Study On Improvements Of Local Mean Decomposition And Its Application For Gearbox Fault Diagnosis

As a kind of key equipment focusing on changing speed and transmitting power, gearbox has been widely used in various fields of industry, whose importance is self-evident. At the same time, it is difficult to estimate the loss caused by the fault(s) of gearbox. Therefore, the research on accurate fault diagnosis of gearbox is useful for improving the security and reliability of the equipment. When the fault occurs, the collected vibration signal contains many vibration sources, and its amplitude and frequency are modulated by the fault impact, which makes the signal nonlinear and non-stationary and accurately extracting the fault features extremely difficult. How to extract the fault features from a complicated vibration signal is to be addressed for accurate fault diagnosis in this research.Many advanced signal processing methods are available for this purpose. Among these methods, local mean decomposition(LMD) is an adaptive time-frequency analysis method, which is suitable for processing nonlinear and nonstationary signals, and thus it is suitable for analyzing vibration signals collected from gearbox. LMD can decompose a multi-component signal into a series of product functions, each of which corresponds to one of the components in the original signal. The analysis on the signal component of interest can be conducted and thus the effects of noise and other disturbances can be avoided. However, there are still some open problems in LMD to be solved, such as the end effects, the selection of sensitive components, and the influence caused by heavy noise. In order to improve the performance of LMD, the problems mentioned above are studied in this research and the main work is as follows:(1) In order to select an appropriate time-frequency method to analyze vibration signals collected from the gearbox, some commonly used signal processing methods, including short time Fourier transform, wavelet transform, Wigner-Ville distribution, Hilbert-Huang transform, and local mean decomposition are compared. Based on their advantages and limitations, the LMD method is selected as the tool of signal processing in this thesis.(2) In order to suppress the distortion caused by the end effects, an adaptive waveform extension based on spectral similarity is proposed. By means of adaptively identifying the waveform that has the most similar spectrum with one end of the signal, this method can naturally extend the original temporal waveform, so that the extended signal can conform to the time-frequency characteristics of the original signal. Accordingly, the problem of the end effects in the normal LMD can be suppressed.(3) Aiming to select sensitive components for subsequent fault diagnosis using the signal components obtained by the improved LMD method, this thesis proposes a sensitive component extraction method based on multiple feature indicators. These indicators in time and frequency domains are used to describe the characteristics of sensitive signal components, so that they can be easily separated from the signal components irrelevant to the fault diagnosis. This method can obtain relatively stable results, superior to those obtained with only a single feature indicator, and thus the adaptive selection of sensitive components is realized.(4) To avoid the influence caused by heavy background noise, an adaptive band-pass filter is designed in this thesis. The optimal parameters, i.e. the center and corresponding band width of the band-pass filter, are determined by the adaptive window superposition in frequency domain, so that the fixed window width in the traditional kurtogram can be adjusted adaptively. Combining with the improved LMD method, the feature signal for fault diagnosis can be accurately extracted from a noisy vibration signal.The results of simulations and experiments demonstrate that, compared with existing methods, the methods proposed in this thesis are effective and applicable to signal processing and fault diagnosis of gearboxes. The proposed methods and results can be viewed as a valuable reference for the application and development of fault diagnosis theory and methodology.