ANSYS Fluent 18.1 vs 19: Different solutions to exact same model?

  • 997 Views
  • Last Post 17 January 2019
eufrat posted this 22 June 2018

Hi all,

I have access to two different ANSYS versions, 18.1 and 19. I often use them parallel to calculate slightly different models at the same time. 

A few days ago I ran the same exact same model on the two different computers for reasons of comparing the calculation time. One computer has far more power than the other one.

Well, I encountered that the solutions were not the same. For example computer A diverges at some point, while computer B does not and keeps on calculating. 

As far as I know, this should be completely impossible. The model is the same, the boundary conditions are the same, so ANSYS should come to ONE solution at computer A and B. Differences in calculation speed might occur, of course.

Both simulations run in double precision. I built it on Computer A, saved it as a case + data file and copied it on Computer B.

I am completely confused. This should not be, so maybe some of you have suggestions what the reason might be for this phenomen.

Thanks in advance,

eufrat

  • Liked by
  • SaisudheerM
Order By: Standard | Newest | Votes
peteroznewman posted this 23 June 2018

Hi eufrat,

The title of this post mentions different solutions, but the content talks about divergence during the iterative solution process.

A solution in my mind would be a pressure value or a velocity value at some point in the model after acceptable convergence was obtained. Those values for a converged solution should be very close for different versions of Fluent.

An improvement to the software would be if a model running in 18.1 diverged and gave me no solution, now converges in 19. I hope that is the case, but it could be the reverse, you don't say which version diverged.

I am more familiar with structural nonlinear models and I have seen differences in solution times between versions, but also in the ability to converge without user intervention. For example, an older version may iterate along, find no convergence for that substep and stop with an error, while a newer version quickly bisects the substep size, resumes iteration and continues to convergence. I'm a little annoyed when a newer version takes longer to solve than an older version, but I believe it is because the newer solver may be more conservative in its iteration strategy to avoid a convergence failure on other, more challenging models.

I'm interested to read what others may say about your observation.

Kind regards,
Peter

eufrat posted this 23 June 2018

Hi Peter,

the version which diverges is the newer one, version 19.

When I compare the solution data of version 19 (before the abort) to version 18.1 they are not the same. Thats why I wrote about different solutions.

I just looked into the convergence criteria and in both versions they are the default values, 1e-3. No difference there..

This is so weird, it drives me crazy..

  • Liked by
  • SaisudheerM
abenhadj posted this 04 July 2018

Hi,

I recommend to help yousrelf out by consistent debugging. Focus one version, read mesh from scratch and setup your problem. Ensure for deep convergence and check the efficiency of partitioning and partition boarders.

 

A.

 

 

Best regards,

Amine

  • Liked by
  • SaisudheerM
SaisudheerM posted this 15 January 2019

hey hi eufrat

Did you got any explanation regarding solution difference in different versions. Even I encountered such problem,Please let me know you any thing

Thank you

Best Regards

sudheer

abenhadj posted this 17 January 2019

Please feel free to create a dedicated thread for your request if you do not receive any answer from eufrat.

Best regards,

Amine

Close