Case B: validation

Contents
  1. Measurement locations
  2. Explore the points
  3. OpenFOAM 12
    1. Measured against modelled
    2. Scores
  4. OpenLB
    1. Scores
  5. FluidX3D
    1. Scores
  6. ML surrogate
    1. Scores
  7. Notes on every run
  8. Leaderboard

Engine status: OpenFOAM 12 scored · OpenLB scored · FluidX3D scored · ML surrogate scored

Nomenclature: symbols, columns and scores

Geometry, reference scales and release

  • \(H\), \(b\), \(D\) (m) — height of the building, cube, tower or tree belt; width of the building; edge of the cubes of Case C: the lengths the cases are drawn and normalised by
  • \(U_H\) (m/s) — reference speed: the mean approach-flow speed at the height \(H\) (each Data page gives the value used)
  • \(U_R\), \(H_R\), \(Z_R\) (m/s, m) — reference speed and height of a workbook: Case E, \(U_R\) at \(Z_R\) = 250 m full scale (1 m in the model); Case M, \(U_R\) and \(H_R\) at the reference point of the tunnel and of each field run
  • \(U_{\mathrm{ref}}\), \(L_{\mathrm{ref}}\) (m/s, m) — reference speed and length of a case’s normalisation: \(U_H\) and \(H\) (\(U_R\) and \(H_R\) in Case M); for the velocity ratios the approach speed at 15.9 m (Case E) or at 500 m (Case F)
  • \(U_b\) (m/s) — Case G: mean approach-flow speed at 9 m, the normalisation of its approach-flow table
  • \(q\), \(c_{\mathrm{gas}}\) (m³/s, ppm) — tracer release of the experiment: volume flow of the released gas, and its tracer content (\(10^6\) ppm when pure); \(Q\) is the source strength given to a solve
  • \(R_b\) (dimensionless) — bulk Richardson number of the heated-floor cases J and L (negative: unstable stratification)
  • \(\mathrm{Re_H}\), \(\mathrm{Re_D}\) (dimensionless) — Reynolds numbers \(U_H H/\nu\) and \(U_H D/\nu\) (\(\nu\): kinematic viscosity of air)
  • \(u_\ast\), \(z_0\), \(\kappa\) (m/s, mm, dimensionless) — friction velocity and roughness length of the log law fitted to the approach flow, and the von Kármán constant (0.41)
  • \(\mathrm{WD}\) (column wind_direction_deg; °) — wind direction (each frame says how it is counted); \(\mathrm{WD_R}\) is the mean direction of a field run at its reference point (Case M)
  • \(x\), \(y\), \(z\) (columns x, y, z; m) — coordinates in the case’s frame (\(+x\), \(+y\) and \(+z\) are the positive directions; each Data page says which way \(+x\) points), \(z\) up from the floor of the tunnel or from the ground
  • \(\mathrm{d}x\) (m) — lattice spacing of the OpenLB and FluidX3D runs, in metres or as a fraction of \(H\) (\(H/64\))
  • \(L_r\) (m) — reattachment length behind the building, read from a wake line at ground level; reported as \(L_r/H\)

Velocity and turbulence

  • \(U\), \(V\), \(W\) (columns U, V, W; m/s) — mean velocity along \(+x\), \(+y\), \(+z\); \(V\) is positive toward \(+y\) in every case
  • \(\lvert\mathbf{U}\rvert\) (m/s) — magnitude of the mean velocity vector, \(\sqrt{U^2+V^2+W^2}\)
  • \(V_{\mathrm{s}}\) (column Vs; m/s) — mean omnidirectional scalar speed of a thermistor (never a vector magnitude)
  • \(\overline{u'u'}\), \(\overline{v'v'}\), \(\overline{w'w'}\) (columns uu, vv, ww; m²/s²) — velocity variances (overbar: time average, prime: fluctuation about it)
  • \(\overline{u'v'}\), \(\overline{u'w'}\), \(\overline{v'w'}\) (columns uv, uw, vw; m²/s²) — Reynolds shear stresses
  • \(k = \tfrac{1}{2}\left(\overline{u'u'}+\overline{v'v'}+\overline{w'w'}\right)\) (column k; m²/s²) — turbulent kinetic energy
  • \(U/U_{\mathrm{ref}}\) (column velocity_ratio; dimensionless) — mean speed over the case’s reference speed (\(V_{\mathrm{s}}/U_{\mathrm{ref}}\) for a thermistor)
  • \(\sigma_{U/U_{\mathrm{ref}}}\) (column velocity_ratio_sd; dimensionless) — standard deviation of velocity_ratio over the observations behind a field value (Case F)
  • \(k/U_{\mathrm{ref}}^2\) (column k_ratio; dimensionless) — turbulent kinetic energy over the square of the reference speed, where only normalised values are published (Case G)

