Skip to content

Showcase

English | Japanese

« Home

A gallery of computed animations and figures, grouped by topic. Each section ends with a link to the detailed page.

GPU liquid cold-plate three-level multi-objective optimization (1D → FVM → LBM)

Unsteady 3D thermal LBM of a staggered pin-fin GPU liquid cold plate (velocity magnitude and temperature, 3.07 M cells)

1D to FVM Pareto front to unsteady thermal LBM verification and Miyabi-G GH200 simultaneous CPU/GPU utilization

Simultaneous CPU + GPU multi-objective optimization of a 700 W-class staggered pin-fin liquid cold plate on one Miyabi-G node (1 GH200, 13.7 min). The 1D model screens 2,000 candidates in 0.1 s to seed Optuna NSGA-II; 64 Grace CPU cores (72% mean utilization) run 116 steady FVM evaluations (snappyHexMesh + buoyantBoussinesqSimpleFoam), while the Hopper H100 GPU (96% mean utilization, 3,933 MLUPS) continuously verifies 15 emerging Pareto designs with 3.07 M-cell unsteady 3D thermal LBM (D3Q19 TRT+Smagorinsky + D3Q7). The knee design (D = 1.02 mm, U = 0.25 m/s) achieves ΔP = 715 Pa, T_max = 52.7 °C, and unsteady hotspot fluctuation T_rms = 0.57 K.

Automotive aero three-level multi-objective optimization (1D → FVM → D3Q27 Cumulant LBM)

Unsteady 3D D3Q27 Cumulant AMR-LBM of Ahmed + diffuser designs (4-level Octree AMR, 5 mm, 13.8 M cells)

1D to FVM to D3Q27 Cumulant LBM Pareto verification against Ahmed 1984 experimental band and Miyabi-G GH200 simultaneous CPU/GPU utilization

Simultaneous three-level optimization of rear slant, underbody diffuser angle, and ride height on the canonical Ahmed body (\(U_\infty = 40\text{ m/s},\; Re_H = 7.68\times 10^5\)) on Miyabi-G (1 GH200 node): 1D aero screening (2,000 designs / 0.1 s) → 64 Grace CPU cores running 32 high-fidelity 440k-cell steady FVM evaluations (5 mm surface + 3 prism layers, \(C_D = 0.183\text{–}0.322\), 78% mean / 91% peak utilization) → Hopper H100 GPU verifying 5 Pareto vehicles with OpenRheo's production D3Q27 Cumulant + WALE + 4-level Octree AMR (5 mm, 13.8 M cells) + Bouzidi + Spalding + Kawai–Larsson friction solver (92% mean utilization, 331 MLUPS). All three fidelities align inside the Ahmed 1984 experimental band (\(C_D = 0.230\text{–}0.285\); optimal 7.2° diffuser: FVM \(C_D = 0.235\) / LBM \(C_D = 0.239,\; C_L = +0.085\)) while catching 3D diffuser stall at 11.7° (\(C_D = 0.304\)).

Vehicle aerodynamics — DrivAer (LBM fed by an FVM solution)

Velocity magnitude on the DrivAer symmetry plane (y = 0): animation of unsteady LBM (D3Q27 cumulant, finest cell 1.1 cm) fed by a steady FVM solution

DrivAer symmetry plane: (a) steady FVM mean, (b) LBM snapshot, (c) time histories at four wake points

Steady FVM (3.43 M cells) gives Cd 0.257 (experiment 0.252). That velocity field becomes the inflow and the wake is solved with LBM (19.82 M cells). → DrivAer fastback

Thermal environment of an apartment living room

Unsteady LBM (wind speed on a vertical section through the seat) Steady FVM temperature and wind speed
Wind speed on a vertical section through the seat (unsteady LBM) Steady FVM temperature and wind speed of the living room

Matches 56 measured surface and air temperatures to RMS 0.63 K. Swinging the circulator raises the turbulence intensity from 34 % to 58 % and lowers SET* from 27.3 to 27.0 °C. → OpenCAE Symposium 2026 talk

rheoMesh boundary-layer meshes

DrivAer centre-plane sections and wall layer counts, left: snappyHexMesh, right: rheoMesh

DrivAer boundary layers Tyre / wheel boundary layers
DrivAer boundary layers Wheel boundary layers

→ rheoMesh

Turbulence — wall-mounted cube (ERCOFTAC case 41)

Animation of the Q-criterion around a wall-mounted cube

→ Wall-mounted cube

Unsteady flow around a duct and grille (an example of multi-level coupling)

An air-conditioning circuit of fan → duct → diffuser → grille, solved part by part with different methods. Parts that do not need spatial distribution use 1D, the diffuser, where a steady solution suffices, uses FVM, and only the grille section, where vortices matter, is solved with unsteady LBM (with AMR).

Circuit layout

Unsteady LBM of the grille section

Velocity magnitude:

Vorticity — vortex shedding in the grille wake is visible:

Pressure:

Division of roles across the three tiers

Tier Target Compute resource
1D model Fan characteristic, pressure loss of straight ducts CPU 1 core
Steady FVM Diffuser (separates but is steady) CPU, a few cores
Unsteady LBM Grille wake (vortex shedding is the main feature) GPU

On GH200 the Grace CPU and the Hopper GPU can be used at the same time, so this division fits as is in 1 node. → Design philosophy §6 · Multi-fidelity (1D ⇄ FVM ⇄ LBM)

Free surface — dam break

Evolution of the water surface for the SPHERIC dam break with an obstacle (MARIN experiment).

SPHERIC dam break, water surface at t = 0.4 s

Dam-break animation:

Time series of pressure and water level are compared with the experiment → Free surface

Heat and phase change — melting and natural convection

Gallium melting (2D) Natural convection (Boussinesq)
Gallium melting Natural-convection animation

Double-distribution-function (DDF) thermal LBM. → Heat transfer and phase change

Multiphase — phase field

A rising bubble solved with a high-density-ratio Allen-Cahn phase field.

→ Two-phase flow (phase field)

Particles and fluid-structure interaction — 3D bifurcation

Particle deposition in a bifurcation, solved with two-way coupling through momentum exchange.

→ Particles and fluid-structure interaction

Overview of validation coverage

Validation summary

Case coverage

All cases are under Categories on Home; validation tiers are in Validation status.