A time series model-based method for gear tooth crack detection and severity assessment under random speed variation

In industry (e.g., wind power), gearboxes often operate under random speed variations. A condition monitoring system is expected to detect faults and assess their severity using vibration signals collected under different speed profiles. A few studies have been reported for condition monitoring of gearboxes under random speed variations, including a novelty diagnostic method and a support vector machine (SVM) based method. However, these methods either are based on the strict assumption that the rotating speed does not vary significantly within a rotating cycle or have the drawback of low classification accuracy. This paper presents a time series model-based method for gear tooth crack detection and severity assessment under random speed variation. Specifically, the rotating speed and phase are considered as covariates in a linear parameter varying autoregression (AR) model for representing impulsive vibration signals. We propose refined B-splines for mapping the dependency between AR coefficients and the rotating phase. The performance of the presented time series model-based method has been validated using laboratory signals. The presented method can assess 93.8% of the tooth crack severity state correctly, which is better than the novelty diagnostic method (74.4%) and SVM-based method (87.7%). (C) 2021 The Author(s). Published by Elsevier Ltd.