Improvement On Lmbp Algorithm And Its' Application On Long-range Fault Diagnosis System

In this paper, I summarize the developing course of fault diagnosis technology and the developing actuality of long-range fault diagnosis technology home and abroad and point out the actuality of fault diagnosis system based on NN and the reason that Neural Networks (NN) is suit to equipment's fault diagnosis. According to the requirement that NN is used in fault diagnosis system, I study the method of mining the training swatches of BP neural networks, realize scientific method on distilling the swatches and the method reduces the dependant on engineers; I ameliorate the method of Levenberg Marquardt Back Propagation (LMBP) algorithm and the ameliorated method improved the training speed of NN. Namely, I select the training swatches with decision tree mining algorithm that is based on maximal information gain and disposes the mean of each watch so as to training the NN more conveniently; initialize the weights and thresholds of NN with the strategy of symmetrical distribution; considering that basic BP algorithm has the limitation of long training time and bad convergence feature, I analyze and compare several usual methods of improving BP algorithm, and then select LMBP algorithm as the equipment's fault diagnosis algorithm through the examples' simulation according to the capability indexes of training epochs and time-consuming, and propose one method of reducing the calculating quantities of LMBP algorithm and improving its' training speed from the point of numerical analysis. At last, I take rotating machine's hydraulic pressure station of Liuzhuang colliery of Xinji group in Huainan city as the simulating object and design one simulating system of long-range fault diagnosis in the laboratory. This simulating system takes ARM S3C44B0X as data collecting board, Vxworks as the operating system of this board, RTL8019S as the internet controller, Borland C++ Builder as exploiting environment of up-machine interface, Matlab as the background calculating software of fault diagnosis algorithm. The test result indicates that this fault diagnosis simulating system of laboratory has better communicational function, higher reliability in software and hardware, stronger validity in fault diagnosis algorithm.