Vous êtes sur la page 1sur 51

Chapter 3

Solutions of Newtonian
Viscous-Flow Equations
Introduction and Classification of Solutions (1/2)
 The equations of viscous flow derived in Chap. 2 are a system of
nonlinear partial differential equations. No general analytical method yet
exists for attacking this system for an arbitrary viscous-flow problem.
 Over past 150 years, a considerable number of exact but particular
solutions have been found which satisfy the complete equations for some
special geometry, many of which are very illuminating about viscous-
flow phenomena.
 Almost all the known particular solutions are for the case of
incompressible Newtonian flow with constant transport properties.
DV
divV = 0; ρ = ∇ˆp + µ∇ 2V , where the total hydrostatic pressure p̂ = p + ρ gz;
Dt
DT
ρcp = k ∇ 2T + Φ
Dt
 Mass and momentum Equations are uncoupled from the temperature and
thus can be solved by V and p independently of T, after which T can be
solved from the energy equation. Note that T itself is not independent of
p or V, since the velocity, which may be pressure-dependent, enters the
energy equation through the terms Φ and (V⋅∇)(T). Here, we consider T
to be very important, both practically and pedagogically.
Introduction and Classification of Solutions (2/2)
 Basically, there are two types of exact solutions of the momentum
equation:
1. Linear solutions, where the convective term V⋅∇ vanishes
2. Nonlinear solutions, where V⋅∇ exist
 It is also possible to classify solutions by the type or geometry of
flow involved:
√ 1. Couette (wall-driven) steady flows
√ 2. Poiseuille (pressure-driven) steady duct flows
√ 3. Unsteady duct flows
4. Unsteady flows with moving boundaries
5. Duct flows with suction or injection
6. Ekman (wind-driven) flows
√ 7. Similarity solutions (rotating disk, stagnation flow, etc.)
 These are the main topics of this chapter, after which we conclude
with the creeping-flow approximation and digital-computer solutions.
Couette Flows (1/7) In honor of M.F.A. Couette (1890)

Steady Flow between a Fixed- and a Moving-Plate (1/3)


Assumptions:
1. steady flow & v = w = 0
2. the total hydrostatic pressure is constant
3. boundary conditions are independent of x or z
(infinite plate)
4. the upper plate moves at a speed U relative to the lower
5. two infinite plates are 2h apart
6. the upper plate is held at temperature T1 and the lower plate at T0
7. constant fluid properties
∂u
Boundary Conditions: Continuity : =0
∂x
1. no slip & no temperature jump d 2u
2. u(−h) = 0 & u(+h) = U Momentum : 0=µ 2
dy
3. T(−h) = T0 & T(+h) = T1 2
∴ the velocity distribution is d 2
T  du 
Energy : 0=k +µ  
U  y dy 2  dy 
u= 1 + 
2  h
∵ buoyancy is neglected ∴ u is independent of T
The shear stress at any point in τ = µ  ∂u + ∂v  = µ du = µ U = const.
 
the flow. Thus the shear stress  ∂y ∂x  dy 2h
is constant throughout the fluid, as is the strain rate.
Couette Flows (2/7)
Steady Flow between a Fixed- and a Moving-Plate (2/3)

The dimensionless shear stress is usually defined in engineering flow as the friction coefficient
τ µ 1
Cf = = =
ρU 2 ρ gh Reh
1
2
However, Churchill (1988) points out that Reynolds number is unsuitable for this nonaccelerating flow, since
density does not play a part. He suggests that one should instead use the Poiseuille number:
2hτ
Po = C f Reh = =1
µU
Clearly, a unit Poiseuille number is more convenient than a varying friction coefficient.

With du/dy = U/2h and boundary conditions known, we can obtain from the energy equation the temperature
distribution:
 T1 + T0 T1 − T0 y  µU  y2 
2
T = +  + 1 − 2 
 2 2 h  8k  h 
The term in brackets [] represents the straight-line distribution which would arise due to pure conduction in
the fluid. The second (parabolic) term is the temperature rise due to viscous dissipation in the fluid.
If T is nondimensionalized by (T1−T0), the dimensionless dissipation parameter arises, the Brinkman
number: µU 2 µcp U2
Br = = = PrEc
k (T1 − T0 ) k c p (T1 − T0 )
Qualitatively, it represents the ratio of viscous effects to fluid conduction effects.
Couette Flows (3/7)
Steady Flow between a Fixed- and a Moving-Plate (3/3)

For low-speed flows, only the most viscous fluids (oil) have significant Brinkman numbers. For example,
take U = 10 m/s and (T1−T0) = 10°C and compare the following numerical values for four fluid:
Fluid µ (kg/m⋅s) k (W/m⋅°C) Brinkman number
Air 1.8 E −5 0.26 7 E −4
Water 1.0 E −3 0.60 1.7 E −2
Mercury 1.54 E −3 8.7 1.8 E −3
SAE 30 Oil 0.29 0.145 20.0
Thus, except for heavy oils, we commonly neglect dissipation effects in low-speed flow temperature analyses.
∂T k µU 2
The rate of heat transfer at the walls: qw = k = (T1 − T0 ) ± ,
∂y ±h
2h 4h
where the (±) refers to lower- and upper-surface, respectively. The first term on the right represents pure
conduction through fluid.
Since many convection analyses result in qw being proportional to ∆T, it is customary to refer their ratio as a
heat transfer coefficient: ξ = qw
T1 − T0 ξL ξ 2h Br
One then nondimensionalize as the Nusselt number: Nu L = = C h ReL Pr = = 1 ± ,
k k 2
where L is the characteristic length of the flow geometry and the (+) means the lower surface.
If Br > 2, both the upper and lower surfaces must be cooled to maintain their temperatures. Note that, since
Nu ≈ 1 for pure conduction, the numerical value of Nu represents the ratio of convective heat transfer to
conduction for the same value of ∆T.
Couette Flows (4/7)
Axially Moving Concentric Cylinders

Assumptions:
1. steady flow & ur = uθ = 0
2. the temperature and pressure are constant
3. boundary conditions are independent of θ or z
4. the inner (r = r0) cylinder moves at a u = U0
or the outer (r = r1) moves at u = U1
5. constant fluid properties

Boundary Conditions: 1 ∂  ∂u 
Axial Momentum : 0= r
no slip r ∂r  ∂r 

u(r0) = U0 or u(r1) = U1
ln ( r1 r ) ln ( r r0 )
∴ the velocity distribution is u = U0 or u = U1
ln ( r1 r0 ) ln ( r1 r0 )
− µU 0 µU1
The shear stress at any point in the flow τ= or τ=
r ln ( r1 r0 ) r ln ( r1 r0 )
Note the difference in curvature for the two cases: If the inner cylinder moves, u(r) is concave,
whereas it is convex if the outer cylinder moves.
Couette Flows (5/7)
Flow between Rotating Concentric Cylinders (1/2)

Consider the steady flow maintained between two concentric cylinders by steady angular velocity of one or
both cylinders.
Andereck et al. 1986
Assumptions:
1. The inner cylinder has radius r0,
angular velocity ω0, and temperature T0.
2. The outer cylinder has radius r1,
angular velocity ω1, and temperature T1.
3. Boundary conditions are independent of θ or z.
4. steady flow, ur = uz = 0 & p = p(r)
5. constant fluid properties

Boundary Conditions:
r = r0 : uθ = r0ω0 T = T0 p = p0 ∂uθ
Continuity =0
∂θ
r = r1 : uθ = r1ω1 T = T1
dp ρ uθ2
r momentum =
The solution to the momentum equation: dr r
r r − r r1 r r 0 − r0 r d 2 uθ d  uθ 
θ momentum +  
uθ = r0ω0 1 + r1ω1 dr 2 dr  r 
r1 r0 − r0 r1 r1 r0 − r0 r1 2
k d  dT   du u 
Energy 0=  r  +µ θ − θ 
r dr  dr   dr r 
Couette Flows (6/7)
Flow between Rotating Concentric Cylinders (2/2)

If this velocity distribution is substituted into the energy equation, the temperature distribution may be found
r14 (1 − ω1 ω0 )  r02   ln ( r r0 )  ln ( r r0 )
2
T − T0
as = Pr Ec  1 −  1 −  +
 r 2   ln ( r1 r0 )  ln ( r1 r0 )
where Pr Ec = µ r0 ω0  k (T1 − T0 )  = Br
2 2
T1 − T0 r14 − r04   
If PrEc = 0, T reduces to the simple heat-conduction. Note that Br is the ratio of viscous effects to conduction
effects. The evaluation of the pressure p(r) is left as HW 1.

