Generalized Autocorrelation Method for Fault Detection Under Varying-Speed Working Conditions

Abstract

Rotating machines often operate under varying-speed working conditions due to the demands of diverse applications and complicated functionalities. Vibration signals of their key components, such as bearings, are analyzed to detect an early fault in case of unexpected failures. Order tracking is a widely used method of handling the nonstationarity of varying-speed vibration signals. However, signals after order tracking are not strictly cyclostationary, i.e., they exhibit accumulated cyclic disturbances easily masked by strong background noises when the bearing fault is at an early stage. Detecting the bearing fault from order-tracked signals with both accumulated cyclic disturbances and strong background noises is very challenging because cyclic disturbances and background noises are two mutually constraining characteristics. Herein, a generalized autocorrelation function (GeACF) is proposed to address this challenge. The autocorrelation function (ACF) is proven to be a special case of the proposed GeACF, which provides more options than ACF for order-tracked signals contaminated with strong background noises. By tuning a free parameter, GeACF can achieve a balanced tradeoff between these two mutually constraining characteristics. The superior performance of the GeACF method over its conventional counterparts is demonstrated through simulation and case studies.