Study On Early Fault Feature Extraction Of Rotating Machinery In Heavy Noise

Heavy rotating machinery occupies an important role in modern industry.Its failure will lead to catastrophic accidents and huge economic loss.Since vibration signals carry the key information of faults,how to extract useful features from vibration signals have been a hot topic.When the mechanical rotating part is damaged,its vibration signals are non-linear,non-stationary and non-gaussian,and the early fault features are weak,strong modulation and large background noise,which make the fault features submerged in the strong background noise and it difficult to extract.In this paper,weak fault feature extraction in strong noise background is studied.Singular value decomposition is widely used in signal denoising.However,for the weak fault signal under the background of strong noise,the fault feature information submerged in the background noise makes it difficult for SVD to obtain the ideal noise reduction effect.This paper proposes a new method to determine the number of effective singular values,which can effectively remove the noise and retain the weak fault feature information in the signal to the greatest extent.Simulation results show that the proposed method is effective in the treatment of weak frequency components in the background of strong noise.As a new time-frequency analysis method,variational mode decomposition has better noise robustness and frequency-domain resolution.It has unique advantages for non-stationary signal processing and has been widely used in bearing fault signal processing.For the problem that decomposition number in VMD needs to be set in advance,this paper proposes to determine the optimal decomposition parameter by the ratio of the total signal energy obtained by decomposition to the original signal energy.And because VMD is essentially a wiener filter,it is difficult to identify the FM components in the signal.For this,the correlation coefficients between the components of adjacent center frequency are used to determine whether they belong to the same frequency modulation component.The FM components in the signal are restored by combining the components which the correlation coefficients are larger than the threshold.The effectiveness of the proposed method is verified by numerical simulation and experimental signals.The typical feature of the vibration response caused by rub-impact fault is that in addition to rotational frequency,some sub-harmonic and super-harmonic components are introduced in the vibration response,and the IF of the fundamental harmonic will periodically fluctuate around the rotation frequency,and the fluctuating frequency of the IF is coincident with the harmonic frequency of the vibration signal.But in the early stage of the rubbing,the intensity of the harmonics are small and the amplitude of the vibration response is weak.Moreover,the vibration signals may contain strong background noise which can completely overwhelm the weak features of the early fault.To extract these fault features accurately,an improved multisynchrosqueezing transform method is proposed in this paper,by iteratively updating the reassignment operator,the number of operations needed to reach the highest concentration degree were reduced.Furthermore,this method is applied to high order instantaneous frequency estimate.By detecting the ridge of the spectrogram,the harmonic components of the signals can be obtained,so as to realize the diagnosis of early rub-impact faults.