Conjugate heat transfer calculator
Hand checks for a CHT simulation. For a hot gas in a pipe or port: the gas-side heat transfer coefficient by Gnielinski and Dittus-Boelter with real air properties, the wall temperatures, the heat lost and the outlet temperature, and what Fluent's Surface Heat Transfer Coef. will print for your reference temperature. For an electronics case: the source term for your chip, the air temperature rise and a heat-sink estimate. Free, no sign-up, runs in your browser.
It opens on the exhaust port and graphics card challenges of Skill-Lync's Ansys Fluent course. The third tab checks the numbers in your report and names the slips Skill-Lync found when it read 1,022 of them.
Gnielinski 1976 · Dittus-Boelter · CoolProp air properties · Fluent User's Guide definitions · Updated 20 Nov 2026
The challenge's boundary conditions: 5 m/s at 700 K at the inlets, a convection wall with h = 20 W/m²K outside. The port's diameter (166 mm is the inlet size most reports quote), wall thickness and length are yours to measure in SpaceClaim; the result is for one straight runner of that size.
The die is the course model's 8 × 8 × 1 mm. The heat sink defaults are an example, not the course geometry: measure your fins. The estimate ignores spreading in the base, the die-to-sink contact and air that bypasses the fins, so treat it as a lower bound on the die temperature.
The case is the one on the Exhaust port tab. Type what your report used; each line says what is right and what is not.
What Fluent prints as Surface Heat Transfer Coef. on the inner wall for each reference temperature, at the runner's mean gas temperature. Only the bulk temperature gives the gas-side h.
- Re 12,400: turbulent, inside both correlations' ranges. Gnielinski 10.9 W/m²K, Dittus-Boelter (n = 0.3) 12 W/m²K.
- Fluent's default air (ρ 1.225 kg/m³, μ 1.7894 × 10⁻⁵ Pa·s) at 700 K would carry 2.43 times this mass flow at Re 56,855, and give an inner h of 17.4 W/m²K instead of 10.9.
- The wall's three resistances per metre: inner film 0.175, wall 0.000046, outer film 0.0904 K·m/W: the gas side carries 66% of the resistance, the outside 34%, the wall almost none.
- Fluent's Surface Heat Transfer Coef. on the inner wall with the reference temperature at ambient (300 K) would read 21.2 W/m²K: that is the outer coefficient you set (20) scaled by r₀/rᵢ, not the gas-side h.
What are the two CHT challenges?
Both are conjugate heat transfer (CHT) projects in Ansys Fluent: the solid and the fluid are meshed together with shared topology, and Fluent solves conduction in the solid and convection in the fluid as one problem, with the wall temperature and heat flux coming out of the coupling instead of being imposed. The graphics card is a project of Skill-Lync's Introduction to GUI based CFD using Ansys Fluent, and so is the exhaust port (its lesson: "Setting up a Conjugate Heat Transfer Simulation of an Exhaust Port"). Skill-Lync fetched 1,613 public student reports on CHT on 20 Nov 2026; 530 distinct exhaust port reports and 492 graphics card reports had enough text to read.
- Exhaust port (457 reports titled "Week 4 - CHT Analysis on Exhaust port"): a port with four inlets and one outlet. Air at 5 m/s and 700 K enters each inlet, the aluminium wall loses heat outside through a convection boundary condition of 20 W/m²K, and the brief asks for the heat transfer coefficient and how you would verify it.
- Graphics card (392 reports titled "Week 6 - CHT Analysis on a Graphics card"): a card with a processor, fins and a base in an enclosure of air at 1, 2.5 and 5 m/s. You choose the processor's power, turn it into a source term, and find the hotspot and the wall heat transfer coefficient.
How do you verify the heat transfer coefficient?
