[afikr]

Hi

I would like to ask if it is possible to prevent cfx solver from creating a .res file.

I am planning to use an interrupt control option for my run.

I am monitoring a value at a plane and would like to use the change of this value as my convergence criteria.

By doing this, I only want to save the .res file for the converged cases based on my convergence criteria.

So if I run 10 cases an once than I will only have the .res files for the only converged cases.

Is this possible?

Thanks

[Opaque]

Output Control/Results File/Option = None

That should do it.

[afikr]

Originally Posted by Opaque

Output Control/Results File/Option = None

That should do it.

Hi Opaque

If I choose this option, will it not create a .res file at all?

[ghorrocks]

Try it and find out.

The only way I can think of to do your original request is to use the "Edit Run in Progress" option. And I am not sure this option is one you can change with edit run in progress anyway.

Why is the results files for the non-converged simulations a problem? If you are running zillions of simulations and you want to manage disk space then write a batch file/shell script which identifies the non-converged simulations and deletes their results files. Or, even easier, do it manually yourself

[afikr]

Hi ghorrocks

I am running a batch run for hundreds of cases on a shared computing cluster.

So, yes, disk space is the issue.

I think it will be easier if I can stop the non-converged cases from being stored as a .res file.

[Opaque]

I assume you have a setup that will converge given a fixed number of iterations.

A simulation can be stopped based on multiple criteria; however, ignoring the residuals criteria is a dangerous approach. There is no quantity in the solution worth using if the residuals are not below a certain threshold (usually 1.E-4 or less).

Using another quantity as a convergence metric implies you have a strong handle on the ability to detect false convergence, and able to set such it is invariant to changes of units, precision, etc.

The only approach I see is to script the simulations, check if the case has converged (extracting the information of interest), and remove the results file if failed to satisfy your specifications. It means you need some disk space to hold the solution before deleting it.

[afikr]

The quantity I am planning to monitor as a convergence criteria is an addition to the residual convergence.

The reason is that the residuals have reached a certain limit and will not go down any further. In most cases, the residuals are fluctuating.

The interrupt control will only take into effect after few hundreds(or more) of iterations. By doing this, I can make sure that residuals have stabilised before using the interrupt control option.

Anyways thank you for your suggestion. I think that is the best way that can be done so far.

What I plan to do now is batch run the cases in smaller amounts at a time.

Thanks

[Opaque]

Have you analyzed the residuals that already refuse to converge?

You can output them, and post-process them in CFD-Post. The fact the residuals do not converge may be related to unsteadiness of the flow, or a poor mesh quality in a particular region of the flow.

It is always important to analyze the source of the problem before committing to something else. The residual theoretically should always be able to go down to round-off given the appropriate precision. There must be something preventing them to do so.

[afikr]

I am simulating a fan/compressor at near stall conditions. That explains the unsteadiness.

At near stall conditions, you cant really rely on residuals alone. That's why I needed to use another variable as a convergence criteria.