Neural Network Meta-Model for FPSO Roll Motion Prediction from Environmental Data

Resumo

The current design process of mooring systems for Floating Oil Production and Offloading units (FP-
SOs) is highly dependent on the availability of the platform’s mathematical model and accuracy of dynamic sim-
ulations, through which resulting time series motion is evaluated according to design constraints. Out of the six

degrees of freedom, roll motion is among the most complex to accurately simulate. We propose a Neural Simu-
lator, a set of neural network surrogate models designed to predict an FPSO’s roll motion statistics directly from

metocean data when subject to different loads. This approach bypasses the need to perform traditional time series
dynamic simulation, as the trained models take measured metocean conditions and directly output the desired roll
motion statistics. This allows for Artificial Neural Networks (ANNs) to be trained through simulation and later
fine-tuned on real FPSO motion. As a result, our proposal presents higher accuracy and reduced computational
time when compared to traditional methods. The ANN surrogate models are trained by real current, wind and wave
data measured in 3h periods at the Campos Basin from 2003 to 2010 and the associated roll response of a Spread
Moored FPSO subject to different drafts, which is obtained through time-domain simulations using the Dynasim

software. Hyperparameter Optimization techniques are performed in order to obtain optimal ANN models special-
ized in different platform drafts. Finally, the proposed models are shown to correctly capture platform dynamics,

providing good results when compared to the statistical analysis of roll motion time series obtained from Dynasim.

We conclude that an ANN surrogate model can be trained directly on real measured metocean conditions and plat-
form roll motion to provide increased accuracy and reduced computational time over traditional methods based on

dynamic simulation. Moreover, the proposed architecture can be integrated into an automated learning framework:
The data-based surrogate models can be continuously fine-tuned and updated with newly measured data, resulting
in improved accuracy over time.