Eventos Anais de eventos
DINAME 2023
XIX International Symposium on Dynamic Problems of Mechanics
Digital Twin of an Offshore Riser Systems for Time Series Prediction Using Deep Learning Models
Submission Author:
Stanley Washington Ferreira Rezende , GO
Co-Authors:
Stanley Washington Ferreira Rezende, Leandro Soares Silva, Aldemir Ap Cavalini Jr, Jose dos Reis Vieira de Moura Jr, Valder Steffen Jr
Presenter: Stanley Washington Ferreira Rezende
doi://10.26678/ABCM.DINAME2023.DIN2023-0069
Abstract
Due to the shortcomings of numerical models and their expensive processing requirements, the use of deep learning models for time series prediction has become feasible. For this reason, the implementation of an API (Application Programming Interface) of the Digital Twin to predict time-domain dynamic responses of a riser structure is proposed in the present contribution. The Long-Short Term Memory (LSTM) and Multilayer Perceptron (MLP) neural networks, employed as prospective prediction models, constitute the foundation of the Digital Twin. The oil platform's actual displacements and rotations caused by the sea's waves are the time series used to create the dynamic reactions of the riser. The forces produced by a finite element model (FEM), using the ANFLEX software developed by CENPES/PETROBRAS, were used as the training output time series. The output of the two neural networks was evaluated and compared with each other both for the training series and for the prediction of series not yet seen by the models. Four metrics were calculated to facilitate comparison, and signal comparisons versus time graphs were plotted.
Keywords
Digital Twin, Oil and Gas industry, Deep Learning Models, Offshore Riser Systems
