An improved sideband energy ratio for fault diagnosis of planetary gearboxes

Abstract Frequency spectrum analysis is an effective and direct way to understand the vibration behaviour of a planetary gearbox (PG). Sideband structures, as a useful and detectable fault information, have drawn widespread research attention to track and indicate the health conditions of the internal gears. However, for planetary gearboxes, different characteristic fault sidebands may mislead the diagnostic decisions, therefore, expert knowledge must be heavily relied on. The sideband related indicators, such as Sideband Energy Ratio (SER) and Sideband Index (SI), which synthesized the amplitudes of characteristic frequencies, have demonstrated their effectiveness in fault diagnosis of a PG. Whereas, the diagnostic mechanisms behind these sideband related indicators were not adequately studied. Moreover, how to properly select the numbers of sidebands to form a reliable indicator so far still has not yet been seriously considered. In this article, the diagnostic mechanisms of SER and SI are reexamined and analyzed via planetary gear vibration signal models with faulty gear. The drawbacks of these indicators are revealed over the theoretical analysis. According to the theoretical derivations, a selection of fewer sideband numbers is applied. A novel indicator, namely an Improved Sideband Energy ratio (ISER), is then proposed for the diagnostic task. The ISER and other sidebands indicators are testified through experimental data under different gear fault scenarios. Three typical intelligent classification algorithms, namely support vector machine (SVM), XGBoost, and deep neural network (DNN) are employed to demonstrate their diagnostic abilities. The results show that the ISER outperforms than SER, SI, and other statistical sideband indicators.