Вы находитесь на странице: 1из 32

Pulsatile Flow Through A Curved Pipe

VIJAYAALAYAN THIVYATHASAN
00466617
THIS IS MY OWN WORK UNLESS OTHERWISE STATED

June 15, 2010

1
Abstract
The main aim of the project is to model blood flow through the human circulatory system. By
modelling the blood vessel as a slowly curving pipe based on a section of the torus ring, I derive an
unsteady form of the Dean Equations and then adapt them to my particular model of blood flow.
Apart from introducing the variable of curvature in a pipe, I will also be introducing Pulsatile Flow
to factor in the effects of the heart as a pumping mechanism.

Contents
1 Introduction 3
1.1 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Derivation of the Unsteady Dean Equations 5


2.1 Solutions to the Unsteady Dean Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Analysis of Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Graphical Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Scaling time with ω10 18


3.1 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 Looking at a large Womersley parameter: α −→ ∞ . . . . . . . . . . . . . . . . . . 20
3.2 Analysis of Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Graphical Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Conclusion 31

2
1 Introduction
In the past Berger et al (1983) have looked into flows in curved pipes, but whilst they have focused
on the complexity that arises through the difference between various degrees of curvature in compari-
son to straight pipes, they have not delved too far into unsteady, and more specifically pulsatile flows
through curved pipes. Pipe flows play an important part in engineering and also in understanding the
mechanics of the human blood supply system. This has been researched by Pedley (1980) who covered
quite extensively amongst other arteries, the aorta, which is highly curved. Though Berger et al have
independently studied the model of a loosely coiled pipe, and unsteady laminar flow in a curved pipe,
they have not combined the two like Siggers and Waters [12], who examined unsteady flows in pipes with
finite curvature. Siggers et al focus on the benefits of understanding blood flow in the arteries in relation
to Atherosclerosis and to this end consider the effects of finite curvature on wall shear stress being as
there is a significant correlation between these properties and Atherosclerosis.

What we are trying to understand in this article is how the flow varies as a result of the pulsatile
flow attributed to the heart and we wish to see how it varies from the models that others have set out.
Initially we will allow for the unknown parameters of amplitude and oscillation, and analyse what limi-
tations our assumptions bear on our equations. Later on we will make a more refined case dealing with
the Womersley parameter and modelling our flow pending the flexibility of our results and whether or
not we will have to constrain our model with further assumptions.

When we model the pumping mechanism of the heart, we introduce three parameters A, ω and the
Dean number K and make the assumption that the pressure G behaves like 1 + A cos(wt) when scaled
appropriately. The amplitude A is the unsteady Reynolds number that Siggers et al look into and ω is
associated with the Womersely number α, a dimensionless expression of the pulsatile flow frequency in
relation to the viscous effects of the flow and is formally defined when we first meet it in our calculations
later on. What we find is that in the absence of any oscillatory motion, i.e. when our amplitude A is
equivalent to zero, we end up with the Steady Dean flow. We also make the assumption that as our pipe
flows along its curvature in the φ direction, it will eventually settle and so we can simplify our model
to an axisymmetric one with no φ dependance, much like the case of two dimensional flow in a straight
pipe.

Looking into a simple model of blood flow and bearing in mind that the curvature we look into is
slight, we can investigate the flow that arises due to the pumping mechanism of the heart. The curvature
is defined by the ratio of the radius of the pipe to the inner torus radius, say ab (see figure in the next
section) and as such it is assumed to be much smaller than 1. While this project focuses on looking at
small curvature, if we considered a highly curved pipe where ab  1 this gives rise to a different flow. The
effect that curvature has on flow appears to be an increase in stability as the critical Reynolds number,
which signifies the threshold between stable flow and turbulence, can be higher by a factor of two or more
(Berger et al). (It is possible to conceive of a case where we consider a model in which a straight pipe
precedes our curved pipe, such that the entry conditions of our pipe are altered, but we do not for the
time being)

The Dean number K is also another parameter, which appears in our non-dimensional Navier-Stokes
equations and it can be said to characterise the flow. The Dean number is proportional to the centrifu-
gal force so by varying the Dean number we shall look at cases of flow, which are driven by different

3
forces. Berger et al make the point that the importance of the Dean number is in its substitution for the
Reynolds number in the specific case where ab  1, which was found by Dean (1928). The case which
I am currently pursuing looks at the case of a small Dean number, permitting expansion in powers of
K and separating equations in terms of leading order. Making approximations as to the relative size of
the Dean number is important in characterizing the type of flow we will be looking at. At small Dean
numbers we get a flow that is similar to Poiseuille flow but with cross-flow due to the centrifugal force,
however slight the curvature. As the Dean number gets larger, we end up with a case where derived
terms in the flow of the curved pipe are less comparable to those found in the flow along a straight pipe.
It is also important to mention that there is an increase in the centrifugal force as K increases, with a
development of boundary layers, which can be shown to be thick in the inner bend and thin in the outer
bend. At small Dean numbers, our flow attains its maximum velocity at angles +/- 90 degrees [12] and
the vortices are symmetric. The importance of our ratio ab must not be understated here either as even
small increments can have effects on the flow resistance that are several times greater but at high Dean
numbers there is a discrepancy such that the flow shows signs of higher resistance when ab  1 (what
Berger et al have classified as loosely coiled) over a highly curved one.

In their analysis of unsteady flow along pipes of slight curvature Siggers et al also deal with the same
parameters that characterise my model of blood flow. Considerations of the Womersley parameter, Dean
number and curvature of the pipe allow them to analyse cases where their defined curvature ab is uniform
and the variation of the Womersley parameter α at extremely low or high values give rise to various
solutions. As well as looking into oscillatory pressure gradients, Siggers et al also make use of sinusoidal
pressure which removes the steady component of the pressure gradient and gives rise to secondary veloc-
ity, allowing for their secondary streaming Reynolds number A which also characterizes the flow. The
value of A is important as it parameterizes the steady streaming element along the core of the pipe.
Similar to my project which deals with expansions in my defined Dean number K, in their calculations
for the asymptotic solutions to an oscillatory pressure gradient, Siggers et al expand in powers of ab and
for the asymptotic analysis for high frequency sinusoidal flow they expand in powers of α−1 .

