Independent project · Thermal simulation

Diamond pin-fin cold plate

CFD reconstruction and validation

I reconstructed a diamond pin-fin cold plate from a research paper and modeled it in ANSYS Fluent. I followed the study's flow conditions to compare my results with its experimental and CFD data, then checked mesh convergence and the use of a half-domain model.

Copper cold plate with diamond pin fins.

1 kW

Applied heat load

0.5 to 4 LPM

Eight flow conditions

5.27M

Selected half-domain cells

<2%

Mesh sensitivity at 3 LPM

Approach

Geometry and simulation setup

The paper also used ANSYS Fluent. I reconstructed its copper Design 1 in SpaceClaim using the reported dimensions and figures, since a complete manufacturing drawing was not available.

During troubleshooting, I traced excessive pressure loss to a pin-spacing error and corrected the minimum passage width from about 0.80 mm to 1.00 mm.

I used a half-domain model to reduce computation time, then checked it against a full-domain simulation at 3 LPM. Pressure drop and thermal resistance agreed within approximately 1%.

Solver: ANSYS Fluent 2025 R2
Physics: Steady water flow with copper-water heat transfer
Turbulence model: SST k-ω
Inlet temperature: 333.15 K (60 °C)
Heated area: Approximately 75 × 75 mm
Applied heat flux: 177,800 W/m²
Comparison data: Published experiment and CFD, digitized from figures

Numerical checks

Mesh convergence

I compared three half-domain meshes at 3 LPM. The 5.27-million-cell mesh differed from the 14.03-million-cell mesh by 0.22% in pressure drop and 1.66% in thermal resistance, so I used it for the flow sweep. It required 62.5% fewer cells than the fine mesh.

Full-domain cold plate mesh overview — Full-domain mesh. The mesh-convergence study used the half-domain model.

Close view of the mesh around diamond-pin passages — Local mesh detail around the pin field.

Pressure-drop and thermal-resistance differences from the fine mesh for coarse, medium, and fine grids — Pressure-drop and thermal-resistance differences relative to the fine mesh at 3 LPM.

Mesh settings

The direct-polyhedral mesh included refinement around the pin field and eight prism layers on the water-side walls. The area-averaged wall y+ remained below 1 throughout the flow sweep.

I used the medium mesh for the remaining simulations because further refinement changed both results by less than 2%.

Results

Thermal resistance and pressure drop

I followed the paper's flow-rate study from 0.5 to 4 LPM in 0.5 LPM steps. I extracted its experimental and Fluent results with WebPlotDigitizer and compared them with my simulations. Higher flow rates improved cooling but increased pressure drop.

Thermal resistance versus flow rate, comparing this model with published experiment and CFD — Maximum-temperature-based thermal resistance. The largest discrepancy occurs at 0.5 LPM.

Pressure drop versus flow rate, comparing this model with published experiment and CFD — Pressure loss increases nonlinearly with flow rate. The model underpredicts the experimental curve near 3 LPM.

Comparison at 3 LPM

At 3 LPM, my full-domain model gave a thermal resistance of 10.40 K/kW and a pressure drop of 1.81 kPa. The experimental values were approximately 9.98 K/kW and 1.99 kPa, respectively, based on the digitized curves. Thermal resistance was closer to the experiment at higher flow rates, with the largest difference at 0.5 LPM.

Static-temperature distribution across the cold plate at 3 LPM — Full-domain temperature distribution at 3 LPM and approximately 1 kW heat input. Legend units: kelvin.

Static-pressure distribution across the water-side surfaces at 3 LPM — Static-pressure distribution from the full-domain model at 3 LPM. Legend units: pascals.

Assumptions and calculation details

Thermal-resistance definition. Rth = (maximum heater-surface temperature − inlet temperature) / applied power. Values in K/W were converted to K/kW for comparison with the paper.

Geometry and material assumptions. The reconstruction used published dimensions and interpreted figures, not manufacturer CAD. The baseline had no meaningful bypass above the pin tops. Water and copper properties were held constant.

Domain check. The plots use the half-domain flow sweep. Full-domain geometry, mesh images, and contours are shown for clarity. The full-domain check was performed at 3 LPM, not at every flow rate.

Outlet treatment. Localized outlet reverse flow persisted, and Fluent's reverse-flow-prevention treatment affected approximately 24% of the outlet area. The reported outputs stabilized, but this remains a limitation. Testing an extended outlet without that intervention would be a useful follow-up.

Reference data. The experiment and published CFD curves were digitized with WebPlotDigitizer from the Pinkus et al. copper and copper-tungsten cold-plate study in the International Journal of Heat and Mass Transfer. I reconstructed and simulated the model; I did not design the original device or conduct the physical experiments.