Discussion Forum

Hi

You are useing "Draw Assembly mode" with contact pairs that are free, that meas that you model is three layers mainly free to slide.

I suppose you should operate in "non assembly mode" for your case but it depends what you are looking for.

You can understand this better if you turn on the Postprocessing and select "Deformed Shape" on.

Are these three objects resting on each other by contact (+ gravity) or are they fixed/glues/welded ?

Not evident to understand as is for me

Good luck
Ivar

Thank you very much for this usefull comment.

I could succesfully simulate a simple 2 layers structure case by removing "use assembly" module.
However, as soon as the structure is a little more complex, I can't get any results again... (cf attached files)

The structure is as follow:
upper part: prescribed displacelement
sides: axial symmetry or symmetry or free
lower part: fixed

the 2 lower layers are attached together (chemically attached)
the top layer is comming from the air and is indenting in the 2 laower layers.

Any further comment would be very usefull.

Best regards,

Hi

Definitively, if you have a structure in two layers that are firmly in contact, type weld, then you should consider them as a composite object with an interiour "continuous" (but non-sliding) boundary. But if you have an objct that is only "in contact", the contact par at this level (hence assembly mode) is required.

This means for your "simple example", combine C=1 + R1 into one "Composite object", then have the sphere R2 as separate object (in draw mode) then use assembly (between the half sphere R2 and the new combined object). This makes a hierarchy of your objects and the interiour boundary is treated as such = continuous interiour boundary, against an contact "pair" boundary. By the way, I would also define the axi symmetry axis of the "ball" as such and not as "free", even if 2D axi mode of COMSOL probably assumes this by default.

There is also another "smoother" ways to simulkate simple contact pairs, see the thread of "problem with contact pair-penalty factor

By the way try to split the boundary of your sphere such that it has a small lengtth, comparable to somewhat greater than te true contact region, it would simplify the convergence, and to allow to have a fine mesh where needed.

Second applying a vertical force/pressure load, rather than a displacement might also influence positively convergence. And I would not constraint it radially at all and select highly non linear problem. Furthermore there is a scaling issue, your convergence criteria is >1E16 for me this means that we are probably close to numericl errors or ill conditions matrices.
I would need still a few hours playing to get it run as is, and I have no time just now, perhaps tonight ;)
But you could try it out too, its en excellent exrercice to learn how to tweak the solvers, check your doc

Good luck
Ivar

Thanks. In fact, to make a composite structure is really a great idea. It works better now although the full displacement range is never reached.
Concerning your idea of applied load instead of constraint displacement, I need to have an idea of forces.
However, on global variable plot, lm2 variable is not recognized, even if I check weak constraints "on".
Is it because large deformations is "on"?

Best regards

Hi,

I'm doing a similar kind of modeling with indentations and I'm having trouble accessing the normal force data. By default, comsol has the data set to von mises stress but I need the normal force in the vertical direction (z) for analysis. If I set the cross-section plot to Body Load in Global z-dir nothing shows up on the graph.

Thanks

Edit: I got it. I had to use boundary integration (is this correct?).

Hi

the lagrance multipliers "lm..." can only by defined on boundaries with Dirichlet conditions (if I'm not totally wrong) and are there to compensate for potential gradient jumps

In general you could estimate thee forces by integrating the "reaction forces" on the fixed boundaries

You should take a systematic look and compare the different forces:

subdomains intergation :
===================
volumic stress quantities
strain energies
body loads
reaction forces
lm

boundaries integration: (identify when to use "compute surface integrals in axi, check your units)
==================
surface traction forces (stress strain related)
edge loads
contact pressure (when applicable)
reaction forces
lm

To understand how they are calculated, you have the doc, but also:
"Physics - Equation systems - Subdomain/boundary Settings - Variables"

Once you catch these subtilities you will better know which one to select in which case: I know it takes time, I have an am passing through this a many times, and each time I jump from one main physics subject to the next I have to repeat again, but there are so many items behind here that one must just really understand where to look,and what to expect

have fun comsoling
Ivar

Small comment on sherical_ind:
I would use "axisymmetric" BC on the axis, in you case I do not beleive it would do anything, as for structural it's implicit, but its good practice to be sure all physics will be correct

Your mesh is really too coarse, at least in the contact region, in 2D it is easy to "box" a region ang get the mesher to refine just the box.
And you seem to build up a singularity on axis I notice, perhaps there is a missin "*r" or 2*pi*r in your physics link somewhere, check carefully the units. As to avoid on axis singularities COMSOL uses the matheamtical trick to multiply many values by "r" and divides later in the solution to stay coherent

Thanks for the quick response. As for the singularity do you mean I should compute the surface integral? (In postprocessing -> Boundary Integration)

Also, should I have the lower boundary of the rectangle "fixed" or prescribed displacement with Rz=0? I would like the rectangle to be able to expand radially but of course, not move vertically.

I seem to be getting convergence issues with larger indentation depths and also when I do not use hyperelastic models.

Hi

You can use Rz=0 as it is fixed "on axis at "r=0", this would be the same as a "roller" condition with a fixed point on axis.

The singularity I noted was when I used a far finer mesh an looked at the von mises stress. it really concetrates at one point on axis, and this I do not like, it makes me believe something is wrong in the settings, but I did not manage toidentify what it was

You should alays be suspicious if you get a stress conctration, it can come from the BC's but also from something wrong in the physics coupling, and in anycase it tends to hide the interesting from your plots

Have fun Comsoling
Ivar

I was not able to solve the stress concentration issue so I decided to try the same problem in 2D geometry. The solver still has problems with convergence. I'll upload the file soon as I do not have it with me right now.

PS: I even tried making the sphere much smaller than the indentation sample but it did not make a difference.

I also decided to try a conical indentation to see what happens. Here is a summary of the problems I am encountering.

Spherical Indentation:

Solver does not converge when I make the model displacement controlled (the variable used in the parametric analysis is for displacement) or force controlled.

Conical Indentation:

Converges for both displacement and force controlled but still does not exhibit the correct behavior. (The values are not just off by a constant).

Only the initialized mesh has been used so far in the models.

The values we plot are from

Postprocessing->Boundary Integration -> Reaction force z-direction vs displacement parameter on #2 in the displacement controlled model.
Postprocessing->Boundary Integration ->Force Parameter vs z-displacement on #2 in the force controlled model.


According to some papers I have, the force v depth for the conical indentation should exhibit behavior like y=ax^2.

I have attached the files below.