AnsweredAssumed Answered

simulation result not symmetric

Question asked by Magnum So on Aug 6, 2014
Latest reply on Aug 18, 2014 by Jared Conway

Like • Show 0 Likes0
Comment • 41

Hi All,

I have tried a static study in simulation to learn thermal expansion

model dimension: 200mmx50mmx50mm with a T leg

material: alloy steel

fixtures: fixed geometry on the T leg bottom face

external loads: temperature 50 Celsius (ref temp at zero strain=20 Celsius)

mesh: standard mesh

The max displacement is 5.676e-002mm.

I expect that the four edges should have the same max. displacement but the probe shows that they have different value. So is there any tolerance in solidworks simulation?

No one else has this question

Outcomes

41 Replies

1 2FirstPrevious

Knut Fiskergard Aug 7, 2014 2:16 AM
Mark Correct
Correct Answer
The difference is very very small. (0,5%?).
I don't do much thermal work, but I think this is what you should consider to be equal.
The result is affected by the mesh (is it absolutely symmetric?) and numerical round-up/-down during calculation and extrapolation of the shown result.
Like • Show 1 Like1
Actions
Magnum So @ Knut Fiskergard on Aug 7, 2014 3:03 AM
Mark Correct
Correct Answer
I don't know how to make sure that the mesh is symmetric. I have tried different method to extrude the above model and found that the number of total nodes and total elements change every time. So I cannot avoid such non-symmetric result, right?
Like • Show 0 Likes0
Actions
Knut Fiskergard @ Magnum So on Aug 7, 2014 3:58 AM
Mark Correct
Correct Answer
I think you should not expect the results to be symmetric - as a hand calc would be.
I don't think Simulation has the tools, as for example Ansys has, to take control of the meshing.
FEA is a numerical approximation. Highly dependent on element size (density), quality, round-offs......
In structural analysis one would check for convergence of the result. How much does the result change if you make the mesh a bit finer in areas with stress concentrations? If the change is less than 3-5%, if that is an acceptabel criteria, the solution is said to have converged and the result is reliable. (you don't want to run with to many elements either)
I would say your results are symetric.

BTW- meshing is not dependent on how you generated the geometry. It is an independent operation.
Like • Show 2 Likes2
Actions
Nicholas Luyster @ Knut Fiskergard on Aug 7, 2014 3:13 PM
Mark Correct
Correct Answer
Hi Magnum,

Knut is right. The mesh will not be symmetrical. SolidWorks Simulation has too much automation to make a symmetrical mesh. In this case, a difference of .5% is effectively zero. FEA is not that accurate.

All the best,

Nick
SolidWorks Simulation Training
Like • Show 1 Like1
Actions
Magnum So @ Nicholas Luyster on Aug 7, 2014 9:45 PM
Mark Correct
Correct Answer
Thanks!
I have tried to increase the model length from 200mm to 2000mm, found that the difference in four edges displacement has increase to be 1.5%
Does it mean that an increase in model scale cause an increase in 'tolerance' in FEA?

In professional field, how to simulate stress issue or thermal issue in large-scale model? Or Solidworks FEA result is just for a reference?
Like • Show 0 Likes0
Actions
Jerry Steiger @ Magnum So on Aug 8, 2014 12:25 AM
Mark Correct
Correct Answer
Magnum,

Most engineers would be very happy with results within 1.5%. Our assumptions about the model and the boundary conditions are usually not good enough to worry about differences that small.

Jerry S.
Like • Show 2 Likes2
Actions
Knut Fiskergard @ Magnum So on Aug 8, 2014 2:59 AM
Mark Correct
Correct Answer
In professional field FEA is the standard method because of the complexity of the tasks, and then add non-linearities (ex. material yielding). But in time you get experienced in different ways of analyzing the result, and always remember the assumptions you made! Change a boundary condition slightly and you could get quite a different result.

It looks like you have added your thermal load on an edge. If this was a structural load, the results would not be reliable close to the edge (st.venant principle). Don't know if it will have a similar effect on a thermal load??