Concentration and temperature

  • \(C\) (column C; ppm) — mean tracer-gas concentration as published (volume fraction \(\times 10^6\) of the released gas diluted in air)
  • \(c_{\mathrm{rms}}\) (column c_rms; ppm) — root mean square of the concentration fluctuation
  • \(c_0\) (ppm) — reference concentration, defined for each case on its Data page: \(c_0 = \dfrac{c_{\mathrm{gas}}\thinspace q}{U_H H^2}\) (\(U_R H_R^2\) in the denominator in Case M)
  • \(c^\ast\) (column concentration_ratio; dimensionless) — dimensionless concentration \(c^\ast = C/c_0\); the value of a solve is \(c\thinspace U_H H^2/Q\) for its mean concentration \(c\) and its source strength \(Q\)
  • \(\mathrm{Sc_t}\) (dimensionless) — turbulent Schmidt number of a solve (0.3, 0.5 and 0.7)
  • \(T\), \(T_{\mathrm{rms}}\) (columns T, T_rms; °C, K) — mean air temperature as published (Cases J and L) and the root mean square of its fluctuation
  • \(\overline{u'c'}\), \(\overline{v'c'}\), \(\overline{w'c'}\) (columns uc, vc, wc; ppm·m/s) — turbulent concentration fluxes
  • \(\overline{u'\theta'}\), \(\overline{v'\theta'}\), \(\overline{w'\theta'}\) (columns uT, vT, wT; K·m/s) — turbulent heat fluxes (\(\theta'\): the temperature fluctuation)
  • \(\theta = \dfrac{T_f - T}{T_f - T_H}\) (dimensionless) — floor-referenced temperature the solves of Cases J and L are scored on: 0 at the floor, 1 in the approach air at \(H\); \(T_f\) is the floor temperature and \(T_H\) the temperature of the approach flow at \(H\)

Scores

  • \(n\) — number of pairs of a measured value \(o_i\) and a modelled value \(p_i\) at one point
  • \(R\) — Pearson correlation coefficient of the pairs
  • \(R^2\) — square of \(R\): share of the measured variance a linear fit to the model explains
  • \(\mathrm{FAC2}\) — fraction of pairs with \(0.5 \le p_i/o_i \le 2\)
  • \(\mathrm{FB}\) — fractional bias \(2(\overline{o}-\overline{p})/(\overline{o}+\overline{p})\), positive when the model is low
  • \(\mathrm{HR}\) (hit rate) — fraction of pairs with \(\lvert p_i-o_i\rvert/\lvert o_i\rvert \le \delta_{\mathrm{rel}}\) or \(\lvert p_i-o_i\rvert \le \delta_{\mathrm{abs}}\), where \(\delta_{\mathrm{rel}} = 0.25\) and \(\delta_{\mathrm{abs}} = 0.05\thinspace U_H\) for a velocity component
  • \(\mathrm{NMSE} = \dfrac{\overline{(o-p)^2}}{\overline{o}\thinspace \overline{p}}\) — normalised mean square error
  • \(\mathrm{NAD} = \dfrac{\overline{\lvert o-p\rvert}}{\overline{o}+\overline{p}}\) — normalised absolute difference
  • \(\mathrm{MG} = \exp\overline{(\ln o-\ln p)}\) — geometric mean bias
  • \(\mathrm{VG} = \exp\overline{(\ln o-\ln p)^2}\) — geometric variance
  • The ratio of \(\mathrm{FAC2}\) keeps its sign, so a model of the other sign than the measurement is a miss; \(\mathrm{FB}\), \(\mathrm{NMSE}\) and \(\mathrm{NAD}\) divide by the means, so they are undefined (shown as —) for a signed component unless the measured and the modelled mean are both positive
  • VDI 3783-9 calls the hit rate \(q\) and the tolerances \(D\) and \(W\); on this site those letters keep their other meanings (\(q\) the tracer release rate, \(D\) the block size of Case C, \(W\) the vertical velocity)
  • The detection limit of a concentration is one value per case, \(10^{-3}\) of the largest measured value. It sets \(\delta_{\mathrm{abs}}\) of the hit rate and the both-below rule of \(\mathrm{FAC2}\), and floors the values in \(\mathrm{MG}\) and \(\mathrm{VG}\).
  • The scores \(\mathrm{FB}\), \(\mathrm{FAC2}\), \(\mathrm{NMSE}\), \(\mathrm{NAD}\), \(\mathrm{MG}\) and \(\mathrm{VG}\) follow Chang & Hanna (2004); the hit rate follows VDI 3783-9.

