Skip to content

p26412/ThermoFOAM-DC

BranchesTags

Repository files navigation

<<<<<<< HEAD

ThermoFOAM-DC v0.1

OpenFOAM-based data-center cooling and hotspot analysis workflow

ThermoFOAM-DC is a reproducible OpenFOAM workflow for studying simplified air-cooled data-center thermal management. Version 0.1 models a rectangular data-center room with rack heat-source zones, floor supply tiles, a return outlet, buoyancy-driven thermal transport, and RANS turbulence modelling.

This v0.1 release adopts RNG k-epsilon as the primary turbulence model, selected from the grid-sensitivity and y+ verification study. Standard k-epsilon and k-omega SST setups are retained as sensitivity cases to compare the effect of turbulence-model choice on the cooling predictions.

v0.1 scope

Included in this release:

  • 3D simplified data-center room geometry
  • six rack heat-source zones using fvModels
  • floor-mounted cold-air supply tiles
  • return outlet
  • foamRun -solver fluid workflow for OpenFOAM Foundation v13 / openfoam-dev style cases
  • RNG k-epsilon primary baseline case
  • uniform structured grid study: grid_07, grid_10, grid_14
  • late-iteration averaging for quasi-steady oscillatory RANS results
  • y+ verification for wall-function consistency
  • standard k-epsilon and k-omega SST sensitivity setups
  • non-uniform mesh-layout sensitivity resources
  • custom dcMetricsPlus post-processing utility
  • curated result tables and plots

Deferred to v0.2:

  • low/high supply velocity cases
  • warmer supply air case
  • high rack heat-load case
  • overloaded rack case
  • design trade-off plots and ranking

Physical model

The v0.1 model uses a single fluid region representing room air. Racks are represented as heat-source cell zones rather than detailed solid cabinets. This is a deliberate baseline simplification.

Main physics:

  • turbulent room airflow
  • temperature transport
  • buoyancy through Boussinesq thermodynamics
  • rack heat sources
  • wall-function RANS turbulence modelling

Primary turbulence model:

RNGkEpsilon

Sensitivity models:

kEpsilon
kOmegaSST

Why late-iteration averages are used

The steady RANS solutions show noticeable quasi-steady oscillation in monitored thermal metrics. Therefore, v0.1 reports late-iteration averaged metrics over:

4000, 5000, 6000, 7000, 8000

The final value at 8000 is treated as a supporting snapshot, not as divine truth. This is a CFD project, not a fortune cookie.

Main engineering metrics

The custom dcMetricsPlus utility extracts:

  • average rack-inlet temperature
  • 95th percentile rack-inlet temperature
  • maximum rack-inlet temperature
  • rack-inlet risk fraction above 305 K
  • average room temperature
  • 95th percentile room temperature
  • maximum room temperature
  • room hotspot fraction above 315 K
  • return outlet area-average temperature
  • return outlet outflow-weighted temperature
  • return outlet outflow flux
  • pressure difference between supply and return

For grid independence, integrated/statistical quantities such as TavgRackInlet, T95RackInlet, T95Room, and hotspot fractions are emphasized over single-cell maxima.

Key v0.1 verification result: RNG k-epsilon

The formal grid study uses the uniform structured grid family and late-iteration averages from 4000-8000.

Grid Cells hEff [m] TavgRackInlet mean [K] T95RackInlet mean [K] Rack risk fraction T>305K TavgRoom mean [K] T95Room mean [K] Room hotspot fraction T>315K deltaP mean [Pa] Role
grid_07 38,220 0.1401 302.957 313.911 0.497 307.129 314.169 0.0319 0.598 coarse reference
grid_10 105,000 0.1000 302.695 313.542 0.481 307.123 313.513 0.0161 0.566 production mesh candidate
grid_14 285,180 0.0717 302.040 311.928 0.449 306.728 312.944 0.00914 0.583 fine-grid reference

The fine grid was checked further because the minimum y+ was locally below 30. Only 8 out of 26,763 wall faces were below y+ = 30, corresponding to 0.0299% of wall faces. Therefore, the fine grid is acceptable as the reference grid for wall-function RANS.

y+ verification: grid_14 RNG k-epsilon

Patch Faces Faces with y+ < 30 Percent below 30 Min y+ Avg y+ Max y+
floor 6526 0 0.0000% 32.1298 91.7744 160.0916
ceiling 6209 2 0.0322% 27.4215 116.0299 183.3349
walls 14028 6 0.0428% 27.4215 89.9197 266.7922
total 26763 8 0.0299% - - -

Recommended mesh interpretation

  • grid_07: coarse reference
  • grid_10: practical production mesh candidate with low runtime
  • grid_14: fine-grid reference with acceptable y+ distribution

For v0.2 parameter studies, use grid_10 for the full sweep and rerun selected important cases on grid_14 for verification.