1.1 Outline
The starting point for this project will be to look at the Navier-Stokes equations in cylindrical co-ordinates
for a section of a torus and to non dimensionalise them. We shall assume that the curvature is slight,
denoted by the ratio of the pipe radius to the radius of the torus, which we have previously mentioned as
a
b  1. In our simplification of these equations we shall further assume that as the pipe is only slightly
curved , that we can treat it much like a straight pipe in the sense that over a certain length travelled
in the φ direction, the flow settles and does not change so we are essentially looking at an axisymmetric
flow. We will then arrive at a parameter K which will characterise the flow, known as the Dean number
and assuming K to be small we shall then expand our velocity and stream function in powers of K so
that we can analyse the flow. Our fluid through the pipe will then need to satisfy the no-slip condition on
the boundary of the pipe. Finally we shall graphically compare the flows and how they change through
time through the two different scalings of time we make.

4
2 Derivation of the Unsteady Dean Equations
In order to derive our starting equations, we look at flow down a slightly curved pipe and model this
using polar co-ordinates (r, φ, z) on a section of a torus ring which is given by (r − b)2 + z 2 = a2 , where
we have chosen b  a and our flow occurs in the φ direction as shown in the figure below:

Figure 1: Diagram of curved pipe

Our velocity u = (ur , uφ , uz ) fulfills the requirements of the axisymmetric Navier-Stokes equation (as we
have decided to obtain solutions which do not have any dependence on φ):

1 ∂ ∂uz
(rur ) + = 0
r ∂r ∂z!
Dur u2φ ∂p  ur 
ρ − = − + μ ∇2 u r − 2
Dt r ∂r r
  
Duφ uφ u r uφ 
ρ + = G(r, z, t) + μ ∇2 uφ − 2
Dt r r
Duz ∂p
ρ = − + μ∇2 uz
Dt ∂z
where G = rb G0 (1 + f (t)). In order to simplify this set of equations we make some assumptions. Given
that the pipe is slowly curving and that b  a with r and z varying over the scale a, we can safely assume
that:
∂ 1 1 1
b  a so that r = b + ax∗ ' b and ∼  '
∂r a b r

Furthermore, to non-dimensionalise our equations, we scale z = az and we take U0 to be a scale of
1
uφ . We then use these scales to obtain a suitable scale for ur and uz such that ur ∼ uz ∼ U0 ab 2 ,
and we scale ur with u∗x . Whilst these scalings are important, what is of most importance is how we
wish to scale our time given the significance of the time driven pressure gradient. Initially we will scale

5
1

t = (ab)
2 ∗
U0 t and later change the scaling depending on any complications that may may arise or in order
to make calculations simpler (Note here that any previous * terms are dimensionless parameters). Upon
substitution into the axisymmetric Navier-Stokes equations, we can re-arrange the resulting equations
and further simplify them with the following definitions for new parameters:

G 0 a2 ρU0 a  a  12
=1 and K=
μU0 μ b
And thus we arrive at the following set of Dean equations:
∂ux ∂uz
+ = 0
 ∂x ∂z
Dux ∂p
K − u2φ = −K + ∇2 ux
Dt ∂r
Duφ
K = 1 + f (t) + ∇2 uφ
Dt
Duz ∂p
K = −K + ∇2 u z
Dt ∂z
(where we have dropped the * terms). The derivation of the Dean Equations is shown in greater detail
by Mestel [8], [9]. Alongside the Dean number K, we have chosen the pressure gradient 1 + f (t) to be
equivalent to 1 + A cos(ω1 t) where we have introduced the non dimensional parameters A and ω1 , both
of which can be defined as we desire in our analysis of the flow. Our velocity here can now be written in
the form U = (U, V, W ) and if we cross differentiate to cancel the pressure terms, we can then introduce a
stream function of the form ψ(x, z) such that U = ψz and W = −ψx . We then have the initial equations:

K(Vt + ψz Vx − ψx Vz ) = 1 + A cos(ω1 t) + ∇2 V
K(Ωt + ψz Ωx − ψx Ωz ) = ∇2 Ω − 2KV Vz

as this is a flow down a pipe of circular cross-section, we can just switch to the (x, z) plane, we can switch
to polar co-ordinates (R, θ) where x = R sin(θ) and z = R cos(θ), which then gives us:

K
(ψθ VR − ψR Vθ ) = 1 + A cos(ω1 t) − KVt + ∇2 V (2.1)
R
K
(ψθ ΩR − ψR Ωθ ) = ∇2 Ω − KΩt − 2KV (VR sin θ + Vθ cos θ) (2.2)
R

Where subscripts denote derivative with respect to the shown variable and where we have defined Ω =
−∇2 ψ . When solving these, we will need to ensure that they fulfill the boundary conditions on R = 1
such that:
V (1) = 0 ψ(1) = 0 ψR (1) = 0
We see that there are three parameters in our problem which characterise the flow. Based on the initial
assumption that the Dean number K is small, we can expand the above equations in terms of K, and
by setting V = V0 + KV1 + K 2 V2 + ... and ψ = Kψ1 + K 2 ψ2 + ... we can then match coefficients, which

6
allow us to then obtain the following equations from equation (2.1):

O(1) : 0 = 1 + A cos(ω1 t) + ∇2 V0 (2.3)


O(K) : 0 = −(V0 )t + ∇2 V1 (2.4)
1
O(K 2 ) : (ψ1θ V0R − ψ1R V0θ ) = −(V1 )t + ∇2 V2 (2.5)
R
We can apply a similar process to equation (2.2) where we then obtain the following two equations:
 
2 cos θ
O(K) : 0 = ∇ Ω1 − 2V0 V0R sin θ + V0θ (2.6)
R
    
cos θ cos θ
O(K 2 ) : ∇2 Ω2 = −Ω1t + 2 V0 V1R sin θ + V1θ + V1 V0R sin θ + V0θ (2.7)
R R

These equations will be solved for each of the various values of V and ψ and by looking at each term, we
can observe how the non linear interactions and their time averages affect the flow compared to that of
a steady flow.

