Sie sind auf Seite 1von 79

Fluid Mechanics and Heat Transport

Peter Ehrhard

WS 2017/18
Fluid Mechanics and Heat Transport

Preface

The lecture “Fluid Mechanics and Heat Transport” (1h), together with the corre-
sponding exercises (1h), during one semester is supposed to bring the master students
in “Process System Engineering (PSE)” onto a common and defined level in both sub-
jects, fluid mechanics and heat transport. This appears to be necessary, since these
students have completed their Bachelor degree at very different universities in the
world. Hence, this course does not start from scratch in both subjects, but instead
assumes that the students have already a reasonable amounts of knowledge in both
subjects. Therefore, the course aims to summarize, to repeat, and to extend the
knowledge of the students to a level, comparable to the level of the Bachelor students
at the TU Dortmund. Naturally, for some students with extensive knowledge in both
subjects, this may simply be a recapitulation. For others, though, it may present a
large amount of new or differently–formulated knowledge. These discrepancies can-
not be avoided, but the course aims to find a compromise here. The presentation
of fluid mechanics and heat transport together in a single course, hereby, appears to
be natural, since transport of momentum and heat have a lot of aspects in common.
Actually, there is likewise a close analogy between heat and mass (species) transport.
The latter, however, is not included in this lecture, since it re–appears at other places
of the PSE curriculum.

Dortmund 2018, Peter Ehrhard.

iii
P. Ehrhard

iv
Fluid Mechanics and Heat Transport

Contents
1 Introduction 1
1.1 Balances on various length scales . . . . . . . . . . . . . . . . . . . . 1
1.2 Molecular transport of momentum and heat . . . . . . . . . . . . . . 2
1.3 Shear stress in fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Relation of viscosity and molecular structure . . . . . . . . . . . . . . 7

2 Balances at 1D flow tubes 10


2.1 Balance of mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Balance of momentum . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Balance of energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Flow out of a container . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Flow with heat addition . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Viscous losses and heat transfer in 1D models 17


3.1 Pressure loss in pipes . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Heat transfer at pipe walls . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Integral balances 27
4.1 Integral mass balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Integral force balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Integral energy balance . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4 Example: flow through a bend . . . . . . . . . . . . . . . . . . . . . . 29

5 Modes of heat transport 35


5.1 Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.1.1 Example 1: Conduction in a wire/rod . . . . . . . . . . . . . . 35
5.1.2 Example 2: Conduction through two plates in contact . . . . . 37
5.2 Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.2.1 Example 3: Radiation between two parallel plates . . . . . . . 42

v
P. Ehrhard

6 Differential conservation equations 44


6.1 Conservation of mass . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.2 Conservation of momentum . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Conservation of heat . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.4 Scaling of the conservation equations . . . . . . . . . . . . . . . . . . 46
6.5 Exact solutions to the Navier–Stokes equations . . . . . . . . . . . . . 48
6.5.1 The Couette flow . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.5.2 The Poiseuille flow . . . . . . . . . . . . . . . . . . . . . . . . 50
6.6 The heat transport between two plates . . . . . . . . . . . . . . . . . 51

7 Boundary–layer equations 52
7.1 Kinematic boundary layer at the flat plate . . . . . . . . . . . . . . . 53
7.2 Thermal boundary layer at the flat plate . . . . . . . . . . . . . . . . 56

8 Reynolds–averaged equations 60
8.1 Conservation of mass . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.2 Conservation of momentum . . . . . . . . . . . . . . . . . . . . . . . 62
8.3 Conservation of heat . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.4 Turbulence models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

9 Collection of formulas 68

10 Nomenclature 72

vi
Figure 1: Vessel with adjacent components.

1 Introduction

1.1 Balances on various length scales

Balances, in general, can be formulated on largely–different length scales; it is just


a question of the size of the control volume. This holds simultaneously for mass,
momentum, heat, or total–energy balances. To clarify this aspect, let us pose an
example, given in figure 1. Here we recognize a vessel (or a reactor), which has an
inflow pipe, an outflow pipe with a pump, and a coil of heating pipes (i.e. a heat
exchanger) for thermal control. In the figure, likewise the mass fluxes entering and
leaving the vessel, ṁin , ṁex , the electrical power supplied to the pump Pel , and the
heat flux entering the heat exchanger Q̇hx are included.
On a macroscopic length scale, let us say a few meters, we could formulate integral
balances. In other words, we could use the entire vessel and all components as control
volume and formulate all fluxes entering and leaving this control volume. These fluxes
could incorporate e.g. fluxes of mass, of momentum, of heat, or of total energy. Based
on such balances, we could infer valuable information on integral quantities such as
the pumping power or the heat flux through the exchanger. For such balances, there
is no need to resolve details inside the control volume, i.e. inside the vessel, the pipes,
the heat exchanger, or the pump. Even the details of phase changes inside the control
volume would be irrelevant, since such effects would simply lead to mass fluxes of a
second phase across the control–volume boundaries.
On a microscopic length scale, let us say a few millimeters or centimeters, we could

1
P. Ehrhard

likewise formulate balances. This typically means formulating balances on small con-
trol volumes of fluid (and/or solid). In the limit of infinitesimally–small control
volumes, this leads to differential equations for the conservation of mass, momentum,
heat, or total energy, so–called transport equations. Solving the transport equations,
provided boundary conditions (BC) and initial conditions (IC) are given, leads to
detailed information inside all components. This could include profiles or fields of
e.g. velocity, pressure, temperature, etc. in either phases of the fluid, or even in the
solid structures.
On the molecular length scale, let us say several angstroms, we could likewise char-
acterize this system. Now the control volume would be a single atom/molecule, or
clusters of atoms/molecules. Balances now would concern the interaction of these
atoms/molecules, i.e. we would have to formulate intermolecular forces or force
potentials. Such considerations are typically conducted by physicists, leading to a
fundamental understanding of the fluid structure or of the molecular transport pro-
cesses. Engineers, instead, are used to treat fluids and solids as continua, whereas
the properties of these fluids or solids, e.g. viscosity or heat conductivity, reflect the
(integrated) molecular details.

1.2 Molecular transport of momentum and heat

Let us pose two simple problems, which are aimed to demonstrate the analogy between
the transport of momentum and heat. Firstly, we place a fluid between two horizontal
solid plates, all at rest at time t = 0. The situation is sketched at the top of figure 2.
If we now, for t > 0, start moving the top plate at velocity u0 = U to the right, we
expect that during an early stage only the fluid adjacent to the top plate will move
with the plate, while most of the fluid in the gap remains at rest. The corresponding
velocity profile u(y) is sketched in the middle of figure 2. As time proceeds, more
and more fluid starts moving in the direction of the top plate, while only the fluid
adjacent to the bottom plate remains at rest. This is because this fluid is in contact
with the bottom plate at rest. The reason why more and more fluid starts moving,
is of molecular nature – it is viscosity. From fluid layer to fluid layer, molecular
forces cause the neighboring atoms/molecules to move. This is molecular transport
of momentum. If we wait long enough, t ≫ 0, we shall see a linear velocity profile
u(y), which is sketched at the bottom of figure 2. While the situation in the middle
of figure 2 is time–dependent, the situation at the bottom of the figure is steady;
this is the so–called Couette flow. Transport of momentum, obviously, proceeds from

2
Fluid Mechanics and Heat Transport

Figure 2: Layer of fluid between two plates.

regions of high velocity to regions of low velocity (downhill), i.e. in our problem from
the top to the bottom plate.
Secondly, we consider the analogue thermal problem. Here, we place a solid plate
(or a fluid at rest) between two defined temperatures. The top of figure 3 gives the
initial situation at t = 0, with both the upper and bottom boundaries at temperature
T0 = T1 = T . Of course, this gives an isothermal solid at temperature T . Similar to
the flow problem in figure 2, we suddenly raise the temperature of the top boundary
to T0 = T + ∆T for t > 0. The solid adjacent to the top boundary will immediately
raise its temperature, while most of the solid layer will remain at temperature T . The
corresponding temperature profile T (y) across the solid layer is sketched in the middle

3
P. Ehrhard

Figure 3: Layer of solid (fluid at rest) between two defined temperatures.

of figure 3. If we wait long enough, t ≫ 0, we finally will see a linear (conductive)


temperature profile T (y) across the solid, as sketched at the bottom of figure 3. Again,
heat is transported from solid layer to solid layer by intermolecular forces – this is heat
conduction. High temperature simply corresponds to an intense Brownian motion of
the atoms/molecules, which will be transferred to the neighboring atoms/molecules.
The situation in the middle of figure 3 is time–dependent, while the situation at the
bottom of figure 3 is steady. Transport of heat, similarly, proceeds from regions of
high temperature to regions of low temperature, i.e. downhill. In our case, the heat
flux is from the top boundary to the bottom boundary of the solid.

Even though, we shall not treat mass or species transport to a larger extend here, it

4
Fluid Mechanics and Heat Transport

Figure 4: Carbon concentration for steel hardening from one side.

remains worth mentioning that molecular species transport behaves also analogously.
In figure 4 we give the time–dependent concentration profile ccarb (y) of carbon, ob-
tained in a steel plate for hardening from one side. Above the steel plate, through
coal a high carbon concentration c0 is established, defining the upper boundary con-
dition for the carbon concentration. As time proceeds, the carbon is diffusing into
the steel, establishing the given profile ccarb (y). For hardening, obviously, it appears
desirable to carbonize particularly regions of high wear, i.e. surface regions. The
inner regions of the steel, in contrast, should be kept at low carbon concentration.
This, after the thermal hardening process, results in a hard surface and a ductile core
of the steel product. Again, the diffusive species transport of carbon into the steel
appears perfectly analogue to the transport of heat into the solid in figure 3.
The analogy between momentum and heat transport by diffusion can be worked out
in more detail. In figure 2, obviously, a force is needed to move the top plate. For
t ≫ 0, this force can be expressed by
U
|F~ | = µA , (1.1)
d
with the plate area A, the thickness of the fluid layer d, and dynamic viscosity µ.
This equation (1.1) results from the shear stress in a Newtonian fluid, which can be
expressed as
∂u
τyx = −µ . (1.2)
∂y
In the given situation, we have a linear profile u(y), leading to du/dy = U/d. The first
index of the shear stress τyx indicates the plane y = const., the second the direction of
the shear stress, namely x. Multiplying the shear stress with the plate area A results
in the force.
The heat flux through the solid in figure 3, for t ≫ 0 can be expressed by
∆T
|Q̇| = λA , (1.3)
d

5
P. Ehrhard

with the thermal conductivity λ and the temperature difference ∆T . This equation
(1.3) results from Fourier’s law of conduction, namely from

∂T
q̇ = −λ . (1.4)
∂y

Again, for the linear profile T (y) we can infer dT /dy = ∆T /d, and we have to multiply
the specific heat flux q̇ (per unit area) with the area of the solid plate, to obtain the
given expression (1.3) for the heat flux Q̇.
Comparing equations (1.1) and (1.3), as well as equations (1.2) and (1.4), demon-
strates the analogy even clearer. Both momentum and heat transport occur in nega-
tive y–direction. τyx can be identified as specific flux of momentum, q̇ as specific flux
of heat. Both fluxes occur downhill, i.e. from regions of high velocity or tempera-
ture to regions of low velocity or temperature. Both molecular transport coefficients,
namely µ and λ, can be modified to obtain so–called diffusivities. These are the
kinematic viscosity ν and the thermal diffusivity κ, defined by
µ
ν= , (1.5)
ρ
λ
κ= . (1.6)
ρcp

Both of these diffusivities are of unit [m2 /s], while dynamic viscosity is of unit [P a s]
and heat conductivity is of unit [W/(m K)]. Hence, the ratio of both diffusivities,
namely
ν
Pr = , (1.7)
κ
is a dimensionless group, the so–called Prandtl number. The Prandtl number is a
pure fluid property, characterizing the ratio of diffusivities for momentum and heat.

1.3 Shear stress in fluids


Equation (1.2) gives the shear stress τyx of a (Newtonian) fluid as function of the
shear rate γ̇ = ∂u/∂y. Precisely this dependency τ = f (γ̇) is usually determined
by a viscosimeter. In figure 5 we give this diagram, whereas a Newtonian fluid,
according to equation (1.2) with µ = constant, is characterized by the straight solid
line through the origin. Dynamic viscosity is nothing else than the slope of this
straight line. Moreover, we have included two curves for shear–thinning and shear–
thickening fluids in form of dashed lines. A shear–thinning fluid has a high viscosity
(slope) at small shear rates γ̇, and develops a decreasing viscosity (slope) for increasing

6
Fluid Mechanics and Heat Transport

Figure 5: Shear stress as function of shear rate.

shear rate. Exactly the opposite is true for shear–thickening fluids. As last example,
a so–called Bingham fluid is incorporated into figure 5 in form of a dotted line. Such
a fluid behaves like an elastic solid as long as a small shear stress τyx < τyield is
present. In other words, in this range any shear stress deforms this fluid elastically,
and the fluid falls back to its initial shape as soon as the shear stress is removed.
For large shear stress, τyx > τyield , the relation between shear stress and shear rate is
linear, similar to a Newtonian fluid. To summarize, the diagram in figure 5 not only
allows to determine the (constant) viscosity µ, it moreover demonstrates whether
viscosity is a function of shear rate. Such fluids, with µ = f (γ̇), are called non–
Newtonian fluids. Typical examples of non–Newtonian fluids are (dense) suspensions
of particles in liquids, (dense) emulsions of non–miscible disperse liquids droplets
in continuous liquids, suspensions of polymers (in liquids), or polymer melts. In
other words, whenever we have a complex structure of a “fluid”, we can expect non–
Newtonian behavior. The whole variety of dependencies τ = f (γ̇) is also termed fluid
rheology.

