I’m running into an issue with under-predicted drag when using a wall-function approach in ANSYS Fluent for a free-surface flow. The case is a partially submerged vertical cylinder, velocities 0.5–3.0 m/s, VOF + k–ω SST. For the purpose of testing my model, I ran all simulations mentioned below at 1m/s. When I fully resolve the viscous sublayer (low y⁺ mesh, y⁺ ≈ 1–4, PRESTO! pressure discretisation), the drag is close to expected (~21 N). However, I can’t use this approach for all velocities because it would either require an excessive number of cells (beyond my available computational resources) or very high aspect ratio cells in the near-wall region, which cause free-surface stability issues in VOF simulations. With the wall function approach (target y⁺ ≈ 30-50), drag drops significantly under the same conditions. Using the same settings as the fully resolved case and only changing the inflation layer to achieve y⁺ ≈ 50 gave a drag of ~9.5 N. After being advised to switch the pressure discretisation from PRESTO! to Modified Body Force Weighted, the drag increased to ~15.5 N (better but still too low compared to the ~21 N baseline). The main loss is in pressure drag; viscous drag changes only slightly and makes up only a small portion (> 4%) of the total drag. I've tried my best to summarise the main geometry, mesh and Fluent settings below.
Geometry The model consists of a rectangular fluid domain containing a partially submerged vertical cylinder of 100mm diameter. The domain is split into two parts:
- The main rectangular region.
- A 400mm diameter cylindrical sub-domain placed coaxially around the cylinder for finer near-cylinder mesh control.
Bodies of Influence (BOIs):
- Cylindrical BOI overlapping the structured fluid sub-domain for near-cylinder sizing.
- Rectangular wake BOI extending downstream of the cylinder.
- Upstream free-surface BOI, extended slightly downstream where deformation is minimal.
- Downstream free-surface BOI, extended slightly upstream to capture upstream deformation.
Meshing
- Structured hexahedral mesh in the cylindrical sub-domain using MultiZone.
- Inflation: first layer height = 0.005 m (y⁺ ≈ 50 at 1 m/s), growth rate = 1.1, 8 layers.
- Edge sizing: 150 vertical divisions in the structured mesh region.
- BOI element sizes: Cylinder = 0.01 m; Wake = 0.02 m; Downstream = 0.02 m; Upstream = 0.04 m.
- Mesh quality: Min element quality = 0.244, Avg = 0.835; Max aspect ratio = 9.92; Avg = 1.84; Max skewness = 0.838; Avg = 0.201; Min orthogonal quality = 0.162; Avg = 0.801.
Fluent
- Mesh converted to polyhedral in unstructured regions.
- Double precision enabled.
- Models: VOF (implicit formulation, implicit body force, interfacial anti-diffusion enabled), k–ω SST turbulence.
- Boundary conditions: Top/bottom/side walls = no-slip; inlet = velocity inlet; outlet = pressure outlet.
- Methods: PISO pressure–velocity coupling; pressure discretisation = Modified Body Force Weighted (after initial PRESTO! tests); volume fraction = Compressive; momentum, TKE, and ω = Second Order Upwind; transient formulation = Second-order implicit.
I have been struggling with this issue for well over a month now and am pretty much stuck on what else I can try. Any help would be greatly appreciated.
What type of prism layers are used in this simulation, and what is the order of the equations used in the solution?
Ala A. K Keelani Hi, thanks for the response!
I have an inflation layer around the cylinder. Outside this, the mesh transitions into a structured hexahedral region around the cylinder and then to unstructured cells further out.
I am using Bounded Second Order Implicit for the transient formulation, and the following settings for spatial discretization:
- Gradient = Least squared cell based
- Pressure = Modified Body Force Weighted
- Momentum = Second Order Upwind
- Volume fraction = Compressive
- Momentum = Second Order Upwind
- Turbulent kinetic energy = Second Order Upwind
- Specific dissipation rate = Second Order Upwind
I couldn't see a way to add images so I have attached a document with all the settings for my model. Hopefully any information I forgot to add is in there.
