensures us to find the least reduction.(3) Adopt BP neural networks where L-M algorithm was introduced to train the best decision table. Because the convergence rate of L-M algorithm is faster than gradient drop method, it can improve BP neural networks' convergence rate by introducing L-M algorithm into BP, so as to shorten training time and improve efficiency.Secondly, the paper makes a visible program on the basis of improved theory method, forming the basic complete computer applied program of fault diagnosis. In the paper, the design of system's friendly user interface use VB language which is facing graphics user, develops program in short period, and takes effect quickly; the data discretization, reduction, neural networks' train and diagnosis use MATLAB language which has high efficiency in programming, a strong ability in value computation and a neural networks toolbox. Meanwhile, the paper makes use of ActiveX automatization and COM techonogy to realize the seamless joint between MATLAB and VB program. Finally, the perper takes a certain fault training data as an object to validate the system's feasibility and validity.