Gear Fault Diagnosis And Identification Based On Higher-order Spectrum

In mechanical condition monitoring and fault diagnosis, feature extraction is the most important issue. In recently years, non-linear, non-stationary, non-Gaussian signal processing techniques attracted more and more attention to meet the needs of accurate diagnosis faults in mechaninal fault feature extraction. The ordinary methods of feature extraction usually hypothesize the vibration signals have the characters of stationary and Gaussian. But the actual signals which we get in practise usually are non-stationary and non-Gaussian distribution, specially when the fault occurs. Higher-order statistical signal processing technique is a powerful tool in processing non-linear and non-Gaussian signals, and it can remove the Gaussian noise from the ordinary signal effectively, and can reserve the phase information and reflect the energy of the signal.For non-linear issue, power spectrum and other traditional signal processing techniques are diffcult to solve these problems ultimately. Higher-order statistical signal processing technique is a useful tool which characterizes the random signal from the higher probability and covers the shortage that the second-order statistics (power spectrum) can't contain the phase information, and can quantitatively describe the non-linear phase couple. Higher-order spectrum has the powerful noise immunization capability, and can remove the Gaussian noise entirely in theory. Generally, the noise in mechanical vibration can be treated as Gaussian noise approximately. So, it is easy to extract the fault feature from the vibration signals with higher-order spectrum. The bispectrum and the bicepstrum are introduced into gear fault diagnosis in this paper, and they are new ways to solve these issues.The main contents are as follows:(1) It presented the HOS theory briefly, and studied the bispectrum and the bicepstrum analysis method, and their algorithms, characteristics, physical significances, and then put the bispectrum into the different gear fault diagnosis situations.(2) Put the bicepstrum into the gear fault diagnosis, and compare with the result of the cepstrum.(3) Associated with the characteristics of the bispectrum, bispectrum-BP neural network was introduced, and with the fault gear signals which picked on the Gear fault experimental table, it has been verified its feasibility and effectibility.Experimental results showed that bispectrum and bicepstrum were feasible and effective for gear fault diagnosis.