Some Special Cases:


1. r0 = ω0 = 0 ⇒ uθ = ω1r = const.
The steady outer rotation of a tube filled with fluid induces rigid-body rotation.

2. ω1 = 0 & r1 → ∞ ⇒ uθ = r02ω0 r
This is a potential vortex, driven in a viscous fluid by the no-slip condition.
Since the total fluid momentum is infinite, it would take infinite time to
generate this flow by steady rotation of the inner cylinder. See figures.

3. r1 − r0  r0 & ω1 = 0 ⇒ uθ ( r0ω0 ) ≈ 1 − ( r − r0 r1 − r0 )
This is a linear Couette flow between effectively parallel plates.

Of interest in viscometry is the moment or torque exerted by the cylinders upon


each other. This moment is independent of r and has the value (per unit depth
(
of cylinder) of M = 4πµ (ω1 − ω0 ) r12 r02 r12 − r02 )
Couette Flows (7/7)
Stability of Couette Flows

All of the solution previously mentioned are exact steady-flow solutions


of the Navier-Stokes equations. They are called laminar flows and have
a smooth-streamline character. It is known that all laminar flows become
unstable at a finite value of some critical parameter, usually the Reynolds
number. A different type of flow then ensues, sometimes still laminar or,
more often, an entirely new fluctuating flow regime called turbulent flows.

The laminar straight-line profile of flow between plates, is valid only up to Reh = Uh/ν ≈ 1500. Above this
value, the pattern changes into a randomly fluctuating flow which has a time-mean S-shaped profile
(Reichardt 1956). This S-shaped profile varies slightly with Reynolds number and increases wall shear and
heat transfer rate by two orders of magnitude.

For example, the axial annular flows are unstable. Experiments (Polderman et al.1986), with the inner
cylinder moving, show fully turbulent flow at a clearance Reynolds number Rec = U0(r1–r0)/ν ≥ 7000. The
probable transition from laminar to turbulent flow is about Rec ≈ 2000. Also, the laminar profiles of rotating-
inner-cylinder flows were valid until a critical rotation rate (Taylor 1923). For r1 − r0  r0 , the critical value

for instability is given by the Taylor number: 3
Ta = r0 ( r1 − r0 ) 0
≈ 1700
ν 2

All laminar flows are subject to instability. Therefore all the exact steady-flow solutions are, in principle,
valid only for a certain finite range of their Reynolds number.
Poiseuille Flows (1/4)
Fully Developed Flows

Consider a straight duct of arbitrary but constant shape. There will be an entrance effect, i.e., a thin initial shear layer and
core acceleration. The shear layers grow and meet, and the core disappears, within a fairly short entrance length Le:
Le
≈ C1 + C2 ReDh where C1 ≈ 0.5, C2 ≈ 0.05, and Dh is a suitable diameter scale for the duct.
Dh
For x > Le, the velocity becomes purely axial and steady, and varies only with the lateral coordinates, that is, v = w = 0 and
u = u(y, z). The flow is then called fully developed. For fully developed flow, the continuity and momentum equations for
incompressible flow reduce to ∂p̂  ∂ 2u ∂ 2 u 
Momentum 0=− +µ 2 + 2 
∂x  ∂y ∂z 

∂u ∂?p ∂p
Continuity =0 0=− =−
∂x ∂y ∂z
These indicate that the total hydrostatic pressure p̂ is a function only of x for this fully developed flow. Furthermore, since u
does not vary with x, dpˆ dx must only be a (negative) constant, that is, ∂ 2u + ∂ 2u = 1 dpˆ = const.
∂y 2 ∂z 2 µ dx
This is the basic equation of fully developed duct flow, the classic Poission equation, subject only to the no-slip condition
uw = 0 everywhere on the duct surface.

Note that the acceleration terms vanish here, taking the density with them. These flow then
are true creeping flows, since they are independent of density, even though the Reynolds
number is not small. There is no characteristic velocity, and no axial length scale either,
since we are supposed to be far from the entrance or exit. The proper scaling should include
µ, dpˆ dx, and some characteristic duct width h. Thus, the dimensionless variables and
µu
( )
equation are y z
y∗ = z∗ = u∗ = 2 ⇒ ∇∗2 u∗ = −1
h h h ( − dp
ˆ dx )
Poiseuille Flows (2/4)
Hagen-Poiseuille Circular Pipe Flows (1/2)

Assumptions:
1. fully developed flow & ur = uθ = 0 ⇒ uz = u = u(r)
2. boundary conditions are independent of θ or z (infinite plate)
3. two infinite plates are 2b apart
4. constant fluid properties

Boundary Conditions:
1. u(R) = 0 no-slip
2. ( ∂u ∂r ) r =0 = 0 ( ds )2 = ( dx )2 + ( dy )2 + ( dz )2 = ( h1dx1 )2 + ( h2 dx2 )2 + ( h3 dx3 )
2
finite value velocity
1  ∂  h2 h3 ∂  ∂  h3 h1 ∂  ∂  h1h2 ∂   1  ∂  ∂u  ∂  1 ∂u  ∂  ∂u  
∵ ∇2 =   +  +  ∴∇2u =   r  +
  + r 
h1h2 h3  ∂x1  h1 ∂x1  ∂x2  h2 ∂x2  ∂x3  h3 ∂x3  r  ∂r  ∂r  ∂θ  r ∂θ  ∂z  ∂z  
1  ∂  ∂  ∂  1 ∂  ∂  ∂  1 d  du 
=  r +   +  r  = r 
r  ∂r  ∂r  ∂θ  r ∂θ  ∂z  ∂z   Coordinate Transformation r dr  dr 

ˆ µ d  du 
dp ( −dpˆ dz )
Momentum 0=− +  r
dz r dr  dr 
∴ u=

(R 2
− r2 )

Thus, the velocity distribution in fully developed laminar pipe flow is a paraboloid of revolution about the
centerline (the Poiseuille paraboloid).
R π R 4  dpˆ 
The total volume rate of flow Q: Q=∫ udA = ∫ u ( 2π rdr ) = − 
section 0 8µ  dz 
Poiseuille Flows (3/4)
Hagen-Poiseuille Circular Pipe Flows (2/2)

Finally, the mean velocity and wall shear stress can be obtained:
Q R ( −dpˆ dz ) 1  du  R  dpˆ  4π u
2

u= = = umax τ w = µ  −  = −  =
A 8µ 2  dr  w 2  dz  R
Even though τw is proportional to mean velocity (laminar flow), it is customary, anticipating turbulent flow,
to nondimensionalize wall shear stress with the pipe dynamic pressure, ρ u 2 .
2

Two different friction factor definitions are in common use in the literature:
8τ w
λ= = Darcy friction factor
ρu 2
2τ λ
C f = w2 = = Fanning friction factor or skin-friction coefficient
ρu 4

By substitution of τw, we can obtain the classic relations:


