Fluent Diverging -- "Divergence detected in AMG solver"

  • 531 Views
  • Last Post 17 January 2019
  • Topic Is Solved
LukeM posted this 16 January 2019

I am having trouble with some UDFs in Fluent. I am using a UDS, and after the simulation runs for 92 time steps, it consistently crashes. The strange part is that all of the residuals look fine -- all normalized residuals are 10^(-2) or lower. I even went down to the AMG solver level (because the error Fluent provided to the console was divergence at that level), but the iterations leading up to the "explosion" seem to have low residuals (again, less than 10^-2). Nothing points to divergence until it happens.

After some playing around, the source term for the UDS seems to be the culprit. When I set that term to zero or just tell Fluent not to solve that UDS equation, the solver can pass through 92 time steps without issue. But when I turn the equation/term back on, Fluent crashes. Another weird issue is that I can significantly change the source term itself (dividing it by two, ramping it up slowly, ramping it up then keeping it steady after time step 80, etc.), but the simulation consistently crashes at time step 92.

What could be causing this issue and how can I solve it?

 

I also attached the console output after the crash (UDS-0 is the problem).

Attached Files

Order By: Standard | Newest | Votes
LukeM posted this 16 January 2019

 For context, my system is simply a box with liquid initially clinging to the sides of the box and gas in the center. My system with the gas in red and the liquid in blue

kkanade posted this 17 January 2019

as ansys employees we can not download attachments. 

i am not sure about settings. but initial guess is you may need refined mesh. 

seeta gunti posted this 17 January 2019

Hi,

Did you check the solution while solution is progressing?

May be if you can check the UDS source term and species-0 in b/n , you will come to know whether solution progressing correctly or not. 

Try reducing the URFs of species and UDS. 

Try running with first order schemes. This may help you .

Regards,

Seeta

LukeM posted this 17 January 2019

I attempted a refined mesh, but the finer mesh actually moved the crash earlier. 

I checked the solution while it was progressing, and all the residuals appeared normal. I lowered the source term in the UDF, and that was unsuccessful. 

I had tried cutting my time step in half, which delayed the crash, but it still happened. I eventually cut the time step by an order of magnitude, and the solution did not crash. I am hoping that was the solution. 

 

Thanks to all for the suggestions. 

Close