Fault Diagnosis Of Gearbox Based On HHT And WNN

Fault diagnosis of gearboxes is of crucial important and has been studied for several decades. Based upon Hilbert-Huang transform (HHT) and Wavelet Neural Network (WNN), a novel method for the fault diagnosis of gearbox fault is proposed in this paper. Hilbert-Huang transform is a new time-frequency analysis method, it is also an adaptive signal processing. The HHT consists of two parts: empirical mode decomposition (EMD) and Hilbert spectral analysis (HSA). EMD method is based on the local characteristics of the signal time scale, complex signals are decomposed into a finite number of intrinsic mode functions (IMFs); Hilbert spectrum and Hilbert marginal spectrum can be obtained by using Hilbert transform to intrinsic mode functions. So that it can effectively extract fault features, identify fault modes, this adaptive decomposition method is very suitable for the nonlinear and non-stationary processes analysis.Base on the nonlinear and non-stationary of gearbox fault vibration signal, to extract the gearbox fault characteristics information, the wavelet packet decomposition is used to the gearbox vibration signals for noise reduction. Then, EMD process and a series of intrinsic mode functions (IMFs) can be acquired by applying Hilbert-Huang transform to some specific frequency band. Finally, we choose some special IMFs to obtain Hilbert spectrum and Hilbert marginal spectrum to extract fault characteristic frequency, so different gear crack fault modes of gearbox can be effectively identified.Another new method of fault diagnosis for gearbox based on hybrid feature extraction and wavelet neural network (WNN) is also proposed in this paper. The time domain analysis, wavelet packet decomposition and wavelet decomposition are applied to extract the fault feature information of vibration signals collected from gearbox. The extracted feature values are regarded as the feature input vector of WNN. The scale parameters, translation parameters, weight values and threshold values in WNN structure are optimized by traditional back-propagation (BP) algorithm. The pattern recognition and classification abilities of WNN are demonstrated through identification and classification for three gear fault modes with different crack sizes. It can be applied to fault diagnosis of rotating machinery very well.