2.1 Solutions to the Unsteady Dean Equations


Looking at our O(1) term we must first note that as V0 is our value for no curvature with no co-efficient
dependence on the Dean number K, then essentially there is no θ dependence, so we can cancel out the
Vθθ term and this simplifies equation (2.3):

∇ 2 V0 = −(1 + A cos(ω1 t))


1
(RV0R )R = −(1 + A cos(ω1 t))
R
(RV0R )R = −R(1 + A cos(ω1 t))
R2
RV0R = − (1 + A cos(ω1 t)) + c
2
R c
V0R = − (1 + A cos(ω1 t)) +
2 R
R2
V0 = − (1 + A cos(ω1 t)) + c log(R) + d
4
If we enforce the boundary conditions we find out that our constants of integration, c = 0 in order for
our V0 to be finite at R = 0 and d is easily worked out by placing R = 1, we then get:
1
V0 = (1 − R2 )(1 + A cos(ω1 t)) (2.8)
4

7
If we now look at equation (2.4) we can derive V0t to give the following:
1 2
∇2 V1 = (R − 1)(Aω1 ) sin(ω1 t)
4
1 1 2
(RV1R )R = (R − 1)(Aω1 ) sin(ω1 t)
R 4
1 3
(RV1R )R = (R − R)(Aω1 ) sin(ω1 t)
4 
1 R4 R2
RV1R = − (Aω1 ) sin(ω1 t) + c
4 4 2
 
1 R3 R c
V1R = − (Aω1 ) sin(ω1 t) +
4 4 2 R
 
1 R4 R2
V1 = − (Aω1 ) sin(ω1 t) + c log R + d
4 16 4

Once again if we place boundary conditions, we get that c = 0 and by working out d we get:
1
V1 = (R4 − 4R2 + 3)(Aω1 ) sin(ω1 t)
64
Aω1 2
= (R − 1)(R2 − 3) sin(ω1 t) (2.9)
64
We must make note of the fact that given our ω1 here is non-dimensional, and if Aω1 is large then our
KV1 may also be large and expansion breaks down, but this is a problem that will be dealt with in the
following section. In order to solve equation (2.5) we need to solve equation (2.6) using what we have so
far calculated (note that as V0 has no θ dependence the final term V0θ disappears):

∇2 Ω1 = 2V0 V0R sin θ


 
1 1 2 −2R
ΩRR + ΩR + 2 Ωθθ = (1 − R2 )(1 + A cos(ω1 t)) (1 + A cos(ω1 t)) sin θ
R R 4 4
1 1 1 3
ΩRR + ΩR + 2 Ωθθ = (R − R)(1 + A cos(ω1 t))2 sin θ (2.10)
R R 4
For all intents and purpose we can treat the t dependent terms as constants as our left hand side only
looks at θ and R derivatives. Also of note is how we can separate our Ω1 term into two functions, one
for θ and another for R, giving us
1
Ω1 = (1 + A cos(ω1 t))2 F (R)G(θ)
4
and if we use this , then we can instantly come to the conclusion that our G(θ) term is sin θ which allows
us to factor out it out and rewrite our equation (10) in the following form (note that the final term on
the LHS changes sign as Gθθ is equal to -sin θ):
1 1
FRR + FR − 2 F = R3 − R
R R

8
We can hazard a guess at the form F (R) will take and write it as
γ
F (R) = δR5 + βR3 − R
24
γ
FR (R) = 5δR4 + 3βR2 −
24
FRR (R) = 20δR3 + 6βR
where δ, β, γ are constants and γ has been written with chosen sign and coefficient for ease of calculations
later on. If we plug these values in the above equation we arrive at:
24δR3 + 8βR = R3 − R
By matching up the coefficients we arrive at our equation for Ω1 :
 
1 2 1 5 1 3 γ
Ω1 = (1 + A cos(ω1 t)) R − R − R sin θ (2.11)
4 24 8 24
As we previously defined ∇2 ψ1 = −Ω1 we get the equation:
1
∇2 ψ 1 = (1 + A cos(ω1 t))2 (γR + 3R3 − R5 ) sin θ
96
This is very similar in form to the equation for ∇2 Ω1 and so is solved in similar fashion until we arrive
with a near complete formula for ψ with two unknown constants  and γ:
 
1 2 γ 3 1 5 1 7
ψ1 = (1 + A cos(ω1 t)) R + R + R − R sin θ
96 8 8 48
These two constants can be worked out using our boundary conditions ψ(1) = 0 and ψR (1) = 0 giving
us our equation for ψ1 :
 
1 1 3 1 1
ψ1 = (1 + A cos(ω1 t))2 R − R3 + R5 − R7 sin θ
96 12 16 8 48
1
ψ1 = (1 + A cos(ω1 t))2 R(4 − 9R2 + 6R4 − R6 ) sin θ
4608
1
ψ1 = (1 + A cos(ω1 t))2 R(4 − R2 )(1 − R2 )2 sin θ (2.12)
4608
Now that we have an equation for ψ1 we can now solve (2.5). One major difference between V2 compared
to the other values of V is that there is a θ dependance giving rise to a more complicated solution. Once
again, (2.5) simplifies due to lack of θ dependance:
1
∇ 2 V2 = − ψ1θ V0R + (V1 )t
R
1 Aω12 2
∇2 V2 = (1 + A cos(ω1 t))3 R(4 − R2 )(1 − R2 )2 cos θ + (R − 1)(R2 − 3) cos(ω1 t)
9216 64
In order to solve this equation we need to solve individually for each component on the RHS, which when
solved using similar techniques as applied to previous equations, we get:
Aω12
V2 = cos(ω1 t)(R6 − 9R4 + 27R2 − 19) + ...
2304
1
... − (1 + A cos(ω1 t))3 (R9 − 10R7 + 30R5 − 40R3 + 19R) cos θ (2.13)
737280

9
Finally we have equation (2.7) to be solved, which can be written in the form (by using equation (2.11)
and the fact that γ = − 32 ):

∇2 Ω2 = −Ω1t + 2(V0 V1R sin θ + V1 V0R sin θ)


 
Aω1 
∇ 2 Ω2 = (1 + A cos(ω1 t)) sin(ω1 t) 3R − 6R3 + 2R5 sin θ
96
 
Aω1
... + (1 + A cos(ω1 t)) sin(ω1 t) (1 − R2 )(R3 − 2R) sin θ
32
 
Aω1
... + (1 + A cos(ω1 t)) sin(ω1 t) (1 − R2 )(R3 − 3R) sin θ
64
 
Aω1
∇2 Ω 2 = (1 + A cos(ω1 t)) sin(ω1 t) (−15R + 18R3 − 5R5 ) sin θ
192
Like before, we can sperate Ω2 into two sperate functions of R and θ such that
 
