Session: 08-06-01 non-presentations

Paper Number: 122498

122498 - Surrogate Modelling of Cfd-Based Non-Linear Hydrodynamics of a Semisubmersible 

Floating Offshore Wind Turbines (FOWT) are a promising technology to harness wind energy in deep-water offshore regions. One of the critical design aspects is the accurate prediction of hydrodynamic loading on the substructures, leading to the precise estimation of the Fatigue Limit State(FLS) and Ultimate Limit State(ULS), ensuring a safe floater design.

The OC5 (Offshore Code Comparison Collaboration Continuation, with Correlation) campaign showed that low-fidelity numerical models based on the potential flow theory, combined with Morison’s drag equation (usually referred to as engineering models) consistently under-predicted the hydrodynamic loading (by about 20%) as these models are limited to linear or weakly non-linear analysis. Furthermore, the semisubmersible design incorporates heave plates attached to the base of its columns to reduce the heave motion by providing supplementary added mass. Heave plates also enhance flow separation and vortex shedding effects that produce viscous damping, and these factors call for high-fidelity simulation tools that accurately capture the non-linear hydrodynamics. Additional studies have revealed that computational fluid dynamics(CFD) more accurately predict the hydrodynamic coefficients, which can be obtained by forced oscillations or free decay simulations. However, Navier-Stokes-based solvers are computationally expensive. In this regard, surrogate models trained on a CFD simulation dataset make it possible to rapidly predict the non-linear hydrodynamic coefficients of semisubmersibles with acceptable accuracy.

In this work, we develop a data-driven surrogate model (DDSM) to predict the added mass and damping coefficients. Three different surrogate modelling approaches are considered - least squares, gaussian process regression, and support vector machines for regression (SVMr). The CFD simulations to generate the training dataset are performed in OpenFOAM. The validated CFD setup in the OC6 phase 1a [1] is used. Forced oscillation simulations are performed to record added mass and damping coefficient matrices for varying amplitude and frequency. In these simulations, a sinusoidal motion is imposed in surge, heave and pitch degrees of freedom (reducing it to a 3DOF model). Out of the CFD database, 80% is used as a training set, and 20% is used as a test set for model validation. The coefficient of determination (R2) and root mean square error (RSME) are considered for model accuracy metrics. 

Firstly, a convergence study will be conducted to understand the minimum training dataset required for a good fit. The training and validation loss will be recorded. The performance of different machine learning (ML) algorithms will be concluded based on the hydrodynamic coefficient prediction. The results from the surrogate model can help improve the overall accuracy of OpenFAST by replacing the WAMIT inputs. 

References
[1] Wang L, Robertson A, Jonkman J, Kim J, Shen ZR, Koop A, Borras Nadal A, Shi W, Zeng X, Ransley E, Brown S. OC6 Phase Ia: CFD simulations of the free-decay motion of the DeepCwind semisubmersible. Energies. 2022 Jan 5;15(1):389.

Presenting Author: Likhitha Reddy Delft University of Technology

Presenting Author Biography: PhD at Delft University of Technology

Authors:

Likhitha Reddy Delft University of Technology