I. Introduction
Fault diagnosis and condition monitoring of rotating elec-trical machines (REM) have been a very active research area in the last decades [1]. As part of predictive maintenance systems, the implementation of fault detection and diagnosis strategies presents numerous advantages, such as reducing repair costs and reducing unplanned downtime of equipment and production processes. In these strategies, different signals from the machines can be used. Among the most widespread are vibrations and electrical measurements such as voltage and current. Vibrations can be acquired through acceleration, speed or displacement sensors in any rotating machine. On the other hand, the electrical variables can be measured in motors or generators through the sensors used for the protection of the machine, without the need to place additional sensors or access the place where it is installed. Numerous methods have been developed for machine diagnosis based on vibration or current analysis. These can be classified into model-based methods, signal-based methods, knowledge-based methods, hybrid methods and active methods [2], [3]. Signal-based methods have had a wide application at the industry level. In addition, these can be sub-classified according to whether they are applied in the time domain, frequency domain or time-frequency domain [2]. The frequency-domain strategies include the full spectrum and, in the time-frequency domain, the full spectrogram.