1.4 Relation of viscosity and molecular structure

Up to now, we have assumed for all fluids, i.e. gases and liquids, that viscosity µ cap-
tures the molecular transport of momentum through the fluid. Of course, the molec-
ular structure of gases and liquids is vastly different. In gases the atoms/molecules

7
P. Ehrhard

Figure 6: Dynamic viscosity of gases and liquids as function of temperature.

are in free flight and undergo collisions from time to time. It is only during collisions,
when an interactions between atoms/molecules can occur, and when an exchange of
momentum can take place. E.g. for air (essentially Nitrogen) at standard conditions
(1 bar, 20 degC) we have a molecule size of d ∼ 10−10 m, a mean speed of the molecules
of c̄ ∼ 500m/s, and a mean free path of ¯l ∼ 10−7m. In other words, a molecule in
average has a free flight of ∼ 1000 diameters, before it collides with another molecule.
According to thermodynamics, the mean speed c̄ of the molecules and the mean free
path ¯l are functions of the variables of state. At high temperatures, the average
speed of the atoms/molecules is large, this statistically-distributed movement of the
atoms/molecules is called Brownian motion. With regard to temperature, e.g., we
have the dependencies

c̄ ∝ T , (1.8)
¯l ∝ √1 . (1.9)
T
This means that increasing temperature leads to more intense and more frequent
collisions. In other words, we can expect more transfer of momentum between the
atoms/molecules. This simply means that increasing temperature T should increase
viscosity µ. This conclusion can, indeed, be confirmed by measurements, qualitatively
sketched in figure 6.
If we consider liquids instead, the situation is different. The atoms/molecules in
liquids are in permanent contact and, hence, permanently interact. The molecular
forces in play are the so–called long–range van–der–Waals forces. These are forces,

8
Fluid Mechanics and Heat Transport

Figure 7: Van–der-Waals force between two (liquid) atoms/molecules.

which occur due to mutual polarization of the electron shells. The (mean) force Fr (in
the r–direction), between two atoms/molecules at distance r, is qualitatively plotted
in figure 7. The first atom/molecule sits at the origin of the diagram (r = 0), the
second is at a distance r. We recognize at large distance r an attractive force between
both atoms/molecules, whereas a repulsive force is present for small distance r. It is
the point F = 0, which defines the (mean) equilibrium distance r ∗ . The Brownian
motion is still present, allowing the liquid molecules to travel through the (dense)
package of molecules at statistically–distributed velocities. The higher temperature,
the more intense this motion will be. In other words, high temperature due to the
Brownian motion tends to loosen the binding between the atoms/molecules. Hence,
we expect less exchange of momentum for higher temperature. For liquids, therefore,
increasing temperature T should lead to decreasing viscosity µ. This likewise can be
confirmed in figure 6. As a final remark, we should note that molecular transport
of heat (by conduction) has the same molecular interactions as basis. Hence, heat
conductivity λ should have a similar behavior as viscosity (cf. figure 6).

9
Figure 8: Flow tube.

2 Balances at 1D flow tubes


We now concentrate on so–called flow tubes. Such a flow tube is sketched in figure 8.
A flow tube is simply a roughly–cylindrical portion of fluid, whereas the axis and the
shell of this flow tube coincides with stream lines. In other words, there is only flow
along the flow tube (in the s–direction), and not across its cylindrical shell. Note,
that a flow tube is simply of fluid, it does not have any walls, and should not be
confused with a technical tube. Along the flow tube, in the s–direction, the cross
section A, pressure p, density ρ, velocity c, and temperature T may vary. However,
within a cross section, at a given s, pressure p, density ρ, velocity c, and temperature
T can be assumed constant. Hence, inside the flow tube, the flow is inviscid, i.e. it is
free of viscous losses. Let us now explore the conservation of mass, momentum and
energy along such a flow tube.

2.1 Balance of mass


Obviously, as there is no flow across the shell, inside the flow tube the mass flux must
be conserved. In other words, considering the cross sections A1 and A2 , we have to
have
ṁ1 = ṁ2 . (2.1)
This can also be formulated as

ṁ = ρcA = constant . (2.2)

10
Fluid Mechanics and Heat Transport

This equation (2.2) provides a relation between ρ(s), c(s), and A(s). The latter,
namely the cross section A(s), usually is known. Equation (2.2) is termed (1D)
continuity equation.

2.2 Balance of momentum


For an inviscid flow under steady conditions, an integral of all forces acting tangen-
tially at the fluid, gives
Zp
c2 dp
+ + gz = constant . (2.3)
2 ρ
This equation (2.3) is usually termed (steady) Bernoulli equation. Even though this
equation represents an integrated momentum balance, it contains terms which can
be interpreted as mechanical parts of the energy balance, as we shall see below. E.g.
the first term can be interpreted as the (specific) kinetic energy of the fluid, and the
third term can be interpreted as (specific) potential energy of the fluid. Further, the
integral term can be interpreted as (specific) work due to compression. This term only
can be determined, if an equation of state is given. Let us discuss several examples.

• p = constant:
For constant pressure along the flow tube, the integral gives simply
Z2
dp
=0; (2.4)
ρ
1

• ρ = constant:
For constant density, the most simple equation of state, certainly valid for liq-
uids, we obtain
Z2 Z2
dp 1 p2 − p1
= dp = ; (2.5)
ρ ρ ρ
1 1

• T = constant:
To evaluate the integral for isothermal conditions, we need to make use of e.g.
the equation of state of an ideal gas, namely of
p
= RT , (2.6)
ρ

11
P. Ehrhard

to obtain
Z2 Z2
dp dρ ρ2
= RT = RT ln (2.7)
ρ ρ ρ1
1 1

Of course, given any other equation of state, as e.g. adiabatic or isentropic change of
state, the evaluation of this integral remains possible.
As equation (2.3) is derived for steady flow, it can also be augmented by an unsteady
term on the left side, which is
Zs
∂c
ds . (2.8)
∂t

2.3 Balance of energy


The balance of energy along the flow tube gives

c2
+ i + gz − q = constant . (2.9)
2
Here, i represents the (specific) enthalpy, which can be expressed by i = cp T , based on
absolute temperature T . The link between the (specific) enthalpy i and the (specific)
inner energy e is
p
i=e+ . (2.10)
ρ
In equation (2.9), the term q represents the heat per unit mass, added to the fluid in
the flow tube. Given the mass flux ṁ through the flow tube, q can be converted into
the total added heat flux Q̇ = ṁq.
Let us now apply the above balances (2.2, 2.3, 2.9) to a few typical problems. These
are, the flow out of a container, and a flow with heating.

2.4 Flow out of a container


We consider the container in figure 9, which is filled with liquid to a level z = h, and
has gas above the liquid. At first, the valve is closed. Let us consider, firstly, a steady
problem, which of course represents an approximation, since both the liquid level and
the pressure in the gas will change (slowly) in time. We place a flow tube (dashed)
from the liquid surface (1) to the exit of the pipe (2). One important question is,
whether this flow tube fulfills all assumptions, necessary for our balances. The flow
tube 1 → 2, in general, clearly does not fulfill these assumptions: While the flow
through most of the container will be approximately inviscid, the flow through the

12
Fluid Mechanics and Heat Transport

Figure 9: Flow out of container.

pipe will certainly have viscous losses. Only if we assume l2 ≪ l1 , we can neglect
the viscous losses in this pipe. Hence, with this additional assumption we obtain for
ρ = constant from (2.2, 2.3)

c1 A1 = c2 A2 , (2.11)
c21 p1 c2 p 2
+ + gz1 = 2 + + gz2 . (2.12)
2 ρ 2 ρ
With our choice of the z–coordinate (cf. figure 9), we can infer z1 = h and z2 = 0.
For the pressures, p1 represents the pressure in the gas above the liquid, and p2 = p∞
holds at the pipe exit. Hence, the equations (2.11, 2.12) simplify to

c1 A1 = c2 A2 , (2.13)
c21 p1 c2 p ∞
+ + gh = 2 + , (2.14)
2 ρ 2 ρ
which should allow to solve for c1 and c2 . These are two equations for two unknowns,
since A1 , A2 , p∞ , ρ, and h are all given. Also p1 must be given, or can at least be
determined from other equations, valid in the gas.
If we open the valve, we have immediately p1 = p∞ . We, moreover, can make use
of the fact A1 ≫ A2 , which via equation (2.13) gives c1 = c2 (A2 /A1 ), i.e. c1 ≪ c2 .

13
P. Ehrhard

Under these circumstances, the above equations (2.13, 2.14) simplify even more and
we obtain the approximation
c2
gh = 2 , (2.15)
2
which gives
p
c2 = 2gh , (2.16)

the so–called Torricelli formula. It is important to note, that this equation (2.16) only
holds due to all the above approximations, and is in no sense the general solution for
the flow out of a container. Another approximation can e.g. be obtained for p1 ≫ p∞ .
This is the outflow due to a pressure difference, in which gravitational effects can be
neglected.
As we already mentioned, strictly spoken the problem under investigation is time–
dependent. Let us, secondly, treat the problem in a transient manner. We still assume
the valve to be open, i.e. we have p1 = p2 = p∞ , and we still neglect viscous losses in
the pipe due to l2 ≪ l1 . Hence, we obtain for the time–dependent problem

c1 A1 = c2 A2 , (2.17)
Z1 Z2
c21 ∂c c22 ∂c
+ gh + ds = + ds . (2.18)
2 ∂t 2 ∂t
sref sref

If we make use of the continuity equation (2.17) to eliminate c1 , and moreover, choose
identical reference points (sref ) within both integrals, we obtain

2 Z2
c22 c2

A2 ∂c
+ gh = 2 + ds . (2.19)
2 A1 2 ∂t
1

The remaining integral, as in most cases, can only be estimated. For this, we assume
that inside the container we have c1 (t) everywhere, while within the short pipe we
have c2 (t). Hence, the integral can be estimated as

Z2 Zl1 lZ
1 +l2
∂c ∂c1 ∂c2 dc1 dc2
ds ≃ ds + ds ≃ l1 + l2 . (2.20)
∂t ∂t ∂t dt dt
1 0 l1

To summarize, the result is an ordinary differential equation (ODE) for c2 (t), namely
"  2 #
c22
    
dc2 l1 A2 A2
1+ l2 + 1− − gh = 0 . (2.21)
dt l2 A1 2 A1

14
Fluid Mechanics and Heat Transport

Figure 10: Flow with heating.

Here, still h(t) is present, which can be linked to c1 (t) via h(t) = h0 − c1 t. It is not
important to solve this equation now, but it appears worth to clarify that with (2.21)
we can even model the complete (inviscid) flow out of the container during most of
the time, based on simple 1D balances.

2.5 Flow with heat addition

Let us concentrate onto a problem, where heat is added to a steady and inviscid flow.
In figure 10 a flow tube of constant cross section A is sketched, whereas the flow tube
rises from left to right, i.e. from z = 0 to z = h. For simplicity, we assume a liquid
in the flow tube and, hence, have ρ = constant. For this situation, the balances of
mass, momentum, and energy yield

ρAc1 = ρAc2 , (2.22)


c21 p1 c2 p 2
+ = 2+ + gh , (2.23)
2 ρ 2 ρ
c21 c2
+ i1 + q = 2 + i2 + gh . (2.24)
2 2
From continuity (2.22), we infer directly c1 = c2 = c, while enthalpy can be expressed
by ii = cp Ti . Hence, we are left with
p1 p2
= + gh , (2.25)
ρ ρ
cp T1 + q = cp T2 + gh . (2.26)

15
P. Ehrhard

We may ask the question, which temperature we expect at the right, if the liquid
enters at the left at a given temperature T1 . This can be deducted from (2.26), with
the result
q gh
T2 = T1 + − . (2.27)
cp cp
This solution is not yet satisfying, as it appears useful to express the added heat via
q = Q̇/ṁ. Moreover, from the momentum balance (2.25), we can infer the hydrostatic
balance
p1 − p2 = ρgh . (2.28)
This physically means, that in this inviscid flow the potential energy builds up a
pressure difference p1 − p2 . In other words, the temperature rise appears independent
of the potential energy, and we simply have


T2 = T1 + . (2.29)
ρcp c A

16
Figure 11: Laminar viscous flow in a pipe.

3 Viscous losses and heat transfer in 1D models


Within the previous chapter, we derived balances for 1D flow tubes. In all these
cases, we assumed pressure, density, velocity, and temperature to be constant within
the cross section of the flow tube. Physically, this implies that we neglect viscous
losses and heat transport within this cross section. It is, indeed, the velocity gradient
and the temperature gradient in a cross section, which causes viscous losses and heat
transport. As we shall see in this chapter, such effects can still be captured by 1D
models.

3.1 Pressure loss in pipes


Let us, firstly, discuss the viscous losses which a fluid experiences within a laminar
pipe flow. Figure 11 gives a sketch of the problem, with a circular pipe of diameter
D (or radius R). As the fluid enters the pipe from the left, it has just at the inlet
a constant velocity within most of the cross section. Just immediately at the wall,
the velocity drops to zero due to the no–slip condition. It takes some length, the
so–called developing length Ldev , before the fluid develops a parabolic velocity profile
u(r). Once this parabolic velocity profile has developed, it is maintained further
downstream. This is called a fully–developed flow.
It is for the fully–developed flow, where we can express the pressure loss based on
the analytical solution. Physically, a viscous flow can only be obtained if the pressure
is decreasing in the flow direction. For our pipe of length L, starting at x = 0 (cf.
figure 11), we shall have a pressure loss of
ρ L
p(x = 0) − p(x = L) = ∆p = u2m λlam . (3.1)
2 D

17
P. Ehrhard

Figure 12: Turbulent (viscous) flow in a pipe.

Here, um is the (1D) cross–sectionally averaged mean velocity, and λlam the so–called
laminar friction coefficient. For the laminar parabolic velocity profile u(r), the friction
coefficient can be calculated with the result
64 um D
λlam = , Re = . (3.2)
Re ν
In equation (3.2) the Reynolds number is defined based on um . By introducing
equation (3.2) into equation (3.1), we can infer the dependencies

∆p ∝ L1 ,
∆p ∝ u1m , (3.3)
∆p ∝ D −2 .

As in all equations we only make use of the mean velocity um , we can maintain the
1D character of our model even in the presence of viscous losses. In other words,
we do not engage fully into the true profile u(r), but instead project the effects of
the velocity gradients into the friction coefficient λlam . It should be noted that a
developing flow, starting at the pipe entrance, causes an additional pressure loss.
Further, for laminar flow the influence of wall roughness appears negligible.
If the flow for Re > 2300 is no longer laminar, we still can use the ansatz (3.1) for
the turbulent pipe flow, i.e. we have
ρ L ūm D
∆p̄ = ū2m λtur , Re = . (3.4)
2 D ν
Here, we simply have replaced the mean velocity um in the cross section and the
pressure drop ∆p by their time–averaged versions ūm and ∆p̄. The time average
appears necessary, as the turbulent flow is strongly time–dependent. Similarly, the

18
Fluid Mechanics and Heat Transport

profiles ū(r) in figure 12 are also time–averaged. Without the time average, each
instantaneous profile u(r, ti ) would be greatly different at each time ti . It should be
noted in figure 12, that the velocity profiles ū(r) are not parabolic as in the laminar
case. Instead, the turbulent profiles are flat in the pipe centre with steep gradients at
the walls. The turbulent friction coefficient λtur cannot be determined analytically.
Instead, it is inferred empirically, based on a large number of experimental data. For
smooth pipe walls, we find
0.3164
λtur = for 2300 < Re < 105 , (3.5)
Re1/4
1 p
√ = 2 log (Re λtur ) − 0.8 for 105 < Re < 3 · 106 . (3.6)
λtur
The first equation (3.5) is called Blasius correlation, it is valid for moderate Reynolds
numbers. The second equation (3.6), the so–called Prandtl correlation, happens to
be implicit and holds for larger Reynolds numbers. Hence, equation (3.6) has to be
used iteratively, to determine λtur for a given Reynolds number. Both correlations
(3.5, 3.6) have the simple dependency λtur = f (Re), as they hold for smooth pipe
walls. For the Blasius correlation (3.5) we can explicitly infer the dependencies

∆p ∝ L1 ,
∆p ∝ ū7/4
m , (3.7)
−5/4
∆p ∝ D .

