Gearbox Digital Twin Data Used in Supervised Learning

• Published in: Journal of Vibration Engineering & Technologies Vol. 12; no. 3; pp. 3087 - 3099
• Main Authors: Sow, Souleymane, Farhat, Mohamed Habib, Chiementin, Xavier, Rasolofondraibe, Lanto, Cousinard, Olivier
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.03.2024
• Subjects: Acoustics; Control; Dynamical Systems; Engineering; Engineering Acoustics; Original Paper; Vibration; Condition monitoring; SVM; Digital twin; Gearbox; Artificial intelligence; Classification
• ISSN: 2523-3920; 2523-3939
• DOI: 10.1007/s42417-023-01035-y

Purpose The industry of the future through digital twins is constantly seeking efficiency, competitiveness, and innovation, particularly in the case of the diagnosis of rotating structures. Their use in the maintenance of systems has been the subject of several research in recent years by studying simulated vibration behaviour of several operating modes, as well as building a substantial database. Method However, the limitations in its implementation lie in the accuracy and reliability of the simulated data, which can lead to a misdiagnosis of the state of the system. The study presented in this paper proposes a method of diagnosis based on the digital twin using a digital and experimental database of a gearbox classified with the Multiclass Support Vector Machine (MSVM). To come up with an outcome, experimental data are collected on the test bench in different modes of operation, then for each of these modes, simulations are made on the numerical model created to generate data, which will be updated on those from the physical model. Results Using a classification algorithm, data from the numerical model are used to train the model and experimental data are used for testing. The results provide an average of 81% of classification accuracy. Conclusion Finally, from the dynamic model of a gearbox developed in this study, several operating modes can be generated and used in case of lack of experimental data, to do a diagnostic.