[Mowgli]

Dear CFDers, (apologies for the long text)



I am attempting to simulate a transient conjugate heat transfer case in OFv6. The case involves a cooling fluid and a quite complex large composite solid, in which heat is generated for some of its parts. I initially approached it using chtMultiRegionFoam, modifying it to serve my equations, and dividing the composite solid body into its constituent solids (25 in number) and defined each of them as a region to solve the fourier's equations with heat gen. I have managed to get my case validated and running.



My problem is now the computation time, mainly because I have material properties varying with temperature (therefore high cost for computing them at each iteration) and many boundaries (100 or so almost all of them using the baffle boundary condition, also modified to take into account temperature variation and in some cases added thermal resistance). I have found this BC to be highly time-costly. Even with access to a cluster and managing to scale my solver depending on mesh parameters, it takes too long to make a meaningful parametric analysis. Eventually I need to extend my model to bigger geometries and the task of waiting weeks for a single simulation is too tedious.



I intend to approach my problem by:
1. Reducing the number of boundaries between the solid regions, by being able to define the composite solid as one solid region instead of several. I will have to ignore the thermal interface resistances then for sure.
Please do correct me if I am wrong here; during discretization of the transient heat conduction equation, the thermal conductivities at the walls/boundaries of the finite volumes are calculated as harmonic mean between the volume centres/nodes which allows the conservation of flux and therefore the multi region idea is congruent to a single region with different material properties. Correct?
I am unable to find the correct way to define different material properties (mine are polynomial and/or power based) within a single mesh in openfoam.
More importantly, lets consider adjacent finite volumes, one involving heat generation and the other not in a one dimensional heat transfer case. If I define my source terms for each properly, can I still solve a single matrix like the standard case without missing any numerical glitches? Are there additional considerations I should be taking into account?



2. Updating material properties at a more convenient point. Since I need a good transient behaviour prediction, I update my cell material properties at every iteration instead of time step. How do I compute the error estimate in my simulations, if I instead update them at time steps? Or is this a no-go at all?



3. Different number of iterations for each time step, to help reduce cost time. But this is fairly straightforward.



If there are any other suggestions, I'd be obliged. Thank you!