FF: Weird behavior on the flow going into AeroDyn

[rthedin]

I'm seeing some weird behavior when comparing standalone OpenFAST and FAST.Farm. Here's my setup, contrasting the two:

• Inflow: TurbSim-generated. Every setting is the same, with the exception of the domain extent (spatial resolution is kept the same) and coherence values
• Turbine model: exactly the same
• FAST.Farm setup has 2 turbines, single InflowWind (Mod_AmbWind=2), curled wake model, and no wake-added turbulence enabled. I'm only looking at the leading turbine here.

The no WAT is for debugging reasons, I plan on turning it on later. I'm expecting the statistics of the turbine quantities to match across the two cases before adding WAT.

First I can look at the actual TurbSim boxes and see if the turbulence is following what I'd expect. For this specific case, I have a reference bts file provided and I'm creating my own, so I compare them. (Not related to the problem at hand- but the only differences are the coherence parameters). The standalone label on the plot below is for the standalone OpenFAST case (grid extents the same as the reference), and the FF single box is the single-InflowWind FAST.Farm case. The box in this case is much larger, but has the same temporal and spatial resolution. Here is the comparison of some basic statistics:

I can pull some numbers so it's easy to compare:

                  grid(ny×nz)    dt  z_hub   meanU   sig_u   sig_v   sig_w  TI_u%
reference           100×55    0.010  2.188   4.190  0.0645  0.0547  0.0548  1.540
standalone LC1.1    100×55    0.010  2.188   4.190  0.0638  0.0535  0.0551  1.524
FF single box       141×117   0.010  2.180   4.190  0.0657  0.0541  0.0556  1.567

So it looks like the turbulence boxes are all okay and consistent. When I run FAST.Farm and compare the leading turbine with the standalone OpenFAST execution, the leading turbine's AeroDyn sees different Wind1Vel* values, which really trickles down and affects the other turbine quantities. Interestingly, there is a large y component of the wind:

It could be due to the wake model, but I think this difference is too large to be from that.

Executed using rc-5.0.1 at f84321a. Tried on andrew-platt@0df2251 but I get a segfault (different issue, still looking into it). All cases executed with OpenMP on.

[jjonkman]

Hi @rthedin,

I'm not sure but using different spatial extents in TurbSim with the same discretization will result in different time series at most points in the grid.

I would first try running standalone OpenFAST with the same TurbSim bts file used in FAST.Farm; do you get the same result from WT1 in that case?

Best regards,

[rthedin]

I agree that using different spatial extents will result in different time-accurate data everywhere. I'm not expecting the time-series to match second by second at all -- I'm just aiming for the statistics to match. And they do. On the first set of plots above (and table), you can see the statistics are pretty much the same.

[rthedin]

Fixed by #3428. With VelInterpCubic as False we get the following. See description on the PR for a thorough description of the issue. Tried with polar and curled wake and results are as expected.