Campbell diagram

[tchatte3]

Dear openfast team,

I am wondering if there is any reference I could be pointed to regards to creating a Campbell diagram in openFAST. Sepcifically, I was looking at a Campbell diagram in the Bladed software and would like to create something similar in openFAST. I understand, I should be starting from the linearization analysis in openFAST, but I am not sure how to compute the frequencies of the various DOF's of interest in the linearization analysis? Should I be using just the PSD's from a time-series of the DOF's ? Any help or guidance regarding this matter would be extremely valuable.

Best Regards,

[ebranlard]

Hi,

We typically run the linearization using a set of matlab scripts (compatible with Octave). The scripts write a set of openfast input files for different operating conditions, run fast, postprocess the linearization files using a Campbell transformation, and then perform a plot. Before plotting, a manual identification of the modes is typically necessary.

I'm currently working on updating these matlab scripts. If you want, you can clone my fork of the matlab-toolbox repository:

    https://github.com/ebranlard/matlab-toolbox

You can try running the script Campbell/example/runCampbell.m. It's only an example, but hopefully by reading the comments you can adapt it to your needs. I hope to merge this into the official matlab-toolbox repository in the coming weeks. I'd recommend doing the linearization without a controller at first (CompServo=0, GenDOF=False) to get familiar with the process.

We are also in the process of doing different updates on the linearization functionalities of OpenFAST (soon we'll have a trim option to automatically find periodic steady states, instead of running for a long time), so things will likely change, but hopefully the matlab scripts will not be affected significantly.

Feel free to report issues in my matlab-toolbox repository since this is not an official release yet.

Emmanuel

[tchatte3]

@ebranlard Thank you very much for pointing me to right direction. The matlab-toolbox would certainly be helpful in this regard.

[tchatte3]

Hi @ebranlard, I am currently facing some issues with the MATLAB-toolbox for Campbell diagram. I was able to fix a few myself, but I am stuck at the final one. I am using MATLAB 2019b

The code fails in postproLinearization() specificially in campbellData2TXT>ShortModeDescr routine. For ShortModeDescr, seems the array is going out of bounds. This is the specific error:

Index exceeds the number of array elements (14).

Error in campbellData2TXT>ShortModeDescr (line **)
Desc = CD.Modes{i}.DescStates(CD.Modes{i}.StateHasMaxAtThisMode);
(inside routine DescCat = ShortModeDescr(CD,i) )
This is happening because CD.Modes is trying to access max(nModesKeep) = 15 in the for-loop.

Any suggestions/guidance on how to overcome this issue?

##############################################################################
Another follow-up question. If I have say 4 operating points (decided by wind speed, rotor rpm & blade pitch: 4 such unique lists avaiable) can I just perform 4 linearization analysis just ar t = 0 (without allowing the systems to evolve in time)? In which case is there any necessity of invoking the controller? Would the controller do anything?

Any guidance would be really helpful, as I am completely new to this field.

Best Regards,
Tanmoy

[tchatte3]

@ebranlard I was able to do a work-around by just running loops from 1 to NModesKeep-1.

Best Regards,
Tanmoy

[ebranlard]

I've submitted a small change that I hope will fix the issue you saw.
ebranlard/matlab-toolbox@d5a06bd

Even at constant rpm (without controller and generator) you need time for the simulation to reach a steady state, since you will likely have transients at the beginning of your simulations. You need time for the tower top deflections, blade deflections, shaft torsion to reach an equilibrium position. These transients will be present with or without a controller.

With a controller on, you also have to wait for the balance between generator torque and rotor speed to take place. You also need to use the simple controller and not the DLL. It's usually more work, that's why I suggested to do linearization without the controller first.

OpenFAST can do linearizations at any time, but for a Campbell diagram, you do want a periodic steady state to be reached.

[tchatte3]

@ebranlard Thank you for the detailed guidance. This is really helpful. I will use your new merge.

PS: The fix you suggested works great. A minor point, with variable nModesKeep, I had to do the initializations

MFreq = zeros(nModesKeep,nOP); 
MDamp = zeros(nModesKeep,nOP);
MDesc = cell(nModesKeep ,nOP) ;

with a temporary nModesKeep = 20, which will later get updated in the for loop

for iOP = 1:nOP
    CD=CampbellData{iOP};
    nModesKeep=min([length(CD.Modes), 20]);
    for i = 1:nModesKeep
        DescCat = ShortModeDescr(CD,i);

Best Regards,
Tanmoy

[tchatte3]

Hi @ebranlard - - A followup question, is there a way I can cotrol the number of output modes to be dumped in the Campbell summary. From the description and the getCampbellData.m it seems that the modes can be collected from the output Channels in the module files we are using. However, when I look at the summary files I see 17 modes and their frequency/damping information are outputted independent of the number of the output channels in Elastodyn/Servodyn say.

