Utilizing t-SNE with various distances for fault isolation in 7-phase electrical machines

Due to their structural resilience, electrical machines with more than three phases are increasingly used in high-power applications (transportation and energy production). However, monitoring them to increase reliability and improve system availability and safety is essential. This work exploits the properties in the frequency domain of electrical currents flowing in 7-phase electrical machines for fault type diagnosis and faulty phase isolation, especially in the case of incipient faults in the latter. These properties are derived from projecting the phase currents in the stationary frames. The fault features are the amplitudes of the fundamental component of the transformed currents (1F). Several distances are used for fault clustering in the t-distributed Stochastic Neighbour Embedding (t-SNE) framework. The clustering results are provided through graphical visualization and two metrics, the Silhouette Score (SS) and Davies–Bouldin Index (DBI), measuring intraclass and interclass distances. The results demonstrate that the proposed features effectively identify the fault type and faulty phase, even when the same incipient fault type affects different phases. Additionally, the Mahalanobis distance performs well in fault type isolation, and the various distance metrics exhibit consistently high performance in faulty phase isolation.