Research On Weak Feature Extraction Of Nonstationary Signals Of Rotating Machinery

Vibration signals analysis is one of the most important methods used for feature extraction and fault diagnosis. However, various kinds of factors, such as the change of the environment and the faults from the machine itself, often make the output signals of the running rotating machinery be non-stationary. Usually, these non-stationary signals contain abundant information about machine faults; therefore, it is important to analyse the non-stationary signals. Furthermore, these vibration signals sampled on the spot often contain amount of noise. If the background noise is too heavy, the useful information will be submerged, so the weak feature extraction of nonstationary signals is one of the current research focuses. With this background, based on time-frequency analysis methods of signal analysis including wavelet tansform, Hilbert-Huang transform (HHT), reassigned wavelet scalogram, reassigned Wigner-Ville distribution spectrogram, and denoising technology inchuding singular value decomposition (SVD), morphology filter, three types of weak feature extraction methods of non-stationary signals are proposed. In addition, condition identification is another key issue. Therefore, SVM is researched and introduced into rotating machinery fault diagnosis combined with these weak feature extraction methods. The main research work and conclusions are as follows:①In terms of waveform-based feature extraction, weak feature extraction method based on wavelet transform is researched in detail. A weak feature extraction method based on parameter optimized Morlet wavelet transform is proposed by using excellent filtering and time-frequency characteristics of Morlet wavelet. Minimum Shannon entropy is used to optimize the Morlet wavelet bandwidth parameter in order to achieve match with the impact component. Then, an abrupt information detection method based on the transitional stage of singular curve of wavelet coefficients'matrix is used to choose the appropriate scale for the wavelet transformation. Furthermore, the proposed method is improved upon and a weak feature extraction method with adaptive Morlet wavelet based on scale periodical exponential spectrum (SPE) is proposed. Modified Shannon entropy is used to optimize central frequency and bandwidth parameter of the Morlet wavelet in order to achieve optimal match with the impact component. Then, SPE spectrum obtained by carrying singular value decomposition (SVD) to the matrix is utilized to choose the appropriate scale for the wavelet transform. Simulation and application results show that the proposed method can be effecitively applied for weak feature extraction.②In terms of spectrum-based feature extraction, weak feature extraction method based on reassigned spectrum is discussed in detail. According to the fact that the resolutions of time-frequency of reassigned scalogram cannot simultaneously attain the best and the readability of time-frequency representation of it would also be reduced when there exist strong noise in a signal, a novel method to improve the readability of time-frequency representation of reassigned scalogram based on parameter optimization and SVD is proposed to overcome this shortcoming. The time-bandwidth product (TBP) of the wavelet basis is optimized by Shannon entropy, so the problem that the resolutions of time-frequency of reassigned scalogram cannot simultaneously attain the best is solved. Then, SVD de-noising is applied to the reassigned wavelet scalogram to reduce the influence of the noise. In order to overcome the shortcoming of reassigned Wigner-Ville distribution spectrogram (RWVDS), a weak feature extraction method based on RWVDS and SVD is proposed. The RWVDS is obtained by using the reassigned algorithm to the WVDS. Then, SVD de-noising is applied to the RWVDS to improve the readability of it. In addtion, in order to improve the precision of wavelet ridge extracted by modulus maximum method based on scalogram, a new method for extracting wavelet ridge based on optimal reassigned wavelet scalogram is proposed. The results of simulation and experiment show that the proposed method is feasible and effective for extracting weak feature of mechanical vibration signals with heavy background noise.③In terms of transient feature extraction, weak feature extraction method based on SVD-morphology filter and HHT is investigated in detail. Due to the influence caused by random noises and local strong disturbances embedded in signal on empirical mode decomposition (EMD) results, a novel integrated SVD-morphology filter method is proposed to overcome this shortcoming. And combining with EMD, a weak feature extraction method is proposed. Firstly, reconstruct the original vibration signal in phase space and decompose the attractor track matrix by singular value decomposition, and then select a reasonable order for noise reduction according to the singular curve. Secondly, filter the de-noised signal by morphology filter. Finally, decompose it by EMD to extract the intrinsic mode functions (IMF) for fault feature extraction. Experimental results show that this method could extract weak feature of rolling bearing effectively, reduce decomposition levels and boundary effect of EMD, and improve the timeliness and precision of EMD.④In terms of fault diagnosis, fault diagnosis method based on weak feature extraction and SVM is researched. Using the excellent multi-class classification performance of SVM and the global search capability of optimization parameters of niche genetic algorithm (NGA), combining the weak feature extraction method of adapative Morlet wavelet or SVD-morphology filter-EMD, a fault diagnosis method based on weak feature extraction and NGA-SVM is proposed. The experimental results of rolling bearing demonstrate the proposed diagosis approach is effective.⑤In terms of software development, a weak feature extraction module of nonstaionary signal of rotating machinery is developed successfully, including wavelet-based sub-module,HHT-based sub-module and spectrum-based sub-module. The module based on time-frequency analysis methods is implemented by using object-orientied programming technology and VC++ 6.0. Finally, the weak feature extraction module is applied to engineering applications and proved to be effective and pratical.At the end of the thesis, the summarization of the thesis and expectation of the feature extraction technology development are presented.