Sieder-Tate Correlation
The Theoretical Basis of the Sieder-Tate Correlation
The Lineage of Turbulent In-Tube Heat Transfer Correlations
The effort to express the heat transfer coefficient of turbulent forced convection in a circular tube as a dimensionless correlation begins with the Dittus-Boelter equation (1930s). Because Dittus-Boelter treats the fluid properties as constant, it carries a known defect: for fluids whose viscosity changes strongly with temperature — oils, glycerine, high-viscosity process liquids — the measurements deviate systematically between heating and cooling. Sieder-Tate (1936) is the classic that absorbed this property-variation effect as a correction term built on the ratio of bulk viscosity to wall viscosity, and it remains in active service today for preliminary heat exchanger design and check calculations.
The Equation and What Each Term Means
In the turbulent regime the Sieder-Tate correlation takes the following form.
$$ Nu = 0.027\, Re^{0.8}\, Pr^{1/3} \left( \frac{\mu_b}{\mu_w} \right)^{0.14} $$
Here \( \mu_b \) is the viscosity at the bulk (mixing-cup) temperature and \( \mu_w \) the viscosity at the wall temperature. Every property is evaluated at the bulk temperature, and only the viscosity-ratio term carries information about the wall. On heating (hot wall) the viscosity near the wall falls, the boundary layer thins and heat transfer increases — \( \mu_b/\mu_w > 1 \) puts the correction on the enhancing side. Cooling does the reverse. The nominal range of validity is \( Re > 10^4 \), \( 0.7 < Pr < 16700 \), \( L/D > 10 \) (fully developed flow).
The Laminar Form, and Which Equation to Choose
For laminar flow (\( Re < 2300 \)) with a long thermal entry length there is the Sieder-Tate laminar equation \( Nu = 1.86\,(Re\,Pr\,D/L)^{1/3} (\mu_b/\mu_w)^{0.14} \), while in the turbulent regime the later Gnielinski correlation (friction-factor based, extended through the transition region to \( 3000 \lesssim Re \lesssim 5\times10^6 \), accuracy around ±10%) is the modern recommendation. A practical way to remember the choice is as a three-way split: "Gnielinski when accuracy and breadth matter, a Sieder-Tate type correction alongside when the viscosity correction of a viscous liquid is the point, Dittus-Boelter for a quick estimate". Whichever you use, state the error band with it (roughly ±20-25% for Sieder-Tate).
Calculation Procedure and a Worked Example
Iterating When the Wall Temperature Is Unknown
μ_w is the viscosity at the wall temperature, right? But the wall temperature isn't known until the heat transfer coefficient is settled — isn't that chicken and egg?
Good catch. Exactly right — strictly speaking it becomes an iteration. The procedure is: (1) compute a provisional h with the viscosity correction set to 1, (2) estimate the wall temperature from the heat flux and the thermal resistances, (3) evaluate μ_w at that wall temperature and recompute h with the correction included, (4) repeat (2) and (3) until the wall temperature stops moving. In practice the correction is raised to the 0.14 power and is therefore insensitive, so two or three passes converge. For a low-viscosity fluid such as water the correction is a few per cent and a single iteration is usually enough.
A Worked Example — How Much Does the Correction Matter for Oil?
Consider heating a mineral oil in a tube (bulk 60°C: \( \mu_b \approx 0.072 \) Pa·s; wall 100°C: \( \mu_w \approx 0.017 \) Pa·s). The viscosity correction comes out as
$$ \left(\frac{\mu_b}{\mu_w}\right)^{0.14} = \left(\frac{0.072}{0.017}\right)^{0.14} \approx 4.2^{0.14} \approx 1.22 $$
In other words a 22% increase in the heat transfer coefficient — for oil the correction is not measurement scatter but a magnitude that drives the design. Switch the same case to cooling (wall 20°C, \( \mu_w \approx 0.28 \) Pa·s) and the correction becomes \( (0.072/0.28)^{0.14} \approx 0.83 \), a 17% reduction. Heating and cooling differ by 40% in total — this figure is the whole reason Dittus-Boelter (no correction) misses on viscous liquids. Water, by contrast (viscosity ratio about 1.7 between 60°C and 100°C), gets a correction of \( 1.7^{0.14} \approx 1.08 \), only about 8%.
