コンテンツにスキップ

forest-building

英語 | 日本語

« Home · « 外部流れ・風工学

AIJ CFD-guideline Case A -- isolated wall-mounted tall building in an atmospheric boundary layer (Meng & Hibi 1998; Tominaga, Mochida, Yoshie, Kataoka, Nozu, Yoshikawa & Shirasawa 2008, JWEIA 96:1749) -- on the FOREST (block-AMR) solver. Cloned from benchmarks/forest_ahmed_body.cpp: the ONLY case-specific changes are the geometry (BuildingGeom AABB prism, 1:1:2), the inflow (analytic ABL power-law U(z)=U_H(z/H)^alpha instead of the ERCOFTAC file reader), the refinement boxes, and the reference scales (Aref=bH, reference speed U_H). Everything else (forest build, Bouzidi wall data, domain BC, MEM force, slice/probe dumps, MRT/RR/HRR collide) is identical.

概要: AIJ ガイドラインの単独高層建物。市街地風環境評価の標準ケースで、風洞実験の風速プロファイルと比較する。

Geometry

forest-building geometry

Fields

この計算が最後の出力ステップで書き出した場をそのまま描いたもの。3D ケースは中央断面。

forest-building p_pa

forest-building umag_ms

Build & run

cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-forest-building

このページの図を作った計算のコスト(Miyabi-G Grace CPU (72 cores, -stdpar=multicore)): 203.2 s wall。

Validation

Tier T2 — 単点比較
QoI velocity_profile
Reference AIJ guidebook Case A/B (Tominaga et al. 2008); Meng & Hibi (1998)
Gate — ゲートなし

Source & run details

View source

Item Value
Target orheo-forest-building
Source native/benchmarks/forest_building.cpp
Lattice —
GPU twin yes (orheo-forest-building-gpu)
Modules ahmed_solver.hpp, building_geom.hpp, forest_inflow.hpp, forest_regrid.hpp, forest_step.hpp
Gallery run 2026-07-27 @ fc5fbb30
Args 8 8000 0.05 1200 0.50 0.27 1 1 .
Wall time 203.19 s
CPU time 7540.10 s
Device Miyabi-G Grace CPU (72 cores, -stdpar=multicore)

Computation log

この計算の実際の出力(先頭と末尾)。

=== forest building (AIJ Case A) ===
nh=8 steps=8000 ulat=0.0500 re=1.2e+03 cs=0.500 alpha=0.270 band=1 max_level=1
building b=0.080m H=0.160m (1:1:2, wall-mounted); dx0=0.02000m  domain blocks=32x22x12 (kNb=4)  tau=0.5010
=== building surface-following forest ===
forest: 8700 leaves, max_level reached=1 (requested 1)
=== building wall data (solid mask + Bouzidi qlink) ===
solid cells=3312 (of 1879200 pool cells)
point probe at world (x=0.410, z=0.070) -> probe.csv
=== time-stepping ===
step      0/8000  t_conv=0.000  Cd=-0.1720  Cl=-133.3570  umax=0.100 | 0.1s  0 step/s  0.0 MLUPS | Mdrift=+0.0 ppm
step   1000/8000  t_conv=0.391  Cd=1.1123  Cl=-133.0805  umax=0.087 | 21.1s  47 step/s  26.4 MLUPS | Mdrift=+0.0 ppm
step   2000/8000  t_conv=0.781  Cd=1.1101  Cl=-133.1377  umax=0.087 | 43.3s  46 step/s  25.7 MLUPS | Mdrift=+0.0 ppm
step   3000/8000  t_conv=1.172  Cd=1.0996  Cl=-133.1805  umax=0.087 | 64.3s  47 step/s  26.0 MLUPS | Mdrift=+0.0 ppm
step   4000/8000  t_conv=1.562  Cd=1.0971  Cl=-133.1977  umax=0.087 | 84.8s  47 step/s  26.3 MLUPS | Mdrift=+0.0 ppm
...

=== done: final Cd=1.0940 Cl=-133.2360 (mean-field QoI = slice profiles) ===
=== wrote slice_y0.csv (instantaneous over 1 steps) ===
=== wrote slice_y0_mean.csv (time-mean over 4001 steps) ===
=== wrote slice_z0125b.csv (instantaneous over 1 steps) ===
=== wrote slice_z0125b_mean.csv (time-mean over 4001 steps) ===
=== wrote slice_z125b.csv (instantaneous over 1 steps) ===
=== wrote slice_z125b_mean.csv (time-mean over 4001 steps) ===
=== wrote slice_zmid.csv (instantaneous over 1 steps) ===
=== wrote slice_zmid_mean.csv (time-mean over 4001 steps) ===