Bibliographic citations
Neira, F., Paredes, G. (2023). Diseño de un sistema de clasificación multiclase para la detección de fallas mecánicas en máquinas rotativas empleando demodulación, autocorrelación y ensamblado stacking sobre señales de vibración obtenidas de la base de datos MAFAULDA [Trabajo de suficiencia profesional, Universidad Peruana de Ciencias Aplicadas (UPC)]. http://hdl.handle.net/10757/670997
Neira, F., Paredes, G. Diseño de un sistema de clasificación multiclase para la detección de fallas mecánicas en máquinas rotativas empleando demodulación, autocorrelación y ensamblado stacking sobre señales de vibración obtenidas de la base de datos MAFAULDA [Trabajo de suficiencia profesional]. PE: Universidad Peruana de Ciencias Aplicadas (UPC); 2023. http://hdl.handle.net/10757/670997
@misc{renati/1296347,
title = "Diseño de un sistema de clasificación multiclase para la detección de fallas mecánicas en máquinas rotativas empleando demodulación, autocorrelación y ensamblado stacking sobre señales de vibración obtenidas de la base de datos MAFAULDA",
author = "Paredes Farfan, Gustavo Alonso",
publisher = "Universidad Peruana de Ciencias Aplicadas (UPC)",
year = "2023"
}
Rotating machines play a fundamental role in the industrial sector, as they are an integral part of multiple applications whose operation is essential to maintain continuous production. To reduce risks related to unforeseen failures in this type of machine, preventive maintenance is a valuable tool. In recent years, in maintenance, research has been carried out for the preventive detection of failures using models based on machine learning. However, current literature tends to use increasingly complex classification models. This paper proposes the use of demodulation techniques, feature extraction in time and frequency, and autocorrelation, applied to the vibrational signals obtained from the MAFAULDA database. This allows the construction of a multiclass model using binary classification models and the stacking ensemble method. As a result, an F1 score of 99.33% was obtained, which is remarkably close to the results presented in research that uses models based on deep learning. This shows that with the right strategy, low-complexity, high-performance models can be built.
This item is licensed under a Creative Commons License