Automatic Detection of Rotor Faults in Induction Motors by Convolutional Neural Networks applied to Stray Flux Signals

The growth in cost and complexity of modern industrial plants leads to decreasing tolerance for performance degradation or system downtimes. In this context, the paper deals about the automatic condition monitoring of the rotor bars of induction AC motors which is performed through two main components. The former is the extraction of the Short-Time Fourier Transform of the motor stray flux during start ups as a fault-related index. The latter is the automatic classification and recognition of the signal through an Artificial Intelligence-based algorithm: a Convolutional Neural Network. This cutting-edge tool is particularly suitable for knowledge-based image recognition problems and its feasible training is here permitted by some data augmentation techniques.