Study of Autoregressive (AR) Spectrum Estimation Algorithm for Vibration Signals of Industrial Steam Turbines

Spectral analysis of the vibration signals of industrial steam turbines provides efficient reference for the characterization and discrimination of turbine faults. Conventional power spectrum estimation methods often exhibit contradiction between variance performance and resolution, leading to poor estimation results. In this study, we investigated Levision-Durbin recursive algorithm, Burg algorithm and periodogram power spectrum estimation algorithm, and also chose Akalke Information Criterion (AIC) to identify the optimal order p. Based on MATLAB, we wrote a simulation program for Autoregressive (AR) spectrum estimation algorithm and designed a graphic user interface, formulating the AR spectrum estimation algorithm program for vibration signals of industrial steam turbines. After field measurement of a steam turbine with sampling number of 400 and frequency of 256Hz, as well as order of 10 and 80, simulation was performed. It was demonstrated that AIC provides efficient reference for the identification of proper order. With the optimal order, AR spectrum estimation algorithm produces good variance performance and resolution, providing reference for the spectral analysis of vibration signals of industrial steam turbines.