Quantities scored

  • \(U\), \(V\) and \(W\) (mean velocity components) over \(U_H\)
  • \(k\) (turbulent kinetic energy) over \(U_H^2\)
  • the speed \(\lvert\mathbf{U}\rvert\) and \(\sqrt{\lvert\mathbf{U}\rvert^2+2k}\), against a thermistor’s \(V_{\mathrm{s}}\)
  • \(c^\ast = C/c_0\) (dimensionless concentration) at turbulent Schmidt numbers \(\mathrm{Sc_t}\) of 0.3, 0.5 and 0.7
  • \(\theta\) (floor-referenced temperature)

Measurement locations

Measurement lines and planes, Case B
Figure 1. Case B, measurement lines and planes: A vertical profiles, \(y/H = 0\); B plane \(z/H = 0.0625\). Positions in multiples of \(H\).

Explore the points

model lowmodel high
Loading the points… Hover a point to see where it sits in 3D and how the model differs from the measurement; click to pin it. This needs JavaScript; the figures on this page show the same scores.

OpenFOAM 12

Status: scored

Measured against modelled

Measured vs modelled, Case B
Figure 2. Case B, measured vs modelled \(U\), \(V\), \(W\) and \(k\), with the 1:1 line and, where the quantity does not change sign, the factor-of-two band. The dots take the colour of the line or plane they were measured on, as in Figure 1.
  • Normalised by \(U_H\). FB is positive when the model is low.
  • Reattachment length behind the building: \(L_r/H\) = 3.27.
  • \(V\) and \(W\) change sign.
  • The model’s \(V\) is 0 on the centre plane (\(y = 0\), a symmetry plane); the tunnel measures small non-zero values there.
Centre-plane profiles, Case B
Figure 3. Case B, vertical profiles of \(U/U_H\) on the centre plane at the first three measured stations behind the building, offset per station.
Interpretation: Case B scores and profiles
  • The streamwise field is right: \(R = 0.96\), FAC2 0.70.
  • \(k\) is under-predicted (FB +0.46, \(R = 0.40\)); Cases A and H show the same under-prediction.
  • \(V\) and \(W\) change sign, so factor-of-two agreement is a harsh measure near zero and FAC2 is low; R and hit rate carry the comparison there.
  • The flat row of model \(V\) ≈ 0 in the \(V\) panel is the symmetry plane: a RANS solution is symmetric, so \(V\) ≡ 0 at \(y = 0\).
  • The model follows the measured profile closely above roof height and in the recovering wake; the near-ground deficit immediately behind the building is where the closure does least well.

Scores

Reattachment length \(L_r/H\) = 3.27.

Score R \(R^2\) FAC2 FB Hit rate
\(U\) +0.959 0.920 0.701 +0.240 0.562
\(V\) +0.910 0.828 0.281 +0.370 0.612
\(W\) +0.929 0.863 0.536 +0.263 0.808
\(k\) +0.399 0.159 0.562 +0.463 0.223

Every score is over the same 224 points.

OpenLB

Status: scored

OpenLB (Eddy3D’s eddy3dWind: Smagorinsky, third-order regularised D3Q19, image 1.9.3-r6, CPU build) solved the experiment with the same fitted inflow as OpenFOAM 12 and probed the same measured points, at \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\) (\(H/12.5\)), 2.6 million cells, 33 min on 24 threads of one node (55 MLUPS). \(H/20\) costs about 16 times the cell-steps.

