Detecting and Predicting Early Faults of Complex Rotating Machinery Based on Cyclostationary Time Series Model

Abstract

Vibration signals of complex rotating machinery are often cyclostationary, so in this paper one novel method is proposed to detect and predict early faults based on the linear (almost) periodically time-varying autoregressive (LPTV-AR) model. At first the algorithms of identifying model parameters and order are presented using the higher-order cyclic-cumulant, which can suppress additive stationary noises and improve the signal to noise ratio (SNR). Then numerical simulations are done and the results indicate that this model is more effective for cyclostationary signals than the classical AR model. In the end the proposed method is used for detecting incipient gear crack fault in a helicopter gearbox. The results demonstrate that the approach can be used to detect and predict early faults of complex rotating machinery by the kurtosis of the residual signal.