Dear Jason,
I fixed the errors you noticed, now the simulation runs correctly.
I inspected the results with the new model compatible with OpenFAST 2.4.0 and they are the same I posted with my initial question about the damping decrease after BeamDyn introduction. Hence structural damping, numerical damping and software implied to run the simulation are not the problem that causes damping decrease.
I attach the current OpenFAST correct model I used for simulations.
Honestly I have no idea to explain or solve this damping decrease.
Thanks for your assistance.
NREL Forum-OpenFASTv2.4.0.7z (139 KB)
Dear Simone,
Given all of the improvements made to BeamDyn over the past several years, I’m very surprised that you are saying the BeamDyn solution does not change between FAST v8 and OpenFAST v2.4. Can you share the OpenFAST v2.4 results for the ElastoDyn-only and ElastoDyn+BeamDyn solutions?
You said earlier that enabling blade flexibility in the ElastoDyn-only solution has very little impact on the free-decay response. Do you see similar levels of blade-tip deflection between the ElastoDyn-only solution (with blade flexibility enabled) and ElastoDyn+BeamDyn solution?
Best regards,
Dear Jason,
I repeated simulations using exactly the same model for both FAST8 and OpenFAST2.4, the differences are only related to the updates introduced with latest versions; in particular ElastoDyn file,ED blade input and ED tower input files, HydroDyn file, MoorDyn file are equal for both softwares, BeamDyn differences are only related to the lines introduced in latest versions.
The simulatios were carried out considering floating(6 platform’s DOFs) completely rigid model, floating(6 platform’s DOFs) flexible[(1st Edgewise+Flapwise mode)+(1st fore aft+side to side tower bending mode)] model using CompElast=1, floating(…) flexible(…) using CompElast=2(hence BeamDyn for rotor).
Most clear differences in the simulations results are related to tip blade deflection during pitch free decay simulations. I attached the figure related to previusly mentioned differences and two folders which includes all the significative figures of the two decay simulations performed by each software.
For what concern surge and pitch free decay, in my opinion pitch decay does not highlight any significative differences between the two softwares neither between the 3 structural models (rigid,flex,flex+BeamDyn); while surge decay highlights damping differences in between the two softwares related to rigid and flexible models, instead flexible+BeamDyn model gives same results for both softwares. In particular, surge free decay with BeamDyn is still less damped than other solutions.
In order to have another comparison parameter, I considered tower base fore aft shear force too. The result related to this parameter highlight high coherence between the 3 models and the two softwares, results are almost equal for all the considered cases indeed.
If you need more data, more parameter inspections or whatever, feel free to ask for that.
Thanks for your assistance, I hope I give you useful material.
Forum Figure OpenFAST.7z (4.24 MB)
Forum Figure F8.7z (2.34 MB)
Dear Simone,
Indeed, the BeamDyn solution is surprising. it definitely looks there is more energy / less damping when BeamDyn is enabled. I can’t really explain this right now.
Have you tried to quantify the different level of damping (using the local minima and maxima) and determine if this is more difference in linear or quadratic damping (via a PQ analysis)? In a free-decay simulation, quadratic damping tends to dominate at large amplitudes and linear damping tends to dominate at low amplitudes. Do the two damping levels approach each other at low amplitudes?
One way to dig into this more is to calculate the natural frequencies and damping through a linearization analysis rather than through free-decay simulations. Linearization functionality for floating wind wind turbines was introduced in OpenFAST v2.4. While this would take some work, I’d hope it could predict the change in damping level between these two models by examining the eigensolution and through the various linearized matrices (mass, stiffness damping), help you better isolate where the change in damping is coming from.
Best regards,
Dear Simone,
I noticed a couple of things in your input files.
First, the timestep in the main .fst file is too large for a simulation using BeamDyn. I would expect this to be on the order of 0.0025 seconds or smaller. This may be causing numerical instability in the coupling of BeamDyn to ElastoDyn – this is noticable in the low speed shaft (LSShftMxa channel).
Second, the twist distribution in the BeamDyn input file is not smooth. BeamDyn is not be able to represent it well with the Legendre Spectral Finite Elements (LSFE) that it uses in calculations. This may lead to some numerical instabilities in the calculations and a poor representation of the blade.
After reducing the timestep for the BeamDyn simulation, I see closer agreement for the Surge decay (I’ll post a plot once the simulation completes).
Regards,
Andy
Dear Andy,
I appreciated your advices. I performed a surge free decay simulation using time step equal to 0.001s which is half the time step you mentioned and it is one order of magnitude less than the previous one I used.
From the new simulation the result obtained is the same I obtained with larger time step.
The only parameter I changed in the simulation was time step, I didn’t modify twist distribution.
Did you change more parameters or did you modify only the time step?
I’ll wait until you post your results before post the PQ analysis performed on the previous data obtained with larger time step.
Thanks for your help and I apologize for the time without giving answer to Dr. Jonkman but I wasn’t familiar with PQ analysis, so I needed more time to make myself clear.
Simone Covre
Dear Simone,
I forgot to mention that I did modify the twist distribution. I set the first 6 keypoints to a twist of 42.8040 degrees (same as keypoint #6 originally). This really isn’t a good fit either and won’t be well represented by the Legendre Spectral elements either due to the discontinuity.
I also increased the order of the element from 4 to 5. This helps a little in achieving a better fit to the twist. With the blade better represented by the Legendre functions, the result is much closer to the ElastoDyn results.
We are in the process of making several updates to BeamDyn that should make this easier for end users:
- Better fitting of the keypoint line using a least squares fit of Legendre spectral functions (github.com/OpenFAST/openfast/pull/576)
- Summary file information with details on the fit to the keypoint line. This will allow for quick plotting to compare how BeamDyn is modeling the blade vs. what was given.
Regards,
Andy
Thanks, Andy! This is very informative.
Best regards,
Thank you very much for the explanation Andy and thank you all for the support.
Simone Covre