Aω1
∇2 Ω2 = (1 + A cos(ω1 t)) sin(ω1 t) F (R)G(θ)
192
where it is quite clear that our G(θ) is sin(θ). After similar calculations as used to find Ω1 we arrive at
the equation
 
Aω1
Ω2 = (1 + A cos(ω1 t)) sin(ω1 t) (−υR − 90R3 + 32R5 − 5R7 ) sin θ (2.14)
9216
where υ is a variable to be determined by boundary conditions on the stream function. We now calculate
ψ2 :
 
2 Aω1
∇ ψ2 = (1 + A cos(ω1 t)) sin(ω1 t) (υR + 90R3 − 32R5 + 5R7 ) sin θ
9216
After similar calculations we get
  
Aω1 υ 3 15 5 2 7 5 9
ψ2 = (1 + A cos(ω1 t)) sin(ω1 t) ηR + R + R − R + R sin θ
9216 8 4 3 80

which when subjected to our boundary conditions, (where we get that η = 12548 and υ = −46), gives us:
  
Aω1 125 23 15 2 5
ψ2 = (1 + A cos(ω1 t)) sin(ω1 t) R − R3 + R5 − R7 + R9 sin θ
9216 48 4 4 3 80
 
Aω1
ψ2 = (1 + A cos(ω1 t)) sin(ω1 t) R(125 − 276R2 + 180R4 − 32R6 + 3R8 ) sin θ
442368
 
Aω1
ψ2 = (1 + A cos(ω1 t)) sin(ω1 t) R(3R4 − 26R2 + 125)(R2 − 1)2 sin θ (2.15)
442368

2.2 Analysis of Flow


If we wish to see how the introduction of a pulsatile pressure affects our flow it may be convenient to look
at the time averages for each of our values and see how important our parameters are in determining the

10
overall nature of the flow. In order to calculate the time average of a variable X, we use the following
formula Z
1 t0 +T 0
hXi = Xdt
T t0
0
Where t is our dummy variable and h i denotes taking the time average our the specified variable. This
general formula can be adapted for our own model to give the following equation
Z 2π
ω 1 ω1 0
hXi = Xdt
2π 0
If we calculate the time average for V0 we get the following
  Z 2π
ω1 1 2
ω1 0 0
hV0 i = (1 − R ) (1 + A cos(ω1 t ))dt
2π 4 0
   ω2π
ω1 1 2 0 A 0 1
= (1 − R ) t + sin(ω1 t )
2π 4 ω1 0
1 2
= (1 − R )
4
Of note is that this is the solution to the steady Dean equations for V0 which indicates that time has no
overall effect on our flow when we are looking at the leading order term. Let us now look to V1 and its
time average
  Z 2π
ω1 Aω1 2 ω1 0 0
hV1 i = (R − 1)(R2 − 3) sin(ω1 t )dt
2π 64 0
   ω2π
ω1 Aω1 2 2 1 0 1
= (R − 1)(R − 3) − cos(ω1 t )
2π 64 ω1 0
= 0
Once again, when compared to the steady solutions to the Dean equations, time appears to have no
contribution to the flow. However, if we look at V2 due to the (1 + A cos(ωt))3 term, we should expect
some lasting contribution. Note that the first term in V2 is just a cos(ωt) term and so its time average
will be 0, so we can focus on the second term.
Z 2π

3
ω 1 ω1 0 0 0 0
(1 + A cos(ωt)) = (1 + 3A cos(ω1 t ) + 3A2 cos2 (ω1 t ) + A3 cos3 (ω1 t ))dt
2π 0
Z 2π
ω 1 ω1 0 3 0 A3 0 0 0
= (1 + 3A cos(ω1 t ) + A2 (cos(2ω1 t ) + 1) + (cos(3ω1 t ) + 3 cos(ω1 t ))dt
2π 0 2 4
We can ignore all the cos(nω1 t) terms for n 6= 0, n ∈ Z as their time averages are equivalent to zero. This
leaves us looking at
Z 2π
ω1 ω1 3 0
= (1 + A2 )dt
2π 0 2
3
= 1 + A2
2

11
If we use this value for calculating our time average for V2 we get the value
 
1 9 7 5 3 3 2
hV2 i = − (R − 10R + 30R − 40R + 19R) cos θ 1 + A
737280 2
If we chose our parameter A = 0 then we have the same solution as we do for the steady Dean equations
but as this is not the case we see that depending on the value we assign to A, the effects of the pulsatile
flow are now being felt with some form of effect. These values combined yield the total time average for
our flow:
 
1 1 3
hV i = (1 − R2 ) − K 2 (R9 − 10R7 + 30R5 − 40R3 + 19R) cos θ 1 + A2 + ...
4 737280 2
We can similarly look at our stream function ψ and note its time average.
  Z 2π
ω1 1 ω1 0 0
hψ1 i = R(4 − R2 )(1 − R2 )2 sin θ (1 + A cos(ω1 t ))2 dt
2π 4608 0
  Z 2π  
ω1 1 2 2 2
ω1 0 A2 0 0
= R(4 − R )(1 − R ) sin θ 1 + 2A cos(ω1 t ) + (cos(2ω1 t ) + 1) dt
2π 4608 0 2
  
1 2 2 2 A2
= R(4 − R )(1 − R ) sin θ 1+
4608 2
As expected, it is clear that depending on our parameter A, unsteadiness has an effect on our flow. We
can also look at the time average for ψ2
  Z 2π
Aω1 ω1 0 0 0
hψ2 i = R(3R4 − 26R2 + 125)(R2 − 1)2 sin θ (1 + A cos(ω1 t )) sin(ωt )dt
442368 0
  Z 2π  
Aω1 ω1 0 A 0 0
= R(3R4 − 26R2 + 125)(R2 − 1)2 sin θ sin(ω1 t ) + sin(2ω1 t ) dt
442368 0 2
= 0

Surprisingly, there is no time-averaged contribution from there ψ2 term.


