Research On Fault Diagnosis Method Of Rolling Bearings Based On Wavelet Packet Transform And Optimized Elman Neural Network

Rolling bearings are widely used in the daily life, industrial production, national defense construction and other areas. The research on rolling bearing fault diagnosis method has certain engineering practice significance. The single diagnosis method such as time-domain and frequency-domain has the shortcomings of low diagnostic accuracy,realizing intelligent diagnosis difficultly and only suitable for rough diagnosis. To make up for these defects, this thesis designs a kind of fault diagnosis method of rolling bearings which combines wavelet packet transform and Elman neural network optimized by Cuckoo search algorithm(simply, CS-Elman). This method consists of two parts: fault feature vector construction and fault type discrimination.The fault feature vector construction method based on wavelet packet transform can effectively deal with the abrupt change and non-stationary signal of rolling bearings.Firstly, the wavelet packet transform algorithm is used to carry out the multi-layer wavelet packet transform of the vibration signal and each layer has a series of frequency bands of the same width. Then, the feature parameter vector is constructed to characterize the fault type from the energy value of each frequency band, and the distribution of feature vector value is different for different fault types. Finally, the training and testing samples are composed of these vectors and fault codes.The fault type discrimination method based on CS-Elman can effectively improve fault diagnosis efficiency and accuracy. Firstly, bird nest position vectors which are randomly generated contain the weights and threshold parameters, and the training error function is used as the objective function. The Cuckoo search algorithm finds the best nest position vector according to the objective function. Then the weights and threshold parameters of Elman network are initialized by the best nest position vector. Finally, the Elman neural network is trained by the standard fault model and then it is inputed intothe fault feature vector, and its output is corresponding to the specific type of rolling bearing fault.This thesis proves the feasibility and efficiency of the fault diagnosis method through the simulation test of rolling bearings fault diagnosis.In addition, the research on the basis of rolling bearing vibration signal data has certain reference value to other mechanical fault diagnosis based on vibration signal.