Repository navigation
Releases
Releases · Eddy3D-Dev/AIJ-Data-Validation
v3.11.3
Changed
- OpenLB stays
not comparableon Cases C to E, now with the wide empty-domain control re-run onopenlb-wind:1.9.3-r6
(the lattice's ground row half a cell under the domain), averaged over two crossings of the 2 m layer. Neither wall law
holds every 2 m point within 5 %: rough law Case C 0.88 (0.78–0.95), D 0.86 (0.79–0.91), E 0.90 (0.85–0.94); power law
C 0.96 (0.886–0.999), D 0.98 (0.928–1.022), E 0.90 (0.86–0.95). Case C's power law does over the first half crossing
(0.954–1.004), not over two. The case pages, the engines page and the README give the1.9.3-r6numbers; the
1.9.3-r4and-r5ones stay beside them.
v3.11.2
Changed
- OpenLB stays
not comparableon Cases C to E, now also on wider, taller and longer empty-domain controls. With Eddy3D's
OpenLB ground wall model and the rough law, in a domain 256 m wide and 320 m tall at
$$\mathrm{d}x = 1.6\thinspace\mathrm{m}$$ , the mean at the cases' 2 m height is within 4 % of the imposed log law (0.23–0.26
with the bounce-back ground), but not every point is within 5 %: Case C 1.02–1.07 (the 9 points over 5 % all within 144 m of
the inlet); Case E 1.012–1.046 averaged over half the time the 2 m layer takes to cross the domain, 0.985–1.0504 over two
crossings and 0.94–1.06 over the second alone; Case D 0.949–1.055 (a pattern across the span: the centre line 0.956, the lines
64 m either side 1.027). These ran onopenlb-wind:1.9.3-r4and-r5, before1.9.3-r6moved the lattice's ground row half
a cell down, and are not re-run on it. The engines page and the README now give the 32 m wide control's1.9.3-r6numbers
too, as the case pages have since 3.11.1 (rough law 0.82–0.86 at 2 m). The meaning ofnot comparableinvalidation.json
now names the control.
v3.11.1
Changed
- The OpenLB runs of Cases A and B are re-run on Eddy3D's
openlb-wind:1.9.3-r6, whose lattice puts its ground row half a cell
under the domain: the bounce-back wall is$$z = 0$$ , where the STL, the inflow and the measured points start. Until then the wall
sat half a cell (0.8 m) up and every building a row short (A's 16 m roof at 15.2 m; B's slab 18.4 m); "2 m" read 1.2 m above
the wall. A:$$U/U_H$$ R +0.76 → +0.78, FAC2 0.43 → 0.47;$$V$$ R 0.65 → 0.77;$$L_r$$ 3.39 → 2.80$$H$$ . B:$$U/U_H$$ R +0.79 → +0.85,
FAC2 0.38 → 0.42;$$V$$ R 0.83 → 0.87. Solved on 24 CPU threads (fp64; 13 and 33 min), re-scored by the same fixtures; pairs and
figures regenerated. The empty-domain controls of Cases C to E are re-run on the same image (their 2 m ratios change: the rough
wall model's first node is the half-cell node now).
v3.11.0
Added
- Grid uncertainty of the OpenFOAM runs of Cases A and E: four meshes, per-point band of the second (Roache's GCI via foamgci; half
the spread where values oscillate).grid_studyin/api/cases/{A,E}/validation.json(null elsewhere), bands in
/api/cases/{A,E}/grid/openfoam12-realizableKE.csv, one figure per page. Run with the eddy-viscosity cap lifted, as
eddy3d-of12-nutcap-lifted; served runs unchanged. API schema 2.6.0.
Changed
- The leaderboard's note on each case page is shorter.
v3.10.2
Changed
- FluidX3D's status on Cases C, D and E is
pending, notnot comparable. Eddy3D's FluidX3D has had an opt-in ground wall
function since 2026-11-05, and in an empty-domain control 96 m wide at each case's own inflow,$$\mathrm{d}x = 1.6\thinspace\mathrm{m}$$
and a steady log-law inflow, every point at the cases' 2 m height is within 5 % of the imposed log law (C 0.99–1.01, D 0.954–0.969,
E 0.98–0.99; 0.20–0.74 with the bounce-back ground). FluidX3D has not been run on these cases. The meaning ofnot comparable
invalidation.jsonnow names OpenLB alone.
v3.10.1
Fixed
- Case C of
submissions/eddy3d-of12-nutcap-liftedis now the fixture's own probes (a re-solve ofTestAijCaseC{0D,1D,2D}ContainerE2Eon
PACE ICE) instead of values sampled from solved fields; the scores move by at most 0.02 (hit rate) and 0.03 (bias).
v3.10.0
Added
- Cases E and L in
submissions/eddy3d-of12-nutcap-lifted, and Case D's 22.5° and 45° headings, from fresh re-solves of the Eddy3D fixtures on
PACE ICE with the cap lifted (a dot in D's probe names, E's 240-minute step limit and L's uncollected probe tables had stopped the first round).
Case E's thermistor-vector hit rate moves 0.412 to 0.550 (before) and 0.300 to 0.438 (after construction); Case L's$$U$$ 0.506 to 0.532,
$$k$$ 0.244 to 0.285 (bias 0.635 to 0.455),$$\theta$$ 0.985 to 0.994; Case D's scalar speed at 22.5° 0.667 to 0.652 and at 45° 0.712 to 0.652.
v3.9.1
Fixed
-
submissions/eddy3d-of12-nutcap-liftedCase B is the served set-up with the cap lifted (it was the measured-inlet run):$$U$$ hit rate
0.562 to 0.540, bias 0.24 to 0.11;$$k$$ hit rate 0.223 to 0.201, bias 0.46 to 0.31.
v3.9.0
Added
- Case D (wind direction 0° only) in
submissions/eddy3d-of12-nutcap-lifted, from a re-solve of the Eddy3D fixture on PACE ICE with the cap lifted:
the thermistor's scalar-speed hit rate moves 0.394 to 0.530, its bias 0.362 to 0.202. The 22.5° and 45° headings did not run (the fixture
stopped at an identifier it rejects). Case E's re-solve timed out after 240 minutes and Case L's probe tables were not kept by the job.
v3.8.0
Added
- Case G (conifer crown) in
submissions/eddy3d-of12-nutcap-lifted, the one case the cap lift makes worse: the$$U$$ hit rate over the
solve's own approach speed falls 0.286 to 0.071 and$$k$$ 's bias goes from −0.09 to −0.83, which suggests the drag values of the sweep
were calibrated with the cap on.