Research On Several Issues Of Wavelet Transform In Rotor Fault Diagnosis

Rotor is the uppermost component of rotating machinery, which is widely used in petroleum, chemical, mechanical, metallurgy and power field, so more attentions have been paid to rotor fault diagnosis. With the development of science and technology, experts home and abroad posed a lot of new methods and new techniques of fault diagnosis, which were applied to engineering. Wavelet transform is the forefront technology of signal processing. This dissertation conducts profound discussion and analysis of the wavelet transformation's application in rotor fault diagnosis.1). Based on studying large number of documents, this dissertation introduces the meaning, content, existing condition and tendency home and abroad of rotating machinery fault diagnosis and summarizes the common methods in fault diagnosis. It introduces the fundamental theory of wavelet, including wavelet transform, MRA(multi-resolution analysis), Mallat algorithm, wavelet packet theory.2). Four operational conditions simulated on the experimental instruments, known as Bently RK4, have been got, including normal, unbalance, radial friction and oil whirl, attaining vibrational signal under different constant velocity conditions, which provide applicable experimental data for further wavelet transform research.3). The dissertation analyzes the selection of wavelet function, the comparison of Daubechies wavelet series, and finally selects db3 wavelet function and verifies by experimental data and achieves a relatively sound inspectional efficiency.4). The dissertation researches wavelet denoising, conducts analysis, comparison and selection of the denoising threshold, applies heuristic threshold method within wavelet soft threshold and conducts denoising processing of noise-affected rotor signals and experimental vibrant signal of the gas turbine, and finally gets a compatible result.5). The dissertation analyzes the wavelet singularity inspection, respectively conducts the inspection on the signal abrupt extent and position, and deduces Lipschitz exponential solution-finding method and thought.6). Based on wavelet packet decomposition, the dissertation acquires the signal features. It obtains the signal features by means of wavelet packet decomposition of data, collected on the experimental platform, proposes the flexible combination thought of wavelet packets, which is closer to actual situation than traditional frequency equal partition method.7). The dissertation researches the fault diagnosis method combined by wavelet packet and neural network, constructing loose wavelet neural network, conducting training and testing of the network, providing the network with the function of rotor fault catalog. Such method boasts excellent engineering applicable value.8). The dissertation researches on programming technique combining VB and Matlab, which has been introduced to rotor fault diagnosis system. It introduces the merit, principle and method of combined programming, based on which, the rotor fault diagnosis software has been programmed, resolving the technique difficult points, and realizing the functions of wavelet denoising, wavelet inspection, wavelet packet feature acquirement, BP network fault diagnosis. Such technique significantly enhances the engineering development efficiency.