Compare the gas-side coefficient with a correlation for turbulent flow in a pipe of the same diameter, at the same mass flow, with air properties at the gas temperature. Gnielinski's correlation (1976) is the better choice: it is valid from Re 3,000 to 5 × 10⁶ and Pr 0.5 to 2,000, and it uses the Petukhov (also credited to Filonenko) friction factor:
Dittus-Boelter is the one in most reports (139 of 530; 7 use Gnielinski). It is stated for Re ≥ 10,000, 0.6 ≤ Pr ≤ 160 and L/D ≥ 10, with properties at the bulk temperature, and the exponent depends on the direction of heat flow:
The exhaust gas is hotter than the wall, so it is being cooled and n = 0.3. For the course case with real air (D = 166.1 mm, 5 m/s, 700 K in), the mean gas temperature over the first metre is 688 K, Re = 12,400 and Pr = 0.709: Gnielinski gives h = 10.9 W/m²K, Dittus-Boelter 12.0 W/m²K. Both are fully developed values. Expect the CFD to be higher near the four inlets, where the boundary layers are thin, and at the bends and the junction, where the flow turns and mixes. A CFD average a little above the correlation is the right sign; one ten times higher points at the mesh or the definition you used.
Fluent's default air at 700 K: 2.4 times the mass flow
Fluent's default material "air" has constant properties: ρ = 1.225 kg/m³, cp = 1006.43 J/kg K, k = 0.0242 W/m K and μ = 1.7894 × 10⁻⁵ kg/m s, the values of air at about 15 °C. Air at 700 K and 1 atm has ρ = 0.504 kg/m³, cp = 1075, k = 0.0518 and μ = 3.42 × 10⁻⁵ (CoolProp). Of the 530 exhaust reports, 406 state the 700 K inlet; 1 switched density to the ideal-gas law and 2 made the other properties depend on temperature. Among the Dittus-Boelter reports, 71 put 1.225 kg/m³ in the text and 3 used a hot-air density.
| 5 m/s at 700 K, D 166.1 mm | Real air | Fluent default air |
|---|---|---|
| Mass flow per inlet | 0.0546 kg/s | 0.133 kg/s |
| Reynolds number | 12,400 | 56,855 |
| h, Gnielinski | 10.9 W/m²K | 17.4 W/m²K |
| h, Dittus-Boelter (n = 0.3) | 12.0 W/m²K | 19.5 W/m²K |
| Inner wall temperature | 432 K | 477 K |
| Gas temperature drop over 1 m | 24.9 K | 14.7 K |
Same velocity, 2.4 times the mass and 4.6 times the Reynolds number. The default air carries more heat capacity, so it cools 1.7 times less per metre, and its gas-side h is 59% higher. The reports' hand calculation with default air (Re 56,855, Pr 0.744, h = 19.52 W/m²K) is arithmetically right and physically the wrong gas. In Fluent, set density to ideal-gas and give μ, k and cp a temperature dependence (kinetic theory, a polynomial or piecewise-linear table), or at least constant values at the mean gas temperature.
Why does Fluent's surface heat transfer coefficient read about 20?
Because of the reference temperature. Fluent's Surface Heat Transfer Coef. is not a property of the flow; it is the wall heat flux divided by the difference between the wall temperature and the temperature in Report > Reference Values:
(q is the combined convective and radiative flux.) Follow the heat from the gas to the room: it crosses the inner film, the wall and the outer film in series, the same heat per metre through each:
If the reference temperature is the ambient, q / (Twall − Tref) on the inner wall is just the conductance of the wall plus the outer film, which you typed in:
For a 5 mm aluminium wall around a 166.1 mm bore with ho = 20 W/m²K that is 21.19 W/m²K, with real air or default air, k-ε or k-ω. One report's four turbulence models gave 21.47, 21.36, 20.99 and 21.01 W/m²K and read the agreement as validation. A Dittus-Boelter value near 20 then looks like a match, and some reports compare with the 20 W/m²K of their own boundary condition. Set the reference temperature to the gas temperature and the same flux gives the gas side (10.9 W/m²K at the mean bulk temperature); the chart in the calculator shows the whole curve. 84 of the 530 reports mention the reference values at all.