How to run the RNG baseline case

From the repository root:

cd cases/01_baseline_RNGkEpsilon
./Allclean
./Allrun

Compile the post-processing utility:

cd ../../src/dcMetricsPlus
wmake

Run metrics:

cd ../../cases/01_baseline_RNGkEpsilon
dcMetricsPlus

Run y+ verification:

foamPostProcess -solver fluid -func yPlus -latestTime
python3 ../../scripts/yplus_fraction.py 8000/yPlus 30
=======
# ThermoFOAM-DC v0.2

**OpenFOAM-based data-center cooling, hotspot, and parameter-study workflow**

ThermoFOAM-DC is a reproducible OpenFOAM workflow for studying simplified air-cooled data-center thermal management. Version 0.2 extends the verified v0.1 RNG k-epsilon baseline into a ten-case cooling parameter study covering supply velocity, supply temperature, rack heat load, overloaded-rack scenarios, pressure-flow proxy metrics, per-rack hotspot analysis, and CFD visualization outputs.

This repository is intended as an open engineering benchmark and research workflow, not a production-certified data-center digital twin.

---
## Version history

- **v0.1.0**: Baseline OpenFOAM data-center cooling workflow with grid sensitivity, y+ verification, turbulence-model sensitivity, and custom hotspot metrics.
- **v0.2.0**: Ten-case parametric cooling study with automated global/per-rack post-processing, pressure-flow proxy analysis, publication-style plots, and CFD contour/profile visualization.

Archived versions:
- v0.1.0: https://github.com/p26412/ThermoFOAM-DC/tree/v0.1.0
- v0.2.0: https://github.com/p26412/ThermoFOAM-DC/tree/v0.2.0


## v0.2 scope

Included in this release:

- simplified 3D data-center room with floor supply tiles, ceiling return, and six rack heat-source zones
- OpenFOAM Foundation v13 / `foamRun -solver fluid` style case setup
- RNG k-epsilon primary RANS workflow
- v0.1 grid, y+, and turbulence-model sensitivity documentation retained
- ten-case v0.2 parameter-study matrix
- automated case generator for C00-C09
- custom `dcMetricsPlus` post-processing utility
- global late-window tables and SVG plots
- optional per-rack table and heatmap generation from `dcMetricsPlus_perRack` outputs
- ParaView Python CFD figure generation
- filled contours with optional isolines
- line-profile CSV/SVG generation for `Ux`, `Uy`, `Uz`, `|U|`, `T`, `p_rgh`, `p`, `k`, `epsilon`, `omega`, `nut`, and `alphat` when available

---

## v0.2 case matrix

| Case | Folder | Description |
|---|---|---|
| C00 | `C00_baseline` | Six racks at 3 kW each, supply velocity 0.75 m/s, supply temperature 291 K |
| C01 | `C01_lowSupplyVelocity` | Baseline heat load, lower supply velocity 0.50 m/s |
| C02 | `C02_highSupplyVelocity` | Baseline heat load, higher supply velocity 1.00 m/s |
| C03 | `C03_warmerSupplyAir` | Baseline heat load, warmer supply temperature 295 K |
| C04 | `C04_highRackLoad` | All racks increased to 5 kW each |
| C05 | `C05_overloadedRack_L2` | Rack L2 overloaded to 8 kW, other racks at 3 kW |
| C06 | `C06_worstCase_lowFlow_highLoad` | Low flow with all racks at 5 kW |
| C07 | `C07_mitigation_highFlow_highLoad` | High-flow mitigation for all racks at 5 kW |
| C08 | `C08_overloadedRack_L2_lowFlow` | Overloaded L2 with low supply velocity |
| C09 | `C09_overloadedRack_L2_highFlow` | Overloaded L2 with high supply velocity |