64 16
λ= Cf = where ReD = ρ uD µ & D = Dh = 2R
ReD ReD
As with Couette flow, the Poiseuille number makes a lot of sense in laminar tube flow, being a pure constant:
2τ D
Po = C f ReD = w = 16
µu
This classic laminar flow solution is in good agreement with experiment, as shown in the figure (Senecal &
Rothfus 1953). The flow undergoes transition to turbulence at approximately ReD ≈ 2000, a value which can
be raised somewhat by taking care to eliminate flow disturbances.
Poiseuille Flows (4/4)
Combined Couette-Poiseuille Flow between Plates
ˆ
dp d 2u
Governing equation: = µ 2 = const.
dx dy
Boundary Conditions:
no-slip: u(+h) = U & u(−h) = 0

u 1 y  y2   dpˆ  h
2
∴ = 1 +  + P 1 − 2  where P =  − 
U 2 h   h   dx  2µU

This is a superposition solution, possible because the nonlinear convective acceleration is zero, of Couette wall-driven flow
(the first term) and Poiseuille pressure-driven flow (the second term). The figure below plots the dimensionless velocity
profiles for various values of the dimensionless pressure gradient P.
Of particular interest is the dash line (P = − 0.25), for which the shear stress µ(du/dy) at the lower wall is zero. For P < − 0.25,
there is a backflow at the lower wall, an indication of flow separation in unbounded shear layers. In terms of the thickness
(2h) of the shear layer, we may write the separation criterion P = − 0.25 in the form dpˆ
( dx ) ( 2h ) µU = 2
2

This is identical in form to the laminar boundary layer separation estimates to be discussed in Chap. 4.

If U = 0 (fixed walls), the solution reduces to pure Poiseuille flow between parallel
plates:  y2   dpˆ  h
2
u = umax 1 − 2  where umax =  − 
 h   dx  2 µ
Similarly, +h 4 Q 2
Q = ∫ udy = humax u= = umax
−h 3 2h 3
6µ 6
Cf = = Po = C f Reh = 6
ρ uh Reh
The Concept of Hydraulic Diameter
For a noncircular duct, the shear stress τw varies around the perimeter. For example, in the equilateral-triangle duct, τw is
zero in the corners and a maximum at the midpoints of the sides.
Therefore we have to define a mean wall shear stress: 1 P
τw =
P ∫ 0
τ w ds
Note that there is no net momentum flux due to the fully developed flow, thus we can equate the net pressure and wall shear
force on the fluid:
ˆ ⇒ τ w =  − 
P A dp ˆ

dx τ w ds = A −dp
0
( )
P  dx 
Compared with Hagen-Poiseuille circular pipe flow:

 du  R  dp
ˆ  A R
τw = µ −  = −  ⇒ =
 dr  w 2  dz  P 2
Thus, for a noncircular duct flow we set :
A 4 × area ( subject to pressure force )
Dh = 4 =
P wetted perimeter ( subject to viscous force )
(
Take the concentric annulus for example, Dh = 4π a 2 − b 2 ) ( 2π a + 2π b ) = 2 ( a − b )
By dimensional reasoning for laminar fully developed flow, we are guaranteed that the fiction
factor of a noncircular duct will vary inversely with Reynolds number based on hydraulic
diameter: C = λ = const. ρ uDh
where Re =
f
4 Reh
h
µ
However, this is the critical flaw. The constant usually does not equal 16 as it did for a circular
pipe. Especially vexing is the fact that the hydraulic diameter concept is not sensitive to
eccentricity as shown in the figure.
Unsteady Flows with Moving Boundaries (1/5)
Assumptions: y
1. two dimensional parallel-flow
v = w = 0 & ∂ /∂z = 0 ⇒ u = u(y, t)
2. the semi-infinite plate is at y = 0
3. the fluid is still for y ≥ 0 & t ≤ 0 x
4. constant fluid properties
5. boundary conditions are independent of x or z ∂u
Continuity =0
Boundary Conditions: ∂ x
1. no slip: u(0, t) = U(t) for t > 0 ∂u ˆ
1 dp ∂ 2u
2. no initial motion: u(y, 0) = 0 for y > 0 x-Momentum =− +ν 2
∂t ρ dx ∂y
Here, we confine ourselves to two cases:
dˆp
1. sudden acceleration of the plane to constant U0 y -Momentum =0
suddenly started or stopped plate (Stokes’ first problem) d y
2. steady oscillation of the plane at U cos(ωt) (Stokes’ second problem)
For this problem, as y → ∞, u should approach to zero velocity for t > 0, thus
∂u ∂u ∂ 2u From the governing equations (continuity & x-
= = 2 =0 momentum), since none of the terms of them is a
∂t y →∞ ∂y y →∞
∂y y →∞
function of x, we know that the pressure gradient
Apply this to x-momentum equation: can only be a function of time, i.e. at a certain time
∂u 1 dpˆ ∂ 2u its value is not a function of position, or its value is
⇒ =− +ν 2 the same at any point of the flow domain.
∂t y →∞ ρ dx y →∞ ∂y y →∞
Thus, dpˆ dpˆ

dpˆ
=0
=0 ⇒ =0
dx dx y →∞ dx
y →∞
Unsteady Flows with Moving Boundaries (2/5)
∂u ∂ 2u
Thus the momentum equation becomes ⇒ Momentum =ν 2
Then we try to change this PDE into a ODE ∂t ∂y
∂ 2u ∂u
∵ ν = const. ∴ = in a dimensional concept
∂y 2
∂ (ν t )
Therefore, we combine the two original independent variables into a new one. Furthermore,
we want the new variable to be dimensionless. In this way, if we also set a new dependent
variable to replace the old dependent variable u, the equation will be dimensionless.
Thus, define η = yα (ν t ) β u
& = f (η )  ∂u   F   1 
τ
U0  µ   ∂y   L2   T  FL−2T 1 L2
From the dimensional analyses, we should take: [ν ] =   = = = −4 2 =
1 ρ
  [ ρ ]  F 
L 3 FL T T
α =1 & β =−  2
L T 

2
Thus, the single dimensionless similarity variable: [ν t ] = L2
η=
y
&
u
= f (η ) [ y] = L
2 νt U0 The coefficient 2 in the denominator is taken to
Therefore make the final ODE more easier to be integrated.
 ∂u df ∂η df η
 = U0 = −U 0
 ∂η 1 y −η  ∂t dη ∂t dη 2t
 ∂t = − =
 2t 2 ν t 2t  ∂u df ∂η df 1
 ⇒  = U0 = U0
 ∂η 1  ∂y dη ∂y dη 2 ν t
=
 ∂y 2 νt  ∂ 2u d2 f 1
 = U0
 ∂y 2 dη 2 4ν t
Unsteady Flows with Moving Boundaries (3/5)
Therefore, the dimensionless momentum equation becomes:
df η d2 f 1 d2 f df
Momentum −U 0 = νU0 ⇒ + 2η =0
dη 2t dη 2 4ν t dη 2 dη
Similarly, the dimensionless boundary conditions becomes f (η = 0) = 1 f (η → ∞) = 0
Thus
d ( f ′) d ( f ′)
∴ f ′ = Ae−η
2
+ 2η f ′ = 0 ⇒ = −2η dη ⇒ n f ′ = −η 2 + C
dη f′
Integrating this equation again, we can obtain f = A∫ e −η 2
dη + B
Then, we can solve the coefficients using boundary conditions
0
⇒ 1 = A∫ e −η dη + B ∴ B = 1
2
∵ f (η = 0) = 1
0

( )
∞ ∞
f (η → ∞) = 0 ⇒ 0 = A∫ e −η dη + 1 ∫
2 2
∵ ∴ A = −1 e − z dz = −1 π 2
0 0

