Session: 16-08-01 Numerical Wave Tank - Intelligent Systems, Automation and Digitalization, Maneuvering

Submission Number: 179811

Neural Network Meta-Models for FPSO Offset Prediction: Integrating Current Depth Profile With Surface Metocean Data 

Predicting the offset motion of a moored offshore platform from the incoming environmental conditions is a challenging endeavor. This horizontal displacement results from complex effects, such as second-order wave drift loads, hydrodynamic drag induced by ocean currents along the hull, mooring lines and risers, wind loads and non-linear restoring forces. Traditional workflows for the design and analysis of such offshore systems rely on sophisticated models and numerical time-domain simulations to accurately reproduce their dynamic behavior. However, this process is time-consuming and can yield inaccuracies due to model simplifications and the inherent complexity of platform dynamics. To address this issue, this study explores the development of a Neural Simulator (NeuroSim) employing Machine Learning models to compute motion statistics of a vessel from metocean conditions. In particular, this paper focuses on incorporating the three-dimensional ocean current depth profile as one of NeuroSim inputs, along with wind and wave/swell parameters, which improves the model’s ability to capture complex interactions between ocean currents and the offshore system that would be unaccounted for if only surface currents are considered. The final data-driven meta-model, obtained through Bayesian hyperparameter optimization, combines Convolutional Neural Networks (CNNs) and Multi-Layer Perceptrons (MLPs). It was trained using data from a spread-moored Floating Production Storage and Offloading (FPSO) unit located offshore on the Brazilian coast, subjected to typical environmental conditions observed in the area. The case-study platform has a large number of mooring lines and risers, and the metocean data were collected from reanalysis datasets. Comparative analyzes revealed that the proposed updated model offers improved accuracy compared to traditional simulation models and demonstrates its potential to complement conventional simulation models.

Presenting Author: Gustavo Alencar Bisinotto Escola Politécnica da Universidade de São Paulo

Presenting Author Biography: Professor of the Mechatronics Engineering Department at Escola Politécnica of Universidade de São Paulo - São Paulo, Brazil. PhD in Control Engineering and Mechanical Automation at the same institution (2021-2024). Laurea Magistrale (equivalent to Master of Science) in the Automation and Control Engineering program at Politecnico di Milano - Milano, Italy (2017-2020). Major in Mechatronics Engineering from the Universidade de São Paulo - São Paulo, Brazil (2014-2020).

Authors:

Gustavo Alencar Bisinotto Escola Politécnica da Universidade de São Paulo
Asaffe Apolinario Duarte Escola Politécnica da Universidade de São Paulo
Rodrigo Augusto Barreira Petrobras Research Center
Anna Helena Reali Costa Escola Politécnica da Universidade de São Paulo
Eduardo Aoun Tannuri Escola Politécnica da Universidade de São Paulo
Edson Satoshi Gomi Escola Politécnica da Universidade de São Paulo