I personally would use some k-omega-SST approach for conjugate heat transfer. This turbulence model is very stable and easy to apply. Due to the bsaic k-omega formulation heat transfer at the fluid-wall-interface heat transfer should be appropriate in general and it should show good behaviour regarding the vortices behind your obstacle. For some more detailed vortex resolution you can also use the SST-SAS model (best applicable for transient 3D cases with second order solution schemes) which is my personal favourite. If you have the time and the computing-capacitiy you can also think about using LES like "WALE" for example.
do you think that k-epsilon is best turbulence model for heat transfer? it is because k epsilon is generally good for flow dynamics but for heat transfer is also good? what about k-w SST model?
I would not use k-epsilon for this kind of problem. It is most valid for free stream turbulence like atmospheric flows. Near the walls k-epsilon behaves quite poor according to my experience and I believe that is where your heat transfer happens.
Currently I am using Fluent for CHT with flow boiling. But for single phase flows you can use OpenFOAM of course which I did in the past. It works very well and quite fast as I remember. If you are familiar with OpenFOAM you can try using it. For complex CHT geometries you can use Ansys meshing and import it into OpenFOAM. Or try OpenFOAM native tools of course. But take care about mesh refinement at the interfaces of fluid and solid. Cell height should be approximately equal on both sides.
For my current simulations with FLuent I use an imported CAD geometry, improved its precision in SpaceClaim and then used Fluent's "Watertight Geometry Workflow" for generating a polyhedral mesh.
Ansys mesh does not have structured mesh. How are tackling with this problem?
I need a high quality mesh. I tried to use ICEM but for my complex geometry, it is difficult.
for your current geometry, you imported and improved CAD file after polyhedral mesh showed the high quality mesh like 0.9 ? what about boundary layer? were you able to reach y+
With polyhedral meshes you have the advantage of very good convergence and high precision solutions with relatively low requirements regarding the number of cells. This is due to the high number of shared cell faces (>6). So you don't necessarily need a structured mesh for accurate solutions i think... And this makes meshing more easy. For refining near the walls you can allways use prism layers to reach y+
I suggest using k-omega with Y+ maintained less than 1 (if your computational resources permit). K-omega is good for understanding near-wall physics. It does not use any wall function approximations, unlike k-epsilon.
Since in your case heat transfer across walls is a matter of concern, I strongly suggest k-mega.
Now do remember one thing, a good turbulence model alone won't give you fantastic results. Hence do take care of your mesh as I believe in your case prismatic layers are needed with very fine refinement. Non-linearities in temperature profile can be estimated only if elements in the prism layers are of order e-6 (i.e in microns).