Application Of Modified Elman Neural Network In Gearbox Fault Diagnosis System

As a key part in many mechanical equipment, gearbox plays a vital role in theconnection and transmission of power. Gearbox failure will be directly related to the operationof the entire device, and even endanger the personal safety. Therefore, research of gearboxfault diagnosis technology plays an important role in industry security and productivity effect.Currently, the improvement of the accuracy requirements of fault diagnosis make thecombination of intelligent field and fault diagnosis more closely and it provides a new andimportant means for fault diagnosis technology. Recurrent neural network has been widelyused in fault diagnosis because of its good dynamic characteristics and Elman artificial neuralnetwork is one of the recurrent neural network.In this paper, through studying the structure and the work process of the gear-box，weconstruct the gearbox fault diagnosis system. The use of wavelet packet theory realize thedecrease of the noise to the measured vibration signals of gearbox. Inspired by the idea ofinformation fusion technology, we extract the mixed feature which can reflect gearboxoperation information as the input parameters of neural network. Based on the carefully studyof the theoretical framework of the Elman neural network, this paper analyzes theshortcomings of Elman network, puts forward a modified Elman network, which is applied tofault diagnosis of engine gearbox. Five fault modes are selected as the studying objects in thegearbox. Modified Elman network, OHF Elman and Elman neural network are established tosimulate. Through a series of training and testing, the results prove that fault diagnosis ofengine gearbox based on the modified Elman neural network has high stability and diagnosesgearbox effectively. It can apply to the fault diagnosis technology of the practice engineeringand provides a more efficient method for gearbox fault diagnosis technology.