Fault recognition of large steam turbine based on higher order spectral features of vibration signals

Abstract

Shaft vibration of large steam turbine has a characteristic of sub-Gaussian signal with predominant components of rotating speed and its harmonics. Most of faults occurred on shaft indicate the change of harmonics with nonlinear coupling reciprocity each other, that makes it difficulty to identify the fault source and extract the fault feature from vibration signals by means of power spectral analysis based on second-order statistical analysis. This paper took an unstable vibration phenomenon with half frequency characteristic of a large steam turbine as an example, and made use of higher order statistics analysis (HOSA) to determine the fault with half frequency characteristics. By comparing the bispectrum and 1(1/2) dimensional spectrum of vibration signals under stable condition with unstable condition, the nonlinear harmonic coupling characteristics of the half frequency component was determined. A method for fault feature extraction was proposed by using the corresponding component in bispectral marginal spectrum and in 1(1/2) dimensional spectrum as feature values to monitor the trend of unstable vibration. And fault recognition and classification were made according to the Fisher criterion. The results show that these feature extraction methods of quadratic phase coupling can clearly reveal the great change caused by abnormal vibration of steam turbine, and reveal the non-Gaussian nonlinear characteristics of vibration signals. The characteristic values are quite sensitive to faults, and they can effectively restrain the Gaussian noise in vibration signals. So they are very suitable for automatic fault diagnosis.