Folding the Result into a Heat Exchanger Calculation
The tube-side \( h_i \) obtained from Sieder-Tate is combined in series with the fouling resistance, the tube-wall conduction and the outside heat transfer coefficient to form the overall heat transfer coefficient \( U \), from which the LMTD or NTU-ε method solves for the exchanger surface area and the outlet temperatures. The bulk temperature is evaluated as the arithmetic mean of inlet and outlet (split the duty into segments when the temperature change is large). Because the ±20% of the correlation maps directly onto the uncertainty in U, state the error band explicitly as the justification for the design margin (area margin).
Guidance for Practical Application
Checklist Before You Apply It
- Check Re — at \( Re < 2300 \) (laminar) move to the laminar equation; at \( 2300 < Re < 10^4 \) (transition) the turbulent Sieder-Tate equation is out of range — use Gnielinski
- Development length — over the first \( L/D < 10 \) from the inlet the entry effect raises h. For short tubes, consider an entry-length correction
- Non-circular cross-sections — the hydraulic diameter \( D_h = 4A/P \) can be substituted (turbulent only; for laminar flow use the exact solution for the specific cross-section)
- Natural convection creeping in — at low velocity and large temperature difference, check the Richardson number. The mixed-convection regime lies outside the scope of forced-convection correlations
- Property evaluation temperature — bulk temperature for the main properties, wall temperature for \( \mu_w \) alone. Confusing this convention is the most frequent mistake of all
Dividing the Work with CFD
In-tube flow can of course be solved with CFD, but the working rule is that "for a plain circular tube, Sieder-Tate/Gnielinski is faster and more trustworthy". CFD earns its place when (1) the geometry has a complex cross-section, a bend or fins, so that no correlation exists, (2) you need the distribution — entry effects, local hot spots, (3) the temperature dependence of the properties is so extreme that the correlation would be an extrapolation. Even then, validate the CFD setup against the correlation on a plain circular tube first before advancing to the real geometry — the same discipline as the flat-plate validation.
Reporting With the Error Band Included
A correlation-based design report should contain: (1) the equation used and confirmation that its range of validity is satisfied, (2) the source of the property data and the evaluation temperatures, (3) the nominal error band of the correlation (±20-25% for Sieder-Tate), (4) the effect of that error on the design quantities (heat transfer area, outlet temperature). A report that gives only the point value "h = 1234 W/m²K" is, given the nature of a correlation, an overstatement dressed up as precision.
How the Tools Handle It
Implementation Status by Tool Class
| Tool class | Implementation | Practical note |
|---|---|---|
| Heat exchanger design tools (HTRI, Aspen EDR, etc.) | Sieder-Tate/Gnielinski-family tube-side correlations are selectable | Always check which correlation is the default. Watch the default fouling factors too |
| 1-D thermo-fluid networks (Flownex, AMESim, etc.) | Correlations built into the heat transfer of pipe elements | Interpolation through the transition region differs from tool to tool |
| Spreadsheets / in-house calculations | The equation implemented directly | Self-check the viscosity-correction iteration and the unit system (Pa·s vs. cP for viscosity) |
| CFD (Fluent, OpenFOAM, etc.) | No correlation used — solved directly | Use Sieder-Tate as the comparison target for validation. The correction effect is not reproduced unless temperature-dependent viscosity is enabled |
What CFD Needs in Order to Reproduce the Viscosity Correction
To capture in CFD the physics that the Sieder-Tate viscosity correction stands for, the temperature dependence of viscosity must be part of the property model (constant-property CFD can only give the Dittus-Boelter-equivalent answer). Furthermore, because the viscosity change near the wall happens inside the boundary layer, a low-Re wall treatment (y+ ≤ 1) is required or the effect is buried in the assumptions of the wall function. When a conjugate heat transfer analysis of an oil system comes out "systematically lower than the correlation", check these two points first — temperature-dependent viscosity and wall resolution.
