Research On Fault Feature Extraction Of Non-stationary Signal Based On HHT

Feature extraction is one of the most important and difficult topic in the fault diagnosis of mechanical equipments. It is directly related to the accuracy of fault diagnosis and reliability of early fault prediction. The fault signals of mechanical equipment are generally non-stationary signals, however, frequency characters of fault signal adopting vibration analysis are main information of fault equipment. Traditional fault feature extraction methods based on the assumption of linear and station coundn't accurately time-frequency various feature of non-stationary signals. So, this paper presents Hilbert-Huang transform (HHT) to extract fault feature of non-stationary signals for mechanical equipment, the method is a new method to deal with nonlinear and non-stationary signals.Three topics are mainly researched in the paper. At first, it coud be found that HHT has tremendous advantanges in dealing with non-stationary signals through simulation study with traditional time-frequency analysis (short time Fourier transform, Wigner-Ville distribution and wavelet transform). Then, three factors could be found that they affect effects of HHT though analysing process of HHT. These factors are construction method of envelopment, stopping criterior of sifting process and Hilbert transform. Then the construction method of envelopment is researched, and the piece cubic Hermite interpolation polynomial (PCHIP) adopted to construct envelopment is presented. By simulation comparision, the improved method is more suited to analyze strong non-stationary signals. The above methods are used to analyze measured vibration data of four conditions for reciprocating compressor respectively, it can be confirmed that HHT based on PCHIP could correctly extract fault featrue of vibration signal of gas valves, which offers a new method for fault diagnosis of gas valves for reciprocating compressor. Finally, a new method of combining EMD with auto-correlation function analysis is presented in the paper could be confirmed to accurately extract amplitude and frequency of weak sinusoidal signal through simulation studies, which affords a new method for extracting weak period signasl submerged in strong background noise.