The other number Fluent offers, the Wall Func. Heat Tran. Coef., is a different quantity. It is built from the turbulent kinetic energy in the wall-adjacent cell and the dimensionless temperature of the wall function, with no reference temperature:
It is not a bulk-to-wall coefficient, and it moves with the near-wall mesh. 131 reports plot both and are puzzled that one reads ten times the other (one table: 27.9 against 205 W/m²K). For validation against a correlation, compute q / (Twall − Tbulk) yourself, with Tbulk the mass-weighted average gas temperature at the same station.
How much does the exhaust cool along the port?
With the resistances above giving a conductance U′ per metre, an energy balance on the gas gives an exponential approach to the ambient (Incropera, Section 8.3, with the overall coefficient in place of h):
For the course case with real air the gas loses 24.9 K in the first metre (1459 W from one inlet's flow), the inner wall sits at about 432 K and the outside at 432 K. The gas-side film carries 66% of the resistance, the outer film 34% and the aluminium wall 0.02%: the wall is nearly isothermal through its thickness, and getting the gas-side h right matters more than the boundary condition outside. The calculator uses properties at the mean gas temperature; Skill-Lync's verifier marches the gas temperature along the pipe with local properties and agrees within 0.2 K.
How do you calculate the source term for the graphics card processor?
Fluent's energy source in a solid cell zone is a power per volume, so divide the power you assume by the volume of the die in cubic metres:
Multiply back to check: source × volume = power. In 150 graphics card reports Skill-Lync could pair a stated power with the source term on the 8 × 8 × 1 mm die. Most are right; at least 10 put 10 to 1,000 times the wrong power into the die, from 0.08 m written for 8 mm, 10e-9 for 1e-9, or 0.1 m for 1 mm. Another 6 kept the power density of a 40 × 40 × 2 mm, 75 W chip on the 8 × 8 × 1 mm die, which injects 1.50 W. The assumed power itself ranges from 1 W to several hundred; 1 W and 10 W are the most common. A real GPU's 75 to 300 W through an 8 × 8 mm die and a small sink at 1 m/s gives die temperatures far above any silicon limit; run the Graphics card tab at that power to see why.
Is my graphics card temperature plausible?
Start with the air. The enclosure inlet passes ρVA of air, and the heat it can carry away is P = ṁ cp ΔT:
For 10 W at 1 m/s through a 65 × 40.5 mm inlet the air warms by only 3.21 K (1.28 K at 2.5 m/s, 0.64 K at 5 m/s). So the hotspot is set by the convection from the fins, not by the air heating up. With a coefficient h on the fins, the fin efficiency and the sink's resistance follow (Incropera's corrected-length fin):
The calculator estimates h from the laminar flat plate over each fin's length, when you have no Fluent value:
| 10 W, air 300 K | 1 m/s | 2.5 m/s | 5 m/s |
|---|---|---|---|
| h, flat plate (W/m²K) | 28.5 | 45.1 | 63.8 |
| Fin efficiency | 0.99 | 0.98 | 0.97 |
| Sink resistance (K/W) | 5.63 | 3.59 | 2.56 |
| Sink base, estimate (K) | 358 | 337 | 326 |
Example sink: 15 aluminium fins (k 202.4 W/m K), 10 mm tall, 0.8 mm thick, 19 mm long, on a 31.5 mm wide base. At 1 m/s the boundary layer at the fins' trailing edge (2.7 mm) is thicker than half the 1.4 mm gap, so the layers merge and the flat-plate h is only a rough guide. Fluent's area-averaged h on the fins is the better input.
