Hey Son ich ngo ,

Can you give some context about what type of multi-scale problem you would like to solve?

Generically I would say that the typical process would be to solve your macro scale model, then use mapFields to supply the boundary conditions from your macro scale model to your miniature model. This will give you 1-way coupling from you macro model to your miniature model, so long as the influence of the miniature model is minuscule compared to your macro model. Otherwise, I would recommend building a very coarse mesh through your macro model, with refinement near your miniature model, and immediately down-stream from it.

You can read more about mapFields here; https://cfd.direct/openfoam/user-guide/v6-mapfields/

Dear Alejandro Vallejo

First of all, I would like to thank for your kind help.

My problem is multiscale modeling of CO2 methanation over Ni/Al2O3 catalyst.

So, the macroscale is a fixed-bed reactor tube, and the microscale is a catalyst particle. Several studies have been done for particle resolved reaction kinetic (Article Advances in fixed-bed reactor modeling using particle-resolv...

Usually, in the large catalyst particle, the diffusion limit (effectiveness of reactions rate or Thiele modulus) affects strongly to the result therefore the particle-scale model is needed. My idea is different with the above article that we assume the whole reactor tube is a porous media, while the diffusion limit in this porous media can be capture by a simple 1D particle model.

Luckily, the geometry of the macroscale is not deformed, so we only need to take the fluid information (T, P, Concentration, etc.) from macroscale to 1D particle model then return the results for effectivenes factors from particle model to the macroscale to adjust the actual volume averaged reaction rate.

For this case, I think we need a stiff ODE or boundary-value problem (BVP) solver library to solve the 1D particle model. Do you have any suggestion for that?

In the next step, I think the mapFields will be useful for more complex problem that need to solve 2 CFD scales at the same time.

Hey Son ich ngo ,

This is a very complex problem. I think there are a number of different approaches that you could take with OpenFOAM.

I think that the approach you highlight might be on the right path, however I would limit the region of porous media to the normalized level of your catalyst bed. (That is unless the entire domain is filled with the catalyst bed) Unless the flow pressure is high enough to significantly disrupt of move particles in the catalyst bed, you can assume for a first level analysis that the porous media is stationary for all intents and purposes. You can then implement your 1D particle model as an additional equation in the solver, to capture that phenomena, although I would suggest bench marking your analysis with experimental values, as I would not trust myself to successfully modify the solver to produce accurate results on the first try.

I think the main question on using the aforementioned method of direct modeling of the catalyst bed (as Lagrangian particles) depends on the size and number of particles. Although I agree that it is computationally expensive, it may be possible to run an axis-symmetric case for your large reactor to decrease the computational costs for directly modeling the particles and their effects.

In any case, it seems like a very interesting problem. Please share how you end up modeling your problem, and any bench marking you may perform.

Dear Alejandro Vallejo ,

Thank you very much.

I have tested the above approach in Matlab for combining particle and reactor scale for a radial perfect mixing plug-flow reactor. Although the result is acceptable comparing with existing experimental data, I would like to investigate more the radial profile by this CFD model.

In OpenFOAM, my idea is employment of a 1D particle scale in a separate scale, then the information of that scale can be coupled with reactor scale via source terms form. Catalyst particles in a pack bed will be considered as porous media. So the objective now is only the boundary-valued problem solver. I am thinking about the Crank-Nicolson method for that.

I will update the progress of my model .