Use of solid elements

Mishal

Hi Guys

I traditionally use shell and beam elements for most of my analysis work ,on occasion I have used solid elements dependent of the model I have developed.

Most of the stuff I deal with will be plate work structures which have fairly think elements (6-16mm thick). Based on what I see on the forum most people use solid elements for analysis, even for instances where I would use shell elements.

Is there any reason for this?
The only legitimate reason I can see is the limitation on CAD, where most people would use fairly low cost software which cannot midsurface geometry thus solids would be the easier route to go...

Sergio

Using shells and beams requires more planning, preparation of the CAD and experience than just pressing a magic button that create a nice (but incorrect) tet mesh. Users tend to be lazy :-)

In the past have worked a lot with engine mounts, doing FEA on rubber and some steel sheets parts, and we use always solid hexas for the sheet metal parts. When I ask to my boss why the company has choosed to model it in that way, he told me that as we work everyday with solid elements, we save a lot of mistakes and considerations due to bad result interpretations (lower/upper side stress...) and get more accuracy with contact situations (even using Abaqus and other solvers).

Add to the count that CCX has several drawbacks with shells and beams, and when you see the expanded results you must understand and convince yourself of the accuracy of some artifacts... and then try to explain to your customer or boss!

Mishal

@Sergio I would assume that you need to a decent amount of elements through the thickness of the solid in order to get accurate results using solid elements, am I correct?

I was once lazy and used solid elements for some sheet metal on the old 32bit Strand7, which developed really poor results and took ages to solve. I was told by the vendor that solid elements should not be used for sheet metal.

prop_design

one thing to keep in mind is ccx expands shells into solids. if you really want to use shell elements, then you would have to use the mecway internal solver. unfortunately, that currently doesn't have a nonlinear option. most of the stuff i do uses solid elements. i can occasionally use shell elements, however, they are still expanded to solids. i generally never use beam elements anymore. in the old days, beam elements are all you could use. shells was state of the art for gov agencies with huge budgets that could afford the supercomputers to run them. it took awhile before fea became mainstream. now, my laptop is more powerful than those supercomputers. whatever element choice you need is available. as you said, it's up to the user to know which to use and when. i don't know if you guys are referring to the users on this forum or not. i have never met any lazy or stupid users on here. so i kind of take offense to that. they have all been very smart and hard working.

Sergio

@Mishal We mesh our parts with mapped quad elements, and then extrude to get three hexa elements in thickness. If the metal/plastic part was just modeled to represent the contact, we use only one element.

Victor

Since this comes up a lot, I thought I'd make a comparison between solids and shells to get a feeling for the error. Here's the well-known Scordelis-Lo roof benchmark but made more extreme with 1/10th the usual thickness, giving it a ratio of radius of curvature to thickness of 1000:1.



It shows that hex20 in a single layer requires about 10-20 times as many elements as quad8 for the same displacement. Bear in mind that this benchmark was designed to be difficult even for shell elements.

As you'd expect from what @prop_design said, both elements perform the same with CCX, but if you use reduced integration (C3D8R and S8R), they seem to be even better than the internal solver's shells.

prop_design

some other notes that are related. using shells is necessary for composites, because that's the only way you can assign element orientations. they are ignored for solids. the other thing i have found very useful is the ccx shell expansion method outputs stress in the zz direction. i see this fail a lot, because the fibers are running in the xy plane. with shells, you can't output stress in the zz direction. so i have come to like the solid expansion that ccx does. however, it also causes a lot of problems when creating models. in any event, using beams and shells would be necessary for very large structures like bridges and buildings. there would simply be too many elements to use solids. things like fishing rods and table/chair legs would be good for beams. it all depends on what you are modeling. this isn't the best program for true shells. mecway recently made a big improvement to the shell formulation. it used to act more like plate elements, even though they were called shell elements. now they act like shell elements. the main issue is no nonlinear solver for them. of course, no stress in zz either. that limit would probably always exist, because it's an assumption of using shells in the first place.

Victor

@prop_design, solids can have element orientation for orthotropic and anisotropic materials, but yea, they don't have composites unless you build up the layers explicitly in the mesh, which isn't really practical for too many layers.

prop_design

@victor the last time i tested solids, you couldn't change the element orientation. if you did, it just used global axes as it normally would. i haven't tested it in several years though. did something change?

update; did a quick test with the current version. it looks like you can define custom element orientations, but not in the same way as for shells. thus, i can't get the layout i would need. i'll email you about it. in any event, what i was originally trying to say is when i use composites i have to use ccx shells (which are expanded to solids).

Mishal

I didn't know that Calculix also expanded plates into solid elements also.
I assume that there is a really poor mesh over the thickness of the element if Calculix is doing this automatically.

It actually makes a lot of sense now regarding the solving time in certain instances.

Am I correct by saying that solid elements can perform well with poor aspect ratios?

Victor

Yes, CCX shells are only 1 solid element in the thickness direction. It shouldn't do much to solving time though because a hex20 solid has 20*3=60 DOFs while a quad8 traditional shell has 8*6=48 DOFs.

Yes, hex20 can perform well with a high aspect ratio as seen in my example above (30:1). With reduced integration, it looks like 100:1 might still be as good as a traditional shell. Tet10 is probably far worse, so I wouldn't use automeshing for this.

prop_design

hi,

i had the same question you did. i did a lot of tests and it didn't seem like it hurt results any. i was surprised that the high aspect ratio expanded shells perform well. i use them for laminates as it's the only way i can model those with this software. you can have as many laminate layers as you like. it makes a new layer of solid elements for each laminate. however, you can do it in a practical manner with 4 to 8 layers. the more layers you add, the more detail you get through the thickness. you can see this by looking at stress plots. four seems to be enough. it helps to keep the model smaller. the biggest issue with this method is it greatly adds to the node count. you need a lot more memory than you would if true shells were used. however, the solids give stress in the zz direction and this seems to be very important in some instances. so it's worth the tradeoffs in my opinion.

to see how much it adds to the model, look at the node count in the preprocessing section. then look at it in the post processing section. mecway displays the counts in the lower left of the screen. you will see the node count is much higher in the post processing section.

anthony

TSpargo

I realize this is an old thread.. I was planning on using shells as well, but they have turned out to be a real pain to get working. If you do use solids instead of shells, how do you adjust for the extra strength you get from the geometry being solid? I know the inner volume doesn't contribute much to the overall strength, but still- all of your results will be optimistic- maybe significantly in some cases.

Victor

I've never heard of modelling a hollow structure as a filled solid by neglecting or adjusting for the internal material's stiffness. Instead, it would usually be thin solid elements forming the walls.

The simplest way is to model the walls as solid in CAD and just make a solid mesh of them. But that's often too inefficient on node count because it makes thin tet elements. You can do better by making a quad-dominant surface mesh then extruding to solid elements so it's hex-dominant and similar to CCX shells expanded to solids.

Shells can be finnicky with CCX but are often the easiest to mesh. What problems were you having?

TSpargo

Self intersecting faces/elements. I'm new to this so trying to keep it simple to start off with. Trying to learning 3D CAD s/w + evaluating 3 different FEA tools, kind of overwhelming.