The machine-readable case matrix is stored in:

```text
metadata/v02_case_matrix.csv
>>>>>>> aeb3467 (Release ThermoFOAM-DC v0.2 ten-case parameter-study workflow)

Repository structure

<<<<<<< HEAD
cases/       OpenFOAM case setups
src/         dcMetricsPlus source code
scripts/     grid, metrics, plotting, v0.2 case generation, and y+ helper scripts
docs/        methodology and publication notes
results/     curated CSV tables and plots
environment/ OpenFOAM version/run notes

Generated OpenFOAM time folders, VTK files, logs, and build outputs are ignored by .gitignore.

ThermoFOAM-DC/ cases/ 01_baseline_RNGkEpsilon/ 02_baseline_kEpsilon_sensitivity/ 03_baseline_kOmegaSST_sensitivity/ parameterstudy/ C00_baseline/ C01_lowSupplyVelocity/ ... C09_overloadedRack_L2_highFlow/ scripts/ create_parameter_cases_v02_10cases.py collect_per_rack_summary.py run_v02_analysis.sh v02_generate_tables_and_plots.py cfd_viz/ src/ dcMetricsPlus/ metadata/ v02_case_matrix.csv inputs/ README.md docs/ results/ tables/ plots/


Generated OpenFOAM time folders, `processor*` folders, logs, VTK output, ParaView `.foam` files, and build outputs are excluded by `.gitignore`.

---

## Requirements

Core CFD workflow:

```text
OpenFOAM Foundation v13 or compatible openfoam-dev style setup
Python 3

Optional visualization workflow:

ParaView with pvpython, or python3-paraview

The table and SVG-plot scripts intentionally avoid NumPy, pandas, and matplotlib so that summary generation works on minimal HPC environments.

Compile the custom metric utility

From the repository root:

source /opt/openfoam13/etc/bashrc
cd src/dcMetricsPlus
wmake
cd ../..

Adjust the OpenFOAM path if your installation is different.

Generate the ten v0.2 cases

The release already includes generated C00-C09 case templates under cases/parameterstudy. To regenerate them from a baseline case:

python3 scripts/create_parameter_cases_v02_10cases.py \
  cases/01_baseline_RNGkEpsilon \
  cases/parameterstudy \
  --overwrite

For the full v0.2 sweep, the recommended production mesh is the grid-10 case from the v0.1 mesh study. If you have regenerated gridStudy_RNG/grid_10, use it as the base:

python3 scripts/create_parameter_cases_v02_10cases.py \
  gridStudy_RNG/grid_10 \
  cases/parameterstudy \
  --overwrite

Run all ten cases

From the repository root:

source /opt/openfoam13/etc/bashrc
./Allrun cases/parameterstudy

Equivalent manual loop:

cd cases/parameterstudy
for case in C*/; do
    echo "Running $case"
    ( cd "$case" && ./Allclean && ./Allrun )
done
cd ../..

Run or reuse dcMetricsPlus

If the simulations are already complete and you already have:

postProcessing/dcMetricsPlus/0
postProcessing/dcMetricsPlus/1000
...
postProcessing/dcMetricsPlus/8000

then you do not need to rerun dcMetricsPlus.

To recompute global metrics over all saved times:

cd cases/parameterstudy
for case in C*/; do
    echo "Metrics $case"
    ( cd "$case" && dcMetricsPlus -time '0:8000' )
done
cd ../..

If your custom utility also writes per-rack metrics to:

postProcessing/dcMetricsPlus_perRack/<rack>/<time>/metrics.csv

then the v0.2 analysis wrapper can automatically collect those into inputs/v02_per_rack_summary_user.csv.

