Abstract:

In condition monitoring, tracking the health, state and characteristics of a hardware equipment is an expensive task. The target of this article is to recommend an alternative way of monitoring the equipment by using artificial intelligence, in order to minimize maintenance costs. An important aspect therein is to detect faults in the rotating equipment, in particular bearings. We will use the vibration signals of the bearing and different techniques within the Digital Signal Processing field (Mel-frequency cepstral coefficients, Fast Fourier Transform, Short Time Fourier Transform) to pre-process the input signal and extract important features in an appropriate way as to train machine learning models. Dimensional reduction techniques (such as Principal Component Analysis) are used to diminish the time and complexity of training. The trained models will be able to recognize di_erent states of the bearing, like fault states (inner, outer, ball) states or a normal state. Finally, we will test the performance of the models on other data sets, di_erent from the one we trained and then save the results to make a statistics regarding the accuracy of the models.