Simulation not converging even when residual is less than tolerance

[Mriganabh]

Hi all,

I am trying to explore the computational time required for coupling solid & beam elements in mastodon. I followed the procedure stated in example 5a. 

The model consist of 5 layers of linear soil with 4 HEX8 elements in each layer. A curve is extended on top of this layered soil block and overlaps with the top-most layer of soil (as shown in the figure-1). This curve is meshed using beam elements.

Figure - 1

The simulation aborts due to non-convergence even though the residual is much lower than the specified tolerance, which is 1e-6(see figure-2).

Figure - 2

I have attached the input files used. Please let me know where am I going wrong.

Thanks,

Mriganabh

[Christopher Wong]

After a good attempt to debug, I cant find anything obviously wrong with you input setup. Maybe try modeling the soil domain as a simple elastic solid, without any of that layering stuff, and see if it runs. Also, I am wondering why your beam is embedded by a depth of 1 node, this is an obvious difference from ex5a, although I don't see why it should cause any problems. Try removing that lower node and see, and only have that one coincident node at the surface, then see if the problem persists.

The problem is you're trying to run a pretty complex model right off the bat, you need to start with something super simple, and iteratively add more levels of complexity. Right now, theres no way I could find where your bug is coming from unless I deconstructed your entire model and put it back together piece by piece. Surely, I would find it then, but this is your model, so you'll have to do that work yourself :)

[Christopher Wong]

Also, notice that, even though your linear iterations are converging, your nonlinear iterations are not. Either the nonlinear relative or absolute tolerances must be satisfied, whichever comes first.

[Swetha Veeraraghavan]

Hi Mriganabh,

As Chris mentioned, the nonlinear residuals are not decreasing as fast. So unless the nonlinear residual are less than 1e-6, the solve is not converged.

How is the beam connected to the solid elements? Are the beam nodes merged with solid elements at the center of the top face? Most often the convergence issue at the first time step is because something is not connected properly in the mesh and this results in rigid body modes.

Regards,

Swetha

On Thu, 4 Jun 2020 at 08:48, Christopher Wong <crswo...@gmail.com> wrote:

Also, notice that, even though your linear iterations are converging, your nonlinear iterations are not. Either the nonlinear relative or absolute tolerances must be satisfied, whichever comes first.

--
https://mooseframework.org/mastodon
https://github.com/idaholab/mastodon
---
You received this message because you are subscribed to the Google Groups "mastodon-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mastodon-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mastodon-users/f2b8e5e1-4630-43c7-93dc-20876a2559d1%40googlegroups.com.

[Christopher Wong]

I checked his nodes, they seem to be connected properly. I could be mistaken though...

On Thursday, June 4, 2020 at 5:36:43 AM UTC-6, Swetha Veeraraghavan wrote:

Hi Mriganabh,

As Chris mentioned, the nonlinear residuals are not decreasing as fast. So unless the nonlinear residual are less than 1e-6, the solve is not converged.

How is the beam connected to the solid elements? Are the beam nodes merged with solid elements at the center of the top face? Most often the convergence issue at the first time step is because something is not connected properly in the mesh and this results in rigid body modes.

Regards,

Swetha

On Thu, 4 Jun 2020 at 08:48, Christopher Wong <crswo...@gmail.com> wrote:

Also, notice that, even though your linear iterations are converging, your nonlinear iterations are not. Either the nonlinear relative or absolute tolerances must be satisfied, whichever comes first.

--
https://mooseframework.org/mastodon
https://github.com/idaholab/mastodon
---
You received this message because you are subscribed to the Google Groups "mastodon-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to mastodo...@googlegroups.com.

[Mriganabh]

Hi Chris and Swetha,

Swetha: Chris is right. I have merged the nodes wherever necessary.

Chris: I tried what you had recommended. I modeled a single layer of soil and had a beam without any overlap, i.e. only the bottom node of the beam was shared. I faced a similar issue.

However, I was able to get the simulation to run by using nodal mass instead of a consistent mass matrix to calculate the inertial forces and moments for the beam elements. Can one of you explain why this happened?

Thanks,

Mriganabh