x6 difference in displacement between internal solver and CCX (beams+shells)?

JMF11

Hello,

I developped the attached model with 1D beams simulating a mast, attached to triangular sail simulated with shells. Mast is clamped at the base. The third corner of the triangular sail is constrained by 0 displacement constraints in X,Y,Z. A uniform pressure is applied to the sail. This a streamlined version of a proposal by Disla

The 1D beams use a square solid bar, which is the only common shape to CCX and the internal solver.

I struggle to understand why I get a huge x6 difference in displacement at the top of the mast between CCX (about 5cm) and the internal solver (about 30cm) (shape of the sail displacement has however similarities). Tens of % would be OK, but it the difference is really a lot.

As all parameters are common between the 2 solvers, I only change the solver and run the simulation.

I checked the "mast" part, removing the "sail" and applying a force at the top of the mast. I get the same displacements with CCX and the internal solver.

Help to understand welcomed. Am I missing something?

The file is attached below,

Best regards,

JMF

prop_design

hi,

it looks like it has to do with mesh density for CCX. i changed the sail to a laminate and that helped a lot. then i did a x2 mesh refinement and it helped some more. the supported element types are different between solvers, do get laminates to work. so i broke the models into separate files. the mast is line3 for ccx and line2 for mecway. the sail when from tri3 to tri6 for both solvers. for the file you posted, it looks like mecway is right. it's at least very consistent with the changes. the displacement is close to 12 inches, it seems. when you said you had a 6x difference. i was seeing a 3x difference in your original model. with ccx being about 3x lower than mecway.

anthony

Victor

A couple of likely problems:

Tri3 elements are the worst kind of shell and may be overstiff in bending with CCX though the internal solver should be OK with simple flat geometry like this. You should normally only use them to fill gaps in quad-dominant meshes. I know you may have been constrained by the linear beam elements but better to use quadratic shells and beams for CCX and linear shells and beams out of necessity for the internal solver as long as the sail is flat. If it's curved, you need quadratic shells.

Unrelated to the differences between solvers, it looks like the sail's behavior will be dominated by membrane action which linear analysis neglects, so use nonlinear.

prop_design

i also made a ccx nonlinear model. it's pretty close to the linear run. i normally see a big difference between the two. you can't run mecway nonlinear models for several releases. hopefully, that will be reintroduced.

i agree with victor though. i run everything in nonlinear. these days, the speed of the solvers is so fast, it's not worth doing a linear model. they usually are not very accurate. at least for the types of things i run. i think this is just a rare case that linear would suffice.

JMF11

Hi, thanks a lot for the feedbacks, and hints.

As I run the free version of the software, as hobbyist, I can' solve the models from prop-design. So I will try redo it within the 1000 nodes limit.

As my testing of the mast alone did not, with line2 was not showing any difference between CCX and the internal solver, I had not digged in using line3 for CCX.

I had tried the tri6 elements, but had the same results/differences as with the tri3. I will try to mesh mostly with quad elements.

@prop_design, what is the intention with moving to laminate instead of an homogeneous shell (if I understand well, your material is 8 layers of the same thing, adding to 2mm) ?

One question on Mecway usage: I did nout found the way yet to assign beam elements automatically for the mast as a whole once the sail is automeshed. So I do it manually by node pairs. Still OK with a coarse mesh, but not with a 1000 nodes mesh. The tutorial explains the process with refinement (1) create a long beam, (2) then refines the beam in let's say 10 segments, which will inherit the beam element. But when I automesh 2D the sail, it replaces all my neatly spaced nodes.

=> How to best proceed to have the mast and sail share the same nodes, with beam elements along the mast line?

And I will run the analysis "non-linear" whenever possible...

Best regards,

JMF

ps: in my exemple, I really see a x6 beween 5cm displacement for CCX and 30cm (12") for the internal solver.

prop_design

hi jmf,

that's odd about you seeing 6x diff. i'm seeing a 3x diff. i'll check again. but i'm sure i checked that a few times. i did change a few things though.

basically, i couldn't find anything wrong with your model. but i saw a big diff in results, like you reported. so when i looked at the model and saw how thin the sail was, that is what led me to try laminate. i had tried just x2 but it failed to solve. the element quality was a problem. so i stuck with the original mesh and ran a laminate. i did four layers at first. it seemed to help a lot so i switched to 8 layers. it's easy to add an remove layers. the basic idea is that more elements through the thickness is usually a good thing. even for solid models. so i wanted to see what would happen. i tried x2 again and it happened to work. that helped get the results even closer.

for the rest of your questions, i'll leave them to victor. he would know best, since he is the developer of the program. i'm just a casual user these days.

anthony

update; yeah i checked again and i'm seeing a 2.92x diff on what i have as your original model. i did do a few small changes. i attached it here. maybe you can open this one (since there are no lams and the mesh is the same). i didn't clear the results, so you shouldn't have to resolve it.

prop_design

update,

just switching the sail to tri6 brings a lot of agreement. for ccx i use line3 for the mast and switch it to line2 for mecway. since meway is giving an error for line3. but they both agree pretty well and you can probably solve this model. the mesh should be small enough. note that i was seeing higher displacements with the laminate sail though. so this model is probably not super accurate. i would go with the ones you can't solve, lol. sorry about that.

