Simulating crossflow
 

Hi all,


I am trying to simulate a (reacting) crossflow situation. The jetstream enters the main flow (100m/s) with 120..150m/s through an inlet duct with d = 4 mm. Boundary layer is depicted with appropriate prism layer cells. It is a steady model and i am using the AMG linear solver. My simulation has trouble to converge though, the main errors i ran into were:

-Every time - velocity turning up and down in the jet duct, no matter how i initialize it
-Every time - residuals are low for the conservation equtions E-1 E-2 but high for the chemistry E4 E5 E6 E7. Chemistry is not working.

-Sometimes - floating point error -> k-eps turbulence


Mass is correct, components are doing the right thing. All property magnitudes pretty much make sense from what i can see in the analysis. Pressure drop in the jetstream is very high (2-3bar on a length of 2cm, i was astonished). Re number in the jetstream (theory) should be between 100000 and 500000.


I am comparing polyhedral and trimmed meshes and different turbulence models but none of the models runs well by now. Please give me some input.


Thank you,

Bernie

Step 1 to complex reacting flows is to not do the reacting part. Did you run your case with non-reacting flow and get satisfactory convergence and sensical results?

Hello, yes I am looking at the non-reacting case first. The result is similar, sensical yes, properly converging no. It seems the energy equation is the one giving me a little trouble https://imgur.com/a/Bx9prEW

Those are steady residuals? It looks like there's an unsteady component to the flow field that the solver is struggling with. Look at some plane sections while it's iterating and see if you can find anything like that.

Yes, it is the velocity in the inlet duct. It turns up and down. But why?

Can you post a picture of the geometry and list the boundary conditions

I solved this problem, is was because I had selected segregated flow. With a coupled model that works well.


I am advancing to the reacting part now. I am trying to use the GRI-Mech 2 mechanism for CH4 and simulate the combustion. My flame is not bruning as much as it should though, just a tiny flame at the edge and very slow combustion.



I am considering two causes:
1) the tables are not appropriate, designed for air which I am not really having.
2) I tried both Laminar flame concept and Eddy Dissipation Concept, cause these models support the complex chemistry table definition. Are they appropriate?




Another thing which is giving me big trouble is the outlet, I can not find a way that it will not be restrictive. Using "Outlet" specifies the mass flux. "Pressure outlet" wants me to give p, T and fractions, which I do not want to specify, cause thats exactly the task of my combustion cfd.





Greetings, Cheet

Some months later! I advanced alot with the model and improved the mesh. Model converges but slow, especially complex chemistry convergence is very slow. Any suggestions to accelerate it? Could an increased CFL number help? Or should I limit the chemistry timestep? Or change the acceleration factor of the Laminar Flame Concept (currently on default = 0.5)? It is a steady model with full chemistry.


I also looked into the Complex Chemistry solver model itself more. There is clustering (selected are temperature, equivalence ratio and chemistry time step by default - should I select species aswell?);
and in-situ adaptive tabulation would be an option? My mechanism has 53 species and 325 reactions.



Thank you,
greets Cheet

Smells like GRIMECH 3.0. =)

Clustering is on/off by default for your version? Turn it on. Clustering is the best way to accelerate it. The default clustering speeds up GRIMECH by a factor of 5x-20x.

Actually that question is already answered, clustering is on for three signals as mentioned previously. Any other suggestions?

I am trying to use better turbulence models, like k-omega or Reynolds Stress.


The model computes one step, and is stuck when trying to compute the second step. Why is that happening?


edit: I am starting to lose my belief, the model runs and converges at a mesh size which is sufficiently coarse so it can be handled locally. When I put the EXACT same model onto the server, regardless of cpu/memory allocation, it will not run after step 1. Any reasons for that?


Bernie