Research On Rotating Machinery Fault Feature Subset Selection Based On The Full Vector Spectrum

Large-scale rotating machinery is often the key equipments in the enterprise, whose main work body is the rotor and other rotating parts. In the petroleum, chemical, metallurgy, power generation and other large and medium-sized enterprises the rotating machinery accounts for about the proportion of 80% of all the machines, including compressors, blowers, steam turbines, generators, rolling machines. As the core device it will cause immeasurable damage and harm to the enterprises and even rob people's lives property in the event of breakdown. So the large-scale rotating machinery's running monitoring and its fault's timely diagnosis are the main concerned themes to the scientists and the technical workers.The rotating machinery running condition monitoring and fault diagnosis base on the useful information which comes from the diagnosed object, such as vibration, noise, speed, temperature, pressure, flow and so on. Between them, the vibration signal contains lots of information, which can help people monitor the device's running condition and diagnose the fault correctly. Fault feature subset selection is the analysis and treatment of the information which comes from the pretreatment dynamic signal, then select the data related to the system status, and handle and diagnose the data next to extract the sensitive parameters related to the system status closely. Extracting the valid feature vector is not only a key procedure in fault diagnosis but also the key factor in putting forward the correct fault diagnosis.With regard to the defect of incomplete and poor real-time of the traditional single-channel data acquisition method, this text combines the full vector spectrum with the rough set theory and wavelet analysis method respectively, putting forward the research on rotating machinery fault feature subset selection based on the full vector spectrum.The full vector spectrum analysis technique bases on rotating machinery homology information fusion which is the common name of vector spectrum analysis and series of extension analysis methods. It can fuse the two or three channels of information from the same section of the rotor. Therefore it not only offsets the deficiency of single channel method:insufficient information, easy-misjudgment and so on but also be compatible with the traditional single channel method. Besides, it has the advantage of high resolution, three-dimensional analysis, the feasibility of instruction of the rotor's vibration intensity and so on.The rough set theory was proposed in the early 1980s by Z.pawlak, it is a kind of mathematical tool processing the fuzziness and the uncertainties. Its basic idea is to achieve knowledge discovery roughly through the rules of dividing database classification and equivalence relations. It has advantage in data reduction. Wavelet analysis is an extension of Fourier analysis and was proposed in 1984. It overcomes the defect of traditional Fourier transform:only consider the sine form vibration energy but not other modes of energy. Besides, it puts low requirement on the input signal and has the good feature of high sensitivity, strong ability to resist noise. Wavelet transform has advantage of time-frequency localization and signal adaptive zoom feature, the ability of multi-resolution analysis which can conduct the signal at different band. It not only extract the characteristics of each band but also retains the corresponding time-frequency characteristics. In all, using wavelet analysis technology is more effective in rotating machinery fault feature subset selection.This article discusses the basic principles and algorithms of full vector spectrum, and combines it with rough set theory and wavelet analysis technique respectively, then puts forward research on rotating machinery fault spectrum feature subset selection based on full vector spectrum-rough set and the comparison study between wavelet-envelope analysis and full vector wavelet analysis in rolling bearing fault feature extraction. In the end, it performs these functions with MATLAB and carries out related experiment to verify it.