Surrogate model based on ANN for the evaluation of the fundamental frequency of offshore wind turbines supported on jackets. (1st January 2023)

• Record Type: Journal Article
• Authors: Quevedo-Reina, Román; Álamo, Guillermo M.; Padrón, Luis A.; Aznárez, Juan J.
• Is Part Of: Computers & structures. Volume 274(2022)
• Publication Date: 2023-01-01
• Subjects: Offshore Wind Turbine -- Jacket structure -- Artificial neural network -- Surrogate model -- Soil-structure interaction -- Pile foundation; Structural engineering -- Data processing -- Periodicals; Electronic data processing -- Structures, Theory of -- Periodicals; 624.171
• DOI: 10.1016/j.compstruc.2022.106917
• Languages: English
• ISSNs: 0045-7949

Abstract

Highlights: ANNs can reproduce the fundamental frequency of OWTs supported on jackets. Evaluation time reduction justifies the cost of training for large-scale executions. Ensemble models increase considerably the performance of the individual ANNs. Abstract: The design of the support structure of offshore wind turbines (OWT) requires the dynamic characterization of the complete structural system, including the soil-foundation subsystem. To minimize resonance phenomena, the structural natural frequencies should be sufficiently apart from those that characterize the different loads. The obtention of natural frequencies can be a computationally expensive procedure, more so if dynamic soil-structure interaction (SSI) phenomena are included. In order to reduce the computational cost, a surrogate model based on Artificial Neural Networks (ANN) is proposed to estimate the fundamental frequency of the assembly formed by the wind turbine, the jacket support structure and the pile foundation. The training dataset is obtained by a finite element substructuring approach in which the dynamic SSI is incorporated through impedance functions obtained from a continuous model. The ability of the proposed ANN-based surrogate model to reproduce the influence that the main variables of the problem have on the fundamental frequency in a sufficiently precise way is shown by comparing its predictions and the results of the finite elements model for new configurations. The high accuracy and significant computational cost reduction justify the use of the surrogate model in applications where a large number of evaluations is required.