Hybrid Time-Frequency Analysis Methods For Mechanical Fault Feature Extraction

The most important and crucial problem in the mechanical fault diagnosis is the feature extraction method of the fault characteristic signal, while it is the very problem most difficult to solve. Because of the complexity of the dynamic signal and the multidisciplinary cross and fusion characteristic of the extracted signal, the feature extraction method has been the most important research direction concerned by the researchers, in which signal de-noising, cross-terms suppressing of time-frequency distribution, instantaneous frequency feature extraction are the main elements. Make use of the non-stationary and non-gaussian signal processing methods merits such as Wavelet Transform (WT), Wigner-Ville Distribution (WVD) and Hilbert-Huang Transform (HHT), the hybrid time-frequency analysis methods for mechanical fault feature extraction are put forward, and this dissertation investigates noise reducing method of the machinery vibration signal and the feature extraction technique of the mechanical equipment thoroughly. Its contributions list as follows:(1) In terms of signal de-noising, a new method of filtering and de-noising based on optimal Morlet wavelet and Singular Value Decomposition (SVD) is put forward.Feature extraction and signal de-noising of the fault signal have been the most important investigation in the signal processing. By reducing the noise of machinery vibration signal, the mechanical fault information can be obtained effectively. This paper analyses the traditional wavelet de-noising method and the filter characteristic of wavelet transform, proposes a new method of filtering and de-noising based on the Morlet wavelet and SVD by using wavelet transform, singular value decomposition technology and the fine time-frequency characteristic of Morlet wavelet. The new de-noising method, which possesses better transient information extraction ability, could reduce the noise and extract the period of the signal effectively and assure the validity of the fault feature recognition.(2) In terms of characteristic improving based on time-frequency distribution, a new feature extraction method based on Adaptive Short-Time Fourier Transform (ASTFT), which could restrain the WVD cross-terms effectively, is put forward. Time-frequency analysis has widely been used for fault feature extraction, while Wigner-Ville distribution based time-frequency method has a supreme defect that there exists interfere of the cross-terms. After investigating the cause of WVD cross-terms, the correlation between auto-terms and cross-terms and the kernel function and component combination for cross-terms suppression, this paper proposes a new fault feature extraction method based on ASTFT to suppress the cross-terms of WVD effectively. Using on bearing fault feature extraction, the new method suppress the WVD cross-terms, reserve the fine characteristics of WVD, and provide an effective analyses method for fault diagnosis.(3) In terms of transient characteristic improving, a new transient characteristic extraction method based on Wavelet Packet transform (WPT) and Hilbert-Huang transform was put forward. After investigating the theory of WPT and HHT, the WPT based de-noising method and the integrity and approximate orthogonality of the EMD, this paper propose a new transient characteristic extraction method based on Wavelet Packet transform and Hilbert-Huang transform, which could eliminate the Modal Mixture of the EMD, decrease the noise interference and the computation of EMD process. The new method is propitious to the fault feature analysis and extraction.(4) In terms of HHT improving, some new algorithms for curve fitting and boundary processing are proposed. Although, as the new theory for adaptive time-frequency analysis, HHT is considered to be experiential method, it's necessary to be consummated. When curve fitting for envelope or mean with cubic spline interpolation, its easily to come into contact with overshoot and undershoot problems. Aim at solve the disfigurement, the subsection interpolation arithmetic based on B spline curve and mixed interpolation curve arithmetic are proposed. Furthermore, it's the effective technique to avoid the aliasing phenomena. In allusion to boundary effect of HHT, After investigating the hypostasis, peculiarity, advantage and disadvantage for some typical boundary process methods, this paper bring forward the improved extending envelope method and edge extrema-powered method, with the methods, it's effectively restrain the boundary effect, and improves the rationality and veracity of the signal feature extraction with HHT.(5) In terms of system exploitation, successfully explore the non-stationary signal analysis system for mechanical fault feature extraction. Investigating the system structure of the parameter-sharing module software, designed the uniform framework of the system module and the apparatus interface, successfully implemented the non-stationary signal analysis system for mechanical fault feature extraction based hybrid time-frequency analysis methods, and proved to be practical and availability by some project applications. There are the summarization of the article and expectation of the feature extraction technology development in the end of article.