[RickJ1]

Hello

I am running a reacting flow of a premixed hydrogen-air mixture into a combustion chamber to simulate a flame in Ansys Fluent. This simulation is steady-state in a 2D-axisymetric geometry. The species its mixture properties are imported through a CHEMKIN file, using the Konnov (2019) mechanism.

When running the simulation through a cluster the results start to differ compared to running on a local computer. When ran through the cluster the flame profile start to get unrealistic and incorrect, but when running the exact same file on the local computer the flame is realistic and correct.
To narrow down the problem, a simulation is ran both on the cluster and the local computer for a cold (non-reacting) flow.

After 3 iterations the residuals of continuity already starts to differ, between the local and the cluster.

The file read on the cluster where the simulations “starts” with is initialized on the local computer.

Attached is a plot showing the difference in continuity between the cluster and the local result on the corresponding iteration.

I was wondering if this difference in continuity might be causing the larger difference when the flow starts reacting. And if so, what causes this difference and how can this be fixed?

Thank you!

[LuckyTran]

The continuity residual is a normalized residual. First it calculates the residuals for the first n iterations (default is 5). Then it normalizes the continuity from then on by the worst residual that it found in the first n iterations.


Because you run on two different machines, you can't expect the raw residuals in the first n iterations to be exactly the same (because the case has to be parallelized differently in each environment). Hence, you can never expect the continuity residuals from iteration n onwards to be the same. Hence you can't compare this delta between them. Anyway, this is why the delta you see is largest at iteration 3 (because 3<5).


All this talk about residuals is of course just a distraction from your real issue, which is your flame on the cluster not looking correct. So why is your cluster giving you bad results? I don't know.

[RickJ1]

Thank you for the explanation!

I was wondering, since the velocity, temperature and other contours do look the same and the residuals are not relevant to compare, if there might be another way to specify whether there is a significant problem/difference with the non-reacting flows which leads to the larger differences in the reacting flow?

[LuckyTran]

Your calculations are different. Any analysis you do after-the-fact to show that they are different will just confirm that they are different, because they are different, which you already knew, because you knew that already. What you need to do is address why they are different in the first place, which you can't do by post-processing. I can look at plots all day, that doesn't change the fact that my result converged to the wrong result.

[RickJ1]

I agree. Yet, the problem is that I can't figure out what is different in the calculations, since I do not change anything.
I started a new thread giving more information on the problem for reactive flows, since I don't want to start a chain of questions.

Cluster and local computer giving different results for reactive flow