Extraction of fault feature values for piston compressor based on bispectrum analysis

Vibration signals of running piston compressor ordinarily include cyclical impulsion components. And they have the typical characteristics of super-Gaussian signals. When the compressor has weak local faults, it probably causes energy and phase changes of the vibration signals. However, the classic analytical methods which base on low-order statistics are often difficult to reflect the changes caused by weak faults, and they are unable to accurately extract fault feature information. So they are not easy to accurately classify faults. This paper propose to use bispectrum which belongs to high-order statistical analysis (HOSA) to analyze vibration signals of the piston compressor, suppress the signal Gaussian random noise of the signals, highlight the vibration changes caused by weak faults, and extract fault feature value from the energy which is from sensitive regions of bispectrum. Through vibration signal analysis of three different types of piston compressor faults, it proves that using bispectrum analysis method can show the differences among various types of vibration failures more clearly. It has high sensitivity when we distinguish failures. Meanwhile bispectrum analysis can effectively suppress Gaussian noise from the measurement vibration signals, and provide more effective characteristic parameters for automatic fault diagnosis.