Alex Baxter from the best established this work step by step why? first thing you must be doing 2D simulation to facilitie work ,then secondly find out some things must be check out such precsion in your type simulation because the kind of simultion it used called multiphase flow and consider the most complex kind of simulation. third thing must be make grid senstivity or grid indepence study to appear every thing more disticly
Ala A. K Keelani thank you for your suggestions. For this particular case, a 2D simulation isn’t feasible, the geometry involves a vertical cylinder piercing a free surface, which can’t be captured in two dimensions. A side view would block the flow entirely, and a top view cannot represent both water and air phases simultaneously, so 3D is required.
On the point about precision: I’ve already used second-order discretisation (momentum, turbulence, transient), controlled Courant numbers, and tested different pressure schemes (PRESTO!, Modified Body Force Weighted). Are there other specific aspects of precision you think I should check in a multiphase setup like this?
I have attempted a mesh refinement study, but it did not behave as expected, instead of approaching grid independence, the drag coefficient and bow wave height kept changing more strongly with each refinement (r ≈ 1.5).
The main challenge remains recovering accurate drag values when using wall functions compared to the fully resolved near-wall approach. If you (or anyone else) has any suggestions for improving wall-function performance in multiphase free-surface simulations, I’d be very interested in your input.
Alex Baxter When I discussed precision, I meant single or double precision because it makes a large difference in this type of simulation. In addition, you should know the distinction among types of prism layers, for example, first layer thickness, total layer thickness, transition layers, aspect ratio, and last aspect ratio. Thank you. Understanding these parameters not only enhances the accuracy of your simulation but also optimizes performance. Ensuring you have the right configurations can significantly impact your results and the reliability of your findings.
Ala A. K Keelani Thank you for clarifying. I am running my simulations using double precision.
For the prism layers, I have tested a range of configurations for both the number of layers and the growth rate. At the moment, I am using 8 layers with a growth rate of 1.1, but I have also tried up to 20 layers and growth rates between 1.05 and 1.2.
The first layer thickness was adjusted to give y⁺ ≈ 60, and the total prism stack transitions smoothly into a structured mesh region around the cylinder. The maximum aspect ratio of any cell is about 10, and the total thickness of the inflation layer is roughly 60 mm (with the cylinder diameter being 100 mm).
I also tested swapping polyhedral cells into the structured region around the cylinder, which increased the drag from ~15.5 N to ~17 N, so the way the prism layers transition into the surrounding mesh seems to have some influence.
Do you think the total inflation thickness relative to the cylinder diameter could be contributing to the drag underprediction, or is it more likely related to the turbulence–free surface interaction?
Yes, I think the both issues related to the accuracy results and something else you must make it, compare your results with someone else and show the difference between you and him versus experimental results.
Ala A. K Keelani Thanks, I’ve already compared against literature and a fully resolved baseline simulation (y⁺≈4) that I ran at 1 m/s where C_D matches the value of ~0.8–1 as reported in literature. The underprediction appears only in the wall-function setup (C_D = ~0.6).
hello, please
Clarify which drag you’re comparing
Separate pressure drag and viscous drag (Reports ▸ Forces… check “Report Viscous/Pressure”).
For a semi-immersed cylinder at high Re, most drag comes from form + waves. If your waves are under-resolved, total drag will be too low even if wall treatment looks okay.
Derrar Benamar Omar Hi, thanks for the response. I forgot to mention but I already broke down the drag into the pressure and viscous coponents. The pressure drag is what contributes to pretty much all of the drag force on the cylinder and is what is reduced when using the wall function approach.
Since my last post I made some progres. I changed the pressure, gradient and momentum discretization settings from the defaults to "Modified Body Force Weighted", "Green-gauss Node Based" and "Third Order MUSCL", respectivley. This helped increase the drag force to ~17.5N (closer to the ~20 N goal).
I also tried to run a mesh sensitivity study to see how drag values evolve with refinement. I surprisingly found that the coarser the mesh, the closer the drag was to the expected value. Additionally, rather than converging with refinement, the results diverge further from the resolved case, and the change in drag grows larger between stages instead of reducing.
This has me quite puzzled, as I expected pretty much the opposite to occur. Do you think this indicates that larger cells are better suited for the wall function approach in this case, or is it pointing to something else affecting the pressure drag recovery?
- Badges
- Science topic
- Similar topics
- Correction