Floating offshore wind is set to play a significant role in the future energy mix. Sustaining cost reductions of floating wind will help bolster industrialization, scale-up, and competitiveness of offshore wind energy.

Physical model testing of floating turbines is key to de-risk design and reduce CAPEX+OPEX costs. However, accurately representing both turbine and platform dynamics is challenging due to scaling limitations—most facilities can simulate either wind or wave forcing, but not both without compromise. A Software-in-the-Loop (SiTL) approach addresses this by omitting the rotor or the floater and applying instead computed aerodynamic or hydrodynamic loads. This setup not only overcomes scaling challenges, but also enables agile testing, such as assessing floating turbine response with different rotors. Therefore, SiTL is a practical and flexible solution for testing floating wind turbines.

Several studies have investigated SiTL applications (e.g., Sauder et al., 2016; Bayati et al., 2017; Chabaud et al., 2018). However, to the best of the author’s knowledge, none have validated a SiTL setup under combined wind and wave conditions. This study addresses that gap by validating a SiTL setup against physical model tests involving both wind and wave forcing.

The SiTL validation campaign was conducted in the DHI deep-water basin in 2025. The facility features a 3D wave maker and a new open-jet wind generator, jointly developed by DHI and DTU Wind. The wind generator can produce vertical wind shear and turbulent wind coherent with the Mann model. The floater was a variant of the TetraSub by Stiesdal Offshore. The turbine was the IEA 22 MW RWT, with scale 1:70.

The experiment comprised two sets of tests: with rotor (reference) and without rotor (SiTL). The controller was either set to provide constant blade pitch and rotor speed, or to react to the external conditions by adjusting the blade pitch and torque. The same open-loop and closed-loop settings were replicated in the SiTL setup.

For the SiTL experiment, the rotor was removed and replaced with a structure able to pull the tower top in 3 degrees of freedom: surge, floater pitch, and floater yaw, using pre-tensioned cables. Floater motions were recorded in real-time and fed to the HAWC2 multi-body software. HAWC2 provided rotor loads in real-time, which were transmitted to the servos that pulled the cables via HexaWire. The wind recorded during the reference experiment was used for the HAWC2 simulation. Unlike other SiTL experiments, the HAWC2 model is the downscaled IEA 22 MW, allowing an exact match with reference results.

The SiTL system was validated under different conditions, including wind-only, wind and waves, uniform or Mann-based turbulence, open-loop or closed-loop. Three setups were compared: reference (with rotor), SiTL and a HAWC2 simulation including floater and hydrodynamic loading. Results show that floater motions matched well in the actuated degrees of freedom, proving the validity and reliability of the developed SiTL setup. Furthermore, the HexaWire system reacted quickly to the rotor loads from HAWC2.

Future work will focus on adding sway and roll degrees of freedom, thus improving the SiTL setup.

Computational Fluid Dynamics (CFD) has become an increasingly valuable tool in the investigation of complex physical processes that are challenging to model using traditional numerical methods. Its applications span a wide range of engineering problems, including the calculation of wind and wave loads on intricate structures, modeling heat and mass transfer, and visualizing fluid flow phenomena. Within Aker Solutions, CFD is actively employed during early-phase, front-end design to address a variety of tasks. To validate the effectiveness of CFD in this context, an internal study was conducted comparing CFD simulation results with those obtained from physical wave model tests.
The focus of this study was a floating wind turbine foundation, characterized as a semi-submersible floater with a central column supporting the wind turbine generator, connected to three outer columns via pontoons and box beams. This concept was subjected to rigorous testing in a wave model basin, where key parameters such as motions, accelerations, and wave interactions were meticulously monitored and recorded. Traditionally, hydrodynamic phenomena like wave run-up, overtopping, and green water are identified and documented through physical wave model testing. However, CFD offers a promising alternative, with the potential to be used both in early-phase concept evaluation and as a complementary tool alongside physical testing.
The numerical model was developed using Simcenter StarCCM+, with careful consideration given to balancing computational efficiency and solution accuracy. Initial simulations, conducted without the floater, focused on capturing wave-induced currents and wave heights to determine optimal mesh requirements. A free decay simulation was then performed to verify the hydrodynamic properties of the floater. For regular wave conditions, attention was given to the spatial and temporal influences on motion response and wave height. An optimization process established the best compromise between accuracy and computational efficiency, specifically regarding the number of cells per wave height and the simulation time step.
Results from the free decay simulations demonstrated less than 5% deviation in rotation around the x-axis compared to the basin model tests. Capturing wave-induced currents required a fine mesh, while wave heights could be adequately represented with a coarser mesh. The optimal configuration—10 cells per wave height and a time step of 0.02 seconds—resulted in a significant reduction in simulation time (over 70%) without compromising accuracy. The numerical results showed strong agreement with experimental data, validating the CFD approach.
In conclusion, the study found that CFD simulations compared favorably with physical model test results for the selected cases, with acceptable simulation times for early-phase concept design. CFD is thus recommended as a valuable tool for early-phase floater design support and as a complementary method to wave model testing, with the potential to eventually replace physical tests in certain scenarios. Delegates will gain insights into CFD model setup, simulation results, and the comparative analysis with wave model tests, equipping them to leverage CFD in their own design processes.

Moored floaters with flexible constraints have natural surge–sway–yaw frequencies well below the dominant frequencies contained in a typical wave spectrum. The low-frequency response is driven resonantly by second-order (difference-frequency) wave loads, producing large slow-drift motions. A key parameter that characterizes these oscillations is the low-frequency damping which dampens out the pole of the response amplitude operator and therefore significantly modifies the response of the floater. At these low frequencies little energy is radiated away so linear potential-flow models systematically overestimate slow-drift motions. In reality, the dissipation mechanism is dominated by viscous effects (form drag, separation, vortex shedding on hulls and moorings) and by wave-drift damping—processes not represented in inviscid theory and difficult to quantify analytically. The state-of-the-art estimation of the low-frequency damping relies on inference from ocean basin data though the current methods still tend to suffer from poor observability and bias.