Victor

@prop_design 6x displacement is dynamic, looks like you changed to static where I also see 3x difference. Good idea to get agreement in static first. I also get good agreement in static using tri6.

@JMF11 line3 is to get proper continuous connection with tri6.

You can typically use a coarser mesh with quadratic elements to end up with both a lower node count and better accuracy than linear elements.

A way to create line elements on shell edges is select the edge faces, extrude 1 layer in any direction, then move the extrudes nodes back onto the original ones and merge. The extruded shell elements get automatically collapsed into line elements.

JMF11

Thanks a lot for all the explanations and the tip for creating line elements on shell edges. I will try this evening (end of leaves today, so now back to normal work).

I will also try to work with quad dominant meshes of quadratic elements.

As the next step is to relax a bit the "free" corner of the sail, to let draft build in (displacement constraint with table).

Best regards,

JMF

prop_design

hi jmf,

your model is interesting to me, so I've been playing around with it. i was wondering if the attached images would be closer to an actual sail. are sails really connected to the entire mast? I don't know anything about sails. The reason I ask is I am having all kinds of weird stuff happen when I try to work with the mesh. I think some of the files I attached previously may have some sort of issue. Even though the results were agreeing better. I think there is something weird going on with ccx as far as the beam to shell connections. So I made a completely new model and still had problems. When I only connect the sail to the mast at two points, things agree and also seem more realistic. I still have to figure out how to connect the sail to the entire mast and be able to trust the ccx results. However, I don't know if that is even realistic or was just your first pass comparison.

notice all the linear models agree pretty well. however, nonlinear is drastically different. that is what i normally see. i couldn't run nl dynamic, as it was taking too long. i also couldn't run modal superposition as it was very slow and when it got to the transfer results to mecway stage it was killing my machine. i had to kill the process after about 30min. i also tried some runs were i added damping. close to a damping ratio of 0.02.

prop_design

ok. i finally figured out what was happening. in the original model only the ends of line2 elements were connected to the shells. so if you converted the beams to line3 for ccx the connections were still the same. so the midpoints of the line3 elements were not connected to the sail. i was connecting those to the sail. so i was getting different results. so basically what i found is how many connection points from the mast to sail makes a big difference. connecting all the nodes is the lowest displacement around 8-9 inches. connecting just the ends of the line2 or line3 is around 10-11in. then connecting just the start and the end of the mast to the sail is a huge displacement of 50+ inches. since mecway doesn't support line3 you can't do a comparison of some ways of building the model. however, when you can do a comparison, they aren't too far apart.

prop_design

on the laminate question. it looks like ccx fails with jacobian errors unless i use 8 lams. i tried lower numbers and it fails. oddly it will run static and nonlinear with any number of lams and shells. so maybe it is doing an additional check for dynamic analysis only.

JMF11

There are different types of sails:
- main sail: fully attached to the mast. It is the type I'm interested in,
- jib: triangular one "attached" to a cable running from the top of the mast to the front of the boat,
- Spinnaker: big very flexible sail for downwind sailing. Big bubble attached to the top of the mast with 2 "free" bottom attachment.

My application for landyacht sails.

What I would like achieve (one can dream :-):



Best regards,

JMF

prop_design

thanks for the clarification. that's cool.

JMF11

I implemented a model based on the advices given above (attached below):
- quad dominant mesh,
- quadatric elements (line3, tri6, quad8)
- non-linear static analysis
- slowly load the shell surface, and then after start to play with the third corner of the sail (relax or tension).

This starts to give interesting results. Difficult for me to confirm the values, but the overall behaviour looks realistic.

I used Victor process to create the line elements on the shell edge. I succeeded to apply the method to generate a line2-tri3-quad4 mesh. Then I converted the elements to line3-tri6-quad8 with the mesh tool.

I failed to apply the method with quadratic elements directly.

Thanks for all the given advices !

Best regards,

JMF

prop_design

hi jmf,

yes, i couldn't use victor's method for quadratic elements either. when you do the move it doesn't work right. not sure why.

Victor

@JMF11, @prop_design, with quadratic elements, you have to also move the midside nodes, not sure if that was the problem.

prop_design

oh thanks. yeah, that is probably it. i didn't select those.