The Fault Diagnosis Virtual System Based On Rough Sets And Neural Networks

At present, utilizing a variety of intelligent fault diagnosis technology combined to achieve fault diagnosis has been a research hotspot in the field of fault diagnosis. neural networks is playing an increasingly important role in the field of fault diagnosis by their own unique advantages, but a large number of redundant information in the training data is severely restricting the efficiency of neural network diagnosis. Moreover, rough sets can delete otiose and irrelevant information. Therefore, the combination of them is inevitable.The paper's purpose is to build fault diagnosis virtual system based on rough sets-neural network and realize its program, we have mainly done the following aspects of work:①Aiming at the existing method on fault diagnosis combined rough set and neural network, based on the analysis of the current advantages and disadvantages of several common methods, bring forward to the following implementation:adopt SOM neural networks to realize the discretization of continuous attributes, adopt Johnson's algorithm in the software of Rosetta to realize attribute reduction, adopt improved BP neural network to realize fault diagnosis.②BP neural network was introduced to L-M algorithm. The algorithm combines the advantages of both of the Gauss-Newton and gradient descent method, so that the network can be adaptive adjust network weights in learning and training process, and can be effective convergence, at the same time, it also greatly enhance the network's convergence speed and generalization ability.③In the paper, the design of system's friendly user interface uses graphical programming language-LABVIEW which is facing graphics user, has strong interaction and the analysis ability of powerful data visualization; the data discretization and neural networks'train and diagnosis use MAILAB language which has high efficiency in programming, a strong ability in value computation and a variety of neural networks toolbox; finally, using an instance of a fault sample data verified the feasibility and practicality of this system.