FAST.Farm: trying to match wake deflection of 2 floating IEA 15-MW against LES

[delloiacono]

Dear all,

I am comparing FAST.Farm against an AMR-Wind LES of two aligned floating IEA 15-MW, focusing on power output and wake center position (both horizontal and vertical).
Here my setup:
setup.zip
The low- and high-resolution domains read by the Ambient Wind module of FAST.Farm are sampled from the LES precursor with a resolution of 10 m and 5 m, respectively. I am using the curled wake model with k_VortexDecay=0, k_vCurl=5 and C_Meander=2.1, following the study of Thedin et. al 2025 for the bottom-fixed IEA 15-MW. All the other FAST.Farm parameters have default values.
Below I am sharing my main findings. Do you think the mismatch between FAST.Farm and AMR-Wind is mainly because of the curled wake model parameters I am using or that something else is wrong in my setup?

Just focusing on the first turbine:

• Horizontal wake deflection: the mean wake center position increasingly shifts with the downstream distance, differently from LES. The case is at zero yaw and zero wind-wave misalignment, but the platform yaw is higher slightly positive in FAST.Farm. Do you think increasing the k_VortexDecay parameter can help? Are there any other possible adjustments?

• Vertical wake deflection: the mean wake center position follows a similar trajectory as a function of the downstream distance, but it is consistently underestimated compared to LES. In contrast the average pitch of the platform in the FAST.Farm simulation is higher than in LES. Which parameters of the wake model can be tuned to improve this behavior? I know that lowering the C_Meander could increase the deflection, but I would also increase the meandering as a side effect. For reference, in Carmo et. al 2024 the curled wake model demonstrated a good ability in matching the vertical wake deflection of a floating NREL 5-MW.

• Meandering: the wake meanders significantly more compared to LES, both horizontally and vertically. Below the std of the horizontal position and a time series as an example.

Side request
Using all default parameters (k_VortexDecay=0, k_vCurl=2 and C_Meander=1.9), FAST.Farm crashes with this error:

T2:FARM_UpdateStates:NearWakeCorrection:Wake model is not valid in the propeller-brake region,
i.e., Ct > 2.0.

Could the reason be that k_vCurl is too low for the inflow I am passing to FAST.Farm? For this floating case I have a quite low turbulence intensity (about 3.5-4%). The rotor-averaged velocity at turbine 2 is significantly lower than in the case with k_vCurl=5. Can be this the main reason why the Ct of the second turbine reaches 2?

• k_vCurl=2

• k_vCurl=5

Thank you very much in advance for your help.

[jjonkman]

Dear @delloiacono,

Thank you for sharing these interesting results. Can you clarify one thing for me before I try to answer your questions: How are you deriving the wake center position for your comparison of wake deflection and meandering? Are you using the SAMWICH toolbox (https://github.com/ewquon/waketracking), and if so, with which settings?

Best regards,

[delloiacono]

Dear @jjonkman,

Yes, I am using the SAMWICH toolbox. In particular, I am detecting the wake centers using a 2D Gaussian distribution with default settings:
`

            tracker = '2-D Gaussian'
            wake = track(x,y,z-z_hub,u, method='Gaussian2D',verbose=False)
            wake.remove_shear(wind_profile=profile)
            yc,zc = wake.find_centers(umin=None,  # default: detected from wake snapshot
                                    A_ref=turb.rotor_area,
                                    A_min=turb.rotor_area/5,  # ad hoc value
                                    A_max=turb.rotor_area*2,  # ad hoc value
                                    AR_max=10.0,  # ad hoc value
                                    verbosity=0,  # > 0 for debugging
                                    )`

Here also a snapshot as an example:

Best regards

[jjonkman]

Dear @delloiacono,

Thanks for clarifying. Here are my responses to your original questions:

