Why does it take ~200 iterations to converge with DTU 10MW turbine?

[passboy666]

Bug description

Hi developers,

I'm running simulations using AMR-Wind with the DTU 10MW wind turbine, and I've noticed that it typically takes about 200 nonlinear iterations to reach convergence in each timestep (or solve cycle), which seems high compared to smaller turbine setups.

Could you help me understand:

Is this expected behavior for such large-scale simulations?

Could it be due to the size of the DTU 10MW rotor, mesh resolution, or actuator model?

Are there any recommended solver settings or relaxation strategies for improving convergence?

Here is some context:

AMR-Wind version: v1.4.0-17-g250778a3

Compiler: Clang 17.0.6

MPI/GPU/OpenMP: All OFF

Actuator model: [e.g., ALM / flexible ALM / ADM — replace with your case]

Wind speed/TSR: [fill in if relevant]

Mesh resolution near rotor: [optional]

Time step size: [optional]

Thanks in advance for your help!

==============================================================================
                AMR-Wind (https://github.com/exawind/amr-wind)

  AMR-Wind version :: v1.4.0-17-g250778a3-DIRTY
  AMR-Wind Git SHA :: 250778a3306e96b46afc117fc722491acaeb6176-DIRTY
  AMReX version    :: 24.03-36-g748f8dfea597

  Exec. time       :: Fri May 10 09:51:15 2024
  Build time       :: May  8 2024 20:36:21
  C++ compiler     :: Clang 17.0.6

  MPI              :: OFF
  GPU              :: OFF
  OpenMP           :: OFF

  No additional third-party libraries enabled

           This software is released under the BSD 3-clause license.
 See https://github.com/Exawind/amr-wind/blob/development/LICENSE for details.
------------------------------------------------------------------------------

Additional context

[hgopalan]

Can you provide the relevant input files?

[passboy666]

Dear hgopalan,

Thank you very much for taking the time to help me review the .inp file. I truly appreciate your support. I will send the file to your email shortly.

Thanks again for your kind assistance!

[passboy666]

Excessive Nodal Projection Iterations and FLLC Not Working with TurbineFastLine (DTU 10MW Case)

Dear AMR-Wind developers and users,

Thank you for all the great work and support you provide to the community.

I am currently running a simulation of the DTU 10MW wind turbine using TurbineFastLine actuator with OpenFAST coupling (AMR-Wind v3.3.1, OpenFAST v3.4), and I’ve encountered two issues I hope to get some help with:

🌀 1. Excessive nodal projection iterations
In my simulation, each time step (Δt = 0.02 s) takes about 40 seconds of real time to compute, and I noticed that nodal projection requires up to 200 iterations per step. This significantly slows down the simulation.

I wonder:

Is this number of iterations expected under certain settings?

Could it be caused by my actuator settings, mesh configuration, or absence of AMR?

Is there any known fix, such as adjusting solver tolerances or modifying actuator mapping settings?

⚠️ 2. FLLC does not appear to be working
I attempted to enable FLLC in my TurbineFastLine setup using the following settings:

Actuator.T0.fllc = true
Actuator.T0.fllc_type = correction
Actuator.T0.fllc_relaxation_factor = 0.2
However, I don't observe any change in the output force distribution or power — and there’s no indication that FLLC is active. I've verified the parameters are correctly defined per the documentation. Could this be a compatibility issue with my current AMR-Wind/OpenFAST versions?

⚡ 3. Overprediction of turbine power output
At a wind speed of 11.4 m/s, the computed power is around 15 MW, while the expected design power is about 11 MW for the DTU 10MW turbine. I’m aware this is a common issue in actuator-based simulations, but:

I’ve tried adjusting epsilon values and the number of blade points — the results didn't improve.

I’m not using AMR adaptive refinement or any controller in OpenFAST.

Even after attempting to enable FLLC, the power remains significantly .

amr.txt

static_box.txt

[hgopalan]

For the 1. point you can lower the default tolerance for the Nodal and MAC projection to avoid hitting the 200 steps.

mac_proj.mg_rtol = 1e-4
mac_proj.mg_atol = 1e-6
nodal_proj.mg_rtol = 1e-4
nodal_proj.mg_atol = 1e-6

2. Can you follow this walk through: https://exawind.github.io/amr-wind/walkthrough/calibration.html and see if it changes the results?

Also what was the source of the DTU_10MW.fst? Did you run a stand-alone OpenFAST to check the results?

[passboy666]

Dear hgopalan,

I apologize for the delayed response.

One weeks ago, I followed the method you kindly provided, and since then the iteration count has reached a reasonable level — thank you again for your help!

However, I still have one critical issue. It seems that the FLLC correction is not being properly activated. At startup, there is no message indicating that FLLC is being used. Currently, I’m using 100 actuator points. That said, I do have a question regarding a point I’ve been analyzing. In my current simulations, the difference in the time-averaged power output between cases with and without blade deformation appears to be quite small. However, I’ve come across several studies suggesting that including blade deformation can lead to a reduction in mean power output by around 10%.

I’m wondering what might be causing this discrepancy in my case. Could it be related to actuator line settings, structural model fidelity, or perhaps other simulation parameters?

I would really appreciate your guidance on this.
Many thanks in advance for your time and help.

Best regards,

[hgopalan]

can you share the inp and fst files. I am not an expert on this topic but someone else may be able to help

[passboy666]

sure！many thanks

DTU_10MW.txt

amr.txt

[mbkuhn]

I looked into the code related to FLLC, and I asked a top user about this as well. There is no output to screen confirming the FLLC correction is on. The lack of a message for that does not indicate that it is not working.

Your second question is a lot more open-ended. I would expect that the simulation parameters, i.e., what the turbine experiences and the ability of the mesh to resolve the flow, would significantly affect the influence of blade deformation. Outside of guessing, you'll have to rely on the expertise of others and on the findings of your study.