A Research Of Fault Diagnosis Algorithm Based On Genetic Neural Network

The modern mechanical equipments are all large and high-precision, and therelationship between each part of them is quite close. A whole machine, even theentire production line will be affected, when a component breaks down. As a mostcommon part of mechanical equipment, the operation state of rolling bearing affectsthe performance of the whole machine directly. In the process of diagnosis, therelationship between the diagnostic mode of rolling bearing and its features is verycomplicated nonlinear relation, so it’s difficult to reflect by the time domain andfrequency domain analysis method comprehensively. Artificial neural network infault diagnosis field shows its huge potential application.This paper is made on the background of the fault diagnosis problems ofprecision motor bearing. After analyzing its fault features, the paper extracts thesuited features to establish the diagnosis model by using the strong nonlinearmapping capability of neural network and the global optimization ability of geneticalgorithm, and discusses its application on fault diagnosis of motor bearing.After analyzing lots of fault diagnosis method based on data driven, this paperpresents an improved fault diagnosis method based on genetic neural network:Firstly, the paper presents a two-step data preprocessing method based on grey roughset, and uses it to reduce the attributes of history data; Secondly, in order to avoid themisleading effect by a special individual at early evolution, the author improved thefitness function; arithmetic crossover algorithm and non-uniform mutation algorithmsuits the float code, and the author added the cross counts and mutation countsrespectively to speed up the algorithm evolution speed..Then the paper presents a fault diagnosis method based on the immune geneticneural network. Based on the theory of self-regulatory mechanism, the paperpresents a self-adapting adjustment method of control parameters based on theexpectation reproduction rate of antibody. And inspired by the elitism strategy, thepaper changes the coefficient of arithmetic crossover algorithm, and the linear operation of the improved algorithm shifted to the larger value of fitness. Theredefined calculation method of antibody’ similarity based on the Euclidean distancemakes the setting of threshold easier to realize.The result of simulation experiment under MATLAB platform is shown that thetwo-step data preprocessing method is feasible and effective; the easier setting ofthreshold improves the generalization ability of the method, and established faultdiagnosis algorithms based on GA-BP and IGA-BP have a better convergent speed,a stronger adaptability, and an accurate classification ability. The diagnosis result isquite good.