The work presented in this paper presents a novel approach for estimating low-frequency added mass and damping from ocean basin data. Wave loads on floating structures are generally difficult to directly measure so the time-histories of hydrodynamic loads are rarely available from ocean basin experiments. The method is therefore formulated such that the wave elevation profile forms the input to the system with the floater motion as the output, both of which are typically measured quantities in a lab.

The approach is similar in nature to Sauder (2021) where the experiments are configured to have identical wave-profiles but different mooring parameters. This configuration, although different across experiments, is represented as a single-input multiple-output (SIMO) system of experiments. From there, the proposed estimator exploits cross-spectral relationships among the outputs that share a common but unknown input. This approach therefore falls under the category of output-only methods. The framework ensures consistent estimates of low-frequency added mass and damping verified with both numerical and experimental data. The method further reduces sensitivity to uninformative frequency bands, improving robustness under experimental conditions where outputs are composed of both wave-frequency and low-frequency responses.

Neural-Network hydrodynamic force predictors for simulating floating offshore wind turbines under operational and extreme waves

Bonaventura Tagliafierro1, Victor Mendoza2, Salvatore Capasso3, José Dóminguez4, Malin Göteman1
1Uppsala University (Sweden), 2Hexicon AB (Sweden), 3Universitat Polytècnica de Catalunya (SPain), EPhysLab – University of Vigo (Spain)

Floating offshore wind turbines (FOWTs) pose significant, among others, structural challenges due to their complex and non-linear interaction between platform motions and environmental loads. Numerical modeling is being used as a tool for understanding and mitigating these interactions, thus supporting their engineering development. However, existing simulation tools are based on low-fidelity models and potential flow theory, limiting their accuracy when the platform motion violates the grounding linear assumption. Our work puts forward a first generation of hybrid solvers that combine machine learning (ML) and physics-based functionalities to resolve time-domain problems for marine structures under combined environmental excitations. The main innovation we present is THYTANN – a transient hydrodynamic tool accelerated by neural networks – that essentially predicts wave force, which is implemented within a dynamic solver for rigid-body dynamics and mooring systems. Such approach guarantees low computational cost compared to other low-fidelity models, with real time factors ranging between 0.01 and 0.1 for a single FOWT, yielding elapsed time for full storm duration estimations in less than an hour.

THYTANN predicts the hydrodynamic force components on an arbitrary hull shape, guaranteeing accuracy even when high nonlinear waves are considered. It is based on a feedforward neural-network (FNN) architecture, trained on a dataset that considers random sea-states simulations using high-fidelity generated data using DualSPHysics – a CFD-based framework. For each body, this dataset links its position, orientation, and velocity, and its corresponding fluid forces. The targeted artificial FNN, which represent one of the most computationally efficient methods in ML, captures information flows in a single direction, moving from a so-called input layer to an output layer, without any loops or backward connections, but still being able to represent nonlinear patterns. We will test this model for a time-dependent external force (unknown to the NN) which represents the wind action.

THYTANN is then embedded within a much wider software framework, which includes a time-integration scheme that promotes modularity of setup and use: each component, such as the mooring lines, the rotor aerodynamics, the hydrostatic and hydrodynamic solvers, is a single call instance, updated at each time step. For this work, we aim to present the model definition and database generation, the implementation of the loss function and a task-based hyperparameter optimization that ensure that the NN enhances the dynamic fidelity of the entire system rather than standalone prediction. Ultimately, as part of the project IM-POWER (Numerical Integrated Model for the POWER output of floating offshore wind farms that are fully grid-connected during sea storms), this work showcases a comprehensive demonstration of the numerical model under-development, applied for the reconstruction of the power output of a FOWT farm under operational and extreme conditions.

The stability of the atmospheric boundary layer (ABL) is an important factor in wind turbine response analysis, as it affects mean wind profiles, turbulence characteristics, and wake behaviour. Existing studies on the sensitivity of floating wind turbine response to atmospheric stability rely on surface-layer profile formulations such as Monin-Obukhov similarity theory (MOST) or empirical power laws. These approaches are limited in their ability to represent stable conditions at the heights relevant to modern wind turbines, frequently overestimating wind shear beyond the shallow surface layer. In addition, stable atmospheres often exhibit local negative shear (low-level jets) and pronounced wind veering – both of which affect the structural response of wind turbines, particularly large rotor systems. More advanced models are therefore required to capture stable boundary layer wind profiles accurately and to reduce uncertainties in response simulations. This study investigates the sensitivity of the global response of a 15-MW floating wind turbine to stable wind profiles derived from analytical models of varying complexity. Wind profiles are generated for combinations of rotor-equivalent wind speed and stability levels using MOST, the Gryning profile, and a recently developed analytical model for stable ABL flow, by Narasimhan. The latter enables modelling of both wind speed and direction profiles and features low-level jets under stable conditions. The selection of models facilitates a systematic assessment of shear, veer, and low-level jet effects while maintaining equal rotor-equivalent wind speed. The wind profiles are used to generate turbulent inflow fields for dynamic response simulations of the IEA 15-MW reference turbine mounted on the VolturnUS-S semi-submersible floater. Substantial differences in platform motion, structural loads, and power output are observed and attributed to variations in the vertical and directional wind speed distributions across the rotor. The findings highlight the limitations of conventional surface-layer models under stable conditions and emphasize the importance of advanced wind modelling approaches to reduce uncertainties in response analyses and support the efficient design of floating wind turbines.