This model does not seem to take long to solve. If you have used Simulation default element size - what happen if you manually set an
element size about half of what you have used?
Like • Show 1 Like1
Actions
Magnum So @ Knut Fiskergard on Aug 8, 2014 5:27 AM
Mark Correct
Correct Answer
I fixed the bottom face, then apply a 30 degree temp. increase in whole model to observe its expansion.
apply default mesh density: four edges have different displacement value (5.673e-001, 5.674e-001, 5.675e-001, 5.678e-001)
apply fine mesh density: two edges in same side have same displacement value (5.677e-001, 5.677e-001, 5.676e-001, 5.676e-001)

Seem that finer mesh density can reduce but not cancel the difference.
Like • Show 1 Like1
Actions
Knut Fiskergard @ Magnum So on Aug 8, 2014 5:56 AM
Mark Correct
Correct Answer
yepsi pepsi - And then the question becomes: How accurate do you need it to be in order to answer the question asked.
Like • Show 1 Like1
Actions
J. Mather @ Knut Fiskergard on Aug 8, 2014 7:34 AM
Mark Correct
Correct Answer

Knut Fiskergard wrote:

BTW- meshing is not dependent on how you generated the geometry. It is an independent operation.

Meshing IS dependent on how you generated the geometry. (in SolidWorks)

My students do an experiment to demonstrate this every semester.
Like • Show 1 Like1
Actions
J. Mather @ Magnum So on Aug 8, 2014 8:17 AM
Mark Correct
Correct Answer

Magnum So wrote:

Or Solidworks FEA result is just for a reference?

What does the F in FEA mean?
What other reference do you have? Actual physical testing/measurements against which you can compare your digital model analysis?
In the physical models that you have tested - is there any measurable variations between tests?
Like • Show 1 Like1
Actions
Knut Fiskergard @ J. Mather on Aug 11, 2014 6:03 AM
Mark Correct
Correct Answer
It does?? Really??? How and why?
Could you please share the experiment that demonstrates this?
This could explain a few of my issues with the Simulation mesher.
Like • Show 0 Likes0
Actions
Knut Fiskergard @ J. Mather on Aug 11, 2014 6:10 AM
Mark Correct
Correct Answer
I sense a retorical question here:
"What does the F in FEA mean"
F stands for Finite - how does that answer Magnum's question?
Like • Show 1 Like1
Actions
Per Engberg Aug 11, 2014 9:53 AM
Mark Correct
Correct Answer
When doing FEM in symmetrical cases it is often advantageous to simplify/reduce the calculation by just running it on one half of the model, using the fact that the symmetry plane will remain flat (frictionless support).
Like • Show 2 Likes2
Actions
Knut Fiskergard @ Per Engberg on Aug 11, 2014 10:21 AM
Mark Correct
Correct Answer
yes, but.....
Excelent answer! Wish I made that point.
But still - no matter how symmetric the input are (if you take the time and effort to analyse the whole part). You must expect "asymetric" results (within numerical round-of accuracy).

And the bigger the part - the more advantageous it is!
Like • Show 0 Likes0
Actions
Jared Conway @ Per Engberg on Aug 11, 2014 3:37 PM
Mark Correct
Correct Answer
this is the way to guarantee symmetric results. also it means half the solve time.
Like • Show 0 Likes0
Actions
J. Mather @ Knut Fiskergard on Aug 11, 2014 7:07 PM
Mark Correct
Correct Answer
Here is the lab experiment.
You should agree that the solid geometry is identical, only difference is how it was created.
Verify that you get different results and then I will come back with discussion on logic.

Others are invited to run this experiment as well to document results.
Attachments
- Lab 1 - Why Different.pdf252.5 KB
Like • Show 1 Like1
Actions
Knut Fiskergard @ J. Mather on Aug 12, 2014 6:11 AM
Mark Correct
Correct Answer
I agree that the final volume is identical.
But I totally disagee with a CAD modelling approach that add geometry within exsiting geomery when you are working with native files. As an intermediate state, finishing up with some bolean feature or similar, to obtain desired geometry is of course ok.
The lab task only have content on 1,5 of the 3 pages. Don't know if that is all or if something went wrong during upload.

I have tried the task, but I do not get different results.
I made 1 sketch and made two extrudes, selecting countours in the sketch (another approach I don't like - find it to be unstable for more complex geomery after making changes to the sketch).
The last feature is set to merge with the first...
Am I missing something from the task?
Like • Show 1 Like1
Actions
J. Mather @ Knut Fiskergard on Aug 12, 2014 8:32 AM
Mark Correct
Correct Answer

Knut Fiskergard wrote:

1. I agree that the final volume is identical.

2. But I totally disagree with a CAD modelling approach

3. I have tried the task, but I do not get different results.

I did not include the entire document as we also run this test in three other programs, I snipped most of it out to avoid irrelevant discussion. I thought I had included all the relevant (trivially simple) information needed to reproduce the behavior. The question was never about modeling approach - I simply reversed engineered a much more complex problem down to as simple an example as I could generate that would demonstrate that differences in modeling technique (I estimate that 90% of what I see I would call "garbage" from a modeling technique perspective and often from a geometry perspective) has an impact on results. When I first started teaching FEA I was confounded why "identical" assignments students were getting different (we will discuss relative significance of differences later) results. I began to analyze modeling technique as part of the process. The fact that you report that did not get different results is interesting as we have run this test hundreds of times in SolidWorks and have never gotten identical results between the two models in the experiment as written. I would be interested in seeing your two files.
Like • Show 1 Like1
Actions
J. Mather @ Knut Fiskergard on Aug 12, 2014 8:28 AM
Mark Correct
Correct Answer
Yes, I was presenting a rhetorical thought question to the OP. Trying to provoke a logical thought process about why the reported results aren't perfectly symmetrical.
A Finite number of elements/nodes are defined.

Imagine being a spider with a different type of "vision" than human. This special spider can only "see" code. It is tasked with building a web (mesh) within a given volume. The hard part for the human, with only the usual human visual experience, is to imagine the difficulty of "seeing" this code and selecting a start point for creating their web and creating it in such a way ("seeing the code") that it will be perfectly symmetrical to the human view of the world.

From the big picture view, does the difference in reported symmetry of results have any any real significance, or, compared to the overall results (and real-world variation) is the difference rather insignificant?
Like • Show 1 Like1
Actions
Knut Fiskergard @ J. Mather on Aug 12, 2014 8:34 AM
Mark Correct
Correct Answer

I might have mix the name on the files, hope that doesn't affect my grade, sir
Attachments
- Lab1-a.SLDPRT.zip60.3 KB
- Lab1-b.SLDPRT.zip65.2 KB
Like • Show 1 Like1
Actions
Jared Conway @ Knut Fiskergard on Aug 12, 2014 1:02 PM
Mark Correct
Correct Answer
i also wouldn't expect the way it was modeled to get different results. orientation might. because i don't know how it decides what order to start meshing faces. that could seed the mesh differently. but again, it would be minimal.
Like • Show 0 Likes0
Actions
Mikael Martinsson @ Knut Fiskergard on Aug 13, 2014 5:15 AM
Mark Correct
Correct Answer
Well, I don't get this lab experiment either and perhaps we're missing some information?
In my trials, the result is the same between the two methods, based on my knowledge of FEA.

But of course, if you don't have the knowledge of the method, and just look at the numbers, it is not exactly the same.

Different thing can affect the results:
- Boundary conditions
- Mesh
- Solver

Boundary conditions effect are obvious.

Solver, well if you use the iterative compared to the direct sparse there will be differences since one is solving the equation directly and the other one makes a guess, verify the result and the guess again until it is "close enough".

Mesh, obviously the size and type of element effects the result. But also how the mesh layouts on the part since SW don't have the possibility to do a mapped mesh.
In my trial I get the results below where you can see that the tets are turned in different direction on different locations.

SW simulation solves the displacement and then extrapolates stress and strain from this value.
Displacement easier converges compared to stress and in my example the displacement is exactly the same between the two results.

The maximums stress is different, but of course this has more to do with the fact that it occurs as a singularity in the sharp inner corner.
It will continue to rise to infinity with finer mesh so it is pointless to look at this result.

SW also modifies the scale based on the maximum value, so to really compare them you need to set the scale similar in both examples.
So in the overall picture, with the same scale settings, I say that the result is the same between these two design methods.
Like • Show 2 Likes2
Actions
J. Mather @ Mikael Martinsson on Aug 13, 2014 7:16 AM
Mark Correct
Correct Answer
When I get I chance I will go back and check my instructions.
I think I snipped out something I should have left in the instructions.

In our original problem - the intention was to show that although exactly the same results were expected - the actual difference in results were not significantly different.
The point of the lab was not to expect 2+2=4 type of answers when doing FEA. We are looking for "ball-park" results. Are we in the ball-park and where?
Like • Show 1 Like1
Actions
Jim Riddell Aug 14, 2014 6:47 AM
Mark Correct
Correct Answer
I have to agree in part with the Professor, I just meshed the identical model (duplicated the tab w/in the program) and ended up with different mesh numbers. Both used curve based mesh with 12 mm size. The first one had 476310 nodes and 247959 elements and the second had 476318 nodes and 247960 elements. There were no differences between the modeling of the "two" parts as it is the same model. Meshing changes.

Does this negate the validity of the FEA? Certainly not. Think of it (as I have heard of students doing) as getting an answer of x.xxx e-15 on one of those fancy TI calculators - that's still 'zero' in a hand calculation.
Like • Show 1 Like1
Actions

simulation result not symmetric

Outcomes

Attachments

Attachments

Related Content

Recommended Content