Cyclic symmetry stepped modes extraction from nodal diameters

hooshsim

Initially, I did have asymmetric design but I made a cyclic assumption so I could run it. Is the output number “1” you referring to NGRAPH or else?

prop_design

no, what i'm referring to is the mecway cyc sym bc popup window. i attached a pic. where it says All you want to clear that out and enter the number 1. above that you enter the number of blades. I'm confused on your end goal. Just be careful. I'm not sure you can just arbitrarily switch to analyzing a symmetric design, when you're really going to build an asymmetric design. At that point, you might be better off just analyzing one fixed blade. One other thing I was thinking about is the center of gravity of your rotor will be way off without adding a lot of dead weight to bring it back to center. The aero loads will also be asymmetric because the blades are closer in that one area. So it's not a design I would ever recommend. I had one guy dead set on building one blade. I couldn't convince him to use two, lol. Whatever you can do with an asymmetric design you can do better with a regular design. Sometimes you don't get a choice on what to do and are forced to do things you know you shouldn't. So not sure what you have going on. However, I wish you luck with it.

ps; you shouldn't need to use any ccx commands. other than forcing pardiso if you want to use that solver. it should default to pastix. if you turn mixed precision off it is more reliable.

disla

¿Why don't you tie the different sectors to build the full model?. It works fine and Frequencies match with the Cyclic model.



The center hole is not really a bug. ND need a special treatment when there are some elements on the axis of rotation. I know because it happens the same when I apply (-y,x,0) BC. If X and Y are =0 the normals are ill conditioned.

disla

Hi Prop,

I have had an insight about your comment:

"it just repeated each blade individually, instead of them being all in one mode, like it should be".

Be sure to add the perturbation parameter for the contacts to keep working the whole analysis.

I'll bet you a beer that you're watching this

prop_design

@disla great idea about using contact on the fan model. i'll give that a try. on the a400m model, i haven't looked at the custom cards in ages. i'll go over them again and see if anything is missing. if you can post the cards you think it should have, that might help. as far as the modes, yeah that looks like what i'm seeing on that model. on the fan model it is a little different, with the extra modes. the a400m model doesn't have extra modes, they just aren't all in the same mode. so if you want 6 modes you have to run 6modes*8blades. which sucks. more memory, more solve time, makes cheap laptop sad.

prop_design

i ran the fan model by rotating one sector and using bonded contact, as @disla suggested. unfortunately, that didn't have any effect on the unexplained modes, when compared to the cyc sym model. there is one other thing i want to try. however, i'll do it with the a400m model, since the mesh is smaller.

prop_design

i believe my ccx cards of ok. i double checked them though. it looks like on the a400m model the extra modes are just the nodal diameters. it sorts by frequency for the full rotor. i think the results are the same as for the cyc sym model, once you take that into account. i don't think running it as complex frequency makes a difference. i double checked it again though. i also rotated the mesh and bonded it. so i used both ideas on the same model. however, the results were the same. i think the one thing it made easier to see though is the nodal diameters. once the mesh is exactly the same for every blade, the results are less confusing in some cases.

disla

There is a way to sort the incompatibility of nodes position between Master and Slave faces for Cyclic symmetry with the internal solver.
Ideally this should be done in the phase of model preparation. It requires to leave the model 0.1º-0.2º short .
Once the step is meshed, that small section that is short can be fulfilled with a slice of the same size build with revolve from the opposite side.
Then the slice is relocated and bonded to the model.
Internal is not multithreaded so I don’t really think this is needed but Anthony commented about having issues with this point.
Tested.

prop_design

i don't quite understand your process there. if you rotate one side to the other, isn't it still misaligned? you mention bonding that side i think? you can message me on here if you want, to explain further.

the last thing i wanted to try was too big to solve with 16gb.

Victor

@disla what a clever workaround bonding a slice of the opposite mesh to make them match! Never heard of that before.

prop_design

oh wait, i think i get what you mean now. yes that's clever, as victor says.

cwharpe

@disla I'm loosely following here. Are you saying you created a gasket of sorts that is bonded between each part? And positioned correctly by mirror/revolve from the opposite pole w.r.t. center hub, maybe from 180 deg position?

disla

Hi cwharpe,

Created a slice of the left face with revolve (1 element per thickness), disconnect slice elements, reposition that slice with rotation on the opposite side, bond and the model is ready for Cyclic Symmetry.

cwharpe

Got it. Thanks.

prop_design

@disla @hooshsim

i redid the fan model, to make it smaller, so that i could solve it the way i wanted. once i did that, the results of the full rotor make a lot more sense. so some of the tricks:

model one sector in cad, then mesh that sector in netgen. use the element quality plot to make sure you have a good mesh.

