search

Exawind / hfm-fsi

Input files and data for HFM FSI Work
0 stars 0 forks source link

Run single regioniii case of iea15mw with consistent blending #67

Open psakievich opened 5 months ago

pscrozi commented 5 months ago

Kevin having MPI issues. Might be build or system. Ganesh has seen a similar issue and excluded nodes having issues. Kevin excluded 5 nodes with issues, but still seeing errors. Probably not going to work by excluding nodes. See conversation in slack. Neil and Nate are seeing similar issues.

psakievich commented 5 months ago

@kevmoor has a single wind speed running in the short queue. He will open a PR on the changes he made to the input files to get it to run, and run for longer if it makes it to the end of the 2 hour submission without erroring out.

kevmoor commented 5 months ago

Kevin to update

pscrozi commented 5 months ago

Kevin had a mistake in the input script such that he wasn't using Phil's new module. With this fixed, Kevin ran over the weekend and ran 40 hours until it timed out. Only difference from what Ganesh ran before the break and now, is that there was a small fix with consistent blending. Kevin can keep running this. It isn't far enough to analyze results. Residuals at 2500 steps in look fine. No reason not to continue this run. Nate and Kevin will discuss restart process and make sure FSI is going fully.

pscrozi commented 5 months ago

Kevin and Nate documented how to do a restart and kicked it off. Currently in the queue.

Assumed that restart time is optional in AMR-Wind. It apparently is.

Matched up checkpoints. How close does it have to be? User manual says it will pull the closest one.

pscrozi commented 5 months ago

The restart was successful. We made it to 20 s. Re were 10^6, but now up to 10^9 and stalled, so may stall soon. Probably the AZ_blend_delta wasn't the root cause of the issues we were seeing. Probably the real issue is either: (1) driving down numerical error in ExaWind; (2) numerical sensitivities in OpenFAST. Are there errors in our simulation strategy, or are there structural sensitivities in the OpenFAST model. This is why we're doing the 5 MW to see if we can do a full power curve with that model. IEA15MW is more sensitive and flexible. A formal verification test for the FSI capability would be ideal. What are the parameters we need to drive down the numerical error.