We can also consider the dimensionless flow rate through the pipe much like Siggers et al (2008) by
considering the equation for each of our Vi and ψ for i = 0, 1, 2 and j = 1, 2. We define
Z 2π Z 1
{X} = XRdRdθ
0 0

where once again {} performs the above operation on our specified X variable and if we want to consider
the average then we divide the above formula by π.
Z 2π Z 1
1
{V0 } = (1 + A cos(ω1 t)) (R − R3 )dRdθ
4 0 0
 2 1
π R R4
= (1 + A cos(ω1 t)) −
2 2 4 0
π
= (1 + A cos(ω1 t))
8

12
Similarly
Z 2π Z 1
Aω1
{V1 } = sin(ω1 t) (R5 − 4R3 + 3R)dRdθ
64 0 0
Aω1 π
= sin(ω1 t)
48
And
Z 2π Z 1
1
{V2 } = Aω 2 cos(ω1 t) (R7 − 9R5 + 27R5 − 19R)dRdθ
2304 1 0 0
Z 2π Z 1
1
... − (1 + A cos(ω1 t))3 R2 (R8 − 10R6 + 30R4 − 40R2 + 19) cos θdRdθ
737280 0 0

17π
= − Aω 2 cos(ω1 t)
3072 1
Using these terms we see that the total dimensionless flow rate through the pipe is:
π Aω1 π 17π
{V } = (1 + A cos(ω1 t)) + K sin(ω1 t) − K 2 Aω 2 cos(ω1 t) + ...
8 48 3072 1
Looking at the stream function terms it is quite simple to realise that as both terms have sin θ, the
integral taken over the whole of 2π averages to zero and so:

{ψ1 } = {ψ2 } = 0

13
2.3 Graphical Interpretation
If we look at the plots for V0 and V1 we see that they are fairly similar given their lack of θ dependence
and as such the only change in the contour plots that they exhibit are the reversal of flow as shown below.
Note that the left hand side of contour plot is the inner bend of the tube whilst the right hand side is
the outer bend of the pipe and the white regions denote positive flow:

Figure 2: The contour plot on the left shows V0 at an amplitude of 1 at time t = π and we must mention that the contour
graph for V0 does not change as we vary time. The contour graph on the right shows the flow for V1 at time t = 3π 2
and of
the same amplitude but at t = π2 the contour graph for V 1 is the same as for V0 . However, unlike the graphs below, we do
not get a third graph for V1

V2 exhibits quite an interesting display in terms of the flow as we alter the parameters of amplitude
and frequency. Not shown below is the fact that as time changes and forces the oscillation, we see a
reversal in the flow much like we did for the V1 contour plots. Also, due to the structure of V2 the value
of ω1 acts as a parameter that can be scaled how we wish. The structure of our pressure gradient is
1 + A cos t and so we realise that the value of A around the value if 1 is important and so by looking at
amplitude values around 1 and varying time we can see how our flow behaves. By keeping the amplitude
at a constant value, we can change the value of time and note the transformation in the flow.

Figure 3: These contour plots show the change in flow as we maintain the frequency ω1 = 1 and amplitude A = 1 whilst
changing the time from 0 to π2 then π. It is quite clear that as we progress through time the flow changes direction and if
we increase our value for time further, we get a repetition in the contour plots as the flow oscillates back and forth. What
we also notice is that at π2 for all these set of graphs as well as for the other set of graphs is that they all exhibit the same
contour plot due to this being the value at which cos is zero and so we have no pulsatile flow but a steady driving gradient.

14
The next set of contour plots looks at an amplitude A = 20:

Figure 4: These contour plots show how the flow is slightly different at a high amplitude yet also shows how it evolves in
time and that the oscillation is exactly the same

Finally, we wish to see how the oscillation behaves over time when we look at an amplitude A < 1.
What we expect and do find is that as the amplitude heads to zero, the flow loses oscillation and
remains more or less constant across time. The transformation and loss of oscillation takes place between
A = 1 × 10−2 and 2 × 10−4 . Here we look at A = 1 × 10−3 :

Figure 5: These contour plots show how the flow changes at a really low amplitude and how there is still pulsation even at
really low amplitude values

Each of the contour plots for V2 are at intervals of π2 from 0 to π. All the plotted graphs are quite
similar with only slight variation due to the amplitude. There is not much emphasis on the frequency ω1
and for each contour plot we have left the value as 1 as for both V0 and V1 the frequency term remains
within the sin and cos term so it only increases the rate of oscillation. However this is not the case for
V2 which has a Aω12 value in the first term and competes with an A3 value in the second term but still
the changes this has on the contour plot is by constraining a ratio between ω1 and A at which we see
certain contour plots. If we increase ω1 we require a higher value of A for our contour plots to remain
invariant in time and essentially give us the same contour plot (as we increase A our contour plots will
only only show the middle graph in Figure 5 as the oscillation about 1 in 1 + A cos t is arbitrary as A is
so large). The time averaged equations are also quite helpful to plot as they give us an understanding
as to how the flow through the curved pipe looks in general with the effect of time being displayed in a
single contour plot without having to look at multiple graphs and having to compare them. We can plot
plot the time averaged equations for the stream function and non-zero velocity values:

15
Figure 6: The contour plot on the left is of V0 . The contour plot in the centre is of V2 and the one on the right is of ψ1

As we plot the values for ψ1 we must note that as out time dependent term remains within a squared
bracket, there is no oscillation due to a negative pressure gradient. But if we look at the contour plots
for ψ2 at a fixed frequency ω1 = 1amplitude A = 1 we see there is a change about t = π:

π 3π
Figure 7: The contour plot on the left is at t = 2
and the one on the right is at t = 2

We see that for V0 and ψ1 there is no oscillation in flow and that for V1 the flow switches between two
fixed forms of flow. V2 on the other hand varies with fixed amplitudes and oscillates about t = π2 whereas
ψ2 oscillates about t = π. These contour plots have been seen before and are going to be a frame of
reference for comparison with other graphs in the next section.

