[jmhaque]

Hello, everyone. I've been trying to simulate turbulent water flow (over a wide range of Re) in a 2D rectangular mini-channel with square ribs of different pitch-to-height ratios on one wall, and compare them to experimental data (given here: https://jopss.jaea.go.jp/pdfdata/JAERI-Tech-97-032.pdf). I am working on Ansys 2023R1.

I had started a thread a while ago in the Ansys Meshing & Geometry subforum to address meshing issues (link: Heat transfer in 2D ribbed channel: ideas to address the mesh). I have conducted a mesh independence study, but the values of average Nusselt number I have obtained are still well below the experimental values. Therefore, I would like some feedback regarding the case settings.

My case is as follows:

Steady state

Turbulence closure: RNG k-epsilon, with enhanced wall treatment and thermal effects checked. I have tried the realizable k-epsilon and SST k-omega models, but they have yielded worse results for a Re of 95000.)
Energy: on

Pressure-velocity coupling scheme: coupled
Spatial discretization:
Gradients: least-squares cell based
Remaining: all second order

Boundary conditions:

Inlet: velocity-inlet. Velocity was calculated from Reynolds number, using the Fluent default properties of water, and the hydraulic diameter of the experimental channel. Turbulence parameters were expressed in terms of the default intensity and the hydraulic diameter, which was taken to be the channel height this time.

Outlet: pressure-outlet at atmospheric pressure, with reverse flow blocked.

The rib spacings and all the sides of the square ribs were assigned a constant heat flux of 0.425 MW/m^2. The walls of the entry and exit sections, and the wall opposite to the ribs were adiabatic. All the walls had the no-slip condition associated.

For post-processing, I played a session file in CFD-Post which calculated local bulk temperatures, wall temperatures, and local Nusselt numbers along the flow direction. The local bulk temp was found by streamwise velocity-weighted integration over lines perpendicular to the channel axis. The local Nu values were then numerically integrated in order to find the average Nu.

I would like some feedback regarding the procedures I have used. Please let me know if other images and data may be required.

Also, I would be grateful for some help regarding the possibility of setting up translationally periodic simulations to investigate this problem. I am particularly concerned that I may need to supply inlet and outlet profiles of velocity, temperature, etc.

[michealkors]

The Nusselt number discrepancy is interesting. Have you tried playing around with the enhanced wall treatment settings? Maybe a finer near-wall mesh in the vicinity of the ribs could help capture the heat transfer better.

As for the periodic BCs, you shouldn't need inlet/outlet profiles for velocity or temperature. The periodicity essentially means the flow "wraps around" from outlet to inlet. There are some good tutorials online for setting up periodic simulations in ANSYS – searching for "ANSYS 2023R1 periodic boundary conditions" should get you started.

[jmhaque]

Thank you very much for your quick contribution to this thread. I've been struggling with the problem for a while.

There have been a few things I did regarding the near-wall treatments and sizing distributions:
1. I tried including pressure gradient effects.
2. In the mesh independence study, the first cell height near the ribs was directly tied to the element number. The tested meshes were all structured mapped quad meshes made using ICEM CFD. I wanted to ensure that the total element number (in turn, the fineness of the overall mesh) scaled with the rib-side cell sizing. For each rib-side cell size, I also tried varying the cell biasing (or growth ratio in ICEM CFD) to vary the node density near the ribs.

Unfortunately, neither of these pushed the Nusselt number values towards the desired values. The maximum local y+ value I had found is around 4. But with my scaling approach, attempting to reduce y+ would mean pushing the total element number to highs my PC may not be able to handle.

As for the periodic simulations, I had set up an initial case according to a tutorial done using an earlier version of Ansys. A fine mesh was used, with distribution similar to the fine mesh tested in the independence study. I specified a constant mass flow rate as part of the periodic conditions. It was calculated using the Reynolds number, using the hydraulic diameter of the 3D experimental channel. (Is this a valid approach to calculating the channel mass flow rate/velocity for the 2D version? I have not found many intuitive resources explaining how fluid flow cases can be reduced from 3D to 2D.) Unfortunately, after 300 iterations, the continuity residual was too high for it to converge, while the rest of the residuals were comfortably below the levels I had set.