Read it the other way to check your run: the die's rise above the inlet air is roughly P times the sink resistance, plus the small air rise. If 10 W gives you 1,000 K, the sink in your model would need a resistance of about 70 K/W, which means the fins are not connected (shared topology off, or a contact gap) or the source is 10 times what you meant. The die itself, silicon at about 149 W/m K, adds little; the PCB, FR-4 at 0.3 W/m K through its thickness, carries almost nothing.
Which slips does the check tab catch?
- Default air at 700 K in the hand calculation or the simulation: 2.4 times the mass flow.
- n = 0.4 for a gas being cooled: Dittus-Boelter uses 0.3 when the wall is colder than the fluid.
- Unit slips in Re: μ written as 1.7892 instead of 1.7892 × 10⁻⁵ in a printed formula, D in mm. The tab recomputes your h from your own inputs.
- Comparing with the boundary condition: 20 W/m²K is what you set outside; the gas-side h is what you are checking.
- Surface HTC with the reference temperature at ambient: it reads back ho ro/ri.
- A source term that is not your power: source × die volume must equal the power you meant.
What this calculator does not cover
- The port's real geometry. Four inlets merging, bends and a junction raise the local h above any straight-pipe correlation. The calculator gives the straight-pipe value to compare with, not the answer.
- Radiation. At 700 K, gas-to-wall and wall-to-surroundings radiation is not small; the course case leaves it off, and so does the calculator. Fluent's surface coefficient includes radiation when the model is on.
- Entry-length and property-ratio corrections. The correlations are for fully developed flow, with properties at the bulk temperature.
- Heat-sink detail. Spreading resistance in the base, the thermal interface, bypass flow and merged boundary layers in narrow fin gaps all raise the die temperature above the estimate.
CHT is where CFD meets thermal design: an exhaust, an electronics enclosure, a battery pack. Both challenges are projects in Skill-Lync's Ansys Fluent course, with mentors who review your setup and your validation.
Frequently asked questions
How do I verify the heat transfer coefficient from a Fluent CHT simulation?
Compare the gas-side h with a correlation for the same pipe. Gnielinski is the better one (valid from Re 3,000), Dittus-Boelter with n = 0.3 when the gas is cooled. Use air properties at the gas temperature, not Fluent's default air. For 700 K air at 5 m/s in a 166 mm port: Re ≈ 12,400, h ≈ 10.9 W/m²K (Gnielinski), 12.0 (Dittus-Boelter).
Why does my Fluent surface heat transfer coefficient come out close to 20 W/m²K?
Because Fluent divides the wall heat flux by (wall temperature minus the reference temperature), and with the reference temperature near ambient that ratio is the outer convection coefficient you set. With h = 20 W/m²K outside a 5 mm aluminium wall it reads 21.19 W/m²K whatever the turbulence model. Set the reference temperature to the bulk gas temperature to read the gas side.
What is the difference between surface heat transfer coefficient and wall function heat transfer coefficient in Fluent?
The surface coefficient is q / (T_wall − T_ref) with T_ref from Report > Reference Values, so it changes when you change the reference temperature. The wall function coefficient is ρ c_p C_μ^¼ k_P^½ / T*, built from the turbulence in the wall cell and independent of any reference temperature; it depends on the near-wall mesh and is not a bulk-to-wall h.
Should I use Fluent's default air for a 700 K exhaust?
No. The default air has constant properties, ρ = 1.225 kg/m³ and μ = 1.7894 × 10⁻⁵ Pa·s, which is air at about 15 °C. Air at 700 K has ρ ≈ 0.504 kg/m³ and μ ≈ 3.42 × 10⁻⁵ Pa·s. At 5 m/s the default air carries 2.4 times the mass flow at 4.6 times the Reynolds number. Use the ideal-gas law for density and temperature-dependent μ, k and c_p.
How do you calculate the source term for the processor in a graphics card CHT simulation?