If the wall is rough, as in almost every technical pipe, the wall roughness ks will have
a considerable influence onto the pressure loss. We can for this situation e.g. make
use of the so–called Colebrook correlation, which is likewise implicit and reads as
 
1 2.51 ks
√ = −2 log √ + . (3.8)
λtur Re λtur 3.71D
Here, we clearly have the dependency λtur = f (Re, ks /D). Note that in all cases (3.2,
3.5, 3.6, 3.8) the friction coefficient λ, as well as the parameters Re and ks /D appear
to be dimensionless. Moreover, in equations (3.1, 3.4) also L/D is dimensionless and
it are the expressions ρu2m /2 and ρū2m /2, which provide the dimension of pressure to
these equations.
Let us, finally, summarize the results for λlam and λtur in figure 13. We do this in form
of the so–called Nikuradse diagram, which plots both Re and λ on logarithmic axes.
This has the advantage, that any power–law function appears as straight line, with
the slope proportional to the exponent of the power law. We recognize in the laminar

19
P. Ehrhard

Figure 13: Nikuradse diagram.

regime (Re < 2300) the solid (and later dashed) curve of dependency λlam ∝ Re−1 ,
which consistently gives a negative slope when plotted on logarithmic axes. The
solid (and later dashed) Blasius curve in the turbulent regime shows the dependency
λtur ∝ Re−1/4 . As the Reynolds number increases further, we recognize the solid
Prandtl curve, smoothly departing from the Blasius curve around Re ≃ 105 . The
Prandtl correlation has a weaker dependency on Re, as obvious from the smaller
(negative) slope in figure 13. The curves obtained for increasing roughness ks /D,
finally, are plotted as dotted lines. These curves are obtained from the Colebrook
correlation (3.8) for increasing values ks /D. It appears that for rough walls with
increasing Reynolds number, the friction coefficient does no longer depend on Re.
This can also be recognized from the Colebrook correlation (3.8), where the first
term in the log function disappears in the limit Re → ∞.

3.2 Heat transfer at pipe walls

Let us, secondly, discuss the problem of heat transfer at the pipe wall. This problem
is closely related to the temperature profile T (r) in the pipe cross section. At the wall,
the fluid sticks to the solid, such that there is no convective transport. If we neglect
radiation in our considerations, immediately at the wall only diffusion is available for

20
Fluid Mechanics and Heat Transport

Figure 14: Temperature profiles in a pipe of (high) constant wall temperature.

heat transport within the fluid. Hence, we can make use of the Fourier law to express
the radial heat flux by
∂T
q̇w = −λ . (3.9)
∂r r=R
Of course, through the solid wall, heat is transported by conduction – this question
will not be addressed here.
In figure 14 the situation is sketched for a given wall temperature Tw > T∞ . We have
the fluid entering the pipe from the left with temperature T∞ . Immediately after
entering the pipe, the fluid is still at T = T∞ in most of the cross section. It is just
the immediate wall vicinity, which has acquired a higher temperature. As obvious
from the most left temperature profile T (r) in figure 14, the temperature gradient
at the wall is very large, which due to equation (3.9) corresponds to a large wall
heat flux q̇w . When the fluid moves downstream along the pipe (for increasing x),
more and more fluid within the cross section heats up, such that the temperature
profile will develop roughly parabolic, with T (r = R) = Tw at the wall and a lower
temperature T (r = 0) > T∞ in the pipe center. Far enough downstream, the entire
cross section will eventually be heated up to a temperature T (r) → Tw . The most
right temperature profile in figure 14 already indicates, that far downstream the
temperature gradient at the wall gets very small, implying a very small wall heat flux
q̇w . For x → ∞, we can even expect q̇w → 0.
The above considerations, based on the temperature profiles T (x, r), are clearly 2D in
nature. It remains, however, desirable to capture the heat transfer by means of a 1D
model. Hence, we introduce a mean temperature Tm (x), based on the cross–sectional
average. The wall heat flux then can be expressed by the simple heat transfer equation

q̇w (x) = −α(x) (Tw − Tm (x)) . (3.10)

Here, α denotes the heat transfer coefficient and q̇w is positive, if it is directed in the

21
P. Ehrhard

positive r–direction. As we can infer from figure 14, all quantities in equation (3.10)
may depend on the axial position x, except for the wall temperature Tw . The ansatz
(3.10) purely relies on the (1D) mean temperature Tm , and all effects caused by the
radial temperature gradient are projected into the heat transfer coefficient α.
As we tend to carry the further discussion in dimensionless form, it appears convenient
to introduce the so–called Nusselt number as
α(x)D
Nu(x) = . (3.11)
λ
Here, λ denotes the heat conductivity of the fluid, and D the diameter of the pipe.
For a fully–developed laminar flow in the pipe, the Nusselt number can be given
empirically, based on a large number of experimental data. These correlations are
 
D
Nu = 3.66 for small values ReP r , (3.12)
x
 1/3  
D D
Nu(x) = 1.077 ReP r for large values ReP r , (3.13)
x x
Within these equations (3.12, 3.13), the product Re P r = P e can be identified as
Péclet number. The Prandtl number P r = ν/κ is a pure fluid property, featuring the
ratio of kinematic viscosity ν and thermal diffusivity κ. While the first correlation
(3.12) does not show a dependency on Péclet number, the second correlation (3.13)
gives Nu ∝ P e1/3 . From the second correlation (3.13) we also recognize a decrease of
the Nusselt number along the pipe, following the dependency Nu ∝ x−1/3 . This result
corresponds to our observations in figure 14, where we already expected a decrease
of the wall heat flux q̇w as we move downstream along the pipe. It often appears
convenient, to work with a heat transfer which is averaged along the pipe length (in
x). For this purpose, we introduce a length–averaged (mean) heat transfer coefficient
αm and a length–averaged (mean) Nusselt number Num by
ZL
1
αm = α(x)dx , (3.14)
L
0

ZL
1
Num = Nu(x)dx . (3.15)
L
0
Essentially, this likewise means that we engage a length–averaged version of the heat
transfer equation (3.10), namely

q̇w,m = −αm (Tw − Tm,m ) , (3.16)

22
Fluid Mechanics and Heat Transport

with the length–averaged (mean) heat flux q̇w,m and the length–averaged (mean) fluid
temperature Tm,m . The correlations (3.12, 3.13), in length–averaged form, are

Num = 3.66 , (3.17)

 1/3
D
Num = 1.615 ReP r . (3.18)
L
As all the above correlations hold for laminar flows (Re < 2300), it is of interest to
have also correlations for turbulent flows. Indeed, in literature we find a correlation,
valid for turbulent flows in the regime Re > 104 , namely
"  2/3 #
(ξ/8)ReP r 1 D
Nu(x) = p 1+ , (3.19)
1 + 12.7 (ξ/8)(P r 2/3 − 1) 3 x

with
ξ = (1.8 log (Re) − 1.5)−2 . (3.20)
This correlation can also be averaged along the pipe, with the result
"  2/3 #
(ξ/8)ReP r D
Num = p 1+ . (3.21)
1 + 12.7 (ξ/8)(P r 2/3 − 1) L

Up to now, we have discussed the case of a given wall temperature. This is one of the
boundary conditions, where mathematically a solution can be inferred. Practically, it
appears difficult to maintain a constant wall temperature. This is the case, because
any heat flux will lead to a (small) change of the wall temperature. Hence, a constant
wall temperature e.g. would require a poorly–conducting fluid in the pipe and a
perfectly–conducting pipe wall. In other words, the assumption of an isothermal wall
is an idealization, which is rarely matched in reality. A second idealization is to assume
that a constant wall heat flux q̇w is present along the pipe. Mathematically, this is also
a boundary condition where a solution can be inferred. Again, this idealization can
be realize in practice to some degree, e.g. by wrapping an electrical heater around
the pipe and insulate the whole wrapped pipe thermally against the environment.
Nevertheless, these idealizations are useful and the true situation in practise is usually
a mixture of these two extremes.
Let us now discuss the situation, present if the wall heat flux is prescribed. Figure 15
gives a sketch of this situation. Again, the fluid enters the pipe from the left at an
inlet temperature of T∞ . Immediately after the entrance, a constant temperature
gradient ∂T /∂r is established at the wall, which is indicated in the figure as dashed

23
P. Ehrhard

Figure 15: Temperature profiles in a pipe of constant wall heat flux.

line. Following equation (3.9), this temperature gradient is directly proportional to


the given wall heat flux q̇w . Otherwise, the temperature profile near the entrance
remains unaffected at T = T∞ over the most part of the inner cross section. If we
move downstream, we recognize an elevated temperature profile, which still features
the identical temperature gradient at the wall. However, the wall temperature has
increased, as has the temperature of the fluid in the entire cross section. Obviously,
due to the given wall heat flux, we permanently have added heat to the fluid along the
pipe, leading to an increase of temperature in the entire cross section. This trend, by
the way, continues if we move even further downstream: the temperature gradient at
the wall remains identical and the wall temperature and the mean fluid temperature
increase further. Still, the heat transfer equation (3.10) can be applied to obtain

q̇w = −α(x) (Tw (x) − Tm (x)) . (3.22)


As we have prescribed the wall heat flux q̇w here, the equation essentially defines the
wall temperature Tw (x) along the pipe. This is clearly different to the situation with
a given wall temperature, where equation (3.10) essentially defines the wall heat flux
q̇w (x) along the pipe.
In literature, we find empirical correlations for the Nusselt number also for a given
wall heat flux. These are valid for a fully–developed laminar flow and read as
 
D
Nu = 4.364 for small values ReP r , (3.23)
x
 1/3  
D D
Nu(x) = 1.302 ReP r for large values ReP r . (3.24)
x x
The averaged versions along the pipe are

Num = 4.364 , (3.25)

24
Fluid Mechanics and Heat Transport

Figure 16: Nusselt number as function of Péclet number (qualitatively).

 1/3
D
Num = 1.953 ReP r . (3.26)
L
For turbulent flow with Re > 104 , the correlations for the local Nusselt number (3.19)
and for the mean Nusselt number (3.21) are both valid also for a given wall heat flux.

Let us summarize the dependencies, obtained for the Nusselt number. In figure 16
the dependency Nu = f (P e) is sketched qualitatively, while we engage logarithmic
axes both for the P e– and for the Nu–axis. We recognize in the lower laminar regime
a constant Nusselt number, i.e. Nu ∝ P e0 , for both constant wall temperature and
constant wall heat flux. In the upper laminar regime, we have for both cases the
dependency Nu ∝ P e1/3 . Strictly, the curves in the laminar regime for constant
wall temperature and for constant wall heat flux are not identical: the curves for
constant wall heat flux are slightly above the curves for constant wall temperature.
Nevertheless, both curves are parallel, indicating a slightly–different constant in the
correlations Nu = f (Re). The dotted and solid curves can be viewed as asymptotes
for small and large values of the Péclet number in the laminar regime. In the turbulent
regime, there is only a single curve for both cases, which follows the dependency
Nu ∝ P e1 . As the abscissa is the Péclet number, the boundary between laminar and
turbulent flow cannot be given without knowing the Prandtl number of the fluid. E.g.
the laminar/turbulent transition will be at P e ≃ 2300 P r, with the Prandtl number

25
P. Ehrhard

in the Range 10−2 < P r < 102 , depending on the fluid. This is why the abscissa has
no scale.

26
Figure 17: Arbitrary control volume.

4 Integral balances
It appears often useful, to apply integral balances to arbitrary control volumes. In
steady situations, it is usually sufficient to know the flow quantities on the surface of
the control volume, rather than within the entire flow field. Hence, valuable informa-
tion as integral mass fluxes, integral forces (due to momentum), or integral energy
fluxes can be inferred. This information, in many cases, answers typical engineering
questions. Let us, therefore, formulate these integral balances for mass, momentum,
and heat, and let us apply these balances to several example problems.

4.1 Integral mass balance

In figure 17 we have sketched the situation at an arbitrary control volume V , which


has a surface A. We assume a flow going through this control volume, indicated in
figure 17 by (instantaneous) streamlines and by two velocity vectors at two exemplary
points on the surface. An integral mass balance at the control volume gives

∂ρ
Z Z
dV + ρ(w ~ · ~n)dA + ḣ = 0 . (4.1)
∂t
V A

The first term in equation (4.1) represents the temporal change of mass inside the
control volume V . A change of mass inside a defined control volume is only possible,
if density ρ varies in time. The second term, by the surface integral balances the flow
entering/leaving the control volume across the surface A. Here, the scalar product
(w~ · ~n) extracts that part of the velocity vector w,
~ which is directed orthogonally to
the surface; ~n is the outward–directed unit normal vector, which is also sketched in

27
P. Ehrhard

figure 17. The last term ḣ, finally, represents an eventual source of mass inside the
control volume. The dimension of all terms in equation (4.1) is [kg/s].
If we assume an incompressible flow, as is the case for e.g. liquids, we have ρ = constant.
Hence, equation (4.1) simplifies to
Z
ρ (w ~ · ~n)dA + ḣ = 0 . (4.2)
A

A similar simplification applies if a steady compressible flow is present, i.e. we have


ρ = ρ(x, y, z) and ∂ρ/∂t = 0. In this case the volume integral in equation (4.1) also
disappears, while the area integral remains as it is in equation (4.1).

4.2 Integral force balance

The conservation of momentum can be formulated in terms of a force balance at the


control volume. Each flow through the surface A causes a so–called momentum force,
which has to be in balance with all other forces, acting at the control volume. Hence,
we have Z
~
X
Fi − ρw( ~ w~ · ~n)dA = 0 . (4.3)
i A

The first term in equation (4.3) reflects the sum of all acting forces. This sum includes
volumetric forces, as e.g. gravitational forces, etc., and it includes surface forces, as
e.g. pressure forces. Moreover, forces acting from the outside at the control volume,
e.g. from the structure, have to be accounted for. The second term is the net
momentum force, due to the flow across the surface A. Once more, the scalar product
(w
~ · ~n) simply couples out the surface–normal component of velocity. It should be
noted, that equation (4.3) is a vector equation, which can be evaluated in one or more
components; the dimension of all terms is [N].

