Authors:
Bryan Puruncajas
1
;
Yolanda Vidal
2
and
Christian Tutivén
3
Affiliations:
1
Mechatronics Engineering, Faculty of Mechanical Engineering and Production Science (FIMCP), Escuela Superior Politécnica del Litoral (ESPOL), Guayaquil, Ecuador, Department of Mathematics, Control, Modeling, Identification and Applications (CoDAlab), Universitat Politècnica de Catalunya (UPC), 08019 Barcelona, Spain
;
2
Department of Mathematics, Control, Modeling, Identification and Applications (CoDAlab), Universitat Politècnica de Catalunya (UPC), 08019 Barcelona, Spain
;
3
Mechatronics Engineering, Faculty of Mechanical Engineering and Production Science (FIMCP), Escuela Superior Politécnica del Litoral (ESPOL), Guayaquil, Ecuador
Keyword(s):
Structural Health Monitoring, Offshore Wind Turbine, Structural Vibration, Data-driven, Convolutional Neural Network.
Abstract:
Structural health monitoring for wind turbines (WT) in remote locations, as offshore, is crucial (Presencia and Shafiee, 2018). Offshore wind farms are increasingly realized in water depths beyond 30 meters, where lattice foundations (as jacket-type) are a highly competitive substructure type (Moulas et al., 2017). In this work, a methodology for the diagnosis of structural damage in jacket-type foundations is stated by means of a small-scale structure -an experimental laboratory tower modeling an offshore-fixed jacket-type WT. In the literature, a lot of methodologies for damage detection can be found (Li et al., 2015). Among them, the vibration-based methods are one of the most prolific ones. However, they are, primarily, focused on the case of measurable input excitation and vibration response signals, with only few recent studies focused on the vibration–response–only case, the importance of which stems from the fact that in some applications the excitation cannot be imposed and
often is not measurable. This work aims to contribute in this area, as the vibration excitation is given by the wind and analyzed by a convolutional neural network (CNN), with a classification accuracy result of 93 %.
(More)