[parth_k]

Hi,

I'm encountering some random spikes like as seen in the figure attached. This nature is only visible with my wall based heat transfer variables dataset. The flow variables show no irregularities. I'm running a CHT 3D simulation for long enough times for it to reach pseudo-steady state using adaptive timestepping and on a super fine mesh. The convergence criteria is maintained at 1e-5 RMS.

Kindly suggest the ways this can be improvised. Residual and the test variable plot attached. Thanks.

[ghorrocks]

This could be numerical error:

If so then tighter convergence tolerance, better mesh quality and/or finer time steps will help. Also make sure you are using double precision numerics, and that all your reference conditions are correctly set.

Or this could be real:

In which case these are little spots which are going unsteady and possibly chaotic. This could be a separation region jiggling about or it could be you are just starting to get some turbulent flow in some areas. If this is the case then you have to live with the variation as it is what is happening!

So do all the validation and verification checks to make sure your numerics are accurate. Once you are convinced your numerics are accurate then the variations are probably real.

[parth_k]

Thanks Glenn for the prompt reply.
I suspect this could be more to do with the numeric setup. The physics of the problem is pretty simplified otherwise and since the data from the flow variables are as expected.

In regard to a tighter convergence as I use adaptive timestepping (considerable high limits of max and min timesteps), what all steps can you enumerate briefly to keep in mind?

Perhaps looking at the heat transfer residuals from the plot above, do I make it to 1e-4 RMS criteria? any variations to consider do you recommend otherwise? (with the convergence target-set as default-0.01, loops iterations are 1-10) I'm surely using a structured super fine mesh for this case.

Thanks.

[ghorrocks]

If you repeat the simulation with the convergence tolerance set to 1e-5 RMS it will show the effect of tighter convergence, and the adaptive time stepping will automatically go to a finer time step, so you will see the effect of that as well. If this finer convergence run gives a similar result that suggests convergence criteria and time step size are OK.