• Horizontal Wake Deflection: We saw something similar in Carmo et al 2024 (https://wes.copernicus.org/articles/9/1827/2024/). The leftward deflection of the wake (when looking downwind) comes from the pitch-induced curled wake shape combined with wake rotation from rotor torque. (You could eliminate this effect in FAST.Farm by setting Swirl = FALSE.) LES should capture the same effect, but it likely also has a Coriolis term (depending on the latitude) that is missing from FAST.Farm and counteracts with opposite wake deflection. (You could eliminate this effect in LES by simulating near the equator.)
• Vertical Wake Deflection: It is nice to see a similar trend between FAST.Farm and LES here, but I'm not sure why there is a mean offset. It is especially odd to me that FAST.Farm is showing a mean deflection at 1D downstream that is below the hub. Could this mean offset be the result in differences in the FOWT model set up, e.g. different hub heights or mean heave displacements? I also believe you are using FAST.Farm within OpenFAST v4.1, which DEFAULTs WAT_ScaleBox to FALSE, whereas I would generally recommend this be set to TRUE because we've seen that wake-added turbulence (WAT) can introduce a mean deflection depending on how the isotropic WAT box was set up. The DEFAULT WAT_ScaleBox was changed to TRUE in OpenFAST v4.2 and newer, so, I either recommend upgrading or changing WAT_ScaleBox to TRUE.
• Meandering: Have you compared meandering for other conditions, such as higher ambient TI? At low ambient TI (below 5%), the meandering from ambient TI is quite small and meandering may be more dominated more by WAT. I'm not sure where you WAT box or other WAT settings came from, but they could be the source of the issue, which I would believe if the meandering matches better at higher ambient TI.
• High Thrust: It looks like the smaller value of k_vCurl results in a stronger wake deficit impinging on T2, which is then likely operating at high tip-speed ratio and high thrust. Regardless, I would recommend using the wake parameters collaborated from Thedin et al 2025 for the IEA Wind 15-MW RWT.

Best regards,

[delloiacono]

Dear @jjonkman,

Thank you for these very useful suggestions. I have looked into these issues in more detail. In particular:

• I tried to use a 1D instead of 2D Gaussian distribution for the wake tracking algorithm, as also done in Thedin et al 2025 (https://wes.copernicus.org/articles/10/1033/2025/wes-10-1033-2025.html) and the results are more reasonable and consistent between FAST.Farm and LES, regardless of FAST.Farm setup.

• Regarding the horizontal wake deflection, Coriolis forcing is switched off also in my LES. I tried to run another FAST.Farm simulation with Swirl=FALSE and the two deflections still have opposite sign, but the error is significantly smaller and the two trajectories look more similar. Could the remaining error (about 7 m at 1D) be due to the grid resolution and/or wake tracking model accuracy, considering that the AMR-Wind grid has size 5 m while the FAST.Farm grid 10 m?

• With 1D Gaussian the vertical wake deflection matches much better the LES, especially in the first diameters downstream. I also ran two simulations, one without WAT (right panel) and one with WAT and WAT_ScaleBox=TRUE (left panel). Although there is an effect, it seems that the main cause of the error was the wake tracking model I was using.

• The WAT does influence significantly the meandering in my case at low TI (here I show only the vertical meandering, but the same is valid also for the horizontal meandering). The meandering with scaled WAT (left panel) is consistently larger compared to LES, while without WAT (right panel) the error is small in the near wake but increases with the distance downstream. I generated my WAT box with the following tool:
  Mann_turb_x64 turbulence generation 64bit DTU Wind Energy, 2014 www.hawc2.dk
  I share with you also my settings:

WAT_settings.zip

Best regards,

[jjonkman]

Dear @delloiacono,

I'm glad to see the FAST.Farm results are now making more sense for your now. See my specific responses below.

• Horizontal Wake Deflection: Yes, the small remaining difference could be the result of spatial-temporal discretization. It could also be the results of time-convergence, e.g., running longer simulations will better average-out the meandering, allowing more accurate deflection outputs. To check this, you could run longer simulations with finer spatial-temporal discretizations.
• WAT: Your settings look OK to me, except that AlphaEpsilon is a bit higher than recommend based on a lengthscale equal to the rotor diameter (240 m): https://openfast.readthedocs.io/en/main/source/user/fast.farm/FFarmTheory.html#ff-wat. That said, this should be fixed by setting WAT_ScaleBox = TRUE. You could also tailor the magnitude of the WAT (e.g., to better match the WAT seen in LES) through FAST.Farm inputs WAT_k_Def and WAT_k_Grad.

Best regards,