16
2.4 Summary
Looking over this section, we have managed to obtain the following important results:
1
V0 = (1 − R2 )(1 + A cos(ω1 t))
4
Aω1 2
V1 = (R − 1)(R2 − 3) sin(ω1 t)
64
1
V2 = Aω 2 cos(ω1 t)(R6 − 9R4 + 27R2 − 19)
2304 1
1
... − (1 + A cos(ω1 t))3 R(R8 − 10R6 + 30R4 − 40R2 + 19) cos θ
737280

1
hV0 i = (1 − R2 )
4
hV1 i = 0
 
1 9 7 5 3 3 2
hV2 i = − (R − 10R + 30R − 40R + 19R) cos θ 1 + A
737280 2
π
{V0 } = (1 + A cos(ω1 t))
8
Aω1 π
{V1 } = sin(ω1 t)
48
17π
{V2 } = − Aω 2 cos(ω1 t)
3072 1

1
ψ1 = (1 + A cos(ω1 t))2 R(4 − R2 )(1 − R2 )2 sin θ
4608
 
Aω1
ψ2 = (1 + A cos(ω1 t)) sin(ω1 t) R(3R4 − 26R2 + 125)(R2 − 1)2 sin θ
442368

  
1 A2
hψ1 i = R(4 − R2 )(1 − R2 )2 sin θ 1+
4608 2
hψ2 i = 0
{ψ1 } = 0
{ψ2 } = 0

What we see from these set of results is that there is only a weak dependence on time and this can
be attributed to our assumption that KAω1  1 and we see from our time averages that the only
modification to the flow is the amplitude A. Our expansion starts to break down when A  1 or when
ω1  1 which leads to KAω1 not being small and so we therefore need to scale time differently when we
non-dimensionalise our flow, which is what we do in the next section.

17
1
3 Scaling time with ω0
Rather than introducing this dimensionless ω1 which may be large and contradict the assumptions the
analysis makes where KV1  V0 or Kω1  1, we can equate our time dependant pressure gradient to
1 + cos(ω0 t), where ω0 is dimensional, and if we scale time differently such that in our derivation of
the Dean equations we have t = ω10 t∗ , we end up with a slightly different set of equations featuring a
parameter α known as the Womersley number
s
ρa2 ω0
α=
μ

The derived set of equations,though more difficult offer a greater understanding of pulsatile flow and
allow for less restrictive assumptions:
K
(ψθ VR − ψR Vθ ) = 1 + A cos(t) − α2 Vt + ∇2 V (3.16)
R
K
(ψθ ΩR − ψR Ωθ ) = ∇2 Ω − α2 Ωt − 2KV (VR sin(θ) + Vθ cos θ) (3.17)
R
Once again, if we match coefficients (of K) we end up with the following set of equations from (3.16) :

O(1) : α2 V0t = 1 + A cos t + ∇2 V0 (3.18)


2
O(K) : α V1t = ∇2 V 1 (3.19)
1
O(K 2 ) : (ψ1θ V0R − ψ1R V0θ ) = −α2 V2t + ∇2 V2 (3.20)
R
and from equation (3.17), we obtain:
 
cos θ
O(K) : 0 = ∇2 Ω1 − α2 Ω1t − 2V0 V0R sin θ + V0θ (3.21)
R
    
cos θ cos θ
O(K 2 ) : −∇2 Ω2 = −α2 Ω2t + 2 V0 V1R sin θ + V1θ + V1 V0R sin θ + V0θ (3.22)
R R

3.1 Solutions
Looking at equation (3.18) it seems appropriate to take particular solutions and combine them with
a general solution in order for us to obtain a V0 that satisfies the necessary conditions set out by the
equation. In this manner we end up with three different equations to solve for equation (3.18):

∇2 V0 − α2 V0t = −1
∇2 V0 − α2 V0t = −A cos t
∇2 V0 − α2 V0t = 0 (3.23)

This gives us a V0 in the form V0G + V0P 1 + V0P 2 where we have V0G is the general solution and is a
function of both t and R whilst V0P 1 is a particular solution and a function of time only and V0P 2 is only
a function of R. The particular solutions are simple to calculate as in either case one of the terms on the

18
left disappear depending on independence of a particular variable and we must also remember that V0 is
independent of θ. For the particular solutions we get:
A
V0P 1 = sin t (3.24)
α2
1
V0P 2 = (1 − R2 ) (3.25)
4
The general solution is more complicated to calculate as it is both a function of t and R. However, if we
model it in the following fashion we can arrive at a modified Bessel equation.
 
V0G = Re F (R)ieit
In which case, when plug this into equation (22), we get;
  
1 2 it
Re FRR + FR − iα F ie = 0
R
As eit cannot equal zero within a finite time period, then the term in the inner brackets must be equivalent
to 0, giving us a special case of the modified Bessel equation of zero order known as Kelvin’s Function of
zero order. As shown by W.W.Bell [1], any equation of the form:
d2 F dF
R2 +R − (ik 2 R2 + n)F = 0
dR2 dR
Has a corresponding solution:
1 1
F = C1 In (i 2 kR) + C2 Kn (i 2 kR)
3 3
= C1 Jn (i 2 kR) + C2 Yn (i 2 kR)
where Jn (R) is known as the bessel function of order n, Yn is called Neumann’s Bessel function of the
second kind of order n and C1 and C2 are constants to be determined by our boundary conditions. In
our equation, n = 0 and our k = α giving our solution for F :
 3   3 
F (R) = C1 J0 i 2 αR + C2 Y0 i 2 αR

To determine our unknown constants C1 and C2 we must use our boundary conditions on V0 , but of
importance is that at 0, Y0 has a singularity giving us an infinite value at R = 0 so C2 = 0. We therefore
end up with a V0 of the form:
1 A h  3  i
V0 (R, t) = (1 − R2 ) + 2 sin t + Re C1 J0 i 2 αR ieit (3.26)
4 α
If we apply the boundary condition, we end up with
A h  3  i
it
0 = sin t + Re C 1 J 0 i 2α ie
α2
Now we cannot be too sure as to whether or not our Bessel function is real or not, so we choose C1
appropriately giving us V0 in the final form:
  3  
1 A A J 0 i 2 αR

V0 (R, t) = (1 − R2 ) + 2 sin t + 2 Re   3  ieit  (3.27)


4 α α J i2 α
0

19
which gives

 3   3  
R A i 2 α J1 i 2 αR
V0R (R, t) = − − 2 Re   3  ieit 
2 α J i α
2
0

If we now look at equation (3.19) we arrive at the solution


h  3  i h  3  i
V1 = Re C3 J0 i 2 αR ieit + (a cos θ + b sin θ)Re C4 J1 i 2 αR ieit + ...

where a and b are constants and we have assumed two possible forms of solutions in the form of some
function of only t and R whilst the second half of the solution is a function of all three variables. If we
consider other functions of only R, θ or t we arrive at contradictions. However, similarly, these general
solutions fall apart under boundary conditions, forcing our constants C3 and C4 to be set to 0. Looking
at the first of our vorticity equations (3.21), we once again separate our equations into particular and
general solutions.