2 2 η
∫e ∫
−η 2 2
∴ f (η ) = 1 − dη or f (η ) = 1 − e − z dz = 1 − erf (η )
π π 0

The solution is well known to erf (η ) erfc(η ) = 1 − erf (η )


be the complementary error function
or probability integral
u
= f (η ) = 1 − erf (η ) ≡ erfc (η )
U0 η η
Unsteady Flows with Moving Boundaries (4/5)
These two complementary solutions (mirror images) are shown in Figures.
The wall shear stress can thus be obtained
du df (η ) dη  2 −η 2  1  −U 0 −η 2
= U0 = U0  − e   = e
dy dη dy  π  2 ν t  πν t
du µU 0
τw = µ =−
dy y =0 πν t
In either case, the plate’s effect diffuses into the fluid at a rate u
proportional to the square root of kinematic viscosity. U0
It is customary to define the shear layer thickness as the point
where the wall effect on the fluid has dropped to 1 percent:
u
= 0.01 for suddenly started case
U0
u
= 0.99 for suddenly stopped case
U0
u
Also, we can estimate the displacement thickness

U0
U 0 δ (t ) = ∫ u ( y, t )dy
0
∞ u ( y, t ) ∞ dy
δ (t ) = ∫ dy = ∫ erfc (η ) dη
0 U0 0 dη

= 2 ν t ∫ erfc (η ) dη δ (t )
0

 ∞ d 
= 2 ν t  η erfc (η ) ∞
−∫ η erfc (η )  dη  integrate by part

0
 0

Unsteady Flows with Moving Boundaries (5/5)
d (erfc(η )) df (η ) 2 −η 2
∵ = =− e Note that the thickness is a measurement of the wall
dη dη π effect. It depends on the kinematic viscosity, not the
∞  2 −η 2  dynamic viscosity. So that the strength of the
∴ δ ( t ) = −2 ν t ∫ η − e dη momentum transport in the fluid is in the order of:
0
 π  Oil > Air > Water
νt ∞
∫ 2η e −η dη
2
=2
π 0 Fluid δ (m)
µ (µPa⋅s) ν (m2/s)
ν t ∞ −η @ 20 °C @ t = 10 s

π ∫0
( )
2
=2 e d η 2
Air 1.8 E 1 1.9 E -5 0.016
Water 1.0 E 3 1.1 E -6 0.003
νt νt
( )

−e −η
2
=2 =2 SAE 30 Oil 2.8 E 5 2.4 E -4 0.086
π 0 π
Similarly, the thermal thickness can be obtained if we impose a temperature to the fluid.
αt k
δ T (t ) = 2 where α = Fluid Air Water SAE 30 Oil
π ρC p
Pr 0.72 7.0 3500
δ (t ) ν
∴ = = Pr Note that the Prandtl number is the ratio of the viscous
δ T (t ) α diffusion rate to the thermal diffusion rate. Therefore, the
strength of the thermal energy transport in the fluid is in the
order of: Oil > Water > Air
Homework
 3-15 free surface flow
 3-16 axially Couette flow between concentric cylinder
 3-18 unstable free surface flow
 3-21 azimuthal Couette flow between concentric
cylinder
 3-19 suddenly started flow
 3-23 radial outflow between two circular plates
 3-28 Ekman spiral
 3-38 noncircular duct flow
Short Firms Lecturer: G.I. Taylor
Deformation of Fluid Element (Re << 1) Rigid Body Rotation

Motion in Very Viscous Fluid Reynolds Number Effects


Chapter 3

Solutions of Newtonian
Viscous-Flow Equations
Introduction and Classification of Solutions (1/2)
 The equations of viscous flow derived in Chap. 2 are a system of
nonlinear partial differential equations. No general analytical method yet
exists for attacking this system for an arbitrary viscous-flow problem.
 Over past 150 years, a considerable number of exact but particular
solutions have been found which satisfy the complete equations for some
special geometry, many of which are very illuminating about viscous-
flow phenomena.
 Almost all the known particular solutions are for the case of
incompressible Newtonian flow with constant transport properties.
DV
divV = 0; ρ = ∇ˆp + µ∇ 2V , where the total hydrostatic pressure p̂ = p + ρ gz;
Dt
DT
ρcp = k ∇ 2T + Φ
Dt
 Mass and momentum Equations are uncoupled from the temperature and
thus can be solved by V and p independently of T, after which T can be
solved from the energy equation. Note that T itself is not independent of
p or V, since the velocity, which may be pressure-dependent, enters the
energy equation through the terms Φ and (V⋅∇)(T). Here, we consider T
to be very important, both practically and pedagogically.
Introduction and Classification of Solutions (2/2)
 Basically, there are two types of exact solutions of the momentum
equation:
1. Linear solutions, where the convective term V⋅∇ vanishes
2. Nonlinear solutions, where V⋅∇ exist
 It is also possible to classify solutions by the type or geometry of
flow involved:
√ 1. Couette (wall-driven) steady flows
√ 2. Poiseuille (pressure-driven) steady duct flows
√ 3. Unsteady duct flows
4. Unsteady flows with moving boundaries
5. Duct flows with suction or injection
6. Ekman (wind-driven) flows
√ 7. Similarity solutions (rotating disk, stagnation flow, etc.)
 These are the main topics of this chapter, after which we conclude
with the creeping-flow approximation and digital-computer solutions.
Couette Flows (1/7) In honor of M.F.A. Couette (1890)

Steady Flow between a Fixed- and a Moving-Plate (1/3)


Assumptions:
1. steady flow & v = w = 0
2. the total hydrostatic pressure is constant
3. boundary conditions are independent of x or z
(infinite plate)
4. the upper plate moves at a speed U relative to the lower
5. two infinite plates are 2h apart
6. the upper plate is held at temperature T1 and the lower plate at T0
7. constant fluid properties
∂u
Boundary Conditions: Continuity : =0
∂x
no slip & no temperature jump d 2u
1. u(−h) = 0 & u(+h) = U Momentum : 0=µ 2
dy
2. T(−h) = T0 & T(+h) = T1 2
∴ the velocity distribution is d 2
T  du 
Energy : 0=k +µ  
U  y dy 2  dy 
u= 1 + 
2  h
∵ buoyancy is neglected ∴ u is independent of T
The shear stress at any point in τ = µ  ∂u + ∂v  = µ du = µ U = const.
 
the flow. Thus the shear stress  ∂y ∂x  dy 2h
is constant throughout the fluid, as is the strain rate.
Couette Flows (2/7)
Steady Flow between a Fixed- and a Moving-Plate (2/3)

The dimensionless shear stress is usually defined in engineering flow as the friction coefficient
τ µ 1
Cf = = =
ρU 2 ρ gh Reh
1
2
However, Churchill (1988) points out that Reynolds number is unsuitable for this nonaccelerating flow, since
density does not play a part. He suggests that one should instead use the Poiseuille number:
2hτ
Po = C f Reh = =1
µU
Clearly, a unit Poiseuille number is more convenient than a varying friction coefficient.