import the mesh into mecway and use the rotate mesh command to make the full rotor

then use bonded contact, to join all the sectors together, as the nodes will be way off at the seams (thanks to @disla for this idea)

lastly, i did a nonlinear prestressed modal with complex freq analysis using ccx

the best way i can find to solve the full rotor is use the (number of modes you want * the blade count). in this case (6 * 7 = 42).

now, when you get the results back, there will be results for all of the nodal diameters. however, they will be sorted from lowest to highest freq only. cyc sym sorts from low to highest, for each nodal diameter. the full rotor will be all mixed up basically.

the last remaining oddity is you will get a random amount of phase shifted duplicate modes. they seem to occur for the higher nodal diameters. on this version of the fan model, i had 18 of those. the interesting thing is i solved for 42 modes. if you subtract out the random modes, you get the 24 modes from the cyc sym analysis.

so it's still a pain to do a whole rotor. however, the above method helps a lot. previously, i was truly getting unexplained modes. i think the main culprit was netgen. however, doing the complex freq analysis should help some too.

if you want the updated models, let me know.

hooshsim

@prop-design
Please share your latest models. Did you run them with paradise solver? I’m still unable to run Pardiso. Have tried other options posted per Victor, still no luck.

prop_design

that's really weird about your pardiso issues. you should be ok to run them with pastix though. i would turn mixed precision off via the environment variable. in some of my files you would have to change PARDISO to PASTIX in certain places. i'll work on posting the updates. it will take about 10 mins or so.

prop_design

here is a link to the latest versions; https://1drv.ms/u/s!AubMNYNIQyqKgnffPAZOPcudy2Lc?e=CdDdPt

i took the old link down to save online storage space. i updated all of the files except for the solve time spreadsheet.

one thing that occurred to me with model size issues is it looks like around 700k nodes will work for any analysis with 16gb of memory. so if you had a full rotor if it's less than 700k you should be fine. an easy way to figure the cyc sym limit is to just divide 700k/# sectors. so that should be your mesh size target for one sector. i think if you output all sectors it will probably still work. it gets a lot slower outputting all sectors, but i think this might keep it from going out of core. my ssd is about 11x less bandwidth than most peoples. my memory is about 3x less bandwidth. so as soon as it hits out of core, it's game over for my laptop. victor has also said you need a patch to get that running for ccx pardiso. however, even if i patched the software it wouldn't help the laptop any. also, the smaller the node count, the more likely you are to be able to animate the mode shapes. i usually can't do that. i can output a video of them though. so that's good.

ps;

the limits i mention are for pardiso. i think pastix is easier on memory. so you may be able to solve a larger model with pastix. although, mecway still needs a lot of memory to output all sectors. my formula takes mecway into account also. the formula is just from my experience on my laptop. so you might find you can use a slightly different formula.

Victor

@hooshsim have you tried compiling the CCX included with Mecway? That's the best option for big models if you don't have enormous amounts (~64GB) of RAM.

hooshsim

@Victor. How do I compile the CCX included with Mecway? Do I need compiler? Thanks!

Victor

Instructions for installing the compilers and other tools and building CCX are in here, reproduced below

%programfiles%/Mecway/Mecway19/ccx/ccx_win64_mkl_pardiso_source_2.19.zip

Instructions for building the following on Windows:
  64 bit CCX with SPOOLES single-threaded
  64 bit CCX with MKL PARDISO multi-threaded and with Out-Of-Core mode available


1) Install MSYS2 64-bit
  From https://www.msys2.org/


2) Update MSYS2
  Open "MSYS2 MSYS" shell and enter this command
    pacman -Syu
  Open "MSYS2 MSYS" shell again and enter this command
    pacman -Sy


3) Extract this archive's ccx folder to msys64/home/<username>/ so that the paths are msys64/home/<username>/ccx/src/...


4) Install tools and compile

  Open "MSYS2 MSYS" shell again and enter these commands
    pacman -S mingw-w64-x86_64-gcc
    pacman -S mingw-w64-x86_64-gcc-libgfortran
    pacman -S mingw-w64-x86_64-gcc-fortran
    pacman -S msys/make
    pacman -S msys/perl
    cd ~/ccx/src
    ./build.sh


5) Find the compiled ccx.exe, ccx_MKL.exe, and their dlls in

  msys64/home/<username>/ccx/src/x64/install

hooshsim

Currently, I have Mecway 18 installed on my machine. Should I install v19 before I follow your suggestion. The link you referring has v19.

Victor

No need. It's exactly the same archive.