How to write immediate results and stop a running simulation?

[M.Brassel]

Does anybody know a way how to force the Moldflow Insight solver writing results and then stop the currently running simulation?

I've got a simulation running on a linux cluster, which gives no intermediate results due to a bug in the solver.

According to the analysis log it should lead to a short shot, as the injection pressure has reached the machine limit after approximately 1.4 s (50 % of filling).

After 2.3 s 72 % are reached.

After 12 s there are still only 72.6 % filled and already 100.00 % frozen.

After 30 s the gates are closed, with only small changes in filling. The flow rate is around 0.01 cm³/s.

After 89s until 130 s there is still no end reached. The flow rate is around 0.001 cm³/s.

But if I cancel the simulation, I will have no results, so I won't see where the flow front hesitates and stagnates.

As this simulation iterates for several hours now, I would like to stop it, look at the results and prepare a better model with changed parameters.

[Javier.Jubierre]

Hi @M.Brassel,

Thanks for posting this issue on our Forum!

If your model is mid-plane or dual domain, you question has an easy answer. In process settings you can define intermediate output files that will write down the current simulation state at predefined time-steps.

This is, however, not possible for 3D simulations. 

I hope tihs helps.

Regards,

Javier

______________________________________________________________

If my post answers your question, please click the "Accept as Solution" button. This helps everyone find answers more quickly!

[bernor_mf · Accepted solution]

Hi,
you cannot force the Insight solvers to output results at a certain analysis step in an ongoing analysis run.
The solver get's the input of when to write intermediate results when starting as described by @Javier.Jubierre in Process settings Advanced options.

If you cancel the simulation, is should keep intermediate result. Well, it does on Windows...
@M.Brassel wrote: "I've got a simulation running on a linux cluster, which gives no intermediate results due to a bug in the solver."
So, it does not write intermediate results at all, for any mesh type, or does not show dynamic update?
For which releases do you see this?

Anyway, as you do not see this intermediate result being written when running on linux due to a bug or defect in software,
I wonder if you can access the analysis log from Windows?
If you can, then you maybe could set up a script that check the analysis log for flow, and when no expected volume increase at a certain expected time
send a message to manually cancel, and yes, results are lost...

Or just do this checking of analysis log manually as initially described in post.

Further, then in a new study set a switch over just before volume stagnates, when no volume increase, reset process settings to get a switch over at that volume, a short packing packing at seen injection pressure when stagnation and no cooling time (and not automatic cooling), which should make analysis complete at give a result. Important not to hit that volume that causes the issue of simulation not converging.
A temperature result might then help to indicate/find the hesitation area. At least you know where end of flow front is, and might be able to
draw conclusions from that.
Not so straight forward, but maybe a way to get some results to be able to make progress?
If this is a defect in linux Insight solver it seems as two simulations needs to be submitted.
A bit awkward...

Would be a nice option to send a "write intermediate result now" signal to solver. And that it works to write intermediate results for solver on linux...

If you get solver warnings in analysis log, you might need to change the solver parameters convergence options.

Hope this helps.

Regards,
Berndt

[M.Brassel]

Hi @Javier.Jubierre,

well that is a nice answer for mid-plane and dual domain simulations.

At least when we know the problem exists.

But in this case it's a 3D-mesh.

The biggest problem lies in the Linux version of the Simulation Job Manager, as this gives no intermediate results or display updates during analysis. Perhaps this will be better with the Moldflow 2017 R2 version?

[M.Brassel]

Hi @bernor_mf,

thanks for your hints on how to do a general problem analysis.

Well in this case it's a 3D mesh model and it is solved with the Linux solver Moldflow 2017 Insight.

I could try it with a Windows solver, if the machine has enough capacity an is not used for other simulations.

The Linux solver gives no intermediate results at all. You always have to wait until an analysis step completes, then you will get the results of the completed analysis step (like fill+pack or warp).

However the analysis log can be viewed inside of Synergy, if the project is opened. If the study is still open, the analysis log is even updated during simulation, as it is with the Windows solver.

If I knew it beforehand that the error would occur after some hour(s) at night, I could do the checking and cancelling by a script, assuming I knew exactly how to analyze the analysis log and I knew in which directory to find it. It's a bit tricky with the new Job Manager, as it creates and leaves lots of empty directories during simulation.

I tried your proposal of setting a V/P switch over near the critical point and a short packing period, as this seemed reasonable. Though I needed two iterations, this helped to see, which regions are filled at which time. So I had an idea how (much) to change the runner system as well as the process settings and valve gate behavior.

Despite the complicated way of doing several analyses to see a first result, I was able to solve this case with your contributions.

[bernor_mf]

Hi @M.Brassel,

you can specify when to write intermediate results for 3D as well, by constant interval or time.

But seems not to work for Insight 3D solvers on linux for 2017.

Regards,

Berndt

[M.Brassel]

You are right. I use these settings myself, especially to get more detailed results. But this changes nothing in the behavior of the Linux solver, that gives these intermediate results only after finishing the analysis step.

[Javier.Jubierre]

Hi @M.Brassel, hi @bernor_mf,

Well, both settings are different.

From one hand, only midplane and Dual Domain, can export output files (you may find the setting on my first reply). From the other hand, we can force the 3D solver to write intermediate results to get a better insight of the evolution one parameter during the simulation. This option, will generate a single output file that will embrace more solutions at predefined time-steps. In any case this option/setting will write temporary output files. This is working on the same way in Linux and Windows OS.

I hope this helps!

Cheers,

Javier

______________________________________________________________

If my post answers your question, please click the "Accept as Solution" button. This helps everyone find answers more quickly!