Frontiers
Extending the Limits — Supercritical, Nanofluids, Non-Newtonian
For supercritical-pressure fluids (sCO2 cycles, supercritical water) the violent property changes near the pseudo-critical point throw the classical correlations badly off, and matching dedicated correlations against variable-property CFD is an active research area. For nanofluids and non-Newtonian fluids (foods, polymer solutions) too, proposals keep appearing for correlations that extend the Sieder-Tate idea of a "viscosity-ratio correction". The common lesson is that a correlation can be trusted only within the range of the database it was developed on — and the importance of checking the range of validity only rises in a new field.
Data-Driven Correlations and Hybrid Forms
In work that regresses in-tube heat transfer from large bodies of experimental and DNS data using machine learning, the mainstream design keeps the dimensionless groups (Re, Pr, viscosity ratio) as the input features. Taking the functional form of a classical correlation as the skeleton and optimising the coefficients on data strikes a good balance between extrapolation safety and improved accuracy, and implementations are beginning to appear in industrial tools. Degradation on fluids and conditions outside the training data is unavoidable, however, so the recommended usage is to run the classical correlation alongside as a sanity-check lower and upper bound.
Convergence with Local Measurement
Infrared thermography and distributed fibre-optic temperature sensing now make the local heat transfer distribution along the tube axis directly measurable, raising the resolution of validation from "a correlation for the mean h" to "verification of the local distribution". Correlating and modelling the local phenomena that used to be buried in the average — entry regions, the region downstream of a bend, the effect of pulsation — is under way.
Troubleshooting
Symptoms, Causes, and Fixes
| Symptom | Likely cause | Fix |
|---|---|---|
| Deviates from the measured h by more than ±25% | Applied in the transition region; entry length; fouling build-up | Check Re and move to Gnielinski. Review the operating history and the fouling factor |
| The sign of the error flips between heating and cooling | The viscosity correction was omitted or inverted (numerator and denominator of μ_b/μ_w) | Re-check the definition of the correction term. Is it >1 on heating? |
| h is off by an order of magnitude | Viscosity units (cP vs. Pa·s); the wrong hydraulic diameter | Check the unit system end to end. Verify that Re is plausible (of order 10^4 to 10^5) |
| CFD comes out systematically lower than the correlation | Constant-property settings; insufficient wall resolution | Enable temperature-dependent viscosity. Move to a y+ ≤ 1 mesh |
| Only the low-flow conditions are far off | Laminarisation, or natural convection creeping in | Check Re and the Richardson number. Switch to the laminar equation or a mixed-convection correlation |
| The predicted outlet temperature does not match the plant | The representative bulk temperature is inappropriate (large temperature change) | Split the pipe run into segments and update the properties and h segment by segment |
When You Cannot Decide Which Equation to Use
Dittus-Boelter, Sieder-Tate, Gnielinski... In one sentence, which one am I actually supposed to use?
In one sentence: "when in doubt use Gnielinski, and for a viscous liquid never forget the Sieder-Tate correction". Gnielinski has the widest range of validity and the best accuracy at ±10%. But for a fluid whose viscosity moves a lot with temperature, whichever equation you use, apply a \( (\mu_b/\mu_w)^{0.14} \) type correction alongside it — the historical contribution of Sieder-Tate lies in that correction term itself, and even though the full equation has aged, the thinking behind the correction is still in service. And if you only need a quick estimate on a low-viscosity fluid such as water, Dittus-Boelter will not cause you trouble in practice. Get into the habit of asking first "how much does this fluid's viscosity move with temperature", and the choice of equation settles itself.
Related: forced convection over a flat plate, index of forced convection articles, iterative coupling in conjugate heat transfer.
Feel the theory hands-on with the interactive simulators in this field
Simulator LibraryRelated Fields
detail
error