Research On Train Rolling Element Bearing Fault Diagnosis Technology Based On Wayside Acoustic Signal

In this thesis, taking the condition monitoring and fault diagnosis of the train rolling bearings as research objectives and the wayside acoustic signals as research objects, in-depth theoretical and applied researches have been implemented aiming at the field of the wayside acoustic signal condition monitoring and fault diagnosis, such as the issues of the distortion of signal spectrum, the extraction of weak signal in mixed signals, and noise reduction of the target signal under strong noise background. These related issues described above in the field of train rolling bearing acoustic signal fault diagnosis have been studied deeply by taking four non-stationary signal processing methods including wavelet transform, the independent component analysis (ICA), the ensemble empirical mode decomposition (EEMD) and spectral kurtosis theory, as well as the resampling and variable sampling technologies.First, from the perspectives of kinematic geometry and acoustics model,the essential reasons for spectrum distortion of the acquired signal with Doppler effect were studied.With the introduction of the resampling technique, an equal-interval resampling method based on the curvature of the frequency offset was proposed for correction of the sampled signal spectrum.Adopting the idea of microelement approximation, correction of the spectrum distortion was achieved, and the application scope of this method in the spectrum correction was discussed. For accurate recovery of the sampled signal spectrum, variable sampling techniques were firstly proposed based on the reasons of spectral distortion, where a point-by-point interpolation technique was used to restore the original signals. Simulated and experimental signals were adopted to verify the effectiveness of these two algorithms, and the results showed that both two algorithms could effectively solve the structural distortion of the spectrum caused by the Doppler effect in respective scope.Secondly, aiming at the facts that the acquired wayside acoustic signals are the mixture of multiple components and the acoustic signal of rolling bearings is extremely weak, the ICA technique and the fast ICA algorithm (FastICA) were introduced. The theory of the constrained ICA was studied, and two following key issues of this theory were discussed:the reference principles of reference signal and the threshold selection. Aiming at the underdetermined problem of the ICA that the number of the sources was unable to be determined, the wavelet transform theory was studied and the redundancy of the continuous wavelet transform (CWT) was analyzed. The method of equal-interval selection of the pseudo center frequency in the main energy region of the frequency domain was proposed to determine the scale of the wavelet transform. It not only solved the correlation of the signals between each other, but also greatly reduces the decomposition scale. The method based on combination of the CWT to select wavelet scales with iso-interval pseudo-center-frequency and the constraint ICA was proposed to extract the weak component from the mixed signals. The verification results of the experimental data fully demonstrated the effectiveness of the method in the extraction of weak shock signal.Finally, spectral kurtosis theory was applied in fault diagnosis of the acoustic signals. The sensitivity of spectral kurtosis method was studied for the weak shock signal in the strong noise background. The performance of noise reduction for the kurtogram algorithm was verified with instances, and the results showed that the noise reduction effect of this algorithm was remarkable. Aiming at the feature that the wayside acoustic signal energy of rolling bearings was weak, two kinds of quadratic filtering algorithms based on the spectral kurtosis were proposed to enhance the signal. One is EEMD algorithm of quadratic noise filtering.The characteristics of the EEMD algorithm in noise reduction were studied. And the policy issue of the noise selection was analyzed. The other quadratic filtering algorithm based on the combination of the CWT and ICA was proposed. According to the characteristics of the quadratic filtering, the basic function of the CWT and the selecting principle of the decomposing scale were proposed. Adapting the failure acoustic signal of the rolling bearing, the quadratic filtering effects of the two algorithms were verified, and the experimental results showed that the fault signal-to-noise ratios of these two quadratic filtering algorithms were significantly improved.In addition, the studies of the thesis are all achieved on the basis of experimental verification. The Doppler effect experimental platform was used to achieve the study of correcting signal spectrum distortion of the wayside acoustic signal caused by the Doppler effect. The acoustic signal based train rolling bearing fault test platform was used to achieve the study of the acoustic signal of the train rolling bearing under different conditions. The experimental results verified the feasibility and effectiveness of the above methods.