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

The ability of neural network for fault diagnosis is in direct ratio to the number of the training examples, so are the complicated degree of neural network fabric and the training time to the quantity of fault training examples. That is to say, good neural network must have complicated neural network fabric. However, the complicated neural network needs a lot of training time to get convergence, which is the main factor that restricts the advanced practicability of it. To deal with the problem is the main objective of this study.This research deals with the training example date by the function of rough set reduction and also eliminates the redundant information of the training examples. These above processes have simplified the training examples and the neural network fabric, which solve the problem concerning on the practicability of neural network. Herein, intelligence hybrid system based on rough sets and neural network for fault diagnosis is established in this paper.The practicality of using rough set to reduce the date was discussed, in this paper, and the interval-valued continuous attribute discretization by applying self-organizing map neural network clustering was proposed, too. This article proposes a simplified indiscernible matrix's method, which reduces the example's condition attribute and eliminates the redundant information of the date. What's more, it provides the methods used to diagnosis compressor's faults based on the intelligence hybrid system by adopting the MATLAB neural network workbox. At last, the functional modules which make up into intelligence hybrid system for fault diagnosis was introduced, including: data acquisition module, date preprocessor module, date reduction module, neural network module, fault diagnosis module and fault expressing module. Here, the article also gives the flow of work of all the above modules.