Divide the power by the volume of the processor in cubic metres. An 8 × 8 × 1 mm die is 64 mm³, which is 64 × 10⁻⁹ m³, so 10 W is 1.5625 × 10⁸ W/m³ and 1 W is 1.5625 × 10⁷ W/m³. Check by multiplying back: the source times the volume must give the power you meant.
Which n do I use in the Dittus-Boelter equation?
n = 0.4 when the fluid is being heated (wall hotter than the fluid) and n = 0.3 when it is being cooled (wall colder). Hot exhaust gas losing heat to the port wall is cooled, so n = 0.3. The equation is stated for Re ≥ 10,000, 0.6 ≤ Pr ≤ 160 and L/D ≥ 10, with properties at the bulk temperature.
Related on Skill-Lync
- Student project: Week 6, CHT analysis on a graphics card
- Pipe flow calculator
- Mixing tee calculator
- Battery thermal model calculator
- 2D heat conduction solver
- CFD courses compared
- All free engineering tools
Sources
- V. Gnielinski, "New equations for heat and mass transfer in turbulent pipe and channel flow", International Chemical Engineering 16 (1976) 359-368: the correlation and its range (3,000 ≤ Re ≤ 5 × 10⁶, 0.5 ≤ Pr ≤ 2,000), with the smooth-tube friction factor (0.790 ln Re − 1.64)⁻² (Petukhov; Filonenko), as restated by F. P. Incropera et al., Fundamentals of Heat and Mass Transfer, Ch. 8.
- Dittus-Boelter: Nu = 0.023 Re0.8 Prn, n = 0.4 heated, 0.3 cooled; Re ≥ 10,000, 0.6 ≤ Pr ≤ 160, L/D ≥ 10, bulk properties (Incropera Ch. 8). R. H. S. Winterton, "Where did the Dittus and Boelter equation come from?", Int. J. Heat Mass Transfer 41 (1998) 809-810, on the McAdams form used today. Laminar fully developed Nu = 3.66 (uniform wall temperature) and 4.36 (uniform flux).
- Ansys Fluent 12.0 User's Guide (afs.enea.it mirror): default air properties (Sections 8.3.2, 8.4.1, 8.5.1, 8.7.1: 1.225 kg/m³, 1.7894 × 10⁻⁵ kg/m s, 0.0242 W/m K, 1006.43 J/kg K); Surface Heat Transfer Coef. h = q/(Twall − Tref) with Tref from Reference Values and q the combined convective and radiative flux; Wall Func. Heat Tran. Coef. = ρ cp Cμ1/4 kP1/2/T*. The default reference temperature differs between sources and releases (288.16 K and 300 K are both quoted), so check the Reference Values panel of your version.
- Air at 1 atm: CoolProp, pseudo-pure air (E. W. Lemmon et al., J. Phys. Chem. Ref. Data 29 (2000) 331; E. W. Lemmon and R. T. Jacobsen, Int. J. Thermophys. 25 (2004) 21), tabulated every 25 K from 250 to 1100 K; interpolation within 0.04% of the full model; within 2% of Incropera Table A.4 at 300 and 700 K.
- Pipe with an external convection: Incropera Section 8.3 (exponential bulk temperature with an overall coefficient). Straight fin with the corrected length Lc = L + t/2, the laminar flat plate Nu = 0.664 Re1/2 Pr1/3 and the Blasius thickness 5x/Rex1/2: Incropera Ch. 3 and 7. Silicon about 149 W/m K; FR-4 0.29-0.34 W/m K through the board (more in its plane); aluminium 202.4 W/m K as Fluent's database lists it in the reports.
- Skill-Lync, 1,613 public CHT student reports, read 20 Nov 2026: boundary conditions, properties, correlations, reference temperatures and source terms; the source-term slips were checked by hand.
- Skill-Lync, scripts/verify/cht.py: re-implements every formula (the wall as a solved network, the outlet temperature by an RK4 march with local properties, the fin by shooting on its ODE) and checks the air table against CoolProp; the page checks itself against it at build time.