How to do a grid sensitivity studie in a case where a pipe cooling down in a big room
 

Hi everybody. I'm now confused about how to do a grid sensitivity studie in a simulation that includes high-speed free jet (not supersonic) and natural convection.

The case is shown in the picture. To make the things easier I took just a pipe to do the study. The case goes like this: first the pipe will be heated to a stable condition with hot air (700 °C) . The hot air will be exhausted by an air exhaust fan. The maximum velocity in this period is about 140m/s. Then the hot air system will be shut down and the pipe will be cooled down because of natural convection.

I think there are at least 6 parameters I need to study:
1. Which kind of element I use. Polyhedral or Trimmed Cell.
2. The size of Element inside the pipe in both options
3. The configuration of prism layer (the thickness of 1st layer, the total thickness)
4. The refinement of the free jet and natural convection
5. The size of the element on the wall of the laboratory (about 6m*4m*3m).
6. The size of the element on the solid side. The influence on heat transfer.

I don't know how to do such a study with so many variables. Can anyone give me a hint? Like which one I should study first and then use the knowledge to study the second?

Thanks a lot.

First I think you have to decide the type of formulation you can adopt. As the problem is unsteady and 3D you have to consider your computational resources. The flow appears having a very large Re number and I don't think you can do a DNS. Maybe also LES will be very expensive.

Hi FMDenaro, thanks for reply. Until today I've been using the RANS-Models (k-epsilon two-layer all y+). I'd also like to point out, which I forgot to write, that this simulation is transient, at least by cooling. So the grid sensitivity study is really important for the next study of time step.

I recommend the following:
1) Start with a simulation of both element types. Use a cell size so that the computational resources are comparable. Therefore you don't need values vor a large t.
2) Increase the mesh size by a factor of 1.5. You see how the element types react to changing of the mesh. Select finally one of the element sizes.
3) Mesh the geometry with common sense and observe the max Courant number. Try, how you can decrease it with only small changes in the mesh. So you find out how different regions should be meshed.
4) make three real simulations (to the simulated time you need) with different mesh sizes. I recommend a factor of 1.5 again.
5) Go back to step 3) if you change the physical properties remarkably.