4.3 Integral energy balance

The integral balance of the total energy has to account for all forms of energy, namely
for kinetic energy, for inner energy, and potential energy. Hence, the most–general
balance reads as
  2  Z  2 
∂ w w
Z
ρ + e + gz dV + ρ + i + gz (w~ · ~n) dA − Q̇ = 0 . (4.4)
∂t 2 2
V A

28
Fluid Mechanics and Heat Transport

Here, e denotes the specific inner energy (per unit mass) and i the specific enthalpy
(per unit mass). Q̇ is the energy, added/removed to/from the control volume per
unit time, due to heat sources, chemical reactions, dissipation, and other molecular
processes. Hence, Q̇ is a flux of energy, the dimension of all terms in (4.4) is [W ].
The added work (per unit time) due to pressure forces is not included in Q̇, instead
it is accounted for in the surface integral of enthalpy i = e + (p/ρ).
From the total energy balance (4.4), we can split off the thermal contributions. For-
mally, this can be done by subtracting the mechanical contributions, which can read-
ily be inferred from an integrated momentum balance. Hence, we obtain the thermal
energy balance

Z Z
(ρe) dV + ρi (w~ · ~n) dA − Q̇ = 0 . (4.5)
∂t
V A

This equation can be simplified even more, if we consider a steady situation, a constant
density ρ, and assume thermodynamic equilibrium and constant fluid properties λ and
cp . For such conditions, we can express the conductive contribution within Q̇ by the
Fourier law to obtain
Z Z
ρcp T (w ~ · ~n) dA − λ (∇T · ~n) dA − Q̇ext = 0 . (4.6)
A A

Consequently, Q̇ext now simply presents the heat flux added from the external to the
control volume.

4.4 Example: flow through a bend

We now apply the above integral balances to a typical engineering problem. In fig-
ure 18 we recognize a bend of 90 degrees, mounted at the end of a pipe system,
which is supposed to release a fluid in a defined direction to the ambient. Typical
questions in such a situation might be associated with the force (or torque), which
such a bend adds onto the pipe system. In figure 18 we have already chosen a control
volume, which is given by the dashed line. There are several options for the choice
of a control volume, but the present choice is dictated by the following ideas: (i)
The control–volume surface follows the contour of the bend, such that there is no
flow across the surfaces A3 and A4 . (ii) The control–volume surface is outside of
the bend wall, meaning that the pressure on the surfaces A3 and A4 is known, i.e.
p3 = p4 = p∞ . (iii) The control volume cuts through the connection of the bend with
the structure above. Hence, forces from the structure above onto the bend have to be

29
P. Ehrhard

Figure 18: Flow through a bend, mounted to another pipe.

considered. (iv) There is flow normally through the surfaces A1 and A2 , the pressure
at the open end is p2 = p∞ . We assume a steady and incompressible flow, i.e. we
have ρ1 = ρ2 = ρ, moreover, velocity can be assumed constant in each cross section.
If we explore the mass balance (4.1), firstly, we realize that there is obviously no
mass source/sink inside the control volume, i.e. ḣ = 0. Together with the above
assumptions, we have Z
ρ (w
~ · ~n)dA = 0 . (4.7)
A
At the surfaces A3 and A4 , we have the velocity vector at each place normal to the
normal vector. Hence, at the surfaces A3 and A4 the scalar product will disappear,
i.e. w
~ 3 · ~n3 = 0, w
~ 4 · ~n4 = 0. Contrary, at the surfaces A1 and A2 , we do have
non–vanishing scalar products, namely
! !
0 0
w~ 1 · ~n1 = · = −w1 , (4.8)
−w1 1
! !
w2 1
w~ 2 · ~n2 = · = w2 . (4.9)
0 0
Hence, we can evaluate the integral in (4.7) and obtain
Z Z Z
ρ (w ~ · ~n)dA = ρ (w ~ 1 · ~n1 )dA + ρ (w
~ 2 · ~n2 )dA , (4.10)
A A1 A2

30
Fluid Mechanics and Heat Transport

with the contributions from A3 and A4 identical to zero. Hence, the result is
Z
ρ (w~ · ~n)dA = ρ (−w1 A1 + w2 A2 ) . (4.11)
A

The first contribution of the integral is due to the mass flux ṁ1 < 0, entering the
control volume through A1 . The second contribution is due to the mass flux ṁ2 > 0
leaving the control volume through A2 . Hence, the sign convention is that a mass
flux is positive, when it leaves the control volume.
For the momentum balance (4.3), secondly, we have to sum all acting forces on the
left side. These are the pressure forces F~pj on all surfaces Aj of the control volume,
and the (bolt) force F~b from the structure above onto the bend. Hence, we have
X
F~i = F~p1 + F~p2 + F~p3 + F~p4 + F~b . (4.12)
i

In general, pressure (or surface) forces are antiparallel to the normal vector and can
be expressed by the integral Z
F~p = − p~ndA . (4.13)
A
The pressure forces in (4.12) are, therefore,
X Z Z Z Z
~
Fpj = − p1~n1 dA − p∞~n2 dA − p∞~n3 dA − p∞~n4 dA . (4.14)
j A1 A2 A3 A4

While the integrals across the surfaces A1 and A2 can be calculated due to constant
pressure and constant normal vector, the integrals across the surfaces A3 and A4
appear difficult: even though the pressure is constant, the normal vectors are not
constant and difficult to compute. However, if we split up the first integral across A1 ,
we can modify (4.14) to the form
Z
~
X
Fpj = − (p1 − p∞ )~n1 dA
j A1
 
Z Z Z Z
−  p∞~n1 dA + p∞~n2 dA + p∞~n3 dA + p∞~n4 dA , (4.15)
A1 A2 A3 A4

which allows to identify a closed integral around the entire control volume at constant
pressure p∞ (in square brackets). This closed integral must be identical to zero. The
result for the sum of pressure forces is
!
0
X Z
F~pj = − (p1 − p∞ )~n1 dA = . (4.16)
j
−(p1 − p∞ )A1
A1

31
P. Ehrhard

Figure 19: Forces acting at the bend.

We collect the sum of all acting forces, and obtain


!
Fbx
F~i =
X
, (4.17)
i
Fby − (p1 − p∞ )A1

with the components Fbx and Fby of the bolt force. Next, we have to determine the
momentum forces within (4.3). For the mass balance, we have already evaluated the
dot products w
~ i · ~ni , and using these results we obtain
Z Z Z
ρw(
~ w ~ · ~n)dA = ρw~ 1 (−w1 )dA + ρw
~ 2 (w2 )dA ,
A A1 A2
! !
0 ρw22 A2
= + . (4.18)
ρw12 A1 0

The complete forces balance, finally, reads as


! ! !
Fbx 0 ρw22 A2
− − =0, (4.19)
Fby − (p1 − p∞ )A1 ρw12 A1 0

which allows to solve for both components of the (unknown) bolt force to obtain

Fbx = ρw22 A2 , (4.20)


Fby = ρw12 A1 + (p1 − p∞ )A1 . (4.21)

It appears worthwhile to discuss all forces in equation (4.19) in terms of their direction.
This is graphically illustrated in figure 19. At the surface A1 , we have two forces
acting, namely a pressure force F~p1 and a momentum force F~j1 , both directed in the

32
Fluid Mechanics and Heat Transport

negative y–direction. At the surface A2 , no (net) pressure forces is present, since we


have collected all pressure forces due to p∞ to form a closed integral. However, a
momentum force F~j2 is present, acting in the negative x–direction. F~b , finally, closes
the force triangle to fulfill the respective force balance.
It should be clear, that the integral momentum balance, in general, applies to both
inviscid and viscous flows. In our example, we have assumed constant velocities in all
cross sections. This corresponds to an inviscid flow. If the velocity profiles, instead,
would reflect a viscous flow with losses, the results would be perfectly–valid also for
this viscous flow. Losses in viscous flows are e.g. incorporated in laminar or turbulent
velocity profiles, or in non–symmetric or even detaching velocity profiles behind the
bend.
Let us, thirdly, explore the thermal energy balance for the bend. As sketched in figure
18, an external heat flux Q̇ext is entering the bend. The thermal energy balance (4.6)
then reads as
Z Z
ρcp T (w ~ · ~n) dA − λ (∇T · ~n) dA − Q̇ext = 0 . (4.22)
A A

The first integral in (4.22) reflects the heat transport by convection, the second in-
tegral reflects the heat transport by conduction, both through the surface into the
control volume. As we have already elaborated on the dot products for the mass
balance, we can make use of these results and obtain for the convective heat flux
Z h i
ρcp T (w ~ · ~n) dA = ρcp T1 (−w1 )A1 + T2 (w2 )A2 . (4.23)
A

The conductive fluxes can likewise be expressed surface by surface, to obtain


" ! ! ! !
∂T1 /∂x 0 ∂T2 /∂x 1
Z
λ (∇T · ~n) dA = λ · A1 + · A2
∂T1 /∂y 1 ∂T2 /∂y 0
A
Z Z #
+ ∇T3 · ~n3 dA + ∇T4 · ~n4 dA . (4.24)
A3 A4

We may assume for simplicity, that the surfaces A3 and A4 are thermally insulated.
For these conditions, the gradients of T3 and T4 will not have a considerable contri-
bution in the normal direction. Hence, the dot products ∇T3 · ~n3 and ∇T4 · ~n4 will
be negligible. The evaluation of the first two terms in (4.24) then leads to
 
∂T1 ∂T2
Z
λ (∇T · ~n) dA = λ A1 + A2 , (4.25)
∂y ∂x
A

33
P. Ehrhard

meaning that there are two conductive fluxes through the surfaces A1 and A2 only.
To summarize, the complete thermal energy balance (4.22) then is
 
  ∂T1 ∂T2
ρcp − w1 T1 A1 + w2 T2 A2 − λ A1 + A2 − Q̇ext = 0 . (4.26)
∂y ∂x

Typically, at least for considerable flow through the bend, we can expect

∂T1
|ρcp w1 T1 | ≫ λ , (4.27)
∂y

∂T2
|ρcp w2 T2 | ≫ λ
. (4.28)
∂x

This physically means, that the convective heat flux through both surfaces A1 and
A2 is much larger than the conductive heat flux through the respective surface. The
ratio of convective and conductive heat flux is usually linked to the so–called Péclet
number, which is defined by
wl convective heat flux
Pe = ∝ , (4.29)
κ conductive heat flux
with the thermal diffusivity κ = λ/(ρcp ), and a velocity and a length scale w, l.
The velocity and length scale can be taken from either surface A1 or A2 . Obviously,
the velocity scale is the respective velocity wi , and the length scale is the respective
diameter of the cross section. The conditions (4.27, 4.28) are fulfilled for P ei ≫ 1
in the respective cross section. Hence, by neglecting the conductive heat fluxes in
(4.26), we obtain
ρcp w1 A1 (T2 − T1 ) − Q̇ext = 0 , (4.30)
where we have already made use of the mass balance w1 A1 = w2 A2 . Equation (4.30)
allows to express the temperature rise due to the external heat flux as

Q̇ext
(T2 − T1 ) = . (4.31)
ρcp w1 A1

34
5 Modes of heat transport
There are three modes of heat transport, namely conduction, convection, and ra-
diation. (i) Conduction is a diffusive transport of heat, which essentially transfers
the Brownian motion of atoms/molecules to neighbouring atoms/molecules by means
of molecular interactions. Hence, for conduction matter appears necessary, whereas
matter may be in a solid, a liquid, or a gaseous state. (ii) Convection is a transport
of heat, associated with the (large–scale) movement of matter. The Brownian motion
of atoms/molecules is transported to another location by a macroscopic displacement
of these atoms/molecules. Hence, for convection macroscopically–moving matter, in
form of solid, liquid, or gas is needed. (iii) Radiation is a transport of heat by means
of an electromagnetic wave. An amount of matter, due to its (intense) Brownian
motion, emits an electromagnetic wave, which then proceeds even through vacuum,
before it is absorbed by another amount of matter. Emission and absorption are
associated with matter; solids, liquids, and gases can both emit and absorb such elec-
tromagnetic waves. The transport of the wave, though, does not need any matter. If
matter is present along the path of the electromagnetic wave, as e.g. liquid or gas,
the electromagnetic wave may be weakened, as part of its energy is absorbed by this
matter.

5.1 Conduction
The general heat transport equation for conduction can be given as
∂T
ρcp = λ∆T + ρq̇ . (5.1)
∂t
Here, ρ denotes density, cp specific heat, and λ heat conductivity of the matter, i.e. of
the solid, liquid, or gas. The term q̇ represents the specific heat flux (per unit mass),
added internally to the flow. The dimension of all terms in equation (5.1) is [W/m3 ].
We shall now focus on two simple examples of heat conduction.

5.1.1 Example 1: Conduction in a wire/rod

In figure 20 we have sketched the problem. A (solid) wire/rod of length L is subjected


to a heat flux q̇1 (per unit area) at its left end (x = 0), and cooled to a temperature
T∞ at its right end (x = L). We shall consider the problem as one–dimensional
and steady. The one–dimensional approximation appears reasonable, as long as the
diameter of the wire/rod is much smaller than its length. The steady assumption

35
P. Ehrhard

Figure 20: Conduction in a wire/rod.

excludes initial transients, which may occur during heating up of the wire/rod after
the heat flux is switched on. If we inspect equation (5.1), we have, obviously, to make
use of its steady and one–dimensional version. Moreover, within the computational
domain 0 < x < L we have no heat sources q̇. Note, that q̇1 defines a boundary
condition at x = 0, as does temperature T∞ at x = L. Hence, as basis we have the
ordinary differential equation
d2 T
0=λ 2 . (5.2)
dx
Obviously, λ 6= 0, such that integration in x gives

dT
= C1 , T = C1 x + C2 . (5.3)
dx
This general solution for T (x) clarifies, that we can expect a linear temperature profile
along the wire/rod. The two constants of integration have to be determined from the
boundary conditions. We have two boundary conditions, which should be sufficient
to determine these two constants. At the left end of the wire/rod, we have the defined
heat flux q̇1 , which can be linked to temperature by using the Fourier law

dT
q̇1 = −λ . (5.4)
dx x=0

Hence, the defined heat flux at the left end of the wire/rod provides a boundary
condition for the gradient of temperature, namely

dT q̇1
=− . (5.5)
dx x=0
λ

The second boundary condition is

T (x = L) = T∞ . (5.6)

36
Fluid Mechanics and Heat Transport

From equation (5.3) with boundary condition (5.5), we infer

q̇1
C1 = − , (5.7)
λ
and from equation (5.3) with boundary condition (5.6), we obtain

q̇1
C2 = T∞ + L. (5.8)
λ
Hence, the solution is
q̇1
T (x) = T∞ + (L − x) . (5.9)
λ
This temperature profile is likewise sketched in figure 20. It is interesting to note,
that the heat flux q̇1 is present at each place x of the wire/rod.

