Research On Rotor System Rub-impact Fault Diagnosis Technology Based On Wavelet Transform

Rub-impact fault between rotor and stator in turbo-generator machinery is a frequent malfunction. With the development of turbo-generator machinery to the direction of higher performance and higher efficiency, the clearance between rotor and stator is reduced and the possibility of rub-impact fault is increased. Rub-impact fault will cause the rotor system happen complicated nonlinear vibration, which is the most important reason why the rotor system occurs instability. Slight rub-impact fault will induce strong vibration and heavy rub-impact fault will induce shaft permanent bow, even it will destroy the entire shaft. So the location of rub-impact fault and feature extraction is essential to the early fault diagnosis like this sort of malfunction. Based on the former researches, the rub-impact signal has been researched, including feature extraction of rub-impact and noise reduction, on the basis of the wavelet analysis tool and Independent Component Analysis (ICA) algorithm. Works of the dissertation are briefly summarized below:(1) The experimental investigation of rub-impact between rotor and stator is performed. Under three circumstances, that is, no rub-impact, slight part rub-impact and heavy part rub-impact, the vibration signals of rotor are gathered. This paper describes the feature of the time-base and the spectrums and the center orbits under above three circumstances. Considering the limitation of FFT method, which can't analyse the unstable and nonlinear vibration signal, the wavelet analysis method is proposed on the basis of the relation between the singularity and the wavelet module maximum to detect the location of the singular points.(2) The traditional methods can't denoise the noise from the rub-impact signal because of the frequency band aliasing phenomenon. The wavelet analysis is introduced and the maximum modulus method is used to remove the interference of random noise based on the different propagation characteristics. The de-noising result indicates that the wavelet method can detect the location of the singular points.(3) As the noise signal is intensive, the wavelet transform module maximum of rub-impact signal will be buried in the random noise signal and the wavelet module maximum method will fail in this circumstance. The fast ICA algorithm is proposed to separate the rub-impact signal from the noise signal. The de-noising result indicates that the fast ICA algorithm is effective to remove the noises and detect the location of the singular points.