With du/dy = U/2h and boundary conditions known, we can obtain from the energy equation the temperature
distribution:
 T1 + T0 T1 − T0 y  µU  y2 
2
T = +  + 1 − 2 
 2 2 h  8k  h 
The term in brackets [] represents the straight-line distribution which would arise due to pure conduction in
the fluid. The second (parabolic) term is the temperature rise due to viscous dissipation in the fluid.
If T is nondimensionalized by (T1−T0), the dimensionless dissipation parameter arises, the Brinkman
number: µU 2 µcp U2
Br = = = PrEc
k (T1 − T0 ) k c p (T1 − T0 )
Qualitatively, it represents the ratio of viscous effects to fluid conduction effects.
Couette Flows (3/7)
Steady Flow between a Fixed- and a Moving-Plate (3/3)

For low-speed flows, only the most viscous fluids (oil) have significant Brinkman numbers. For example,
take U = 10 m/s and (T1−T0) = 10°C and compare the following numerical values for four fluid:
Fluid µ (kg/m⋅s) k (W/m⋅°C) Brinkman number
Air 1.8 E −5 0.26 7 E −4
Water 1.0 E −3 0.60 1.7 E −2
Mercury 1.54 E −3 8.7 1.8 E −3
SAE 30 Oil 0.29 0.145 20.0
Thus, except for heavy oils, we commonly neglect dissipation effects in low-speed flow temperature analyses.
∂T k µU 2
The rate of heat transfer at the walls: qw = k = (T1 − T0 ) ± ,
∂y ±h
2h 4h
where the (±) refers to lower- and upper-surface, respectively. The first term on the right represents pure
conduction through fluid.
Since many convection analyses result in qw being proportional to ∆T, it is customary to refer their ratio as a
heat transfer coefficient: ξ = qw
T1 − T0 ξL ξ 2h Br
One then nondimensionalize as the Nusselt number: Nu L = = C h ReL Pr = = 1 ± ,
k k 2
where L is the characteristic length of the flow geometry and the (+) means the lower surface.
If Br > 2, both the upper and lower surfaces must be cooled to maintain their temperatures. Note that, since
Nu ≈ 1 for pure conduction, the numerical value of Nu represents the ratio of convective heat transfer to
conduction for the same value of ∆T.
Couette Flows (4/7)
Axially Moving Concentric Cylinders
Assumptions:
1. steady flow & ur = uθ = 0
2. the temperature and pressure are constant
3. boundary conditions are independent of θ
or z
4. the inner (r = r0) cylinder moves at a u = U0
or the outer (r = r1) moves at u = U1
5. constant fluid properties

Boundary Conditions: 1 ∂  ∂u 
Axial Momentum : 0= r
no slip r ∂r  ∂r 

u(r0) = U0 or u(r1) = U1
ln ( r1 r ) ln ( r r0 )
∴ the velocity distribution is u = U 0 or u = U1
ln ( r1 r0 ) ln ( r1 r0 )
− µU 0 µU1
The shear stress at any point in the flow τ = or τ=
r ln ( r1 r0 ) r ln ( r1 r0 )
Note the difference in curvature for the two cases: If the inner cylinder moves, u(r) is
concave, whereas it is convex if the outer cylinder moves.
Couette Flows (5/7)
Flow between Rotating Concentric Cylinders (1/2)

Consider the steady flow maintained between two concentric cylinders by steady angular velocity of one or
both cylinders.
Andereck et al. 1986
Assumptions:
1. The inner cylinder has radius r0,
angular velocity ω0, and temperature T0.
2. The outer cylinder has radius r1,
angular velocity ω1, and temperature T1.
3. Boundary conditions are independent of θ or z.
4. steady flow, ur = uz = 0 & p = p(r)
5. constant fluid properties

Boundary Conditions:
r = r0 : uθ = r0ω0 T = T0 p = p0 ∂uθ
Continuity =0
∂θ
r = r1 : uθ = r1ω1 T = T1
dp ρ uθ2
r momentum =
The solution to the momentum equation: dr r
r r − r r1 r r 0 − r0 r d 2 uθ d  uθ 
θ momentum +  
uθ = r0ω0 1 + r1ω1 dr 2 dr  r 
r1 r0 − r0 r1 r1 r0 − r0 r1 2
k d  dT   du u 
Energy 0=  r  +µ θ − θ 
r dr  dr   dr r 
Couette Flows (6/7)
Flow between Rotating Concentric Cylinders (2/2)

If this velocity distribution is substituted into the energy equation, the temperature distribution may be found
r14 (1 − ω1 ω0 )  r02   ln ( r r0 )  ln ( r r0 )
2
T − T0
as = Pr Ec  1 −  1 −  +
 r 2   ln ( r1 r0 )  ln ( r1 r0 )
where Pr Ec = µ r0 ω0  k (T1 − T0 )  = Br
2 2
T1 − T0 r14 − r04   
If PrEc = 0, T reduces to the simple heat-conduction. Note that Br is the ratio of viscous effects to conduction
effects. The evaluation of the pressure p(r) is left as HW 1.

Some Special Cases:


1. r0 = ω0 = 0 ⇒ uθ = ω1r = const.
The steady outer rotation of a tube filled with fluid induces rigid-body rotation.

2. ω1 = 0 & r1 → ∞ ⇒ uθ = r02ω0 r
This is a potential vortex, driven in a viscous fluid by the no-slip condition.
Since the total fluid momentum is infinite, it would take infinite time to
generate this flow by steady rotation of the inner cylinder. See figures.

3. r1 − r0  r0 & ω1 = 0 ⇒ uθ ( r0ω0 ) ≈ 1 − ( r − r0 r1 − r0 )
This is a linear Couette flow between effectively parallel plates.

Of interest in viscometry is the moment or torque exerted by the cylinders upon


each other. This moment is independent of r and has the value (per unit depth
(
of cylinder) of M = 4πµ (ω1 − ω0 ) r12 r02 r12 − r02 )
Couette Flows (7/7)
Stability of Couette Flows

All of the solution previously mentioned are exact steady-flow solutions


of the Navier-Stokes equations. They are called laminar flows and have
a smooth-streamline character. It is known that all laminar flows become
unstable at a finite value of some critical parameter, usually the Reynolds
number. A different type of flow then ensues, sometimes still laminar or,
more often, an entirely new fluctuating flow regime called turbulent flows.

The laminar straight-line profile of flow between plates, is valid only up to Reh = Uh/ν ≈ 1500. Above this
value, the pattern changes into a randomly fluctuating flow which has a time-mean S-shaped profile
(Reichardt 1956). This S-shaped profile varies slightly with Reynolds number and increases wall shear and
heat transfer rate by two orders of magnitude.

For example, the axial annular flows are unstable. Experiments (Polderman et al.1986), with the inner
cylinder moving, show fully turbulent flow at a clearance Reynolds number Rec = U0(r1–r0)/ν ≥ 7000. The
probable transition from laminar to turbulent flow is about Rec ≈ 2000. Also, the laminar profiles of rotating-
inner-cylinder flows were valid until a critical rotation rate (Taylor 1923). For r1 − r0  r0 , the critical value

for instability is given by the Taylor number: 3
Ta = r0 ( r1 − r0 ) 0
≈ 1700
ν 2

All laminar flows are subject to instability. Therefore all the exact steady-flow solutions are, in principle,
valid only for a certain finite range of their Reynolds number.
Poiseuille Flows (1/4)
Fully Developed Flows

Consider a straight duct of arbitrary but constant shape. There will be an entrance effect, i.e., a thin initial shear layer and
core acceleration. The shear layers grow and meet, and the core disappears, within a fairly short entrance length Le:
Le
≈ C1 + C2 ReDh where C1 ≈ 0.5, C2 ≈ 0.05, and Dh is a suitable diameter scale for the duct.
Dh
For x > Le, the velocity becomes purely axial and steady, and varies only with the lateral coordinates, that is, v = w = 0 and
u = u(y, z). The flow is then called fully developed. For fully developed flow, the continuity and momentum equations for
incompressible flow reduce to ∂p̂  ∂ 2u ∂ 2 u 
Momentum 0=− +µ 2 + 2 
∂x  ∂y ∂z 

∂u ∂?p ∂p
Continuity =0 0=− =−
∂x ∂y ∂z
These indicate that the total hydrostatic pressure p̂ is a function only of x for this fully developed flow. Furthermore, since u
does not vary with x, dpˆ dx must only be a (negative) constant, that is, ∂ 2u + ∂ 2u = 1 dpˆ = const.
∂y 2 ∂z 2 µ dx
This is the basic equation of fully developed duct flow, the classic Poission equation, subject only to the no-slip condition
uw = 0 everywhere on the duct surface.

