Wind CFD: the RANS approach

The 02 | Outdoor workflow (Atmospheric Boundary Layer → Outdoor Case → Run → Probe) solves pedestrian-level wind with steady Reynolds-averaged Navier–Stokes (RANS) in OpenFOAM 12. Eddy3D writes the case — geometry, mesh controls, boundary conditions, turbulence model, numerics — and OpenFOAM solves it. The physics is OpenFOAM's; this page explains what Eddy3D chooses on your behalf, what you can change, and where the approach stops being trustworthy.

When steady RANS is the right tool

Eddy3D offers several ways to get a wind field. Steady RANS is the workhorse: it gives the time-averaged flow around buildings at a cost a design team can afford for 8–16 wind directions.

The model Eddy3D writes

Equations

OpenFOAM's incompressibleFluid solver module solves the steady, incompressible RANS equations with an eddy-viscosity closure:

∇·U = 0
∇·(U U) = −∇p + ∇·[(ν + ν_t)(∇U + ∇Uᵀ)]

U is the mean velocity, p the kinematic pressure, ν the molecular and ν_t the turbulent (eddy) viscosity, which the turbulence model supplies. Pressure–velocity coupling is SIMPLE. Whether the run is steady or transient is decided by the time-derivative scheme, which Run Settings writes for you (see Transient runs).

Turbulence models

Run Settings → Turbulence Model. All four are OpenFOAM 12 RAS models.

The atmospheric boundary layer (inflow)

The Atmospheric Boundary Layer component sets the inlet as the equilibrium neutral boundary layer of Richards & Hoxey (1993), which OpenFOAM's atmBoundaryLayer conditions implement:

u*        = κ · U_ref / ln((Z_ref + z0) / z0)
U(z)      = (u* / κ) · ln((z − z_g + z0) / z0)
k         = u*² / √C_μ                          (constant with height)
ε(z)      = u*³ / (κ · (z − z_g + z0))
κ = 0.41,  C_μ = 0.09

• U_ref at Z_ref — the reference speed and its height (default 10 m).
• z0 — aerodynamic roughness length of the terrain upstream. One value, or one per wind direction (z0 per Direction), for example from Land Cover Roughness, which derives z0 from OpenStreetMap land cover over the upstream fetch.
• Station versus site. A weather station measures at 10 m over open terrain. Feeding that speed straight into an urban profile tells the model the urban 10 m wind equals the open-country one, which overstates pedestrian wind by up to about 2×. Wind Conditions (with Site Terrain set) applies the EN 1991-1-4 §4.3 terrain transfer (×0.54 for category IV, ×0.84 for III), and Velocity Amplification Factors applies the same factor to the hourly series. The default is "None": no transfer.

Boundary conditions

Using the inlet's z0 on the ground keeps the approaching profile from changing along the empty fetch, the "horizontal homogeneity" condition. Check it with the ABL_Verification template: some decay is normal, a profile that changes shape completely is not.

Domain

Two shapes, both on Outdoor Case:

• Box Domain. Automatic extents follow COST 732 best practice: 5 H upstream, 15 H downstream and 5 H to the sides and top (H = tallest building). The inlet is one fixed face, so an 8-direction study builds and meshes one rotated box per direction.
• Cylinder Domain. One mesh serves every direction. The side wall is split into segments, and each direction decides which segments are inlet and which are outlet (Kastner & Dogan 2020). An 8-direction study meshes once instead of eight times. Each direction is still solved from scratch.

Outdoor Case checks the frontal blockage against the 3 % limit of the ASCE/SEI CWE Prestandard. Model the surrounding buildings within about 240 m of the study area before trusting results near the edge of the context.

Mesh

blockMesh writes a uniform background mesh and snappyHexMesh refines it towards buildings, terrain and any refinement regions. Each refinement level halves the cell size and costs about eight times the cells in that region. Cell Size shows the size each level actually produces (for example L3 · 1.25 m), and the Wind Study banner estimates the cell count as a range before you mesh, because a single number would claim more precision than the estimate has. Height Bands refines by height above ground without any geometry: finest near pedestrian level, coarser above.

Numerics

Run Settings → Numerics is a point on a triangle with three corners: Accuracy, Robustness and Speed. The default sits towards Robustness. It sets the convection schemes, under-relaxation, corrector counts and linear-solver tolerances together. A few other controls matter on their own:

• Warm-up Iterations — run the first iterations with first-order upwind and heavy relaxation, then switch to the chosen numerics. The default (−1) picks this from the numerics point.
• Potential Flow — initialise velocity and pressure with a quick potentialFoam solve instead of a uniform field. Measured on test cases, this and other "better initial fields" do not reliably converge faster; the effect changes sign between cases.
• Velocity cap — every wind case carries an fvConstraints limit of |U| ≤ 100 m/s. It keeps a diverging start finite and logs when it bites. A run whose maximum |U| sits at the cap has diverged; it has not converged.
• The pressure equation always uses the GAMG solver; the Laplacian scheme defaults to limited 0.5.