Generate v0.2 tables and publication SVG plots

Use a late-window average to avoid startup transient contamination:

bash scripts/run_v02_analysis.sh \
  --cases-root cases/parameterstudy \
  --late-start 4000

Expected global outputs:

results/tables/v02_global_timeseries.csv
results/tables/v02_global_summary.csv
results/tables/v02_global_summary_compact.csv
results/tables/v02_flow_balance_table.csv
results/tables/v02_heat_balance_diagnostic_table.csv
results/tables/v02_main_findings.csv
results/plots/publication/Fig02_TavgRackInlet_mean_std.svg
results/plots/publication/Fig03_T95RackInlet_mean_std.svg
results/plots/publication/Fig04_rackRiskFraction_mean_std.svg
results/plots/publication/Fig05_roomHotspotFraction_mean_std.svg
results/plots/publication/Fig06_pressureFlowProxy_mean_std.svg
results/plots/publication/Fig07_pressureProxy_vs_T95RackInlet.svg

If per-rack data are available, additional outputs include:

results/tables/v02_per_rack_summary.csv
results/tables/v02_per_rack_hotspot_ranking.csv
results/plots/publication/Fig08_perRack_Tavg_heatmap.svg
results/plots/publication/Fig09_perRack_T95_heatmap.svg
results/plots/publication/Fig12_perRack_risk_heatmap.svg

A root-level helper is also provided:

./Allpostprocess_v02 cases/parameterstudy 4000

ParaView figure generation

Create .foam files:

bash scripts/cfd_viz/make_foam_files.sh cases/parameterstudy

Generate optional derived CFD fields at time 8000:

source /opt/openfoam13/etc/bashrc
bash scripts/cfd_viz/write_cfd_fields_all_cases.sh cases/parameterstudy 8000

Generate main ParaView CFD figures:

export PVPYTHON_BIN=pvpython
bash scripts/cfd_viz/run_pv_figures_all_cases.sh cases/parameterstudy 8000

Generate velocity/scalar contours with contour lines and profile plots:

export PVPYTHON_BIN=pvpython
PV_NUM_CONTOUR_LINES=18 PV_CONTOUR_LINE_WIDTH=1.5 \
bash scripts/cfd_viz/run_velocity_figures_all_cases.sh \
  cases/parameterstudy \
  8000 \
  results/plots/cfd_velocity_with_isolines

This produces filled contours, isolines, and profile plots for available fields such as:

Ux, Uy, Uz, |U|, T, p_rgh, p, k, epsilon, omega, nut, alphat

aeb3467 (Release ThermoFOAM-DC v0.2 ten-case parameter-study workflow)

Current limitations

  • racks are represented as heat-source zones, not detailed server cabinets <<<<<<< HEAD
  • no porous rack resistance in v0.1
  • no experimental validation yet
  • transient RANS was tested but was too expensive for this v0.1 workflow
  • LES/hybrid RANS-LES are deferred to later high-fidelity work

Roadmap

v0.2

Add RNG k-epsilon design parameter studies:

  • low supply velocity
  • high supply velocity
  • warmer supply air
  • higher rack heat load
  • one overloaded rack
  • optional worst-case and mitigation cases

v0.3

Add porous rack resistance and better rack airflow representation.

v0.4

Add validation or benchmark comparison with published data-center cooling datasets.

v1.0

Automated data-center cooling case generation, parametric sweeps, and design ranking.

  • no porous rack pressure-loss model in v0.2
  • no experimentally validated data-center benchmark comparison yet
  • pressure-flow proxy is a relative engineering metric, not a full fan-energy model
  • per-rack post-processing depends on the availability of dcMetricsPlus_perRack outputs
  • CHT, server fan curves, CRAH/CRAC coil models, and raised-floor plenum modelling are deferred

Recommended interpretation

v0.2 is suitable as a GitHub engineering/research workflow showing:

  • reproducible OpenFOAM case setup
  • parametric CFD case generation
  • automated hotspot metrics
  • late-window statistical reporting
  • per-rack hotspot comparison when available
  • CFD visualization and line-profile extraction

For journal submission, v0.2 should be extended with validation, uncertainty analysis, selected fine-grid reruns, and improved physical realism.

License

GPL-3.0-or-later.

aeb3467 (Release ThermoFOAM-DC v0.2 ten-case parameter-study workflow)

About

OpenFOAM workflow for data-center cooling CFD, hotspot metrics, grid/y+ verification, and thermal-management studies.

Topics

data-center heat-transfer cfd openfoam hvac energy-efficiency thermal-management rans electronics-cooling

Resources

Readme

License

View license
Activity

Stars

6 stars

Watchers

0 watching

Forks

3 forks
Report repository

Releases 1

ThermoFOAM-DC v0.2.0: Ten-case data-center cooling parameter study Latest
Jun 11, 2026

Packages

 
 
 

Contributors

Languages