Automatic detection of corrosion in ball bearings of soft-started induction motors, obtaining the persistence spectrum of the stray-flux signals

This work, presents a novel method to automatically detect and classify the presence of corrosion in ball bearings of soft-started induction motors. The method relies, first, in obtaining the persistence spectrum images of the part of the stray-flux signals in which the soft-starter is in operation. Then, those images are used as input for a convolutional neural network, which automatically differentiates not only between healthy or faulty bearings, but also between different levels of corrosion in the ball bearings. For the study, a soft-starter of a well-known brand is used to start an induction motor, setting different combinations of the variable parameters of that device. Also, different levels of load are applied to the motor. The proposed method proves its potential, achieving a high accuracy.