∇2 Ω1 − α2 Ω1t = 2V0 V0R sin θ


   3  
1 A AR J 0 i 2 αR

= sin θ − (R − R3 ) − 2 R sin t − 2 Re   3  ieit  + ...


4 α α J0 i 2 α
  3   3  
 2 i 2 α J1 i 2 αR
A
... − sin θ  2 (2 sin t)Re   3  ieit  + ...
α J0 i α 2

  3   3  
A i 2α J 1 i 2 αR

... − sin θ  2 (1 − R2 )Re   3  ieit  + ...


2α J0 i 2 α
   3    3   3  
 2 J i 2 αR i 2α J i 2 αR
A 0 1
... − sin θ  2 2Re   3  ieit  Re   3  ieit 
α J i2 α J i2 α
0 0

Taking into account a generalised answer and the first two particular solutions we get:
h  3  i 11 1 3 1
 
AiR it

it 5
Ω1 = Re C5 J1 i 2 α ie + R − R + R sin θ − Re ie sin θ
4 24 8 16 α4

The remaining four particular solutions pose complex answers which we can’t fully analyse nor draw any
useful conclusions from so we will have to consider an alternative route.

3.1.1 Looking at a large Womersley parameter: α −→ ∞


In order to allow our calculations to proceed at a reasonable pace we make the assumption that our
parameter α is large, considering the high frequency limit. Letting α −→ ∞ allows us to simplify our

20
Bessel functions to its asymptotic form. We can also reduce our Laplacian operator to the two possible
forms:
1
∇2 Ω − α2 Ωt becomes either ΩRR + ΩR − α2 Ωt or ΩRR − α2 Ωt
R
Owing to the fact that as our α is large, in order for our equation to balance in orders of magnitude we
require ΩRR to be in balance with our α2 Ωt term and we realise that as our R term is of O(1), this leads
∂2
to the assumption that ∂R  R1 ∂R

and so we can justify the neglect of some terms in the ∇2 operator.
Note that in either case we have dropped the R12 Ωθθ term being as it is comparatively the smallest value
in the differential equation. These assumptions that we have used concerning the Laplacian operator are
derived from the fact that oscillatory flow of a viscous fluid along a solid wall leads to the formation of
a particular boundary layer known as the Stokes boundary layer and these vorticity oscillations change
exponentially across the small boundary layer they are confined to. Further more, we only use these
assumptions on the terms which are important along the boundary layer. We also have the asymptotic
form of the Bessel function to consider in order to simplify our calculations:
 3   
(i + 1) (i+1)
√ αR ∼ e 2 αR as α −→ ∞

J0 i 2 αR ≡ J0
2
3 1 d h  3 i i 12 d h (i+1) i i 12 (i + 1) (i+1)
√ αR √ αR
J1 (i 2 αR) = − 3 J0 i 2 αR ∼ e 2 = √ e 2 as α −→ ∞
i 2 α dr α dr 2

Where we have used the fact that J1 (bR) = − 1b dr d


[J0 (bR)]. As the term α(i+1)

2
appears often in the
calculation, we shall assign it the label λ. Under this assumption we can rewrite V0 as:
1 A A h i
V0 (R, t) = (1 − R2 ) + 2 sin t + 2 Re eλ(R−1) ieit
4 α α
Using our new assumptions we now have the following set of equations to solve:
 h i
2 A A λ(R−1) it
Ω1RR − α Ω1t = sin θ − 2 R sin t − 2 Re Re ie + ...
α α
 2 !
A h i
λ(R−1) it
... + sin θ (2 sin t)Re λe ie + ...
α2
 h i
A 2 λ(R−1) it
... + sin θ (1 − R )Re λe ie + ...
2α2
 2 !
A h i h i
λ(R−1) it λ(R−1) it
... + sin θ 2Re e ie Re λe ie
α2

As well as:  
1 3
∇2 Ω1 − α2 Ω1t = sin θ (R − R)
4
Which we have already solved in the previous chapter and can substitute in directly. Now looking at each
equation separately, we can obtain particular solutions as well as a general solution, all of which can be
analysed with greater ease. Our general solution is of the form:
h √ √  i
Ω1G1 = sin θ Re C6 e iα(R−1) + C7 e− iα(R−1)) ieit

21

However, owing to the fact that α is large, the term e− iα(R−1) grows exponentially as we head away
from the surface and towards the centre (as R 6 1) and so we require C7 = 0. Our particular form of the
solution is:
 
1 3 5
 A  it 
Ω1P = sin θ γR + 3R − R − 4 Re Re + ...
96 α
  
A 2λ iR λ(R−1) it
... + 2 sin θ Re − + e ie + ...
α (1 − α2 )2 (1 − α2 )
 2   
A λeit ie−it λ(R−1) it
... + sin θ Re − + e ie + ...
α2 1 − 2α2 λ
    
A i 2λR i 1 4iλ2 2 λ(R−1) it
... + 2 sin θ Re − − + + R λe ie + ...
α 2(1 − α2 ) (1 − α2 )2 2(1 − α2 ) (1 − α2 )2 (1 − α2 )3
... +Ω1 P6

for γ an unknown constant. In order to solve the final particular solution, we need to make use of the
fact that:
1 ˉ ]
Re[X]Re[Y ] ≡ Re[(X + X)Y
2
for two complex functions X and Y, where X ˉ is the complex conjugate of X. So we now end up having
to solve
 2
A ˉ
2
Ω1RR − α Ω1t = sin θ Re[(eλ(R−1) ieit − eλ(R−1) ie−it )eλ(R−1) ieit ]
α2
 2 √
A
= 2
sin θ (Re[λie2λ(R−1) ie2it ] + Re[λe 2α(R−1) ])
α

which can be solved to give the following particular solution:


 2  
A λ √2α(R−1) iλ
Ω1P6 = sin θ Re e + e2λ(R−1) e2it
α2 2α 2 2(2 − α2 )

But we must not forget to also include the general solution involving e2it :
h √ √  i
Ω1G2 = sin θ Re C8 e 2iα(R−1) + C9 e− 2iα(R−1)) ie2it

and for the same reasons regarding exponential growth as we head away from the surface, we set C9 = 0.
Now we have to use our original definition for Ω1 in order to solve for ψ1 whilst amending this definition
in accordance to the assumptions we made at the start of this subsection (note that the value of γ is

22
different from the previous section):
h √ √ i
ψ1RR = sin θ Re C6 e iα(R−1) ieit + C8 e 2iα(R−1) ie2it + ...
 