Note that the acceleration terms vanish here, taking the density with them. These flow then
are true creeping flows, since they are independent of density, even though the Reynolds
number is not small. There is no characteristic velocity, and no axial length scale either,
since we are supposed to be far from the entrance or exit. The proper scaling should include
µ, dpˆ dx, and some characteristic duct width h. Thus, the dimensionless variables and
µu
( )
equation are y z
y∗ = z∗ = u∗ = 2 ⇒ ∇∗2 u∗ = −1
h h h ( − dp
ˆ dx )
Poiseuille Flows (2/4)
Hagen-Poiseuille Circular Pipe Flows (1/2)

Assumptions:
1. fully developed flow & ur = uθ = 0 ⇒ uz = u = u(r)
2. boundary conditions are independent of θ or z (infinite plate)
3. two infinite plates are 2b apart
4. constant fluid properties

Boundary Conditions:
1. u(R) = 0 no-slip
2. ( ∂u ∂r ) r =0 = 0 ( ds )2 = ( dx )2 + ( dy )2 + ( dz )2 = ( h1dx1 )2 + ( h2 dx2 )2 + ( h3 dx3 )
2
finite value velocity
1  ∂  h2 h3 ∂  ∂  h3 h1 ∂  ∂  h1h2 ∂   1  ∂  ∂u  ∂  1 ∂u  ∂  ∂u  
∵ ∇2 =   +  +  ∴∇2u =   r  +
  + r 
h1h2 h3  ∂x1  h1 ∂x1  ∂x2  h2 ∂x2  ∂x3  h3 ∂x3  r  ∂r  ∂r  ∂θ  r ∂θ  ∂z  ∂z  
1  ∂  ∂  ∂  1 ∂  ∂  ∂  1 d  du 
=  r +   +  r  = r 
r  ∂r  ∂r  ∂θ  r ∂θ  ∂z  ∂z   Coordinate Transformation r dr  dr 

ˆ µ d  du 
dp ( −dpˆ dz )
Momentum 0=− +  r
dz r dr  dr 
∴ u=

(R 2
− r2 )

Thus, the velocity distribution in fully developed laminar pipe flow is a paraboloid of revolution about the
centerline (the Poiseuille paraboloid).
R π R 4  dpˆ 
The total volume rate of flow Q: Q=∫ udA = ∫ u ( 2π rdr ) = − 
section 0 8µ  dz 
Poiseuille Flows (3/4)
Hagen-Poiseuille Circular Pipe Flows (2/2)

Finally, the mean velocity and wall shear stress can be obtained:
Q R ( −dpˆ dz ) 1  du  R  dpˆ  4π u
2

u= = = umax τ w = µ  −  = −  =
A 8µ 2  dr  w 2  dz  R
Even though τw is proportional to mean velocity (laminar flow), it is customary, anticipating turbulent flow,
to nondimensionalize wall shear stress with the pipe dynamic pressure, ρ u 2 .
2

Two different friction factor definitions are in common use in the literature:
8τ w
λ= = Darcy friction factor
ρu 2
2τ λ
C f = w2 = = Fanning friction factor or skin-friction coefficient
ρu 4

By substitution of τw, we can obtain the classic relations:


64 16
λ= Cf = where ReD = ρ uD µ & D = Dh = 2R
ReD ReD
As with Couette flow, the Poiseuille number makes a lot of sense in laminar tube flow, being a pure constant:
2τ D
Po = C f ReD = w = 16
µu
This classic laminar flow solution is in good agreement with experiment, as shown in the figure (Senecal &
Rothfus 1953). The flow undergoes transition to turbulence at approximately ReD ≈ 2000, a value which can
be raised somewhat by taking care to eliminate flow disturbances.
Poiseuille Flows (4/4)
Combined Couette-Poiseuille Flow between Plates
ˆ
dp d 2u
Governing equation: = µ 2 = const.
dx dy
Boundary Conditions:
no-slip: u(+h) = U & u(−h) = 0

u 1 y  y2   dpˆ  h
2
∴ = 1 +  + P 1 − 2  where P =  − 
U 2 h   h   dx  2µU

This is a superposition solution, possible because the nonlinear convective acceleration is zero, of Couette wall-driven flow
(the first term) and Poiseuille pressure-driven flow (the second term). The figure below plots the dimensionless velocity
profiles for various values of the dimensionless pressure gradient P.
Of particular interest is the dash line (P = − 0.25), for which the shear stress µ(du/dy) at the lower wall is zero. For P < − 0.25,
there is a backflow at the lower wall, an indication of flow separation in unbounded shear layers. In terms of the thickness
(2h) of the shear layer, we may write the separation criterion P = − 0.25 in the form dpˆ
( dx ) ( 2h ) µU = 2
2

This is identical in form to the laminar boundary layer separation estimates to be discussed in Chap. 4.

If U = 0 (fixed walls), the solution reduces to pure Poiseuille flow between parallel
plates:  y2   dpˆ  h
2
u = umax 1 − 2  where umax =  − 
 h   dx  2 µ
Similarly, +h 4 Q 2
Q = ∫ udy = humax u= = umax
−h 3 2h 3
6µ 6
Cf = = Po = C f Reh = 6
ρ uh Reh
The Concept of Hydraulic Diameter
For a noncircular duct, the shear stress τw varies around the perimeter. For example, in the equilateral-triangle duct, τw is
zero in the corners and a maximum at the midpoints of the sides.
Therefore we have to define a mean wall shear stress: 1 P
τw =
P ∫ 0
τ w ds
Note that there is no net momentum flux due to the fully developed flow, thus we can equate the net pressure and wall shear
force on the fluid:
ˆ ⇒ τ w =  − 
P A dp ˆ

dx τ w ds = A −dp
0
( )
P  dx 
Compared with Hagen-Poiseuille circular pipe flow:

 du  R  dp
ˆ  A R
τw = µ −  = −  ⇒ =
 dr  w 2  dz  P 2
Thus, for a noncircular duct flow we set :
A 4 × area ( subject to pressure force )
Dh = 4 =
P wetted perimeter ( subject to viscous force )
(
Take the concentric annulus for example, Dh = 4π a 2 − b 2 ) ( 2π a + 2π b ) = 2 ( a − b )
By dimensional reasoning for laminar fully developed flow, we are guaranteed that the fiction
factor of a noncircular duct will vary inversely with Reynolds number based on hydraulic
diameter: C = λ = const. ρ uDh
where Re =
f
4 Reh
h
µ
However, this is the critical flaw. The constant usually does not equal 16 as it did for a circular
pipe. Especially vexing is the fact that the hydraulic diameter concept is not sensitive to
eccentricity as shown in the figure.
Unsteady Flows with Moving Boundaries (1/5)
Assumptions:
1. two dimensional parallel-flow
v = w = 0 & ∂ /∂z = 0 ⇒ u = u(y, t)
2. the semi-infinite plate is at y = 0
3. the fluid is still for y ≥ 0 & t = 0
4. constant fluid properties
5. boundary conditions are independent of x or z ∂u
Continuity =0
Boundary Conditions: ∂ x
1. no slip: u(0, t) = U(t) = U0 for t > 0 ∂u ˆ
1 dp ∂ 2u
2. no initial motion: u(y, 0) = 0 for y > 0 x-Momentum =− +ν 2
∂t ρ dx ∂y
Here, we confine ourselves to two cases:
dˆp
1. sudden acceleration of the plane to constant U0 y -Momentum =0
suddenly started or stopped plate (Stokes’ first problem) d y
2. steady oscillation of the plane at U cos(ωt) (Stokes’ second problem)
For this problem, as y → ∞, u should approach to zero velocity for t > 0, thus
∂u ∂u ∂ 2u From the governing equations (continuity & x-
= = 2 =0 momentum), since none of the terms of them is a
∂t y →∞ ∂y y →∞
∂y y →∞
function of x, we know that the pressure gradient
Apply this to x-momentum equation: can only be a function of time, i.e. at a certain time
∂u 1 dpˆ ∂ 2u its value is not a function of position, or its value is
⇒ =− +ν 2 the same at any point of the flow domain.
∂t y →∞ ρ dx y →∞ ∂y y →∞
Thus, dpˆ dpˆ

