Eventos Anais de eventos
DINAME 2023
XIX International Symposium on Dynamic Problems of Mechanics
Fault Identification in Rotating Systems using Convolutional Neural Networks
Submission Author:
Carlos Alberto Alves Viana , SP
Co-Authors:
Carlos Alberto Alves Viana, Diogo Stuani Alves, Tiago Machado
Presenter: Tiago Machado
doi://10.26678/ABCM.DINAME2023.DIN2023-0012
Abstract
The application of Machine Learning methods and models in vibration analysis of rotating machines has become an important milestone in the era known as Industry 4.0. Several models are used to identify faults and monitor assets. In this paper, a CNN (Convolutional Neural Network) is tested for classification of ten machine faulty condition classes. The experimental acceleration data – in the time and frequency domains – are converted to images through the vibration images technique. The results pointed to the conclusion that the vibration images in the frequency domain present better contrasts between the types of faults due to its inherent feature extraction characteristics, resulting in an accuracy of 99.4% compared to 97.1% in the domain of time.
Keywords
Mechanical Vibration, Deep learning, Fault classification, Machine condition monitoring, Predictive maintenance
