Research On Time-frequency Analysis Technology Of Full Information Fusion For Rotating Machinery Fault

The structure and function of rotating machinery are becoming more and more complex,and the requirements for the safety and stability of the mechanical system are becoming more and more strict.Therefore,the fault diagnosis of rotating machinery has become a research hotspot in the field of process monitoring.When the rotating machine fails,it will produce dynamic non-stationary signals.Time-frequency analysis technology can effectively process non-stationary signals,extract the internal information of complex mechanical data,which is more advanced than using stable sine wave decomposition signal in spectrum analysis.The quality of feature extraction of vibration signal determines the validity of diagnosis and prediction methods in the online monitoring system.The early fault signal of rotating machinery is weak and easy to be interfered by environmental noise and other structural components,so it is difficult to extract fault features.Aiming at the limitations of incomplete and inaccurate vibration data of single channel signal,based on the full vector spectrum theory of full information fusion,this paper studies the feature extraction of weak fault under strong background noise by time-frequency analysis technology,and solves the problem of fault sensitive feature extraction and data fusion of non-stationary multi-channel signal.The main work and research results are as follows:(1)The fault diagnosis algorithm based on improved harmonic wavelet and fractal is studied.The signal of rotating machinery is processed by improved harmonic wavelet with Gaussian envelope,and the weak fault feature is extracted by G-P correlation dimension.The validity of the algorithm is verified by the simulation experiment.The data collected on the rotor experimental platform is calculated and the correlation dimension after the improved harmonic wavelet processing can identify the fault very well,with high fidelity and good stability,which is superior to the traditional correlation dimension algorithm and the harmonic wavelet fractal algorithm.(2)The fault diagnosis algorithm of full vector frequency band entropy(FV-FBE)is studied.The short-time Fourier transform(STFT)is used to calculate the frequency band entropy(FBE).According to the minimum principle of FBE,the band-pass filter bandwidth and center frequency of dual channel signal are designed adaptively.The filtered dual channel signal is demodulated with full vector Hilbert envelope to obtain full vector envelope spectrum for fault diagnosis.The research shows that FV-FBE algorithm can reduce the low-frequency discrete noise components and extract the weak fault features comprehensively.(3)The fault diagnosis algorithm of full vector AR spectral kurtosis is studied.The auto regression model(AR)is used to pre-whiten the bearing signal,which retains the transient impact and steady-state noise of the signal.The adaptive band-pass filter based on the improved harmonic wavelet is used to design the spectral kurtosis.The filtered dual channel signal is demodulated by the full vector Hilbert envelope,and the full vector envelope spectrum is obtained for bearing fault diagnosis.The research shows that the full vector AR spectral kurtosis algorithm can effectively extract the weak fault of bearing,which is better than the traditional spectral kurtosis algorithm.