5.1.2 Example 2: Conduction through two plates in contact

In figure 21, the second exemplary problem is sketched. We have two stacked plates
in (perfect) thermal contact, which are subjected to a constant heat flux q̇0 at the
top plane (z = 0), and experience a constant temperature T∞ at the bottom plane
(z = −L1 − L2 ). Both solid plates have, in general, an arbitrary heat conductivity λi ,
and are infinitely extended in both the x– and the y–direction. Once more, we assume
the problem to be steady. We further recognize, that there is no heat source/sink
within the computational domain −L1 − L2 < z < 0; q̇0 is entering at the boundary
at z = 0 and, therefore, defines a boundary condition. Given these assumptions, the
governing equation can be inferred from (5.1) for both plates as
 2
∂ Ti ∂ 2 Ti ∂ 2 Ti

0 = λi + + . (5.10)
∂x2 ∂y 2 ∂z 2

As both boundary conditions are not dependent on x and y, and since end effects can
be neglected due to the infinitely–extended plates, we cannot expect a dependency
of the temperature field on x and y. With T = T (z), equation (5.10) simplifies for
λ1 , λ2 6= 0 to
d2 Ti
=0, (5.11)
dz 2
valid in both solid plates. This happens to be a simple one–dimensional ordinary
differential equation, similar to the previous example (cf. equation (5.2)). The two
different solid plates are treated in a straight–forward fashion, since we simply have
to apply (5.11) in two different regions in z. From the previous example, we know

37
P. Ehrhard

Figure 21: Conduction through two plates in contact.

how to integrate equation (5.11) and how to infer the general solutions in both plates.
The result from this integration is
dT1
= C3 , T1 = C3 z + C4 , (5.12)
dz
dT2
= C5 , T2 = C5 z + C6 . (5.13)
dz
We now have to formulate boundary conditions, to determine the four constants of
integration. Two boundary conditions appear analogous to the previous example,
namely

dT1 q̇0
= , (5.14)
dz z=0 λ1

T2 (z = −L1 − L2 ) = T∞ . (5.15)

There are, obviously, two additional boundary conditions needed, to determine all
four constants. For additional conditions, we explore that both temperature and
heat flux have to be continuous at the contact plane (z = −L1 ). Hence, we have the
conditions

T1 (z = −L1 ) = T2 (z = −L1 ) , (5.16)



dT1 dT2
λ1 = λ2 . (5.17)
dz z=−L1 dz z=−L1

From equation (5.17) we can infer, that the heat flux q̇0 is not only present at the top
plane (z = 0), but also at the contact plane (z = −L1 ). This is not surprising from
a physical point of view. Indeed, this heat flux is present in all planes z = contant of
these stacked plates. Hence, boundary condition (5.17) could be posed at any other
plane, as e.g. at the bottom plane (z = −L1 − L2 ).

38
Fluid Mechanics and Heat Transport

The determination of the constants, now appears straight forward. From equation
(5.12) with boundary condition (5.14), we obtain
q̇0
C3 = , (5.18)
λ1
and from (5.13) with boundary condition (5.15), we obtain

T∞ = −C5 (L1 + L2 ) + C6 . (5.19)

The condition (5.17) at the contact plane with equation (5.12) gives
q̇0
C5 = , (5.20)
λ2
and condition (5.16) with equations (5.12, 5.13) provides
q̇0
C4 = (L1 + L2 ) . (5.21)
λ1
Hence, the solutions are
q̇0
T1 (z) = T∞ + (z + L1 + L2 ) , (5.22)
λ1
q̇0
T2 (z) = T∞ + (z + L1 + L2 ) . (5.23)
λ2

Again, both temperature profile appear to be (piecewise) linear in z. Depending


on the ratio λ1 /λ2 , the gradient of temperature in both solids can be different. In
figure 22 three typical profiles are given: For λ1 = λ2 we have λ1 /λ2 = 1, and we
obtain a single straight line through both plates (dotted profile), featuring an identical
temperature gradient in both plates. For λ1 > λ2 , we get λ1 /λ2 > 1, meaning that
the heat conductivity in the top plate is better than in the bottom plate. Hence,
the gradient dT1 /dz (in the top plate) will be smaller than the gradient dT2 /dz (in
the bottom plate). This can be inferred from a comparison of both solutions (5.22)
and (5.23). The corresponding profile is given in figure 22 as solid line. Inversely, for
λ1 /λ2 < 1, the temperature profile features a larger gradient in the top plate. Such a
temperature profile is given as dashed line in figure 22.

39
P. Ehrhard

Figure 22: Qualitative temperature profiles across two stacked plates.

5.2 Radiation
To discuss the emission and absorption of heat by radiation, we first have to introduce
the so–called black body. A black body is an idealized body, which emits at a given
temperature a characteristic maximum radiative heat flux. On the other hand, this
black body absorbs all radiation, irradiated from the ambient onto its surface. Of
course, in practise such a body does not exist and we have to engage the so–called
emissivity ǫem and absorptivity ǫab to capture the behaviour of real bodies, subject
to radiation.
Radiation, in general, occurs over a range of wave lengths, depending on the actual
temperature. The emitted and absorbed heat flux, therefore, can be obtained from
an integral over the entire range of relevant wave lengths. If we have a black body at
temperature T , over all wavelengths it emits the heat flux

q̇em,bl = σT 4 . (5.24)

Here, σ = 5.67 · 10−8 W/(m2 K 4 ) is the Stefan–Boltzmann constant, and T is the


absolute temperature of the black body. Obviously, the emitted specific heat flux
is of dimension [W/m2 ]. To account for the surface properties of a real body, the
emissivity ǫem can be employed, which is dimensionless and has to be in the range
0 < ǫem < 1. Hence, the emitted heat from a real body amounts to

q̇em = ǫem σT 4 . (5.25)

40
Fluid Mechanics and Heat Transport

Figure 23: Radiative heat fluxes around a real body.

In figure 23 the emitted heat flux (to one side) at such a real body is sketched.
Obviously, the emitted radiation also depends on direction. Hence, to capture the
emission to one half–space, we need to integrate the directed radiation over a semi–
spherical segment. This integration can likewise be incorporated into ǫem .
On the other hand, a black body absorbs all irradiated radiation. If we have a radia-
tive flux q̇∞ from the ambient, all of this flux arriving at the black–body surface will
be absorbed. Again, q̇∞ represent the radiative flux due to all relevant wavelengths.
A real body, depending on its surface properties and on the irradiation angle, will
only absorb some part of this radiative flux, while other parts are reflected or trans-
mitted through the body. Hence, all these effects can be incorporated by integration
of the directed radiation over a semi–spherical segment and can be imaged into the
absorptivity ǫab . Similar to the emissivity, the absorptivity ǫab is dimensionless and
has to be in the range 0 < ǫab < 1. The absorbed radiative heat flux for the real body
then is
q̇ab = ǫab q̇∞ . (5.26)
Both, emissivity ǫem and absorptivity ǫab , can be found in tables for a wide range
of materials, surface roughness, and temperatures. This information on ǫem and ǫab
in such tables is given either for the surface–normal direction or, integrated, for the
semi–spherical segment.

41
P. Ehrhard

Figure 24: Radiation between parallel plates.

5.2.1 Example 3: Radiation between two parallel plates

If we have two parallel plates at temperatures T1 and T2 , with both plates large and
the distance between the plates much smaller, the radiative fluxes can be computed
in a straight–forward manner. This situation is sketched in figure 24. If both plates
are black bodies, the specific radiative heat flux is simply governed by the emissions
from both plates, which are
q̇em,1,bl = σT14 , (5.27)
q̇em,2,bl = σT24 . (5.28)
Since all irradiated fluxes are absorbed by both (black) plates, the heat flux is simply
Q̇12,bl = σA T14 − T24 ,

(5.29)
with the plate area A. To account for the real character of both plates, we have
to account for the emissivities of both plates, namely for ǫem,1 and ǫem,2 . Since all
radiation will be finally absorbed, after several reflections, the absorptivities do not
enter the problem. Hence, for real plates one finds
σ
A T14 − T24 ,

Q̇12 = (5.30)
K0
1 1
K0 = + −1. (5.31)
ǫem,1 ǫem,2

42
Fluid Mechanics and Heat Transport

The heat fluxes Q̇12,bl and Q̇12 are taken positive, if they are directed in the positive
x–direction (cf. figure 24). The constant K0 depends on geometry, it can also be
found for other simple geometries as e.g. for two coaxial surfaces.

43
6 Differential conservation equations

If we want to describe a flow field or a thermal field on the continuum level in all de-
tails, we have to deal with the differential conservation equations and have to integrate
these within the field of interest, while adequate initial and boundary conditions need
to be applied. Hence, firstly, we shall for general coordinates summarize these dif-
ferential equations. Secondly, we shall apply an adequate scaling to these equations,
which will help to formulate these equations in dimensionless form, and in parallel,
allow to judge on the magnitude of several terms in these equations. Thirdly, we shall
infer several (exact) solutions to these equations for simple situations.

The differential conservation equations will be given under the following assumptions:

• The fluid is assumed to be incompressible, i.e. ρ = constant holds;

• The fluid is assumed to be in thermodynamic equilibrium. This means, for


simple fluids, we can treat the fluid as Newtonian with constant viscosity µ.
Similarly, the fluid can be treated as a so–called Fourier fluid with constant
conductivity λ;

• The fluid is assumed to have constant properties, otherwise.

6.1 Conservation of mass

Conservation of mass is ensured by the so–called continuity equation

∇w
~ =0. (6.1)

Equation (6.1) is a linear, scalar, partial–differential equation. Herein, ∇ denotes the


divergence operator, applied to the velocity vector w. ~ In cartesian coordinates, with
w
~ = (u, v, w), we obtain e.g.

∂u ∂v ∂w
∇w
~= + + . (6.2)
∂x ∂y ∂z

This operator can be formulated within all common coordinate systems, the result is
readily available in mathematical handbooks.

44
Fluid Mechanics and Heat Transport

6.2 Conservation of momentum

Conservation of momentum is ensured by the so–called Navier–Stokes equations


 
∂w
~
ρ +w~ · ∇w
~ = −∇p + µ∆w ~ + ρf~ . (6.3)
∂t

Note that equation (6.3) is a vector equation. We likewise have the divergence opera-
tor ∇w~ present on the left side of the equation, within the so–called convective term.
By the way, the term w~ · ∇w~ inside the convective term appears to be nonlinear. This
is why the Navier–Stokes equations are considered to be nonlinear partial–differential
equations. On the right side of equation (6.3), we recognize a gradient operator ∇p,
applied to the scalar pressure. For cartesian coordinates, e.g., this gives
∂p
 
∂x
∂p
∇p =   . (6.4)
 
∂y
∂p
∂z

Moreover, the Laplace operator ∆w ~ is present on the right side of equation (6.3) in
the so–called viscous (or diffusive) term. E.g. for cartesian coordinates, this operator
gives
 ∂2u ∂2u ∂2u 
∂x2
+ ∂y2 + ∂z 2
 ∂2v ∂2v ∂2v 
∆w ~ =  ∂x2 + ∂y2 + ∂z 2  . (6.5)
∂2w ∂2w ∂2w
∂x2
+ ∂y 2
+ ∂z 2

As already mentioned above, both gradient and Laplace operators can be found for
all other common coordinate systems in mathematical handbooks. Finally, the term
f~ in equation (6.3) reflects a volumetric force, per unit mass. Such a force may arise
e.g. due to gravitational, electrical, or magnetic fields.

6.3 Conservation of heat

Conservation of heat can be ensured by the so–called heat–transport equation


 
∂T
ρcp +w
~ · ∇T = λ∆T + ρq̇ . (6.6)
∂t

Equation (6.6) is a scalar and nonlinear partial–differential equation, due to the non-
linear term w
~ · ∇T on the left side, within the so–called convective term. Within this
nonlinear term ∇T represent the gradient operator, applied to the scalar temperature.

45
P. Ehrhard

In cartesian coordinates, it is

 ∂T

∂x
∂T
∇T =   . (6.7)
 
∂y
∂T
∂z

On the right side of equation (6.6), within the so–called conductive (or diffusive)
term, we have the Laplace operator ∆T , applied to the scalar temperature. This, e.g.
for cartesian coordinates, gives

∂2T ∂2T ∂2T


∆T = + + . (6.8)
∂x2 ∂y 2 ∂z 2

Finally, on the right side of equation (6.6), with q̇, we have a heat source/sink per
unit mass.

6.4 Scaling of the conservation equations

If we inspect the nonlinear equations for the transport of momentum (6.3) and heat
(6.6), it appears obvious that analytical (exact) solutions of these equations will
be possible only in very few simple cases. Hence, since these equations have been
formulated, it has been a concern to simplify them. This, however, is only possible if
we can judge which terms are important (or large) and which terms are negligible (or
small). Or, to cite Albert Einstein: Everything should be made as simple as possible,
but not simpler. A formal tool to allow for such a judgement is the scaling of the
equations. Scaling aims to transform all dimensional variables of various magnitudes
into dimensionless variables of a magnitude of order one. This appears possible, if
we e.g. non–dimensionalize the x–coordinate with the correct length scale in the x–
direction, say x0 . Hence, in dimensionless form we obtain X = x/x0 and have X ∼ 1,
meaning X is of order one. If we repeat this process for all variables, we shall have all
dimensionless variables of order one, and the magnitudes of all terms will be collected
in coefficients in front of the dimensionless variables. These coefficients will also be
dimensionless, they are called dimensionless groups. Let us apply such a scaling to
the above conservation equations (6.1, 6.3, 6.6).

