Design of a Floating Offshore Structure by a Deep Neural Network

Resumo

The design process of an offshore system requires a suitable dynamic model of the floating structure, accord-
ing to the geometry and environmental conditions. However, the choice of the hydrodynamic model is directly

associated with the computational cost since dozens of simulations are necessary during the design process. In
this work, the assessment of different parameters allowed to accelerate the design analysis model applying Deep
Neural Networks instead of the hydrodynamic model. A two-phase study assessed the main parameter’s calibration
to the Deep Neural Network (DNN), with a forced 1 DoF Mass-Springer-Damper (MSD) system and a validated
hydrodynamic model for heave motions of a semi-submersible platform. The use of simplified models in this study
allowed a large volume of cases to compose the dataset. Systematically varying the number of layers in the DNN
and the number of neurons in each layer, the authors selected the combination with the minimum mean squared
error. The response surface provided by the best network configuration provides a response surface that is useful
for optimization tasks of future works.