polito.it
Politecnico di Torino (logo)

Application of Artificial Intelligence algorithms for the classification of rotating systems health status

Luca Cibrario

Application of Artificial Intelligence algorithms for the classification of rotating systems health status.

Rel. Cristiana Delprete, Luigi Gianpio Di Maggio. Politecnico di Torino, Corso di laurea magistrale in Mechatronic Engineering (Ingegneria Meccatronica), 2022

[img]
Preview
PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (5MB) | Preview
Abstract:

Intelligent Fault Diagnosis (IFD) refers to the application of machine learning theories to machine fault diagnosis and seems to be a promising way to releases the contribution from human labor, since it automatically recognizes the state of health of machine. The work presented in the thesis has as objective the identification, training, application and evaluation of an artificial intelligence algorithm suitable for assessing the state of health of a spherical roller bearing rotating system through the analysis of its vibration path. The first step consists in the creation, using the approach of nonlinear multi-body models, of the dynamic model of a real bearing, which can simulate its behavior in presence of localized defects on the inner or outer raceway. The interaction between bearing elements is based on the theory of elastic contact elaborated by Hertz; the decrease in deformation of the rolling elements due to the thickness of the lubricant film and damping effects of the fluid itself according to Elastohydrodynamic Lubrication (EHL) theory are also included in the model. Once the database of vibration paths, related to different rotation speeds, radial loads and the presence or absence of localized defects on the tracks, has been created, it can be analyzed in time and frequency-domain to extract the necessary features to train the artificial intelligence algorithm. At the same time, this operation is also carried out for vibration paths obtained experimentally under conditions similar to those of the numerical data. The last step in the research consists in choosing the best algorithm among those belonging to supervised learning, such as the Support Vector Machine (SVM) and the k-Nearest Neighbors (kNN), evaluating their performance on both numerical and experimental data and a finally choosing the features of greatest impact through the Shapley values method.

Relators: Cristiana Delprete, Luigi Gianpio Di Maggio
Academic year: 2022/23
Publication type: Electronic
Number of Pages: 117
Subjects:
Corso di laurea: Corso di laurea magistrale in Mechatronic Engineering (Ingegneria Meccatronica)
Classe di laurea: New organization > Master science > LM-25 - AUTOMATION ENGINEERING
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/24470
Modify record (reserved for operators) Modify record (reserved for operators)