46
Fluid Mechanics and Heat Transport

In the absence of a concrete problem, one possibility for scaling is

(x, y, z)
(X, Y, Z) = ,
l0
(u, v, w)
(U, V, W ) = ,
u0
p
P = , (6.9)
ρu20
T − Tc
Θ = ,
Th − Tc
t
τ = .
(l0 /u0 )

Here, l0 is a characteristic length scale of the flow problem, e.g. a pipe diameter or the
length of an object in a flow. The characteristic velocity scale u0 of the flow problem
may be e.g. a mean velocity in a pipe or the far–field velocity around an object in
the flow. The pressure is scaled with the inertial pressure scale p0 = ρu20 . This is
reasonable, as long as inertia plays a major role in the flow problem; it corresponds
to the case of large Reynolds numbers. If, instead, viscous effects dominate the flow
problem, there is a second option to engage the viscous pressure scale p0 = µu0 /l0 .
Finally, Tc is the lowest, and Th the highest temperature in the (thermal) problem,
this could e.g. be a wall temperature. Given a length and a velocity scale, a transport
time scale t0 = l0 /u0 is readily inferred. If we introduce the above scales (6.9) into the
conservation equations (6.1, 6.3, 6.6), we have e.g. to make use of ∂x = l0 ∂X, ∂u =
l0 ∂U, ∂T = (Th − Tc )∂Θ, etc., to replace the dimensional quantities by dimensionless
quantities, and obtain

∇W ~ = 0, (6.10)
" #
∂W~
~ · ∇W
+W ~ = −∇P + Re−1 ∆W~ + F r −1~ef , (6.11)
∂τ
 
∂Θ ~
+ W · ∇Θ = P e−1 ∆Θ + Q∗ . (6.12)
∂τ

Within these equations (6.10, 6.11, 6.12) the operators ∇ and ∆ have to be executed in

47
P. Ehrhard

dimensionless coordinates. Further, a number of dimensionless groups arises, namely

u 0 l0 inertial forces
Re = ∝ ,
ν viscous forces
u20 inertial forces
Fr = ∝ ,
|f | l0 volumetric forces
u 0 l0 convective heat flux
Pe = ∝ , (6.13)
κ conductive heat flux
q̇ l0 added heat flux
Q∗ = ∝ .
cp u0 (Th − Tc ) convective heat flux

Re is the Reynolds number, F r a (generalized) Froude number, whereas the direction


of the volumetric force (as e.g. gravity) is represented by the unit vector ~ef , P e is
the (thermal) Péclet number, and Q∗ is the dimensionless version of the added heat
flux q̇.
With regard to the Reynolds number, we can readily infer the limit Re → ∞, which
means that inertial forces dominate over viscous forces. Inside equation (6.11), this
leads to the disappearance of the viscous term, multiplied by Re−1 . Hence, we obtain
the so–called Euler equations. Conversely, the limit Re → 0 means that viscous forces
dominate over inertial forces. Hence, after a multiplication of equation (6.11) with
Re, the convective term disappears. Hence, we have the so–called Stokes equations
left. It should be noted, that in this case pressure should be non–dimensionalized
by a viscous pressure scale, such that the pressure term survives within the Stokes
equations.
The limits of the heat transport equation with regard to the Péclet number are, firstly,
P e → ∞. This means that the convective heat flux dominates over the conductive
heat flux. Hence, in equation (6.12) the conductive term disappears, as it is multiplied
by P e−1 . Secondly, the limit P e → 0 indicates that we have heat transport mainly
by conduction. Hence, in equation (6.12) the convective term disappears, and we are
left with the pure heat conduction equation, which was already discussed in chapter
5 (cf. equation (5.1)).

6.5 Exact solutions to the Navier–Stokes equations

Indeed, the Navier–Stokes equations do not allow for many analytical (exact) solu-
tions, due to their complexity and non–linearity. Let us discuss a family of solutions
for a fluid between two infinite, parallel plates (cf. figure 25). This, obviously, defines
a two–dimensional problem in (x, y), if we simply have one axis of the coordinate

48
Fluid Mechanics and Heat Transport

Figure 25: Fluid between two infinite, parallel plates with (a) the velocity profile of the Couette flow
and (b) the velocity profile of the Poiseuille flow.

system parallel to any movement (of fluid or plate). We further assume for simplic-
ity that we have no volumetric forces acting onto the fluid (f~ = 0), and that we
have a steady and fully–developed flow. The latter assumptions imply ∂u/∂t = 0,
∂v/∂t = 0, as well as ∂u/∂x = 0 and ∂v/∂x = 0. The governing equations, derived
from equations (6.1, 6.3), are therefore

dv
= 0, (6.14)
dy
du ∂p d2 u
ρv = − +µ 2 , (6.15)
dy ∂x dy
∂p
0 = − . (6.16)
∂y

Here, we have already made use of u = u(y) and v = v(y), which is due to the
fully–developed flow. From equation (6.16) we further can infer p = p(x), and from
equation (6.14) we find v = constant. Since we have solid plates at y = ±h, there will
be no flow through the plates, and hence, the boundary conditions are v(y = ±h) = 0.
This provides the general solution v = 0 for the entire fluid layer. Hence, equation
(6.15) can be simplified further to obtain

d2 u 1 dp
= . (6.17)
dy 2 µ dx

Twice a simple integration in y gives the general solution for u(y), namely

1 dp 2
u(y) = y + C1 y + C2 . (6.18)
2µ dx

49
P. Ehrhard

To determine the constants C1 and C2 , we need to provide two boundary conditions.


Obviously, there are two means to drive a flow between the two plates: (i) we move
one plate and keep the other plate at rest, (ii) we keep both plates at rest and apply
a pressure gradient. Let us discuss, what solutions will occur for these two cases in
the following paragraphs.

6.5.1 The Couette flow

The first option to drive a flow between the plates involves moving e.g. the up-
per plate, say with velocity up , and keep the second (lower) plate at rest. Hence,
kinematically we have the boundary conditions

u(y = −h) = 0 , (6.19)


u(y = +h) = up . (6.20)

Moreover, the application of a pressure gradient is not necessary. Hence, we also have
dp
=0. (6.21)
dx
If we apply the boundary conditions (6.19, 6.20) and condition (6.21) to the general
solution (6.18), we can determine the constants in a straight–forward manner and
obtain the specific solution
up h y i
u(y) = +1 , (6.22)
2 h
v = 0. (6.23)

This corresponds to a linear shear profile, which is already sketched in figure 25. This
flow is termed Couette flow.

6.5.2 The Poiseuille flow

The second option to drive the flow involves a pressure gradient as well as two plates
at rest. Hence, kinematically we have the boundary conditions

u(y = −h) = 0 , (6.24)


u(y = +h) = 0 . (6.25)

Further, to drive a flow in the positive x–direction, we have to have a high pressure
at the left entrance of the gap, and a low pressure at the right exit of the gap. This
corresponds to a pressure gradient
dp
<0. (6.26)
dx

50
Fluid Mechanics and Heat Transport

We introduce the boundary conditions (6.24, 6.25) into the general solution (6.18) to
determine the constants, and obtain the specific solution

h2 dp  y 2
 
u(y) = −1 , (6.27)
2µ dx h
v = 0. (6.28)

This solution corresponds to a parabolic velocity profile, which is already sketched in


figure 25. This flow is termed plane Poiseuille flow, as there exists also a solution for
a circular pipe, which is termed axis–symmetric Poiseuille flow.

6.6 The heat transport between two plates


Even though the following discussion will not lead to an analytical solution for the
temperature field, it remains interesting to inspect the governing equation for the
thermal problem. For the flow, with equation (6.18), we have found a general solution,
which is of character u(y), v = 0. If we assume for simplicity a steady temperature
field with no heat sources, i.e. q̇ = 0, we find from the heat transport equation (6.6)
immediately  2
∂2T

∂T ∂ T
ρcp u(y) =λ + , (6.29)
∂x ∂x2 ∂y 2
as governing equation for T (x, y). On the left side, we recognize the convective heat
transport by the flow, the right side is associated with the conductive heat transport in
both directions. Even for such a simple flow, there is obviously no simple way to solve
for the thermal field. Typically, a separation ansatz would be applied, to separate
the dependencies of temperature on x and y. This, however, already represents an
approximation and does not maintain the full variety of solutions. Further, boundary
conditions for the temperature an all four edges of the fluid domain are needed.

51
Figure 26: Airfoil with magnified (kinematic) boundary layer.

7 Boundary–layer equations

As already discussed in section 6.4 on scaling, approximations for the full conservation
equations (6.1, 6.3, 6.6) are highly welcome. One important approximation can be
derived from the existence of so–called boundary layers. If we have e.g. an airfoil at
high ambient fluid velocity (cf. figure 26), we expect at sufficient distance from the
surface of the airfoil to see the ambient velocity u∞ . However, at the airfoil surface,
i.e. at the wall, we have the no–slip condition. Hence, adjacent to the airfoil, there
exists a fluid layer, across which the tangential fluid velocity drops from u∞ to zero.
For large Reynolds numbers, this layer appears to be thin, i.e. δ(x) ≪ l, and is
called (kinematic) boundary layer. These circumstances can be employed to simplify
the governing mechanical equations (6.1, 6.3) substantially. Finally, if a thermal
problem is considered, let’s say the airfoil is heated, for large ambient velocities u∞ ,
the fluid layer, across which the hot fluid temperature at the airfoil surface drops
to the ambient temperature, appears likewise thin. This layer, for small thickness
δth (x) ≪ l, is termed thermal boundary layer. Analog to the mechanical case, this
observation will be helpful to simplify the heat transport equation (6.6).

For the airfoil in figure 26, the curvature of the wall is typically small, such that
to first approximation the treatment of the kinematic and thermal boundary layers,
adjacent to a flat plate, will be useful in understanding the effects and in deriving the
equations of boundary layers. Hence, within the next two sections, we shall discuss
these boundary layers in all necessary detail.

52
Fluid Mechanics and Heat Transport

7.1 Kinematic boundary layer at the flat plate


Within figure 27, we have sketched the problem of a boundary layer at a flat plate.
We assume a Newtonian fluid of constant fluid properties, particularly the fluid is
incompressible (ρ = constant), and we assume the boundary layer and the flow to be
steady and plane in (x, y). Further, we have no volumetric forces acting at the fluid,
i.e. f~ = 0. Hence, from the general mechanical conservation equations (6.1, 6.3) we
infer for the given boundary–layer problem at first

∂u ∂v
+ = 0, (7.1)
∂x ∂y
 2
∂ u ∂2u
  
∂u ∂u ∂p
ρ u +v = − +µ + , (7.2)
∂x ∂y ∂x ∂x2 ∂y 2
 2
∂ v ∂2v
  
∂v ∂v ∂p
ρ u +v = − +µ + . (7.3)
∂x ∂y ∂y ∂x2 ∂y 2

The objective is to derive simplified mechanical equations, valid within the bound-
ary layer. In chapter 6 we have already learned, that the method of scaling allows
to judge on the magnitude of the terms, present in the conservation equations for
mass and momentum. In the case of a boundary layer, we have considerably more
information about its geometry, namely it appears to be a thin and extended layer of
fluid, attached to the solid plate (cf. figure 27). This information can be employed
for the scaling to great advantage. Instead of using homogeneous length and velocity
scales as in (6.9), we introduce for the boundary layer the scaling
x y
X= , Y =,
l εl
u v
U= , V = , (7.4)
u∞ v0
p
P = .
ρu2∞

The scaling (7.4) reflects our best knowledge about the boundary layer, as sketched
in figure 27. We expect it to be of length l in the x–direction, and to be of small
thickness δ ∼ εl in the y–direction. Here, ε ≪ 1 is a small number which expresses
the magnitude of the boundary layer thickness, based on its length. Across the
boundary layer, we expect the wall–tangential velocity component u to vary from zero
to u∞ , while we suspect a different order of magnitude for the wall–normal velocity
component v. Therefore, we introduce a general (unknown) velocity scale v0 for v,
and hope that the equations will tell us what v0 should be. Further, thin kinematic
boundary layers appear typically for large Reynolds numbers, meaning that inertial

53
P. Ehrhard

Figure 27: Kinematic boundary layer at the flat plate.

forces will dominate viscous forces, at least in the outer region of the boundary layer.
Hence, an inertial pressure scale p0 = ρu2∞ , based on the present far–field velocity
u∞ , appears suitable.
Firstly, we substitute the scales (7.4) into the continuity equation (7.1), and obtain

∂U v0 ∂V
+ =0. (7.5)
∂X εu∞ ∂Y

As the first term is of order one, the equation can only be fulfilled if the second term
is also of order one. Since ∂V /∂Y ∼ 1 holds, we have to have
v0
∼1. (7.6)
εu∞

As v0 is an arbitrary velocity scale, which simply has to fulfill the order–of–magnitude


relation (7.6), we choose v0 = εu∞. This physically confirms, that the wall–normal
velocity component v is much smaller than u∞ .
We now introduce the scaling (7.4) into the first Navier–Stokes equation (7.2) and
obtain

ρu2∞ ρu2∞ ∂P 2
∂2U
   
∂U ∂U µu∞ 2∂ U
U +V =− + 22 ε + . (7.7)
l ∂X ∂Y l ∂X εl ∂X 2 ∂Y 2

This equation tells us, firstly, that within the viscous term |ε2 ∂ 2 U/∂X 2 | ≪ |∂ 2 U/∂Y 2 |
holds, such that we can neglect the first contribution. Secondly, we recognize that
both the convective term and the pressure term are of identical order of magnitude,

54
Fluid Mechanics and Heat Transport

namely ∼ ρu2∞ /l. Hence, it appears natural to divide equation (7.7) by (ρu2∞ /l) to
obtain
∂U ∂U ∂P ∂2U
U +V ≃− + Re−1 ε−2 2 . (7.8)
∂X ∂Y ∂X ∂Y
Here, we have introduced the Reynolds number Re = u∞ l/ν. Again we have a
choice, this time with regard to the viscous term: (i) if we pick Re−1 ε−2 ≪ 1, we
shall loose the viscous term, (ii) if we pick Re−1 ε−2 ≫ 1, we shall loose the inertial
term. However, both inertial and viscous forces are important ingredients of the this
boundary–layer problem. (iii) Only if we pick Re−1 ε−2 ∼ 1, we can keep both forces
at play in equation (7.8). Hence, we make the choice Re−1 ε−2 = 1, which fulfills the
order–of–magnitude relation in (iii) and essentially picks an appropriate length scale
in y, namely y0 = εl. This leads to

δ
∼ ε = Re−1/2 , (7.9)
l
which allows to infer several important dependencies, namely

δ ∝ Re−1/2 ,
δ ∝ l1/2 , (7.10)
δ ∝ x1/2 .

