Method Research Of Fault Diagnosis Based On Rough Set Theory And PSO-BP Neural Network

As the modern production equipment increasingly become large-scale, preciseness andautomatization, equipment failure is showing the characteristic of diversity, randomness andhysteresis. Besides, most faults are caused by many factors and theses factors are interactional,which causes the traditional fault diagnosis and detection is more difficult to apply thelarge-scale and complicated system. Therefore fault diagnosis techology has graduallyfocused on the study of artificial intelligent hybrid technology, for example, the fusion ofArtificial Neural Network and Rough Set has been vastly applied.This intelligent algorithm with the combination of RS and Particle Swarm Optimized BPNeural Network has been introduced to the field of fault diagnosis. This essay would, with thegoal of diagnostic accuracy, using practical cases as the objects of study, aim to verify thefeasibility of this intelligent algorithm in the application of fault diagnose. The main pointsare as the following:First of all, the paper analyzes the importance of equipment fault diagnosis, summarizesthe common theory and method of fault diagnosis, and analyzes the applicable range andadvantages and disadvantages of various diagnosis methods, which lead to the application ofrough set theory in fault diagnosis. According to above analysis and study, this paperproposes an intelligent algorithm for fault diagnosis which combines rough set theory, BPneural network and particle swarm optimization algorithm.Secondly, it narrated the principle and the process of learning algorithm for RS theory,BP neural network, PSO separately.And then, analyzing the possibility of combining RS and ANN，it designed the faultdiagnosis model of RS and PSO-BP neural network. According to the order of the faultdiagnosis model, it studied the methods of the discretization of continuous attributes andattribute reduction. Given that these two methods would influence the results of handling faultfeatures, by contrastive analysis, this essay firstly proposed to study the sample data bydiscretization of continuous attributes with self-organizing map network which has strongclustering, and then proposed to reduce the sample data to get the minimum reductionattribute core by reduction method which is highly accurate and convenient to calculate, basedon an improved differential matrix. Besides, it designed the diagnosis system of PSO-BPneural network and algorithm process.At last, fault diagnosis to rolling bearing was conducted by the fault diagnosis system ofRS, PSO-BP neural network and RS-PSO-BP network. Simulation study and contrastiveanalysis to the diagnosis system of BP neural network, PSO-BP network and RS-PSO-BPnetwork were conducted as well. Simulation result showed that intelligent fault diagnosis method based on RS and Particle Swarm Optimized BP Neural Network has high accuracy,short training time and simple network structure.