ISHM for fault condition detection in rotating machines with deep learning models

The electromechanical impedance-based SHM method (ISHM) aims to correlate changes in vibration signatures with physical phenomena. At the same time, monitoring of rotating systems is necessary for economic and safety reasons. Thus, the structural health monitoring of rotating machines is commonly assessed by using vibration sensors together with a SHM technique, such as the ISHM approach. As a result, a large amount of data have to measured; consequently, both machine and deep learning techniques have become relevant for fault detection purposes. It is worth mentioning that previous studies used the ISHM technique associated with CNN models for monitoring the structural condition of beams. In this sense, the main objective of this work is to contribute to the topics of SHM and artificial intelligence, demonstrating another potential application of convolutional neural networks to support the diagnosis of structural damage of rotating systems by using the ISHM approach. For this aim, structural condition of a rotor supported by two ball bearings, with two disks, and one pulley was monitored by considering four different health conditions and three different operating speeds. Then, a 6-layer 1D-CNN model was formulated individually for the three PZT sensors attached to the rotor shaft. As input data, all sample points of the measured impedance signatures were considered. The results from this implementation demonstrate the potential of the procedure conveyed as shown by a minimum accuracy of 92.22% for all evaluated PZT patches.