Measured vs modelled U and V, Case B, one row per engine
Figure 4. Case B, measured vs modelled \(U\) and \(V\) on the same points, one row per engine: OpenFOAM 12, OpenLB and FluidX3D. FluidX3D’s hollow points are the near-ground heights it is not scored on.
Quantity n FAC2 FB hit rate R OpenFOAM FAC2 / hit rate / R
\(U/U_H\), all points 224 0.42 +0.22 0.28 0.85 0.70 / 0.56 / 0.96
\(U/U_H\), points ≥ 2 cells above the floor 95 0.61 −0.00 0.53 0.93 —
\(V/U_H\) 224 0.21 — 0.59 0.87 0.28 / 0.61 / 0.91
\(W/U_H\) 224 0.32 — 0.70 0.80 0.54 / 0.81 / 0.93
  • OpenLB meets the ground with a plain no-slip (bounce-back) wall at \(z = 0\) and carries the roughness only in the inlet profile. In the undisturbed column at the windward face its \(U\) is 0.61 of the inlet log law at the first node (half a cell up), 0.88 one cell up, 1.14 at 1.5, 1.10 at 2, 1.05 at 3 and 1.01 from 4.
  • The slab’s wall faces sit at \(z = 0\) and \(19.2\thinspace\mathrm{m}\) (the 20 m slab quantised to whole cells, 432 solid nodes). Until image 1.9.3-r6 the solver’s lattice started at the domain’s floor, which put the wall half a cell up; this run replaces the one made with that lattice.
  • The resolved \(k\) is 13 % of the measured one; \(k\) is not compared.
  • Reattachment length: \(L_r = 2.63\thinspace H\); the field fits best sampled \(H/4\) downstream (R 0.86 against 0.85).
  • Control: with the lateral sign flipped, \(V\)’s R becomes −0.87; \(U\) scrambled across the points drops R to 0.08.
  • OpenLB’s STL reader segfaults when lattice nodes sit on its octree planes; Eddy3D moves such a lattice by \(10^{-4}\thinspace\mathrm{d}x\) (LBMCaseXml). With the ground row half a cell under the floor this case is clear of them as configured.
Interpretation: OpenLB on Case B
  • The near-zero model \(U\) near the floor is the no-slip wall; above two cells FAC2 rises to 0.61 and FB vanishes.
  • The solve is nearly steady: a steady log-law inflow at \(H/12.5\) resolves almost no turbulence, so \(k\) is not compared, and with nothing to entrain the wake runs displaced.
  • Its frame is right.

Scores

Reattachment length \(L_r/H\) = 2.63.

Score R \(R^2\) FAC2 FB Hit rate n
\(U\) +0.849 0.721 0.424 +0.220 0.277 224
\(U\), points ≥ 2 dx +0.929 0.864 0.611 -0.001 0.526 95
\(V\) +0.869 0.755 0.214 +1.046 0.594 224
\(W\) +0.796 0.634 0.321 +0.549 0.701 224
\(k\) (resolved) +0.449 0.202 0.049 +1.533 0.018 224
\(V\), lateral sign flipped (control) -0.869 0.755 0.179 +3.823 0.509 224
  • Per-point values: /api/cases/B/pairs/openlb.csv
  • Reproduce: Eddy3D fixture TestAijCaseBOpenLbContainerE2E (OpenLB eddy3dWind, image 1.9.3-r6, CPU build, \(\mathrm{d}x = 1.6\thinspace\mathrm{m}\)); figure validation/plot_cases_ab_engines.py

FluidX3D

Status: scored

FluidX3D (Eddy3D’s GPU lattice-Boltzmann engine: D3Q19 with Smagorinsky LES) solved the experiment at \(\mathrm{d}x = H/64\) (3.125 mm, 121 million cells) with the same fitted log-law inflow, averaging over 500 \(H/U_H\) after 100 of spin-up. The run has an empty-domain control on the same lattice, and a point is scored only at a height where that control holds the imposed log law within 5 %: 88 of the 224 points. Its row is in Figure 4.

Quantity, comparable heights n FAC2 FB hit rate R OpenFOAM on the same points: FAC2 / hit rate / R
\(U/U_H\) 88 0.64 +0.26 0.57 0.95 0.78 / 0.68 / 0.97
\(W/U_H\) 88 0.48 — 0.32 0.85 0.78 / 0.69 / 0.95
  • The run meets the ground with FluidX3D’s bounce-back wall; it predates Eddy3D’s opt-in ground wall function. In the empty domain the three lowest heights read 0.77, 0.58 and 0.93 of the imposed log law. Those points are scored separately (velocity.*.excluded) and drawn hollow in Figure 4.
  • Empty-domain control, \(U\): FAC2 0.56, R 0.81. The solved field: FAC2 0.64, R 0.95.
  • \(V\) is not a headline score: all 88 comparable points lie on the centre plane, where the mean \(V\) is zero by symmetry.
Interpretation: FluidX3D on Case B
  • The excluded points are the floor again.
  • The solved field beats its approach flow on every \(U\) statistic.
  • Its frame is right.