dpˆ
=0
=0 ⇒ =0
dx dx y →∞ dx
y →∞
Unsteady Flows with Moving Boundaries (2/5)
∂u ∂ 2u
Thus the momentum equation becomes ⇒ Momentum =ν 2
Then we try to change this PDE into a ODE ∂t ∂y
∂ 2u ∂u
∵ ν = const. ∴ = in a dimensional concept
∂y 2
∂ (ν t )
Therefore, we combine the two original independent variables into a new one. Furthermore,
we want the new variable to be dimensionless. In this way, if we also set a new dependent
variable to replace the old dependent variable u, the equation will be dimensionless.
Thus, define η = yα (ν t ) β u
& = f (η )  ∂u   F   1 
τ
U0  µ   ∂y   L2   T  FL−2T 1 L2
From the dimensional analyses, we should take: [ν ] =   = = = −4 2 =
1 ρ
  [ ρ ]  F 
L 3 FL T T
α =1 & β =−  2
L T 

2
Thus, the single dimensionless similarity variable: [ν t ] = L2
η=
y
&
u
= f (η ) [ y] = L
2 νt U0 The coefficient 2 in the denominator is taken to
Therefore make the final ODE more easier to be integrated.
 ∂u df ∂η df η
 = U0 = −U 0
 ∂η 1 y −η  ∂t dη ∂t dη 2t
 ∂t = − =
 2t 2 ν t 2t  ∂u df ∂η df 1
 ⇒  = U0 = U0
 ∂η 1  ∂y dη ∂y dη 2 ν t
=
 ∂y 2 νt  ∂ 2u d2 f 1
 = U0
 ∂y 2 dη 2 4ν t
Unsteady Flows with Moving Boundaries (3/5)
Therefore, the dimensionless momentum equation becomes:
df η d2 f 1 d2 f df
Momentum −U 0 = νU0 ⇒ + 2η =0
dη 2t dη 2 4ν t dη 2 dη
Similarly, the dimensionless boundary conditions becomes f (η = 0) = 1 f (η → ∞) = 0
Thus
d ( f ′) d ( f ′)
∴ f ′ = Ae−η
2
+ 2η f ′ = 0 ⇒ = −2η dη ⇒ n f ′ = −η 2 + C
dη f′
Integrating this equation again, we can obtain f = A∫ e −η 2
dη + B
Then, we can solve the coefficients using boundary conditions
0
⇒ 1 = A∫ e −η dη + B ∴ B = 1
2
∵ f (η = 0) = 1
0

( )
∞ ∞
f (η → ∞) = 0 ⇒ 0 = A∫ e −η dη + 1 ∫
2 2
∵ ∴ A = −1 e − z dz = −1 π 2
0 0

2 2 η
∫e ∫
−η 2 2
∴ f (η ) = 1 − dη or f (η ) = 1 − e − z dz = 1 − erf (η )
π π 0

The solution is well known to erf (η ) erfc(η ) = 1 − erf (η )


be the complementary error function
or probability integral
u
= f (η ) = 1 − erf (η ) ≡ erfc (η )
U0 η η
Unsteady Flows with Moving Boundaries (4/5)
These two complementary solutions (mirror images) are shown in Figures.
The wall shear stress can thus be obtained
du df (η ) dη  2 −η 2  1  −U 0 −η 2
= U0 = U0  − e   = e
dy dη dy  π  2 ν t  πν t
du µU 0
τw = µ =−
dy y =0 πν t
In either case, the plate’s effect diffuses into the fluid at a rate u
proportional to the square root of kinematic viscosity. U0
It is customary to define the shear layer thickness as the point
where the wall effect on the fluid has dropped to 1 percent:
u
= 0.01 for suddenly started case
U0
u
= 0.99 for suddenly stopped case
U0
u
Also, we can estimate the displacement thickness

U0
U 0 δ (t ) = ∫ u ( y, t )dy
0
∞ u ( y, t ) ∞ dy
δ (t ) = ∫ dy = ∫ erfc (η ) dη
0 U0 0 dη

= 2 ν t ∫ erfc (η ) dη δ (t )
0

 ∞ d 
= 2 ν t  η erfc (η ) ∞
−∫ η erfc (η )  dη  integrate by part

0
 0

Unsteady Flows with Moving Boundaries (5/5)
d (erfc(η )) df (η ) 2 −η 2
∵ = =− e Note that the thickness is a measurement of the wall
dη dη π effect. It depends on the kinematic viscosity, not the
∞  2 −η 2  dynamic viscosity. So that the strength of the
∴ δ ( t ) = −2 ν t ∫ η − e dη momentum transport in the fluid is in the order of:
0
 π  Oil > Air > Water
νt ∞
∫ 2η e −η dη
2
=2
π 0 Fluid δ (m)
µ (µPa⋅s) ν (m2/s)
ν t ∞ −η @ 20 °C @ t = 10 s

π ∫0
( )
2
=2 e d η 2
Air 1.8 E 1 1.9 E -5 0.016
Water 1.0 E 3 1.1 E -6 0.003
νt νt
( )

−e −η
2
=2 =2 SAE 30 Oil 2.8 E 5 2.4 E -4 0.086
π 0 π
Similarly, the thermal thickness can be obtained if we impose a temperature to the fluid.
αt k
δ T (t ) = 2 where α = Fluid Air Water SAE 30 Oil
π ρC p
Pr 0.72 7.0 3500
δ (t ) ν
∴ = = Pr Note that the Prandtl number is the ratio of the viscous
δ T (t ) α diffusion rate to the thermal diffusion rate. Therefore, the
strength of the thermal energy transport in the fluid is in the
order of: Oil > Water > Air
Uniform Suction on a Plate
Assumptions:
1. 2D steady flow at velocity U at y = ∞
2. the pressure is constant
3. neglect the gravitation force
4. boundary conditions are independent of x
(infinite plate at y = 0)
5. the plate is porous
6. constant fluid properties
Boundary Conditions:
1. no slip only in x-direction u(0) = 0
2. uniform suction at the wall v(0) = vw = const.
The governing equation can be obtained from the x-momentum equation:
u =u ( y )  2 u =u ( y ) 
 ∂u steady
∂u ∂u  ∂p̂
p̂ = const.
∂ u ∂ 2u 
+µ 2
neglected
x-Momentum ρ +u +v = ρ gx − + 2
 ∂t ∂x ∂y  ∂x  ∂x ∂y 
   
