[Weiqiang Liu]

hello all,

I am doing a catalytic methane combustion simulation. The gas mechanism is GRI 3.0. the surface mechanism is Dutchman mechanism.

the reaction model is species transport with laminar finite rate. After almost 200000 iterations with 3 days' computation, all the residuals are lower than 1e-4 though with some fluctuations. Some of the results are consistent with open literature however some other results are not.

I am wondering is the calculation speed normal regarding to the similar case? The setting of the model is very simple without UDF.

I patched a 2000k temperature to the whole computation domain (including both solid and fluid domain) in order to 'ignite' the mixture. on the other hand, the mole fraction of water and carbon dioxide were patched in the rear part the channel.

I am using parallel mode of fluent with 8 nodes.

A lot of work has been published regarding to the similar case. Can somebody tell me is this calculation speed normal and is my method of igniting the mixture right?

weiqiang

[LuckyTran]

Yes that works for ignition.

Calculation time seems to be on the right order of magnitude if you have 1 million cells or more. But yes, it sounds about right.

[Weiqiang Liu]

Originally Posted by LuckyTran

Yes that works for ignition.

Calculation time seems to be on the right order of magnitude if you have 1 million cells or more. But yes, it sounds about right.

My case only has 50000 cells. So the calculation time is not normal, right? Some authors mentioned that they got their results with a powerful workstation for 30 days.

[LuckyTran]

50k cells might just be too few to receive any benefit of parallelization. Run a quick test with 1 vs 2 vs 4 vs 8 cores. Just do like 10-100 iterations to time it.

[Weiqiang Liu]

Originally Posted by LuckyTran

50k cells might just be too few to receive any benefit of parallelization. Run a quick test with 1 vs 2 vs 4 vs 8 cores. Just do like 10-100 iterations to time it.

I've done test before. calculation speed is much faster with 8 cores than 1 cores.
you mean 8 cores might have no difference with 2 or 4 cores?

[LuckyTran]

If 8 runs faster than 1 then okay.

Are you complaining your simulation is too fast or too slow? To me, it feels a little slow for 50k cells (by some integer factor like 1/2 or something).

[Weiqiang Liu]

Originally Posted by LuckyTran

If 8 runs faster than 1 then okay.

Are you complaining your simulation is too fast or too slow? To me, it feels a little slow for 50k cells (by some integer factor like 1/2 or something).

I think it's too slow. now after 4 days' iteration, the maximum temperature is still decreasing. some xy plots have a lot of discontinuities when I post-process the case.

I am wondering what is the suitable convergence criteria for my case. The normal criteria like 'energy residual is below 1e-6 and the other residuals are below 1e-3' is apparently not enough.

[LuckyTran]

Do you have any warnings like max something limited or min something limited? Those can slow down a simulation by a lot because they stall the convergence of the calculation at the inner iteration level.

Originally Posted by Weiqiang Liu

I am wondering what is the suitable convergence criteria for my case. The normal criteria like 'energy residual is below 1e-6 and the other residuals are below 1e-3' is apparently not enough.

Residuals should never be used as convergence criteria anyway until you have a lot of experience for your specific problem. Make solution monitors and look at them.

[Weiqiang Liu]

Originally Posted by LuckyTran

Do you have any warnings like max something limited or min something limited? Those can slow down a simulation by a lot because they stall the convergence of the calculation at the inner iteration level.






Residuals should never be used as convergence criteria anyway until you have a lot of experience for your specific problem. Make solution monitors and look at them.

there is no such warning during simulation.

yes, I tried to build a monitor to judge convergence. however, I did not know which parameter is appropriate. The mass balance is reached very early. I am wondering can I set a monitor which check the outlet temperature. since the outlet temperature is decreasing continuously, once the outlet temperature is steady, the whole case can be judged converged?

[LuckyTran]

Originally Posted by Weiqiang Liu

since the outlet temperature is decreasing continuously, once the outlet temperature is steady, the whole case can be judged converged?

That's a sometimes a necessary but not always sufficient condition. If the outlet temperature is not steady then obviously the solution isn't converged. But even when the outlet temperature is steady, that does not mean there are not oscillations distributed throughout the domain.


Depending on what your sought-after-parameter is, you may or may not care at all whether the outlet temperature is steady or not. Good CFD users think about these and decide what their sought-after-parameter is before they press the run button. Don't wait until 3 days later and 200k iterations to start thinking about what is convergence.

[Weiqiang Liu]

Originally Posted by LuckyTran

That's a sometimes a necessary but not always sufficient condition. If the outlet temperature is not steady then obviously the solution isn't converged. But even when the outlet temperature is steady, that does not mean there are not oscillations distributed throughout the domain.


Depending on what your sought-after-parameter is, you may or may not care at all whether the outlet temperature is steady or not. Good CFD users think about these and decide what their sought-after-parameter is before they press the run button. Don't wait until 3 days later and 200k iterations to start thinking about what is convergence.

yes, my sought-after parameters are the maximum temperature and the surface coverage of pt(s). however they just keep changing. I think the case has not converged yet.

but like you said GRI 3.0 is designed to give good results and computation stability. why are the residuals fluctuating so strongly?

[LuckyTran]

Well first I don't care that the residuals fluctuate. What matters is how the solution behaves.

9 times out of 10 it is a mesh quality issue. The other 1 time out of 10, the user has tweaked a setting they shouldn't have.


GRI is designed for good results in chemical kinetics. Not necessarily CFD.


I see from your other post that now your mixture is not ignited at all so there's no point in any further discussing.

[Weiqiang Liu]

Originally Posted by LuckyTran

Well first I don't care that the residuals fluctuate. What matters is how the solution behaves.

9 times out of 10 it is a mesh quality issue. The other 1 time out of 10, the user has tweaked a setting they shouldn't have.


GRI is designed for good results in chemical kinetics. Not necessarily CFD.


I see from your other post that now your mixture is not ignited at all so there's no point in any further discussing.

yes, the temperature keeps decreasing until to room temperature. I don't know what is the right way to ignite the mixture.

[LuckyTran]

It is an art. You have to learn how to do it for each specific problem.


It is like learning how to light a grill or start a real fire. You try many different ways until you figure it out.

[Weiqiang Liu]

Originally Posted by LuckyTran

It is an art. You have to learn how to do it for each specific problem.


It is like learning how to light a grill or start a real fire. You try many different ways until you figure it out.

thanks for your encouragement!