Scores

Reattachment length \(L_r/H\) = 2.63.

Score R \(R^2\) FAC2 FB Hit rate n
\(U\), comparable heights +0.947 0.896 0.636 +0.260 0.568 88
\(W\), comparable heights +0.845 0.714 0.477 +0.114 0.318 88
\(U\), all points +0.905 0.820 0.638 +0.172 0.411 224
\(U\), approach flow alone (control) +0.807 0.651 0.557 -0.541 0.398 88
  • Per-point values: /api/cases/B/pairs/fluidx3d.csv
  • Reproduce: Eddy3D fixture TestAijCaseBFluidX3DE2E (Eddy3DCli.Tests/Validation/TestAijVelocityFluidX3DE2E; FluidX3D GPU lattice Boltzmann, D3Q19 Smagorinsky LES, \(\mathrm{d}x = H/64\), 100 + 500 \(H/U_H\) spin-up + averaging, with an empty-domain control on the same lattice); figure validation/plot_cases_ab_engines.py

ML surrogate

Status: scored

Eddy3D’s Wind Predictor (the public Yel-2.0 network) predicts \(\lvert\mathbf{U_{\mathrm{mean}}}\rvert/U_{\mathrm{ref}}\) at one height, 1.8 m, so the case is scaled ×144 to put its measurement plane there; the prediction is read at the tunnel’s own sensor coordinates. How the surrogate is run is on the ML surrogate page.

Measured vs predicted, ML surrogate, Case B
Figure 5. Case B, measured vs predicted pedestrian-height speed, ML surrogate, with the 1:1 line and the factor-of-two band.
Quantity n FB FAC2 R hit rate OpenFOAM, same points and quantity
\(\lvert\mathbf{U}\rvert/U_H\), \(z = H/16\) plane 122 +0.82 0.36 −0.24 0.17 FB +0.12, FAC2 0.84, R +0.87, hit rate 0.57
  • FB is positive when the model is low. Reversing the wind changes the predicted field by at most \(1 \times 10^{-4}\) \(U_{\mathrm{ref}}\): the prediction does not depend on the wind direction.

Scores

Score R \(R^2\) FAC2 FB Hit rate
\(\lvert\mathbf{U}\rvert\) (\(z = H/16\) plane) -0.236 0.055 0.361 +0.821 0.172
\(\lvert\mathbf{U}\rvert\), lattice cut at ±250 m +0.010 0.000 0.459 +0.463 0.189
wind reversed (control) -0.236 0.055 0.361 +0.821 0.172

Every score is over the same 122 points.

Notes on every run

  • Normalisation: velocities by \(U_H\), \(k\) by \(U_H^2\) (the benchmark's own convention; Case E by \(U\) at 15.9 m, as it is published); hit rate \(\delta_{\mathrm{rel}} = 0.25\), \(\delta_{\mathrm{abs}} = 0.05\) in those units

Leaderboard

Rank Run Source Hit rate, \(U\) Hit rate, \(V\) Hit rate, \(W\) Hit rate, \(\lvert\mathbf{U}\rvert\) Hit rate FAC2 R \(R^2\)
1 OpenFOAM simpleFoam (Eddy3D Outdoor case writer) (eddy3d-of12-measured-inlet) community 0.571 0.607 0.821 0.625 0.656 0.578 +0.936 0.876
2 OpenFOAM 12 Eddy3D 0.562 0.612 0.808 0.625 0.652 0.585 +0.936 0.876
3 OpenFOAM simpleFoam (Eddy3D Outdoor case writer) (eddy3d-of12-nutcap-lifted) community 0.540 0.638 0.723 0.634 0.634 0.580 +0.934 0.872
4 FluidX3D Eddy3D 0.411 0.545 0.527 0.424 0.477 0.491 +0.875 0.766
5 OpenLB Eddy3D 0.277 0.594 0.701 0.321 0.473 0.376 +0.837 0.701

All runs scored by validation/score_submissions.py. Ranked by mean hit rate (VDI 3783) over the tasks; bold: best per column. Not ranked: the ML surrogate (no point ids) and negative controls. Community rows are self-reported (Validate your own solver).