Research On Weak Signal Feature Extraction Of Mechanical Equipment In Heavy Noise

The bad scene environment, changing working condition and nonlinear characteristics ofthe production equipment, making the measured vibration signals tend to have a characteristicof strong background noise, nonlinear, and early weak fault information. Due to thesecharacteristics, it is disadvantaged to extract the incipient fault of mechanical equipmentaccurately. Aiming at the incipient fault feature submerged in heavy noise, this thesis focuseson the method which based on the singular value decomposition (SVD) and the stochasticresonance (SR) technology of weak signal extraction and detection in heavy noise.SVD as a nonlinear filter method is widely used for signal de-noising and detection;however, facing to the incipient fault feature submerged in heavy noise, the de-noising abilityof single SVD system is limited. For this reason, a novel integrated of incipient faultdiagnosis method is presented based on the principles of Local mean decomposition (LMD)and difference spectrum of singular value. Original nonlinear signals are decomposed byLMD and a group of Product functions (PFs) are obtained; however, it is difficult to extractfault frequencies in strong noise. In order to identify the fault pattern, a Hankel matrix of a PFwas constructed and decomposed with SVD. Accordingly, difference spectrum of singularvalues is determined. On the basis of difference spectrum theory, the number of helpfulsingular values can be selected correctly; filtered signal of some PFs component was analyzedby using envelope spectrum. The effectiveness of the proposed methodology wasdemonstrated with the simulated data and the bearing inner-race defective signal measured ofinner races.When detecting a weak signal submerged in strong noise through the SR, adjusting thesystem parameters and noise intensity can be conducted to make SR do well. Consequently, amethod based on SVD used to adjust the noise intensity of SR is put forward. In this method,the original signal is preprocessed and reconstructed by means of SVD, and then we search for a component signal. In the component signal, the components of the characteristic signalmatch noise strength. Then the component signal is processed with the non-linear bistablesystem to obtain SR response. The measured vibration signals from Engineering is usuallygreater frequency signal containing many frequency components and direct currentcomponents. These components have a negative influence to development of SR. For thisprolem, a method enhanced weak signal detection based on Ensemble empirical modedecomposition (EEMD) and Cascaded bistable stochastic resonance (CBSR) is designed.Firstly, an original signal is decomposed by EEMD, using the high-pass filter characteristicsof the EEMD, The IMF components and false components smaller than the characteristicfrequency are weeded out, then the remaining IMFs are used to create a synthesized signalafter processed by Elliptic high-pass filter, by processing the synthesized signal with CBSRsystem, characteristic frequency is acquired. Simulation signal and bearing inner-racedefective signal are analyzed; the results show that the proposed method is effective forreducing low-frequency noise and direct current components effect to stochastic resonance.Aiming at the detection problem of the weak signal in heavy noise, a method enhancedweak signal detection based on cascaded piecewise-linear stochastic resonance is proposed. Inthis scheme, a new model of piecewise-linear stochastic resonance that avoids the saturationphenomenon in classical bistable system when detecting a weak signal, is applied, and thusthe cascaded piecewise-linear stochastic resonance are selected, which can not only removehigh frequency noise efficiently but also enhance the energy of low frequency signals.Simulation signal and bearing inner-race defective signal are analyzed; the results show thatthe proposed method is effective for raising the signal-to-noise ratio. Compared with cascadedbistable system, it is more convenient to adjust less correlated parameters.