Convergence

• Convergence (default 1e-4) stops the run once the initial residuals of p, U and k (plus ω for SST) are all below the tolerance. ε is never in that list. Its residual stalls around 10⁻² even when the field has settled, because of how the wall functions fix ε next to walls, so it would hold every k-ε run to the iteration limit.
• Auto-stop on plateau ends a run cleanly once every residual has been flat for about 100 iterations. Some wind directions, often the oblique ones, settle into a small limit cycle instead of converging; their residuals never reach the tolerance however long they run.
• Live Residuals shows the run's progress, rate and estimated time to finish. Field monitors (default on) also log the domain's maximum |U| every iteration. This is the earliest sign of a cold-start overshoot.
• Flat residuals are not proof of a settled field. Eddy3D tests the residual history for stationarity (the KPSS test) and can run a grid-convergence study (Roache GCI) through the foamgci toolkit.

Transient runs (URANS)

Choose PIMPLE as Solver Algorithm on Run Settings to make the same case transient. The controls change meaning: you set Duration and Time Step in seconds, a Max Courant number, and an Averaging Window for the time-averaged fields. Iteration counts, the residual stop and the plateau stop do not apply to a transient run and are not written. The component says so on the canvas, because treating 1000 iterations as 1000 seconds would give a wrong answer that looks right.

From a solved case to results

• Probe samples velocity (or any field: Cp, a pollutant, TKE) at points, usually a grid about 1.5–2 m above ground. Wind Field Viewer and Streamlines draw it.
• Velocity Amplification Factors turns each direction's U / U_ref into an hourly field by scaling with the weather file's hourly speed and direction. This is the annual wind every comfort assessment reads. Its optional TKE input gives a gust estimate for gust-referenced criteria.
• Wind Comfort classifies that annual field against Lawson, Davenport, NEN 8100 or Wellington criteria, for the year, a season or a month.
• With Pressure Coefficient on, Run Settings writes facade Cp. Airflow Network Cp exports it to EnergyPlus.

Validation

Eddy3D checks that the cases it writes let OpenFOAM reproduce measured data. The benchmarks run through the same case writer as the canvas. Headline results (2026-09-27, OpenFOAM 12; the full tables and their caveats are in the repository's VALIDATION.md):

These are typical RANS strengths and weaknesses. Mean speeds away from the wake are reliable; wake lengths and concentrations in the wake are not.

Limitations

• Steady RANS gives the mean, not the gust. Gust-referenced criteria rely on a gust estimate from the modelled turbulence, not on resolved gusts. For peak values use LES (11 | LBM).
• Wakes are typically too long and recirculation too weak with k-ε closures. Results directly behind buildings deserve the least trust.
• Not every direction converges to a tolerance. Some, especially oblique ones, end in a limit cycle. Check the residuals and the field monitors before reading a direction's result.
• This is not a structural-load tool. Eddy3D results are for pedestrian comfort, ventilation and dispersion screening. They do not demonstrate compliance with a wind-load standard.
• The physics belongs to OpenFOAM. Cite OpenFOAM, the turbulence model, and the Eddy3D method papers you relied on (see Methods & Citations).

References

• Richards, P. J., & Hoxey, R. P. (1993). Appropriate boundary conditions for computational wind engineering models using the k-ε turbulence model. Journal of Wind Engineering and Industrial Aerodynamics, 46–47, 145–153.
• Franke, J., Hellsten, A., Schlünzen, H., & Carissimo, B. (Eds.) (2007). Best practice guideline for the CFD simulation of flows in the urban environment. COST Action 732.
• Tominaga, Y., et al. (2008). AIJ guidelines for practical applications of CFD to pedestrian wind environment around buildings. Journal of Wind Engineering and Industrial Aerodynamics, 96, 1749–1761.
• Shih, T.-H., et al. (1995). A new k-ε eddy viscosity model for high Reynolds number turbulent flows. Computers & Fluids, 24(3), 227–238.
• Launder, B. E., & Spalding, D. B. (1974). The numerical computation of turbulent flows. Computer Methods in Applied Mechanics and Engineering, 3(2), 269–289.
• Yakhot, V., et al. (1992). Development of turbulence models for shear flows by a double expansion technique. Physics of Fluids A, 4(7), 1510–1520.
• Menter, F. R. (1994). Two-equation eddy-viscosity turbulence models for engineering applications. AIAA Journal, 32(8), 1598–1605.
• EN 1991-1-4:2005+A1:2010, Eurocode 1: Actions on structures — Part 1-4: Wind actions.
• ASCE/SEI, Prestandard for Computational Wind Engineering.
• Kastner, P., & Dogan, T. (2020). A cylindrical meshing methodology for annual urban computational fluid dynamics simulations. Journal of Building Performance Simulation, 13(1), 59–68.