Hi

We are running a combustion case with a industrial low NOx vortex burner.

How does CFX calculate the amount of NOx resultant at combustor exit?

Thermal and prompt NOx?

Are relative values comparable and trustworthy between testcases?

The actual result levels are highly doubtful (0.2 ppm, where tests show 4 ppm) but perhaps we can compare different cases to each other at least?

Please help me out

Regards, Henrik

Hi,

Combustion modelling is not an easy topic. I trust you have extensively verified and validated the model on suitable test cases before applying it to the burner?

Glenn Horrocks

[Luk_Fiz]

Hi,
I have very similar problem with my actual work.
I am trying to simulate gas combustor, very similar to that one from tutorial. At the point I am interested in obtaining a repeatable and well converged solution, after that maybe some data about physicial measurements will be presented.
After all, I have managed to perform test for different combustion models and so on, but there is a big problem about mesh independency for NOx. The NOx mass fraction in the outlet can hesitate by factor of 100 (!!!) going from say 1.5 mm mesh to 1 mm mesh (entire volume of combustor has about 10^6 elements so I think it is well-represented). Starting automatic mesh refinement gives nothing.
The problem is, that as I refine mesh, that high temp region becomes longer and thiner. The volume of NOx production also changes. However the result of volumeInt (No. production rates) in [mol/sec] is very stable through different meshes but for some unexplained reason the no.mass fraction at the outlet differs, while all other componenets and moments are very well converged. The NOx is only problem.

I dont know if it is something numerical (the NOx concentration in such small burner is low anyway) or I am doing something wrong.
If someone has any comment I would be very gratefull.

Luk

[ghorrocks]

Your statement about mesh sensitivity is wrong. It is very common for people to think that they have zillions of elements so therefore my mesh is fine. But you just stated that a small change in mesh size makes a huge change in results! This means of course you are miles away from mesh independence. You will have to run finer meshes.

I am no expert on combustion modelling but I suspect the flame front is quite a thin layer in the model. As you resolve this layer thinner and thinner the temperatures go higher and therefore you get more NOx. This is just my guessed mechanism of how NOx could be very sensitive to mesh size.

[Luk_Fiz]

Yes I agree with You. However my statement that I have many elements and still situation is unclear was related to hardware resources-the problem of unability of simulation of such a small burner on relatively good PC. I agree that number of elements is very relative and in general no one can said about a system that it has "many" elements.

Anyway - I am stupid - no argue. The problem of discrepancy of "production" and "leaving the domain" of NOx was in imbalances. After running simulation for a long time all is well. I mean - still there is a big change in NOx between meshes but rate of NOx production ~ rate of NOx outflow.

[Luk_Fiz]

After two days of hard working I must confirm that at the moment I am unable to obtain mesh independent solution (in terms of reaction rates) of methane combustion in combustor similar to that one from tutorials. I am using EBU, and other things like in tutorial.

The problem is, that any time I refine mesh the "reaction front" (which physically can be very thin- I know) becomes numericaly thinner. I cannot refine mesh further because of memory limitations. Right now I want to play around mesh refinement by automatic but I am not sure for which quantities to refine.

Of course there are different methods (like PDF or others) but this is very sad finding

Luk

[ghorrocks]

Yes, it looks like you have shown your computer resources are not adequate to do your analysis. Time to try getting a bigger computer or more parallel licenses.

This is why CFD so often ends up running on supercomputers.

[Luk_Fiz]

Hi,
I am still working...
I made up a thin cross section through the domain to make it 2D (domain is not symmetric but I am despered to see what is going on - it allows for increase of "symmetric" numer of nodes). Performing some study I found out, that more or less mesh-converged solution under these conditions can be achieved with reaction zone divided into 0.1mm elements. In this case, NOx at the outlet stabilizes.

It seems that something is very wrong down here, however the proof is not straightfroward. I can show at least few papers that declares calculations of fe. furnaces in which NOx level and other quantities was in very good level of consistency with measurements. In none of them combustion volume (which can has capacity of hundreds of cubic meters) was divided into (0.1mm)^3 cells!!It would be not possible on any of existing supercomputrs.

I am not sure what but as I said - something is wrong. Maybe someone can share his experiences?

Luk

[ghorrocks]

Maybe a different combustion model would not be as sensitive to mesh size? I have no experience with combustion modelling so I am just guessing.

[Luk_Fiz]

Does not matter - Your experienced notices holds on my self-counsious

I am still thinking that models, combustion, libraries are the same. One thing that is on my mind is mixing problem. In my case, combustion is lead with very reach flame, but with not very good mixing. It seems, that as the mesh becomes more and more refined, ther "mixing front" becomes thinner and thinner. It allows for flame elongation, and since CH4 is still available the flame is running longer and longer. This produces more NOx by mixing O witn N under high temperature over longer distance.
Anyway it is still not - mesh converged case and searching through the net I was unable to find a test case so "meshed".

Luk