The dependencies in (7.10) confirm, firstly, our initial assumption, that large Reynolds
numbers should lead to thin boundary layers; in fact, for Re → ∞ we have δ → 0.
Secondly, the boundary layer growth along the plate, following a square–root law.
We now introduce the scaling (7.4) into the second Navier–Stokes equation (7.3) and
obtain
2
∂2V
   
2 ∂V ∂V ∂P −1 2∂ V
ε U +V =− + Re ε + . (7.11)
∂X ∂Y ∂Y ∂X 2 ∂Y 2
Again, we can neglect one contribution in the viscous term due to |ε2 ∂ 2 V /∂X 2 | ≪
|∂ 2 V /∂Y 2 |. More importantly, as Re−1 = ε2 , we find in summary

∂P
≃0, (7.12)
∂Y
whereas all terms of magnitude ∼ ε2 have been neglected. This physically means, that
the pressure inside the boundary layer is essentially given by the far–field pressure
P (Y → ∞) = P∞ . Inside the boundary layer, we have the pressure

δ ∂P
P ≃ P∞ + ≃ P∞ . (7.13)
εl ∂Y

55
P. Ehrhard

As this statement holds at each place X along the plate, this likewise means
∂P
≃0 (7.14)
∂X
in equation (7.8). In dimensionless form, we therefore can summarize the so–called
boundary–layer equations at the flat plate as
∂U ∂V
+ = 0, (7.15)
∂X ∂Y
∂U ∂U ∂2U
U +V ≃ , (7.16)
∂X ∂Y ∂Y 2
P ≃ P∞ . (7.17)

These equations can, of course, be transformed back to their dimensional form, once
we know which terms are negligible. The dimensional form is
∂u ∂v
+ = 0, (7.18)
∂x ∂y
∂2u
 
∂u ∂u
ρ u +v ≃ µ 2 , (7.19)
∂x ∂y ∂y
p ≃ p∞ . (7.20)

Of course, to solve either (7.15–7.17) or (7.18–7.20), we need to pose boundary con-


ditions at the plate, e.g. for u(y = 0), v(y = 0), ∂u/∂y(y = 0), or ∂v/∂y(y = 0), and
in the far field, e.g. for u(y → ∞), v(y → ∞), ∂u/∂y(y → ∞), or ∂v/∂y(y → ∞).
As the mechanical equations in this boundary–layer approximation appear to be
parabolic and steady, conditions in planes x = constant and initial conditions are
usually not needed.

7.2 Thermal boundary layer at the flat plate


Within figure 28, the situation with a thermal boundary layer at a flat plate is
sketched. We assume that for large ambient velocity u∞ , a thin thermal bound-
ary layer develops, across which temperature drops from the plate temperature Tw to
the ambient temperature T∞ . The heat transport equation (6.6) provides the general
basis for this problem. We assume a Fourier fluid of constant thermal properties and
no heat sources/sinks present (i.e. q̇ = 0). If further, the flow and the thermal field
are steady and plane in (x, y), the governing equation is
 2
∂2T
  
∂T ∂T ∂ T
ρcp u +v =λ + . (7.21)
∂x ∂y ∂x2 ∂y 2

56
Fluid Mechanics and Heat Transport

y
0 T¥ Tw
T

dth(x) T(y)
x
l
Figure 28: Thermal boundary layer at the flat plate.

Obviously, the thermal boundary layer in general has not to be identical in thickness
with the kinematic boundary layer. Therefore, we introduce a modified scaling with
regard to the y–direction and the v–component of velocity. This reads as
x y
X= , Ȳ =
,
l εth l
u v
X= , V̄ = , (7.22)
u∞ εth u∞
T − T∞
Θ = .
Tw − T∞
This scaling is completely analog with the scaling (7.4) for both, coordinates and
velocity. The temperature scale is based on the highest (Tw ) and the lowest (T∞ )
temperature in the problem, such that we have Θ ∼ 1.
We introduce the scaling (7.22) into the governing equation (7.21), and obtain
2
∂2Θ
 
∂Θ ∂Θ −1 −2 2 ∂ Θ
U + V̄ = P e εth εth + . (7.23)
∂X ∂ Ȳ ∂X 2 ∂ Ȳ 2
Here, we have defined the Péclet number by P e = u∞ l/κ, whereas the thermal
diffusivity κ is given by κ = λ/(ρcp ). Firstly, within the conduction term on the right
side of the equation, we recognize ε2th ∂ 2 Θ/∂X 2 ≪ ∂ 2 Θ/∂ Ȳ 2 . Hence, we can neglect
the first contribution within the conduction term. Secondly, as the convective term
on the left side of the equation is of order one, with P e−1 ε−2
th in front of the conductive
term, we have a choice: For P e−1 ε−2 th ≪ 1, the conductive heat flux will disappear,
−1 −2
and for P e εth ≫ 1 the convective heat flux will disappear. As both contributions
to the heat transport are important in the present boundary–layer problem, the only

57
P. Ehrhard

reasonable choice is P e−1ε−2


th ∼ 1. This keeps both, the conductive and the convective
heat flux present in equation (7.23). Once more, the specific choice P e−1 ε−2 th = 1
fulfills the above order–of–magnitude relation, and picks the thermal length scale
according to
δth
∼ εth = P e−1/2 . (7.24)
l
Between the Péclet number P e and the Reynolds number Re, there exists a connec-
tion, namely
u∞ l u∞ l ν
Pe = = = Re P r . (7.25)
κ ν κ
In equation (7.25), the Prandtl number P r, as ratio of kinematic viscosity and thermal
diffusivity of the fluid, represents a pure fluid property. P r ∼ 1 holds for standard
fluids as air (P r ≃ 0.7) or water (P r ≃ 7). Further, P r ≪ 1 is valid e.g. for liquid
metals of low viscosity and excellent thermal conductivity, and P r ≫ 1 is valid e.g.
for highly–viscous oils of poor thermal conductivity. Using (7.24), we can infer several
dependencies, namely

δth ∝ P e−1/2 ,
δth ∝ l1/2 , (7.26)
δth ∝ x1/2 .

Moreover, with P e = P r Re and (7.10) we find


δth
∝ P r −1/2 . (7.27)
δ
Hence, for standard fluids with P r ∼ 1, both the kinematic and the thermal boundary
layers are of similar thickness. For e.g. liquid metals (or P r ≪ 1), from equation
(7.27) we find δth ≫ δ, i.e. the thermal boundary layer is much thicker than the
kinematic boundary layer. Inversely, for highly–viscous oils (or P r ≫ 1), we have
δth ≪ δ, i.e. the thermal boundary layer is even thinner than the kinematic boundary
layer. Obviously, for P r ≪ 1 we cannot guarantee a thin thermal boundary layer and
have to be careful with regard to the validity of the boundary–layer approximation.
We may summarize the heat transport equation in dimensionless form, valid within
a thin thermal boundary layer, as
∂Θ ∂Θ ∂2Θ
U + V̄ ≃ , (7.28)
∂X ∂ Ȳ ∂ Ȳ 2
or transformed back to the dimensional form
∂2T
 
∂T ∂T
ρcp u +v ≃λ 2 . (7.29)
∂x ∂y ∂y

58
Fluid Mechanics and Heat Transport

Of course, to solve either (7.28) or (7.29), we need to pose boundary conditions


at the plate, e.g. for T (y = 0) or ∂T /∂y(y = 0) and in the far field, e.g. for
T (y → ∞) or ∂T /∂y(y → ∞). As the heat transport equation in this boundary–layer
approximation appears to be parabolic and steady, conditions in planes x = constant
and initial conditions are usually not required.

59
8 Reynolds–averaged equations

To capture turbulent flows, it is – of course – possible to use the time–dependent and


three–dimensional conservation equations, as given in chapter 6. It is important to
note that turbulence is inherently time–dependent and three-dimensional. Turbulent
simulations, based on these basic equations are termed direct numerical simulations
(DNS). There remain, however, severe problems: (i) there is no chance to solve the
conservation equations (6.1, 6.3, 6.6) analytically; (ii) for a numerical solution, we
have to resolve, on the one side, the length scale of the problem L, e.g. with L being
the width of the flow channel, and on the other hand, the smallest vortices in the
turbulent flow, which are of the size of the so–called Kolmogorow length η. If we
express the ratio of the macroscopic length L and the Kolmogorow length η, we find

L Lu′ 3/4
= ∝ Re3/4 , (8.1)
η ν

with the characteristic velocity fluctuation amplitude u′ and the kinematic viscosity
ν of the fluid. Hence, it appears obvious that, with increasing Reynolds number, we
obtain an increasing ratio of largest and smallest length scale, to be resolved. For
a turbulent flow at e.g. a Reynolds number of Re = u0 L/ν = 106 and a moderate
degree of turbulence T u = u′/u0 = 10%, we have L/η ≃ 5623. In other words, to
resolve e.g. the channel width L adequately, we would need about 104 nodes across
the channel in the simulations. As this is only one direction, the total number of
nodes in all three directions is of the order ∼ 1012 . Together with the necessity for
time–dependent simulations, this presents a formidable numerical task, not to say,
it is simply impossible to solve this numerical problem even on the most advanced
computers.
Hence, since the years of Osborne Reynolds, it appeared necessary to introduce so–
called turbulence models. The first step, hereby, is to make use of the so–called
Reynolds ansatz, to decompose all quantities into a (temporal) mean contribution
and a fluctuation contribution. Hence, this ansatz applied to all quantities reads as

u(x, y, z, t) = ū(x, y, z) + u′ (x, y, z) ,


v(x, y, z, t) = v̄(x, y, z) + v ′ (x, y, z) ,
w(x, y, z, t) = w̄(x, y, z) + w ′ (x, y, z) , (8.2)
p(x, y, z, t) = p̄(x, y, z) + p′ (x, y, z) ,
T (x, y, z, t) = T̄ (x, y, z) + T ′ (x, y, z) .

60
Fluid Mechanics and Heat Transport

Figure 29: Velocity component u(x0 , y0 , z0 , t) at a fixed position, as function of time for a turbulent
flow.

Here, e.g. the temporal mean of u is defined by


t+t0
1
Z
ū(x, y, z) = u(x, y, z, t)dt , (8.3)
t0
t

and by definition, the temporal mean of e.g. the fluctuation u′ is


t+t0
1
Z
u′ (x, y, z) = u′ (x, y, z, t)dt = 0 . (8.4)
t0
t

The statements (8.3, 8.4), of course, hold for all quantities in equation (8.2). The
situation for the example u is sketched in figure 29. If we measure the velocity
component u at any fixed position (x0 , y0 , z0 ) in the turbulent flow, we shall see a
stochastic signal as sketched in the figure. The (temporal) mean ū is independent
of time and appears as (dashed) straight line; the (turbulent) fluctuation u′ is the
difference (u − ū). The basic idea behind the decomposition (8.2) is, to allow access
to the (temporal) mean quantities ū, v̄, w̄, p̄, and T̄ , which are of primary engineering
interest. In contrast, the fluctuations u′ , v ′ , w ′ , p′ , and T ′ are not in the main focus
of engineering applications and, hopefully, can be modeled in their impact onto the
(temporal) mean flow.

8.1 Conservation of mass


We now have to introduce the ansatzes in (8.2) into the conservation equations,
starting with equation (6.1) for the conservation of mass, and obtain at first
∂ ū ∂u′ ∂v̄ ∂v ′ ∂ w̄ ∂w ′
+ + + + + =0. (8.5)
∂x ∂x ∂y ∂y ∂z ∂z

61
P. Ehrhard

Equation (8.5) can be averaged (in time) to obtain

∂ ū ∂u′ ∂v̄ ∂v ′ ∂ w̄ ∂w ′
+ + + + + =0. (8.6)
∂x ∂x ∂y ∂y ∂z ∂z

In (8.6) all expressions with ū, v̄, and w̄ are already independent of time, and the
(temporal) mean does not modify anything. In contrast, for the (temporal) mean of
the fluctuation terms, e.g.
∂u′
=0 (8.7)
∂x
holds, if the stochastic process is ergodic. We already concluded that e.g. the fluctu-
ation u′ is free of a mean (cf. equation (8.4)). A stochastic process is called ergodic,
if also the temporal derivatives, and for convective transport also spatial derivatives,
are free of a mean. For such an (idealized) stochastic process, we obtain from (8.6)

∂ ū ∂v̄ ∂ w̄
+ + =0, (8.8)
∂x ∂y ∂z

a continuity equation for the (temporal) mean flow. From the difference of (8.5) and
(8.8), we further infer
∂u′ ∂v ′ ∂w ′
+ + =0, (8.9)
∂x ∂y ∂z
an instantaneous continuity equation for the turbulent fluctuations. The above mean
continuity equation (8.9) in cartesian coordinates, can also be given in vector notation,
to allow for its application within other coordinate systems. The results is

∇w
~ =0. (8.10)

8.2 Conservation of momentum

Next, we introduce the ansatzes in (8.2) into the Navier–Stokes equations (6.3), which
ensure the conservation of momentum. We shall explore this for the x–component of
these equations, and find at first
 
∂ ′ ′ ∂ ′ ′ ∂ ′ ′ ∂ ′
ρ (ū + u ) + (ū + u ) (ū + u ) + (v̄ + v ) (ū + u ) + (w̄ + w ) (ū + u ) =
∂t ∂x ∂y ∂z
 2 2 2

∂ ∂ ∂ ∂
− (p̄ + p′ ) + µ (ū + u′ ) + 2 (ū + u′) + 2 (ū + u′ ) + ρfx . (8.11)
∂x ∂x 2 ∂y ∂z

62
Fluid Mechanics and Heat Transport

Again, equation (8.11) can be averaged in time, whereas several types of terms occur,
as e.g.
∂ ū ∂ ū
= ,
∂t ∂t
∂u′
= 0,
∂t
∂ ū ∂ ū
ū = ū ,
∂x ∂x
∂u′ ∂u′
ū = ū =0,
∂x ∂x
∂ ū ∂ ū
u′ = u′ =0, (8.12)
∂x ∂x
∂u′
u′ 6= 0 ,
∂x
∂ 2 ū ∂ 2 ū
= ,
∂x2 ∂x2
∂ 2 u′
= 0.
∂x2
Due to the assumed ergodic process, all (temporal) averages of the fluctuations and
their temporal and spatial derivatives are zero. It should be noted though, that
the temporal derivative of the mean quantities may still be non–zero, due to slow
variations in time. Slow variations in time may not disappear from averaging over a
short period t0 (cf. equation (8.3)). Further, all mean quantities and their (spatial)
derivatives can be factored out and can be treated as constants. Care needs to be
taken with products of fluctuations and their derivatives: Even though both, the
(temporal) average of the fluctuation, and the (temporal) average of the derivative of
a fluctuation, are zero, their product involves a nonlinear operation, and therefore,
the (temporal) average of that product is not zero. This can be understood from the
simple sine function sin (ωt), which averaged in time is zero. However, the product
of two sine functions, i.e. sin2 (ωt), averaged in time is not zero! In other words,
the (temporal) mean of the product of two mean–free functions, due to the nonlinear
operation, in general is not zero.
After performing the (temporal) average of equation (8.11), and obeying the conjunc-
tions in (8.12), we arrive at
 2
∂ ū ∂ 2 ū ∂ 2 ū
  
∂ ū ∂ ū ∂ ū ∂ ū ∂ p̄
ρ + ū + v̄ + w̄ =− +µ + + 2 + ρfx
∂t ∂x ∂y ∂z ∂x ∂x2 ∂y 2 ∂z
 