∴ the velocity distribution is u = U 1 − e yvw ν ( G.E.
)
ρ vw
du d 2u
=µ 2
Physically, vw must be negative (wall suction), dy dy
otherwise u would be unbounded at large y.
The boundary layer thickness defined to be the point where u = 0.99U, δ = −4.6ν vw ,
is independent of y or U. Since the convection toward the wall exactly balances the
tendency of the shear layer to grow due to viscous diffusion.
Assumptions:
Similarity Solutions (1/5)
1. z-direction is infinite
2. The velocity distribution v in the x-span is finite,
thus there will be a stagnation point on the plate.
3. set this stagnation point as the origin of the coordinate
4. two dimensional ideal flow (steady & incompressible)
5. constant fluid properties
6. consider a particular solution: u = ax & v = −ay
where a is a positive constant proportional to U0/L
Objective: U0 & L: characteristic velocity & length
To understand the flow field near the stagnation point. continuity G.E.
∂u ∂v
Boundary Conditions: + =0
∂x ∂y
no-slip condition on the plate
x-Momentum
u(x = 0) = v(y = 0) = 0
The governing equation can be satisfied only if:  ∂u ∂u  ∂ˆp  ∂ 2u ∂ 2u 
ρ  u + v  = − + µ  2 + 2 
 − ρ ( ax )
2  ∂x ∂y  ∂x  ∂x ∂y 
 p̂ = + g ( y)
 −ρ y -Momentum

2 ∴ p̂ = (a x
2 2
)
+ a 2 y 2 + const.
 ∂ 2v ∂ 2v 
− ρ ( ay )  ∂v ∂v 
2 2
 ∂ˆp
 p̂ = + f ( x) ρ  u + v  = − + µ  2 + 2 
 2  ∂x ∂y  ∂y  ∂x ∂y 
ρ
or p0 = ˆp +
2
( u 2
)
+ v 2 = const. ⇒ Bernoulli’s equation

⇒ The given velocity distribution is for an inviscid flow.


Similarity Solutions (2/5)
Thus the streamline is given as:
?i j kˆ
 
V × ds = 0 ⇒ u v 0 = ( udy − vdx ) kˆ = 0
dx dy 0
dx dy dx dy
⇒ = ⇒ = ⇒ n x = − n y + C
u v ax −ay
⇒ n ( xy ) = C ∗ ⇒ xy = const. ⇒ family of hyperbolas

Thus, streamlines can be drawn as shown in the figure.


Note that the velocity distribution used can not satisfy the no-slip boundary condition.
Hiemenz (1911) modified the stream function to vary with y to account for viscous
effects so that the no-slip condition can be satisfied.
∂ψ No-slip at the wall
v=− = − Bf ( y )
∂x u y =0 = 0 ⇒ f ′(0) = 0
∂u ∂v ∂u ∂ψ
+ =0 ⇒ − Bf ′ ( y ) ⇒ u= = Bxf ′ ( y ) v y =0 = 0 ⇒ f (0) = 0
∂x ∂y ∂x ∂y
This modified velocity distribution must satisfy the Navier-Stokes equations. First, we
substitute u and v into Navier-Stokes equations to determine the required form of the
pressure term.
Similarity Solutions (3/5)
x-Momentum
 ∂u ∂u  ∂ˆp  ∂ 2u ∂ 2u  ∂ˆp
ρ u + v  = − + µ  2 + 2  ⇒ ρ ( Bxf ′ )( Bf ′ ) + ( − Bf )( Bxf ′′ )  = − + µ ( Bxf ′′′ )
 ∂x ∂y  ∂x  ∂y  ∂x
 ∂x
y -Momentum
 ∂v ∂v  ∂ˆp  ∂2v ∂2v  ∂ˆp
ρ u +v  = − +µ 2 + 2  ⇒ ρ ( − Bf )( − Bf ′ ) = − + µ ( − Bf ′′ )
 ∂x ∂y  ∂y  ∂y  ∂y
 ∂x
∂p̂ 2 ν  2 df ν df ′  
2 1 df
2
ν df ′ 
∵ = − ρ B  ff ′ + f ′′  = − ρ B  f +  = − ρ B  + 
∂y  B   dy B dy   2 dy B dy 
1 ν 
∴ p̂ = − ρ B 2  f 2 + f ′  + g ( x )
2 B 
∂p̂  

∂x
= ρ B 2 x  f ′′′ +
B
ν
( )
ff ′′ − f ′2  = g′( x) Consider the flow far from the region
 from y -momentum of nonzero vorticity (y → ∞), the
  

from x -momentum velocity u should not depend on y.
∴ f ′′′ +
B
ν
( ff ′′ − f ′ ) = const.
2
u y →∞ = Bxf ′ y →∞ ≠ f ( y ) ⇒ f ′ =1
∴ f ′′ = f ′′′ = 0

f ′′′ +
B
ν
( ff ′′ − f ′ ) = const. = −
2
ν
B The similarity method has transformed the
original PDE into an ODE, and leaves only one
independent variable, that is, the coordinate y.
Similarity Solutions (4/5)
Then, we have to nondimensionalize this equation to eliminate the dimensional
constant B and ν. Generally, similarity solutions require an infinite spatial and
temporal extent. However, the body length scale L does not exist. Rather, the proper
length and velocity scales are ν B and ν B . Therefore, the appropriate
dimensionless variables can be defined: ∂ψ
y B u = = x ν BF ′ (η )
η= =y ∂y
ν B ν ⇒
∂ψ
ψ = Bxf = x ( Bf ) = xF (η ) ν B v = − = − F (η ) ν B
∂x
Thus we rewrite the ODE as a dimensionless form:
Note that the term Bf has the
ν dimension of velocity.
∵ Bf = F (η ) ν B ∴ f = F (η )
B
df df dη  ν  B
f′= = = F′ = F′
dy dη dy  B  ν
df ′ df ′ dη B B
f ′′ = = = ( F ′′ ) = F ′′
dy dη dy ν ν
df ′′ df ′′ dη  B  B  B 
f ′′′ = = = F ′′  =   F ′′

dη dy  ν 
dy  ν ν 

⇒ F ′′′ + FF ′′ − F ′2 = −1
Similarity Solutions (5/5)
The dimensionless boundary conditions are u = v = 0 at the wall (η = 0) and u = Bx
at a large distance from the wall:
Note that this nonlinear equation has
F (0) = F ′ (0) = 0 & F ′(∞) = 1
no analytical solution yet. However, a
The numerical solutions of this dimensional solution from numerical calculations is
analysis are shown in the plot: available.
ψ = xF ν B ∝ F
∂ψ
u= = x ν BF ′ ∝ F ′
∂y
du du dη  dF ′  ν
τw ∝ = =x νB  ∝ F ′′
dy y =0
dη dy η =0  dη  B
In the plane flow, the stagnation points
actually form a stagnation line (z = 0).
However, in an axisymmetric flow,
stagnation is a true point.
Due to the continuity equation in cylindrical
geometry, the velocities become
u = Bx & v = −2By.
Following from the same procedure, an similarity solution can be determined.
Creeping Flow
The basic idea of creeping flow, developed by Stokes (1851) is that density (inertia)
terms are negligible in the momentum equation when Reynolds number is very
small.
High Re ⇒ inertial force dominate
DV * 1 *2 * p − p0
Re
1 ⇒ = −∇* p* + ∇ V where p* =
Dt* Re ρU 2
∂V *
∂t*
(

)( )
+ V * ⋅ ∇* V * = −∇* p*

pressure
 force
rate of change of momentum convective force

Low Re ⇒ viscous force dominate


DV * p − p0
Re  1 ⇒ Re = −∇* p* + ∇* 2V * where p* =
Dt* µU L
* * *2 *

 p = ∇
V
pressure force viscous force
Homework
 3-15 free surface flow
 3-16 axially Couette flow between concentric cylinder
 3-18 unstable free surface flow
 3-21 azimuthal Couette flow between concentric
cylinder
 3-19 suddenly started flow
 3-23 radial outflow between two circular plates
 3-24 uniformly porous cylinder
 3-26 similarity solution (uniformly porous cylinder)
Short Firms Lecturer: G.I. Taylor
Deformation of Fluid Element (Re << 1) Rigid Body Rotation

Motion in Very Viscous Fluid Reynolds Number Effects

Vous aimerez peut-être aussi