Inquiry about FAST.Farm Simulation Parameters and VTK Post-processing Configuration

[6Mr-Y]

Hello,I am currently conducting simulations for the IEA 15 MW wind turbine using FAST.Farm, and I have encountered several technical issues during the calculation process. I would appreciate your professional advice on the following questions:
1.Regarding the parameter Mod_AmbWind in the .fstf input file. I have consistently set its value to 2 in all my previous simulations. Could you please explain whether switching the value to 2 or 3 will bring significant discrepancies to the final simulation results?
2.For turbulent wind field generation via .bts files, is it mandatory to use six distinct random seeds to generate six independent .bts files for each wind speed case?
Taking the hub-height wind speed of 10.59 m/s as an example, does it require six unique values assigned to RandSeed1 in the .inp file to create six separate .bts files, followed by six corresponding FAST.Farm simulation runs?
I only adopt a single random seed for all my simulations so far. I would like to know how much impact a single seed will impose on the reliability of my results.
3.I intend to analyze wake velocity deficit by exporting VTK files for post-processing in ParaView. However, when running a 4000-second simulation, the exported VTK files consume extremely large storage space, requiring nearly 4TB of memory to load.
Could you kindly recommend proper configuration adjustments to reduce the memory footprint of VTK outputs, while still guaranteeing valid wake velocity deficit analysis?
4.Determination of the time step DT in the .fst file: For the IEA 15 MW turbine model, I configure the blade structural solver as CompElast = ElastoDyn combined with BeamDyn. At present, I set the global time step DT = 0.001 s. I wonder what criteria and principles should be followed to select a proper DT value for this coupled structural solution framework.
Thank you very much for your time and support. I look forward to your reply.Best regards

[jjonkman]

Dear @6Mr-Y,

Here are my answers to your questions:

1. FAST.Farm has distinct low- and high-resolution domains because they generally require quite different spatial extents and different spatial-temporal discretizations. With Mod_AmbWind = 2 in FAST.Farm, the turbulent wind inflow must use a fine spatial-temporal discretization to be sufficient for the high-resolution domains, but a large enough size to be sufficient for the low-resolution domain. It is generally very computationally expensive in TurbSim to generate turbulent wind inflow that meets both needs. Mod_AmbWind = 3 was developed to get around this issue. Given this, we intend to deprecate the Mod_AmbWind = 2 option in a future release.
2. The goal of any analysis with turbulent inflow should be to ensure that the results are statistically converged for the application. While mean values tend to converge quickly, standard deviations, and high-order statics (such as mins/maxes) take longer to converge. Obtaining convergence be achieved by running multiple shorter simulations with different seeds or a longer simulation. Given the computational expense of the inflow generation and ability to run inflow generation and FAST.Farm simulations in batch mode, it is typically more efficient to run multiple shorter simulations.
3. It would help if you shared your FAST.Farm input file to see how you have set up the VTK visualization output already. But certainly, outputting only slices through the domain rather than the full domain and writing only a subset of time steps instead of all time steps should help. Also, if your goal is to look specifically at the wake deficits instead of the flow field, you could also output that directly without the VTK visualization.
4. The choice of OpenFAST time step (DT) has been discussed in other OpenFAST questions in this repository and our forum, e.g., see: https://forums.nlr.gov/t/ptdm-implementation/4543/15.

Best regards,

[6Mr-Y]

Thank you very much for your previous reply @jjonkman . After reviewing your advice, I still have several follow-up questions for further confirmation:
1.You mentioned that Mod_AmbWind = 2 brings extremely high computational cost. May I take it that there is negligible difference in simulation accuracy between Mod_AmbWind = 2 and Mod_AmbWind = 3? Currently, I adopt Mod_AmbWind=2 to compute the full-domain turbulent wind field, and a single run takes more than 4 days to finish. Could you confirm that Mod_AmbWind=2 is still a valid choice for my simulation if I have no restriction on computation time and hardware cost?

2.My total simulation duration is set to 4000 seconds with tstart = 400 seconds, which leaves an effective runtime of 3600 seconds. Under this setup, is it unnecessary to perform six repeated simulations with distinct random seeds for average results? Running six cases will result in an unaffordable time cost for me.
3.Regarding my VTK input parameter file as shown in the figure, the main content I want to study is about the deficit in wake velocity, and I hope to produce an effect chart as follows




Looking forward to your feedback.
Best regards,

[jjonkman]

Dear @6Mr-Y

Here are my responses:

1. I can see why your TurbSim simulation would be computationally expensive! Your set-up should be reasonable (though expansive) for a single wind turbine (or row of turbines), but you can imagine how unreasonable it would be if you had a whole farm to simulate! Mod_AmbWind = 3 is meant to be accurate enough while being more practical for farm-wide simulations.
2. A single 1-hour simulation for a given set of conditions is quite common for wind energy applications, although experience has shown that more seeds or longer time (from 60 minutes up to say 100 minutes) is preferred to ensure statistical convergence.
3. The setting of WrDisWind = TRUE with WrDisDT = DEFAULT would generate many large files. I would suggest increasing WrDisDT (if suitable for your application) and switch to visualization slices rather than the full flow domain to save time/space.

Best regards,

[6Mr-Y]

Thank you for your reply @jjonkman . I will follow your advice for the simulation. Thanks again