Bearing Failure Classification with Low Complexity Neural Network

In this work, an AI enabled fault classification system for roller bearings based on an approach tailored for embedded implementation is presented. The fault classification system is based on vibration signal analysis and exploits a low computational cost neural network (NN) along with a novel low-complexity preprocessing technique, which allows for minimizing the dependence of the classifier design on the bearing rotational speed and on the vibration signal sampling frequency. The proposed fault detection system was trained with emulated data and tested with emulated data and real world example, both related to different working conditions with respect to those used for training the NN, providing very promising results.