LAUSR

As the ‘aorta’ of mechanical equipment, a planetary gearbox (PG) is prone to suffer from failures and even bring disastrous results due to the tough service environments. Frequency spectrum analysis is the most commonly used conditional monitoring method in engineering scopes and prior guidance for fault diagnosis and reliability analysis to be relied on. The sideband energy ratio (SER), which synthesized some characteristic frequencies, has shown its effectiveness in fault diagnosis to some extent. However, the key parameters to form the SER are empirically selected and fixed, limiting its sensibility and extensibility. To this end, this paper proposes a dynamic sideband energy ratio (DSER) scheme to diagnose gear faults of a PG under different operational conditions adaptively. The SER matrix is constructed based on different sideband numbers and bandwidth setting values for an operational condition. The SER matrix will then be fed to two machine learning algorithms: the deep neural network and the support vector machine, to get a fault classification accuracy map. Finally, the optimal sideband number and bandwidth can be obtained based on the highest classification accuracy to get the DSER. The trained DSER can directly be applied to the remaining data of the PG. Experimental studies demonstrate that the DSER is more outperform the SER in diagnosing Sun and planet gear faults under different operational conditions. More importantly, DSER has the potential to determine the security working domain and, therefore, promote the connection between fault diagnosis and abundant reliability analysis methods.