Application Research On New Methods Of Nonstationary Signal Feature Extraction And Diagnosis Techniques

The dynamic signals of mechanical equipment often possess nonstationaritiesdue to occurrence of fault, variance of operation and inherent nonlinearity ofequipment. If the nonstationary signals are effectively denoised, decomposed anddemodulated, the useful components are extracted, and the interference from othercomponents are restrained, the signal to noise ratio and the quality of diagnosticinformation will be improved. Aiming at rotating machinery and reciprocatingmachinery, this research focuses on practical diagnosis techniques based on principleof nonstationary signal feature extraction.Aiming at the problems that adiabatic elimination stochastic resonance in smallparameters can not detect weak signal in large parameters in engineering applications,and present researches mostly focus on single frequency, a new numerical methodcalled the step-changed stochastic resonance is proposed in this thesis. The newmethod can detect the multi-frequency weak signal in large parameters which isoverwhelmed in heavy noise efficiently. And when it is used as the pretreatment of thewavelet analysis, it also can decrease the weak signals' distortion induced by theheavy noise in wavelet analysis, and improve the reliability of the wavelet analysis inweak signal detection under the condition of low signal to noise ratio.Empirical mode decomposition (EMD) method is a new tool for the nonlinearand nonstationary signals analysis. It is based on the local feature of the signals, andcan adaptively decompose signals into several intrinsic mode functions (IMFs)according to its characteristic time scale. This thesis discusses the EMD-basedtime-frequency analysis method's theory and basic algorithm. After the signal isdecomposed, the IMFs can be used to calculate the Hilbert spectrum to gain theinstantaneous frequency, and show the exactly changes of the frequency. Finally, thesignal is expressed by the energy distributing on the time-frequency plane in the formof Hilbert time-frequency spectrum.The nonstationarities of the vibration signals of rotating machinery usuallybehaves as its statistical feature is seasonal cyclostationary. Based on the analysis ofcyclic autocorrelation demodulation and its performance, the method of EMD-basedcyclic autocorrelation demodulation in specific frequency band is proposed. Themethod can decrease cross-term interference caused by multi-modulator andmulti-carrier and improve the reliability of the analysis. As the application of theEMD-based cyclic autocorrelation demodulation theory in fault diagnosis, the methodis used to extract fault feature from vibration signals of rolling element bearingsuccessfully.As the carrier of the key technology, some practical techniques of thedevelopment of the vibration signal analysis system based on LabVIEW aresummarized. A simple effective modification algorithm for the vibration signalsintegration in frequency domain is proposed. This algorithm provides the technicalsupport for the integrality and accuracy of equipment dynamic information.