School of Mechanical Aerospace and Civil Engineering

Introduction

MSc Fluid Mechanics In a turbulent ow the mean temperature equation can be written as

Turbulent Heat Flux Modelling

T. J. Craft
George Begg Building, C41
Contents: Navier-Stokes equations Inviscid ows Creeping Flows Boundary layers Transition, Reynolds averaging Turbulence and modelling Mixing-length models One- and Two-equation models Turbulent heat transport Near-wall ow modelling Linear EVM weaknesses Flow management

(Uj ) = + t xj xj

uj xj

(1)

where = / is the thermal diffusivity, with the molecular Prandtl number. To close the system, we need to approximate the turbulent heat uxes, ui . Here we consider relatively simple methods for approximating ui , which mirror the eddy-viscosity approach for modelling the Reynolds stresses, ui uj . A similar situation arises for other transported scalars in turbulent ows (eg. species concentration). Similar modelling practices to those outlined here are usually adopted for them.

Reading: F.M. White, Fluid Mechanics J. Mathieu, J. Scott, An Introduction to Turbulent Flow P.A. Libby, Introduction to Turbulence P. Bernard, J. Wallace, Turbulent Flow: Analysis Measurement & Prediction S.B. Pope, Turbulent Flows D. Wilcox, Turbulence Modelling for CFD Notes: http://cfd.mace.manchester.ac.uk/tmcfd - People - T. Craft - Online Teaching Material

Turbulent Heat Flux Modelling

2011/12

2 / 11

Eddy-Diffusivity Models

When considering the dynamic eld, an analogy was drawn between the Reynolds stresses and viscous stresses. This led to the eddy-viscosity approach of modelling ui uj = t

Extending the above ideas to the turbulent heat uxes, we might approximate these by ui = t xi where the eddy-diffusivity, t , is taken as t = t /t and t is the turbulent Prandtl number.

(3)

Ui Uj + xj xi

+ (2/3)k ij

(2)

Again, t is not a property of the uid, but in principle depends on the local ow and turbulence conditions. In practice, t is usually taken as a constant of around 0.9 for near-wall ows. In free ows a slightly lower value (around 0.7) is often more appropriate. Notice that with this form the turbulent heat transport depends directly on the corresponding mean temperature gradient. A constant temperature in one direction implies no turbulent heat transport in that direction.

with t being the eddy (or turbulent) viscosity.

We noted that t is not a property of the uid, but depends on local ow conditions. Apart from in the viscous sublayer, t is generally much larger than . We examined a number of modelling approaches for approximating t , ranging from mixing-length (zero-equation) schemes to two-equation models.

Turbulent Heat Flux Modelling

2011/12

3 / 11

Turbulent Heat Flux Modelling

2011/12

4 / 11

Turbulent Heat Fluxes in Simple Shear Flow

Clearly the model prediction of u = 0 is not an accurate representation of reality. However, if streamwise gradients are relatively small, misrepresenting u may not have too serious an effect, at least in some non-buoyant ows. The reason for this can be seen from the boundary layer form of the mean temperature equation: U

In a simple shear ow, with U(y ), (y ), the eddy-diffusivity model gives

=0 x v = (t /t ) y
u = (t /t )

U(y) (y) y x

U(y)

(y)

In equilibrium conditions experiments suggest t 0.7 0.8 and |u /v | 1.1 in an homogeneous free shear ow.

+V = x y y

v y

(4)

Measurements at higher strain rates show t 1.1 and |u /v | 2.2.

In this situation only the cross-stream heat ux, v , is particularly inuential.

Turbulent Heat Flux Modelling

2011/12

5 / 11

Turbulent Heat Flux Modelling

2011/12

6 / 11

Buoyancy-Affected Flows

Horizontal Buoyant Flows

In a buoyancy-affected ow we saw an extra generation term appears in the k transport equation: Gk = gi ui (5) With the above eddy-diffusivity model we relate the turbulent heat uxes directly to the corresponding temperature gradients. To understand the model behaviour we thus examine two cases: one where the temperature gradient and gravitational vector are aligned, and one where they are not.
g

In the (stable) situation shown, using the eddy-diffusivity model for v , we now get Gk = g v = g(t /t )

U(y)
g

(y)

(6)

/ y is positive, g is negative, so Gk is negative as we would expect.

If the temperature gradient were reversed, Gk would change sign, again as expected. Although qualitatively showing the correct sign, such simple models are often not particularly accurate in a quantitative sense, particularly in stably stratied ows.

U(y)
g

(y)

T hot

T cold

y x
y x

Turbulent Heat Flux Modelling

2011/12

7 / 11

Turbulent Heat Flux Modelling

2011/12

8 / 11

Vertical Buoyant Flows

The GGDH Heat Flux Model

g

Now consider a vertical ow as shown. We still have

T hot

An improved turbulent heat ux model is often provided by the generalized gradient diffusion model of Daly & Harlow (1970). In this, we take ui = c k uu i j xj

T cold

Gk = g v = g(t /t ) y

(7)
y x

(8)

The constant c is typically taken around 0.3. In a simple shear ow considered earlier, we now have u = c v = c k uv y k 2 v y
y x U(y) (y)

However, / y is now rather small (the dominant temperature gradient is normal to the wall), so Gk is also small. From the simple shear ow discussion earlier, we expect the model to underpredict the magnitude of the streamwise heat ux (v in this case). Here, this underprediction can be expected to lead to an underestimation of the magnitude of Gk .

Note that reliable values of the individual Reynolds stress components are now needed.
2011/12 10 / 11

Turbulent Heat Flux Modelling

2011/12

9 / 11

Turbulent Heat Flux Modelling

If using the linear EVM formulation for the stresses the heat ux expressions become k U u = c t y y v = (2/3)c k2 y

The GGDH generally gives a better representation of the turbulent heat uxes than the simple eddy-diffusivity model (in the above example, u is non-zero now). However, to realize a signicant improvement a better underlying model for the Reynolds stress components than the eddy-viscosity representation is often needed.

Turbulent Heat Flux Modelling

2011/12

11 / 11