When I am comparing with another software Bladed" I see that modes like Rotor 1st edgewise collective" exists, but for the matlab toolbox I see only progressive and regressive modes, but no collective modes. Is there a way to tap those extra variables from the toolbox, or they need to be added as matrices for eigenfrequency analysis separately?

Best Regards,
Tanmoy

[ebranlard]

Hi, Thanks for the updated tip above, I'll include that.

The number of modes is related to the number of states of your system, typically half. When you use ElastoDyn you'll typically have between 12 and 17 modes based on which DOF are activated. You can see the states listed in a .lin file. The matlab code attempts to perform an identification of the full-turbine modes, based on which states are excited, which is quite a difficult task, which is why this is typically done manually for now. If you use BeamDyn you will have more modes, but the identifications will be quite difficult. We are currently working on some visualization, but the process is quite involved.

The first collective edgewise mode is usually combined with the drive train torsion, since the collective edgewise motion of the blades will induce torsion of the shaft. This is why it doesn't appear.

I hope that helps,

Emmanuel

[tchatte3]

@ebranlard Thank you. Yes, I was seeing that the .lin files the number of continous states is 30 X 30 matrices, so it does make sense. So, is there a way to decouple the edgewise collective mode from the drive train torsion? Or any rule of thumb to compute the frequencies of edgewise collective from the drivetrain torsion mode?

Best Regards,
Tanmoy

[ebranlard]

Your edgewise collective frequency will be close to the individual blade edgewise frequency but will increase with increased rotational speed. You could run with a fixed drivetrain (GenDOF=False, DrTrDOF=False) to get these frequencies out at different RPM.

[tchatte3]

@ebranlard , OK this makes complete sense, Thanks a lot.

Best Regards,
Tanmoy

[tchatte3]

Hi @ebranlard : A follow up question related to the interpretation of the modes in the Campbell analysis. (CampbellDataSummary.xlsx)

I read the mode shapes signed magnitudes, but I believe I am not interpreting them correctly. For the NREL 5MW turbine case, when I am using No Servo for operating points for OP: 3 mps wind speed, the .xlsx file says that mode number 2 corresponds to Tower F-A mode 1, @ wind speed 3 mps.

However, when I look at

Mode 2 signed magnitude in column H, I guess this is the mode max state for the different degrees of freedom and not really the mode shape. I guess, to properly extract the mode shapes I need to tap out the Eigenvalue_save vectors from the eigenanalysis in the tool-box, is that correct?

Best Regards,

[ebranlard]

Hi @tchatte3,

I think your interpretation is correct, but the thing that can be confusing is that the first column lists states. When using EalstoDyn these states have an easy interpretation since they correspond to the amplitude of a shape function, and each shape function induces some predefined displacements on each node of a body, in a single direction. If you were running Beamdyn, you'll have hundreds of states for the blades, each representing a displacement in x y or z of a node of the blade. In this table, you'll have hundreds of lines, and you'll see that the states that corresponds to x-displacements of the blade nodes have a strong amplitude (indicating some flapping of the blade).

What you can read from the table is that, for this mode 2, there are some important flapping amplitudes and tower fore-aft amplitude. This is likely expected for what we call the "tower fore-aft mode" of the full structure. Given the low frequency, we can identify this as a tower mode, and not the 1st collective or the 2nd collective flap. I personally like to look at the "Summary.txt" file to identify the modes. It is usually not enough, and you need to look at the continuity from one operating point to the next.

We are currently working on some mode shape visualization (#373), and we'll release this soon. This process requires to run OpenFAST twice, so it is more involved and ressource-heavy. You can try compiling this version, and using the matlab script "runCampbell_Trim.m" in my repository. You will need "paraview-python" if you want to generate avi files. Otherwise, you can open the vtk files using a regular paraview software.

You can find some documentation below, but this is out of sync with my version of the matlab-toolbox. My version of the matlab-toolbox basically tries to automate the steps mentioned in the documentation. You can read it to see what's happening behind the scene, but you won't need to follow the steps mentioned or download any of the files:
https://github.com/OpenFAST/r-test/blob/pullrequest/bjonkman-linear/glue-codes/openfast/5MW_Land_ModeShapes/vtk-visualization.md

I'd appreciate your feedback on running the script "runCampbell_Trim".

Thanks

[tchatte3]

Hi @ebranlard , thanks for the detailed response. This is indeed very helpful. I see, ok I will checkout #373 and use the runCampbell_Trim option that you suggested.

Best Regards,
Tanmoy