∂u ′ ∂u ′ ∂u′
−ρ u ′ +v ′ +w ′ . (8.13)
∂x ∂y ∂z

63
P. Ehrhard

This equation appears to be similar to the first (x–direction) of the standard Navier–
Stokes equations (6.3), formulated for the mean flow with ū, v̄, w̄, and p̄. However,
on the right side of the equation we obtain additional terms, so–called correlations,
involving the fluctuations. These are (spatial) derivatives of turbulent stresses, or
so–called Reynolds stresses. By applying the continuity equation (8.9), we can infer
the identity

∂u′ ∂u′ ∂u′ ∂ ′2 ∂ ′ ′ ∂


u′ + v′ + w′ = (u ) + (u v ) + (u′ w ′ ) , (8.14)
∂x ∂y ∂z ∂x ∂y ∂z

which allows to bring equation (8.13) into its final form, namely
 2
∂ ū ∂ 2 ū ∂ 2 ū
  
∂ ū ∂ ū ∂ ū ∂ ū ∂ p̄
ρ + ū + v̄ + w̄ =− +µ + + 2 + ρfx
∂t ∂x ∂y ∂z ∂x ∂x2 ∂y 2 ∂z
 
∂ ′2 ∂ ′ ′ ∂ ′ ′
−ρ (u ) + (u v ) + (u w ) . (8.15)
∂x ∂y ∂z

The Navier–Stokes equations in the y– and the z–direction can be treated in analog
fashion. Hence, finally we arrive at the so–called Reynolds–averaged Navier–Stokes
equations (RANS), which are given in vector notation as
" #
∂w
~
ρ +w
~ · ∇w ~ + ρf~ + ∇Ttur ,
~ = −∇p + µ∆w (8.16)
∂t

with the turbulent stress tensor


 
(u′2 ) (u′ v ′ ) (u′ w ′ )
Ttur = −ρ  (v ′ u′ ) (v ′2 ) (v ′ w ′)  . (8.17)
 

(w ′ u′ ) (w ′v ′ ) (w ′2 )

The turbulent stresses in Ttur are physically caused by the turbulent fluctuations
u′ , v ′ , and w ′ , which over relatively–large distances can transport momentum very
efficiently. In contrast, the molecular stresses (due to viscosity µ) can only transport
momentum to the neighboring molecules, i.e. over molecular distances. In summary,
in free turbulence, with u′ , v ′ , w ′ relatively large, the most important transport of
momentum is by the turbulent fluctuations – the molecular transport (by viscosity
µ) can be neglected. However, at walls we can expect u′ , v ′ , w ′ → 0, due to the no–
slip boundary condition. Hence, at walls turbulent fluctuations are not available and
momentum transport is mainly by molecular effects, i.e. by viscosity µ.

64
Fluid Mechanics and Heat Transport

8.3 Conservation of heat

As for the mechanical equations, it appears straight forward to introduce the Reynolds
ansatzes in (8.2) into the heat–transport equation (6.6). In a first step, we obtain
 
∂ ′ ′ ∂ ′ ′ ∂ ′ ′ ∂ ′
ρcp (T̄ + T ) + (ū + u ) (T̄ + T ) + (v̄ + v ) (T̄ + T ) + (w̄ + w ) (T̄ + T )
∂t ∂x ∂y ∂z
 2 2 2

∂ ∂ ∂
=λ (T̄ + T ′ ) + 2 (T̄ + T ′ ) + 2 (T̄ + T ′ ) + ρq̇ . (8.18)
∂x 2 ∂y ∂z

Again, equation (8.18) is averaged in time, and in analog fashion as for the terms in
(8.12), the terms can be simplified. After this procedure, we arrive at

∂ 2 T̄ ∂ 2 T̄ ∂ 2 T̄
   
∂ T̄ ∂ T̄ ∂ T̄ ∂ T̄
ρcp + ū + v̄ + w̄ =λ + + 2 + ρq̇
∂t ∂x ∂y ∂z ∂x2 ∂y 2 ∂z
 
∂ ′ ′ ∂ ′ ′ ∂
−ρcp (u T ) + (v T ) + (w T ) . (8.19)
′ ′
∂x ∂y ∂z

This equation is of similar form as the original heat–transport equation (6.6), except
that here the (temporal) mean quantities ū, v̄, w̄, and T̄ appear. Moreover, additional
terms due to the turbulent fluctuations are present – these are (spatial) derivatives
of turbulent heat fluxes. Similar to the turbulent transport of momentum, due to
the fluctuating velocities, an exchange of heat over relatively–large distances occurs.
This turbulent transport of heat, in developed turbulence, is much more efficient than
the molecular transport of heat by conduction, since the conduction process involves
heat transport only between neighboring molecules. The situation, once more, is
different at walls, where the wall presence damps turbulent fluctuations, such that
immediately at the wall u′ , v ′ , w ′, T ′ → 0 holds. Hence, at the wall all turbulent heat
fluxes disappears, and only the molecular transport by conduction is left.
Let us summarize the heat transport equation (8.19) in vector form, such that it can
be used in coordinate systems different from the cartesian. Hence, we obtain
 
∂ T̄ ~ ,
ρcp +w
~ · ∇T̄ = λ∆T̄ + ρq̇ + ∇Q̇ tur (8.20)
∂t

with the turbulent heat flux vector


 
(u′ T ′ )
~
Q̇ = −ρcp  (v ′ T ′ )  .
 
(8.21)
tur

(w ′ T ′ )

65
P. Ehrhard

8.4 Turbulence models

If we inspect the derived equations (8.10, 8.16, 8.20) for turbulent flow and heat trans-
port, we realize that the equations with one continuity equation, three Navier–Stokes
equations, and one heat transport equation, present five equations. On the other
hand, we have five unknown mean quantities, namely ū, v̄, w̄, p̄, and T̄ . Moreover,
nine turbulent correlations are present in the stress tensor Ttur and three turbulent
correlations in the turbulent heat fluxes Q̇tur . Symmetry arguments as u′v ′ = v ′ u′ ,
u′ w ′ = w ′u′ , and v ′ w ′ = w ′ v ′ , further allow to reduce the number of unknown correla-
tions in the stress tensor Ttur to six. This leads to a stress tensor, which is symmetric
with regard to the diagonal. Nevertheless, we are left with 5+6+3=14 unknowns –
a problem which simply cannot be solved. This is the so–called closure problem of
turbulence.
There is only one chance to solve the above problem, which is to introduce addi-
tional models – so–called turbulence models. Generally, for all turbulent correlations
an ansatz is needed, which links these correlations to the mean quantities or their
derivatives. In most cases, such an ansatz relates the correlation to the gradient of
a mean quantity. One example is the so–called concept of eddy–diffusivity, which
appears to be analog to the molecular diffusivity ansatz. For a two–dimensional case,
e.g., the ansatz is
dū
τyx = −ρu′ v ′ = −ηtur . (8.22)
dy
Here, ηtur , in analogy to molecular transport, is termed turbulent viscosity or eddy
diffusivity. In contrast to the molecular viscosity, however, ηtur is not a fluid property,
but associated with the intensity of turbulent fluctuations in the field.
Another example of a turbulence model is the so–called mixing–length model of
Prandtl. E.g. for a two–dimensional case, the turbulent shear stress is expressed
by
2
dū dū
τyx = −ρu′ v ′ = ρl , (8.23)
dy dy
where l is the so–called mixing length. Again, it is important to note that l is not
a constant nor a fluid property. Similar to the eddy diffusivity, it depend on the
turbulent fluctuations in the flow field. To be more precise, it characterizes over
which distances turbulent fluctuations can transport momentum. Such information
is usually only available from experiments, and not from theoretical considerations.
Finally, there are numerous other turbulence models in literature. These are on one
side turbulence–parameter models without, with one (e.g. Spalart–Allmaras model),

66
Fluid Mechanics and Heat Transport

or with two (e.g. k–ǫ model, or k–ω model) additional transport equations. The k–ǫ
model appears to be the most–widely applied model within industrial applications, it
has been tested and validated for many types of applications, and it is present in many
commercial CDF codes. Finally, so–called Reynolds–stress models solve algebraic or
differential equations for all relevant stresses.

67
9 Collection of formulas
• Equations along flow tubes
assumptions: one–dimensional (mean) velocity and temperature,

mass conservation:
ṁ = ρcA = constant , (9.24)

conservation of momentum (steady):

Zp
c2 dp
+ + gz = constant , (9.25)
2 ρ

conservation of momentum (non–steady):

Zp1 Z1 Zp2 Z2
c21 dp ∂c c2 dp ∂c
+ + gz1 + ds = 2 + + gz2 + ds , (9.26)
2 ρ ∂t 2 ρ ∂t
sref sref

conservation of energy:

c2
+ i + gz − q = constant , (9.27)
2

• Mechanisms of heat transport


assumptions: constant properties,

molecular conduction (1D, steady):

∂T
q̇cond = −λ , (9.28)
∂x
convection (1D, steady):

q̇conv = −ρcp u(T − T0 ) , (9.29)

radiation (emitted from real body):

q̇em = eem σT 4 , (9.30)

68
Fluid Mechanics and Heat Transport

• Pressure loss and heat transfer in pipes


assumptions: one–dimensional (mean) velocity and temperature,

laminar flow, pressure loss:


ρ L
∆p = c2m λlam , (9.31)
2 D

64 cm D
λlam = , Re = , (9.32)
Re ν
turbulent flow, pressure loss:
ρ L
∆p̄ = c̄2m λtur , (9.33)
2 D

0.3164 c̄m D
Blasius: λtur = , Re = , (9.34)
Re1/4 ν

1 p
Prandtl: √ = 2 log (Re λtur ) − 0.8 , (9.35)
λtur
 
1 2.51 ks
Colebrook: √ = −2 log √ + , (9.36)
λtur Re λtur 3.71D
laminar flow, heat transfer, Tw given:
 
D
Nu = 3.66 for small values ReP r , (9.37)
x
 1/3  
D D
Nu(x) = 1.077 ReP r for large values ReP r , (9.38)
x x

α(x)D ν
Nu(x) = , Pr = , (9.39)
λ κ
laminar flow, heat transfer, q̇w given:
 
D
Nu = 4.364 for small values ReP r , (9.40)
x
 1/3  
D D
Nu(x) = 1.302 ReP r for large values ReP r , (9.41)
x x
turbulent flow, heat transfer, all boundary conditions:
"  2/3 #
(ξ/8)ReP r 1 D
Nu(x) = p 1+ , (9.42)
1 + 12.7 (ξ/8)(P r 2/3 − 1) 3 x

69
P. Ehrhard

ξ = (1.8 log (Re) − 1.5)−2 , (9.43)

• Integral fluxes at control volume


assumptions: incompressible, steady, all properties constant, Fourier fluid,

mass: Z
ρ(w
~ · ~n)dA + ḣ = 0 , (9.44)
A

momentum: Z
F~i −
X
ρw(
~ w~ · ~n)dA = 0 , (9.45)
i A

heat: Z Z
ρcp T (w
~ · ~n)dA − λ (∇T · ~n)dA + Q̇ext = 0 , (9.46)
A A

• Differential equations
assumptions: incompressible, all properties constant, Newtonian & Fourier
fluid, dissipation neglected, for all coordinates,

continuity eqn.:
∇w
~ =0, (9.47)

Navier–Stokes eqs.:
 
∂w
~
ρ +w
~ · ∇w ~ + ρf~ ,
~ = −∇p + µ∆w (9.48)
∂t

heat transport eqn.:


 
∂T
ρcp +w
~ · ∇T = λ∆T + ρq̇ , (9.49)
∂t

• Boundary–layer equations
additional assumptions: flat plate, outer flow parallel to plate, 2D, steady, no
force term, no heat source term,

70
Fluid Mechanics and Heat Transport

continuity eqn.:
∂u ∂v
+ =0, (9.50)
∂x ∂y
Navier–Stokes eqs.:
∂2u
 
∂u ∂u
ρ u +v ≃µ 2 , (9.51)
∂x ∂y ∂y

p ≃ p∞ , (9.52)
heat transport eqn.:
∂2T
 
∂T ∂T
ρcp u +v ≃λ , (9.53)
∂x ∂y ∂y 2

• Reynolds-averaged differential equations


additional assumptions: isotropic turbulence, mean flow and T-field steady,

continuity eqn.:
∇w
~ =0, (9.54)
Navier–Stokes eqs.:

~ + ρf~ + ∇Ttur ,
 
ρ w~ · ∇w
~ = −∇p + µ∆w (9.55)

 
(u′2 ) (u′ v ′ ) (u′ w ′ )
Ttur = −ρ  (v ′ u′ ) (v ′2 ) (v ′ w ′)  , (9.56)
 

(w ′ u′ ) (w ′v ′ ) (w ′2 )
heat transport eqn.:

ρcp w
 ~ ,
~ · ∇T = λ∆T + ρq̇ + ∇Q̇ (9.57)
tur

 
(u′ T ′ )
~
Q̇ = −ρcp  (v ′ T ′ )  ;
 
(9.58)
tur

(w ′ T ′ )

71
10 Nomenclature

A cross–sectional area,
c velocity,
cp specific heat,
D (inner) pipe diameter,
e specific inner energy (per unit mass),
g gravitational acceleration,
ḣ mass source,
p
i=e+ ρ
specific enthalpy (per unit mass),
ks depth of roughness,
L pipe length,
ṁ mass flux,
Nu Nusselt number,
Pe Péclet number,
p pressure,
Pr Prandtl number,
q added heat (per unit mass),
q̇ added heat flux (per unit mass),
q̇∞ radiative flux within ambient,
q̇w wall heat flux (per unit area),
q̇cond conductive heat flux (per unit area),
q̇conv convected heat flux (per unit area),
q̇em emitted heat flux by radiation (per unit area),
Q̇ added heat flux,
Q̇tur turbulent heat flux,
R
R= m specific gas constant,
R gas constant,
Re Reynolds number,
s tangential coordinate (flow tube),
sref coordinate of reference point,
t time,
T temperature,
Tw wall temperature,

72
Fluid Mechanics and Heat Transport

(u, v, w) velocity components,


(x, y, z) cartesian coordinates,
z vertical coordinate,

f~ specific forces (per unit mass),


F~i forces acting at control volume,
~n normal vector (pointing out of control volume),
w~ velocity vector,

α heat transfer coefficient,


ǫem emissivity of body,
ǫab absorptivity of body,
κ = λ/(ρcp ) thermal diffusivity,
λ heat conductivity,
λlam ,λtur pressure loss coefficients,
µ dynamic viscosity,
ν = µ/ρ kinematic viscosity,
ρ density,
σ Stefan–Boltzmann constant,
Ttur turbulent stress tensor,

sub– and superscripts:


◦m cross-sectional average,
¯◦ temporal average,
◦′ fluctuating quantity.

73

Das könnte Ihnen auch gefallen