Research On Key Technology Of Nonstationary Fault Diagnosis Based On Full Vector Spectrum

When rotating machinery failure, vibration signals showed non-stationary characteristics. Although traditional non-stationary signal analysis method could extract fault feature, the results were incomplete and imprecise because of single data. In order to overcome the limitations due to incomplete and imprecise information, two mutually orthogonal sensors were usually installed on the same section of a rotor in the rotating machine field. The dissertation introduced full vector spectrum technology based on information fusion. Some new basic algorithms of analysis methods were proposed, which involved full vector wavelet analysis, full vector higher-order spectrum analysis and full vector empirical mode decomposition analysis. The methods solved the data fusion of multi-channel nonstable signal, and extracted the fault feature.The research based on the prejects of "Basic research on full vector spectrum technology system construction and fault diagnosis"(National Natural Science Foundation of China, Grant No.50675209), and the "The engineering application research on full vector spectrum and fault diagnosis "(the Outstanding Talent Innovation Foundation of Henan, Grant No.0621000500). The theories and methods of non-stationary signal which are from two different sensors were studied. The primary researches of the dissertation include four parts:The first part is wavelet analysis method based on the full vector spectrum technology and its application in fault diagnosis; the second part is higher order statistics method based on the full vector spectrum and its application in fault diagnosis; the third part is empirical mode decomposition based on full vector spectrum and its application in fault diagnosis; the fourth part is to construct full vector spectrum technology system. The main achievements and precise contents of the above three primary parts are summarized as the follows.(1) Research on full vector wavelet transform and full vector wavelet packet and its application in the fault diagnosis of rotating machines. Aiming at the non-stationary vibrating characteristics of rotating machine, a new method of full vector wavelet packet-envelope was presented by merging vibration signal from two mutually orthogonal sensors. Compared to the new method, the limits of the traditional wavelet packet-envelope and full vector Hilbert envelope analysis method were demonstrated by showing their application to simulated data and actual signals. Data fusion of vibration signals was considered as evidences for the validity of the new technique. Results show that the new approach is more effective. Furthermore, this part discussed the compatibility of the new method.(2) Considering multi-source information on the same section of a rotor, research on vector-bispectrum analysis method. The drawback of using general bispectrum taking individual source information alone is integral to display the nonlinear properties. The analysis conclusions of vector-bispectrum are more complete and reliable. In view of a large number of background noise from engineering equipment, a new method for combining wavelet packet with vector-bispectrum was put forward. Compared to the vector-bispectrum analysis, the proposed method can improve the signal-to-noise ratio.(3) The Wigner trispectrum was studied in the application of non-stationary fault vibration signals. The Wigner trispectrum analysis method based on one single source could cause some fuzzy results because of ignoring information of the relationship between the two sensors. Vector Wigner trispectrum was proposed by combining full vector spectrum technology with Wigner trispectrum, and applied to rotating machine fault diagnosis. Simulation and experimental data show that the proposed method provided a more reliable and complete results.(4) Research on full vector empirical mode decomposition analysis method based on full vector spectrum. To deal with the error caused by different vibrating amplitude, merged data was analyzed by performing full vector EMD method. It fused the double channels information and inherited the superiority of the traditional EMD. For superposition of multi-component signal, the results of full vector empirical mode decomposition were affected by the relative size of signal frequencies. Full vector ensemble empirical mode decomposition based on data fusion was raised to solve the issue of mixed mode in the empirical mode decomposition analysis method. The characteristic frequencies could be clearly obtained. Both simulated data and actual data collected from engineering equipment are used to verify the effectiveness of the proposed method. According to the modulating characteristics of each intrinsic mode function, two analysis methods of full vector EMD demodulation and full vector EEMD demodulation were given.(5) Based on the above theoretical study, the non-stationary fault diagnosis system based on full vector spectrum was constructed. The overall framework of fault diagnosis system based on full vector spectrum was devised. The major functions and technical features of the system were introduced.