Automatic classification of stator asymmetries and insulation thermal damages in induction motors, applying persistence spectrum and a convolutional neural network to the stray-flux signals

The aim of this work is to present a new methodology to automatically detect and classify stator asymmetries in induction motors through the study of the stray-flux signals. Moreover, insulation damages caused by overheating are also considered in this paper. The proposed method relies on the obtention of the persistence spectrum of the steady-state stray-flux signals, using the resulting images as input for a convolutional neural network to automatically classify not only the different types of faults, but also different levels of stator asymmetry. The results prove that, even when different types of power supply are used (grid-fed or variable frequency drive) or different levels of load are applied to the motor, this new method is able not only to detect, but to separate between levels of stator asymmetry achieving a high accuracy.