Robust vibration-based fault diagnosis of a planetary gearbox under environmental and operational uncertainties

Abstract

Vibration-based fault diagnosis of planetary gearboxes can effectively prevent many undesired failures and thereby reduce the maintenance costs of large-scale engineering systems. However, this type of fault diagnosis is often challenging due to various uncertainties, such as the uncertain operating conditions that affect the vibration characteristics of the gearbox. To cope with the uncertainty-related challenges of vibration-based fault diagnosis of planetary gearboxes, this thesis presents three research thrusts: 1) quantitative definition of the stationary operating condition of a gearbox, 2) data-efficient fault diagnosis using autocorrelation-based time synchronous averaging (ATSA), and 3) tooth-wise fault identification using a health data map (HDmap), without the use of an encoder system. The first research thrust presents a class-wise fault diagnosis methodology to solve the challenges that arise from the uncertain operating conditions of a gearbox. In the proposed method, the operating condition of the gearbox is quantitatively divided into multiple classes in such a way that the vibration signals in each class are homogeneous. The second research thrust presents a data-efficient time synchronous averaging (TSA) method for a planetary gearbox. To enhance the signal-to-noise ratio, conventional TSA for a planetary gearbox extracts the vibration signals using a narrow-range window function, which requires a significant amount of stationary vibration signals. However, in practice, stationary vibration signals are rarely obtainable due to the uncertain operating conditions of the system. In this research, an autocorrelation function is used to extend the range of the window function to enable reliable fault diagnosis, even with a small amount of stationary vibration signals. The third research thrust proposes an original idea for tooth-wise fault identification of a planetary gearbox. The proposed method is based on a health data map that can be used even with uncertain vibration characteristics. The two-dimensional health data map can sketch the health data corresponding to every pair of gear teeth to isolate the location of the faulty gear tooth. In addition, a Hilbert transform-based phase estimation technique is employed for an encoder-less health data map that is suitable even under the slightly varying rotational speed.