[pattim]

Hello Everyone: In chemistry cases (reactingFoam) the Arrhenius equations are solved by an ODE solver - I'm wondering how these outputs are coupled to the flow equations? The interest is this: it might be useful to be able to partition CPUs in chemistry-space somehow, since chemistry often dominates over flow, which is a grid phenomenon, and CPUs can currently only be partitioned over grid-space.

In order to improve my understanding, is there a write-up on the ODE-PDE coupling in OF?

Thank You!
Patricia

[dlahaye]

Papers by Cuoci, Contino, Parrente and others. Should do proper homework in order to give a better answer. Sorry ...

[pattim]

Originally Posted by dlahaye

Papers by Cuoci, Contino, Parrente and others. Should do proper homework in order to give a better answer. Sorry ...

Thanks for the references. CFD is a big field, and chemistry in CFD is not that popular. In particular, solver structure (SIMPLE vs. PISO) is even less-discussed as related to ODE source-term coupling of interphase transport of chemical species (sources/sinks).

[dlahaye]

Please allow me to try to answer two posts.

Chemically reacting flows and combustions are huge fields of research. I collected some papers on the simulation of chemically reacting flow simulations in OF in a Dropbox folder that I am happy to share. I am happy to support a write-up on PDE-ODE coupling to help clarify these matters.

I can image that deflagration/detonation simulations are hard to make converge. I do agree that non-reactive flows might not be the ideal starting guess for the simulations. Possible one can include more intermediate step by freezing alternatively the flow and the chemistry or by simplifying the chemistry.

[Numericer]

Originally Posted by dlahaye

Please allow me to try to answer two posts.

Chemically reacting flows and combustions are huge fields of research. I collected some papers on the simulation of chemically reacting flow simulations in OF in a Dropbox folder that I am happy to share. I am happy to support a write-up on PDE-ODE coupling to help clarify these matters.

I can image that deflagration/detonation simulations are hard to make converge. I do agree that non-reactive flows might not be the ideal starting guess for the simulations. Possible one can include more intermediate step by freezing alternatively the flow and the chemistry or by simplifying the chemistry.

Dear Domenico,



Can you please share the link to the Dropbox folder?



I want to understand how ODE solution is tied to the conservation equations.



Are the ODE for species concentrations solved at every grid point in the domain, stored in the source term and then supplied to the PDE's?

[dlahaye]

Greetings,

I copied the papers to the drive

https://mega.nz/folder/nUsSxQjB#r9XIuDz_6mKHma820wI1CQ

There are at least two ways to solve for the species mass fraction.

The first is not to worry about the ODE solvers at all. In this approach, the outer PIMPLE iteration takes care of the time advancement. Example is reactingFoam.

The second approach is to implement a splitting method in which transport and chemistry receive separate treatment. In this approach, the chemistry is solved by an ODE. See papers by Cuoci among others.

[pattim]

Hmmm... chemistry is usually "stiff" - so an good ODE solver is usually needed. I wonder if the outer Pimple loop is good enough to converge the chemistry? (converging the flows is relatively easy.) It must be good enough because the reactingFoam examples usually converge for arbitrarily complex Arrhenius reaction sets and temperatures/pressures, although I've never tried converging through shocks or strong gradients, and I don't think there's any example cases of that.

[dlahaye]

Dear Pattim,

Thank you for raising your thoughtful comments.

I agree that chemistry can be stiff. No doubts here.

I, however, do not see any ODE-solver reactingFoam being called from reactingFoam.

I do instead see the various mass fraction field Y_i being solved within the PIMPLE loop.

Do deal with complex problems, more complex solvers have been developed. The work by Cuoci and co-authors is a good example.

Does this help?

[pattim]

Yes, you're right - it all depends on the solver. A very good non-ODE solver can deal with stiff chemistry terms. Solvers have come a long way in the last couple of decades. I'm looking over those papers. Thank you very much for uploading them!!

I've been using reactingFoam for quite complex hydrocarbon combustion (DANDIA CHEMKIN-style) - it's basic research on yields of PAH under fuel-rich conditions. It might be useful if I could include evaporation of fuel droplets, but I don't think the solvers care good enough (yet) to do that well.

[dlahaye]

The repository still lacks reference on the implementation on the TDAC reduction algorithm, the soot modeling, the flamelet modeling and the ignition. I will be happy to add in case any one sends suggestions.

Cheers, Domenico.

[pattim]

Originally Posted by dlahaye

The repository still lacks reference on the implementation on the TDAC reduction algorithm, the soot modeling, the flamelet modeling and the ignition. I will be happy to add in case any one sends suggestions.

Cheers, Domenico.

Thank you for your kind comment. I thought there were one or two names given as references for TDAC? Maybe I'm wrong. I've simply used chemkin-style detailed chemistry - and was successful, although I haven't tried high-speed or compressible flows. If by "soot-modeling" you mean detailed Arrhenius chemistry - I don't think that's known science of high-temperature gas-phase macromolecules (i.e., detailed Arrhenius model) as of now. I think current Arrhenius models only go up to, maybe C15 aromatic molecules? (And that's thousands of reactions.)

I think ignition is only really a problem with this approach (detailed Arrhenius) at high MACH numbers. There are some (expensive) fully-coupled PDE/ODE detailed Arrhenius models available that can do high MACH numbers chemical reactions and combustion. The regular suite of aerospace flow solvers can't tackle that with sufficient reality, however.

It would be very instructive and useful to have a compressible version (relatively low MACH numbers... 1-3?) of the TDAC examples. (TDAC+choked nozzle? It would be worth testing how high in MACH number the solver will continue to converge.) Fuel droplet evaporation with gas-phase Arrhenius chemistry would be another useful model to explore (below MACH1). I'm told these sorts of things can be difficult to code for a fully-coupled solvers, but if the existing OF solvers used here (PISO?) are as good as they appear to be for combustion, it may be doable?