A visualized classification method via t-distributed stochastic neighbor embedding and various diagnostic parameters for planetary gearbox fault identification from raw mechanical data

Planetary gearboxes are widely used in various types of machinery and play an important role in the transmission. Since the structure of planetary gearbox is more complicate than the fixed shaft gearbox, fault identification of planetary gearbox is challenging. Detection and diagnosis methods based on the analysis of raw mechanical vibration signals of the planetary gearbox have been studied widely because of the intrinsic advantage of revealing mechanical failure. However, the effectiveness of published studies for visualizing various types of planetary faults simultaneously are not satisfying. In this paper, several parameters that have been proved to be able to indicate the feature of the planetary gearbox vibration signals in different operation states are used to extract comprehensive fault information. Then, t-Distributed Stochastic Neighbor Embedding (t-SNE) is used to reduce the dimensionality and realize the visualization of fault feature to identify multiple types of faults. Experiments containing different types and levels of faults were performed to obtain raw mechanical data. The effectiveness of the method for visualization of planetary gearbox faults is verified by a multi-level comparative analysis.