Abstract:

In this paper, we apply topological data analysis (TDA) method for the processing of time-domain stator current signals of an induction motor under various fault conditions, and show that it can effectively reveal data features related to the fault condition. We show that classification accuracy of machine learning models for motor faults can be greatly improved when trained with TDA processed data, in comparison with models trained with time-domain stator current data. As a mathematical tool, TDA is effective in the development of data-driven fault detection and classification for motor applications.