Adaptive state detection of gearboxes under varying load conditions based on parametric modelling

Abstract

Conventional vibration monitoring techniques are unable to provide accurate state analysis of a gearbox under varying load condition. This paper proposes a novel technique for state detection of gearbox, which fits a time-varying autoregressive model to the gear motion residual signals applying a noise-adaptive Kalman filter, in the healthy state of the target gear. The optimum autoregressive model order, which provides a compromised model fitting for the healthy gear motion residual signals collected under various load conditions, is determined with the aid of a specific model order selection method proposed in this study. Consequently, a robust statistical measure, which takes the percentage of outliers exceeding the three standard deviation limits is applied to evaluate the state of the target gear, where the standard deviation of autoregressive model residuals takes its maximum in all tested gear motion residual signals for model order selection. The proposed technique is validated using full lifetime vibration data of gearboxes operating from new to failure under four distinct load conditions. The investigated load conditions include: (1) constant load, (2) one jump from 100 to 200% nominal torque level, (3) one jump from 100 to 300% nominal torque level, and (4) constant changed to sinusoidal. In each application, the specific model order selection and comparison of the proposed gear state indicator with three counterparts proposed in recent studies are addressed in detail. The Kolmogorov–Smirnov test is also performed as a complementary statistical analysis. The results show that the proposed technique possesses a highly effective and robust property in the state detection of gearbox, which is independent of varying load condition as well as remarkable stability, early alarm for incipient fault and significant presence of fault effects. The proposed gear state indicator can be directly employed by an on-line maintenance program as a reliable quantitative covariate to schedule optimal maintenance decision for rotating machinery.