1 3 5
 A  it 
... + sin θ γR + 3R − R − 4 Re Re + ...
96 α
  
A 2λ iR λ(R−1) it
... + 2 sin θ Re − + e ie + ...
α (1 − α2 )2 (1 − α2 )
 2   
A λeit ie−it λ(R−1) it
... + sin θ Re − + e ie + ...
α2 1 − 2α2 λ
    
A i 1 4iλ2 2λR iR2 λ(R−1) it
... + 2 sin θ Re − + + − + λe ie + ...
α 2(1 − α2 ) (1 − α2 )2 (1 − α2 )3 (1 − α2 )2 2(1 − α2 )
 2  
A λ √2α(R−1) iλ 2λ(R−1) 2it
... + sin θ Re e + e e
α2 2α2 2(2 − α2 )

Allowing for the various combinations of general solutions that satisfy our equation and choosing those
which will help satisfy our boundary conditions, we arrive at the following set of solutions:
 
C6 √iα(R−1) it C8 √2iα(R−1) 2it
ψ1 = sin θ Re e ie + e ie + ...
iα2 2iα2
... +R sin θ Re[C12 ieit + C13 ie2it ] + ...
    3 
1 γ 1 1 A R it
... + sin θ R + R3 + R5 − R7 − 4 Re e + ...
96 8 8 48 α 6
  
A 2λ iR 2i 1 λ(R−1) it
... + 2 sin θ Re − + − e ie + ...
α (1 − α2 )2 (1 − α2 ) λ(1 − α2 ) λ2
 2   
A eit ie−it 1 λ(R−1) it
... + sin θ Re − + 2 e ie + ...
α2 1 − 2α2 λ λ
    
A i 1 4iλ2 4 − 2λR 1 λ(R−1) it
... + 2 sin θ Re − + + + e ie + ...
α 2(1 − α2 ) (1 − α2 )2 (1 − α2 )3 (1 − α2 )2 λ
 2 2  
A i(λ R − 2λ(R + 1) + 2) 1 λ(R−1) it
.. + 2 sin θ Re e ie + ...
α 2(1 − α2 ) λ3
 2  
A λ √2α(R−1) i 2λ(R−1) 2it
... + sin θ Re e + e e (3.28)
α2 4α4 8λ(2 − α2 )

23
We now need to solve for the unknown constants that are left in the solutions by enforcing our boundary
conditions on ψ1 :
 
C6 it C8 2it
ψ1 (1, θ.t) ≡ 0 = sin θ Re ie + ie + ...
iα2 2iα2
... + sin θ Re[(C12 ieit + C13 ie2it ] + ...
    
1 γ 5 A 1 it
... + sin θ + + − 4 Re e + ...
96 8 58 α 6
  
A 2λ i 2i 1 it
... + 2 sin θ Re − + − ie + ...
α (1 − α2 )2 (1 − α2 ) λ(1 − α2 ) λ2
 2   
A ie2it 1 1
... + sin θ Re − − + ...
α2 1 − 2α2 λ2 λ
 2  
A λ i 2it
... + sin θ Re + e + ...
α2 4α4 8λ(2 − α2 )
    
A i 1 4iλ2 4 − 2λ 1 it
... + 2 sin θ Re − + + + ie + ...
α 2(1 − α2 ) (1 − α2 )2 (1 − α2 )3 (1 − α2 )2 λ
  2  
A i λ − 4λ + 2
... + 2 sin θ Re 2
ieit
α 2(1 − α ) λ3

and on ψR :
 
C6 it C8 2it
ψ1R (1, θ, t) = sin θ Re √ ie + √ ie + ...
iα 2iα
... + sin θ Re[C12 ieit + C13 ie2it ] + ...
    
1 3γ 23 A 1 it
... sin θ + + − 4 Re e + ...
96 8 58 α 2
  
A 2λ2 i(λ − 1) 1 it
... + 2 sin θ Re − + ie + ...
α (1 − α2 )2 (1 − α2 ) λ2
 2  
A ie2it 1
... + sin θ Re − − + ...
α2 1 − 2α2 λ2
 2  
A λ i 2it
... + sin θ Re √ + e + ...
α2 2 2α3 4(2 − α2 )
    
A i 1 4iλ2 2(1 − λ) it
... + 2 sin θ Re − + + + ie + ...
α 2(1 − α2 ) (1 − α2 )2 (1 − α2 )3 (1 − α2 )2
  
A i (λ − 2) it
... + 2 sin θ Re 2
ie
α 2(1 − α ) λ

24
By matching coefficients, (powers of eit ), we end up with the following set of equations:
 
C6 iA A 1 i(1 + λ) 2λ − 1 4iλ
+ C 12 = − + + + = D1
iα2 6α4 α2 λ λ3 Y λY 2 Y3
 
C iA A i (1 + 2λ) 4iλ2 )
√ 6 + C12 = − 4 + 2 − + = D2
iα 2α α λ2 Y Y2 Y3

where Y = (1 + α2 ) and the values D1 and D2 have been assigned for ease of reference as they do not
contain any variables in them. We can solve this to get values for both C6 and C12 :

iα2
C6 = √ [D1 − D1 ]
1 − iα
1 h √ i
C12 = √ D2 − iαD1
1 − iα

Now looking towards the coefficients of e2it we have:


 2   
A 2iα2
1−λ 2λ − 1
C8 = √ −
α2 λ(1 − 2α2 ) 8λ(2 − α2 )
1 − 2iα
 2 " √ ! √ !#
A 1 (1 − λ) 2iα 1 (1 − 2λ) 2iα
C13 = 2 2
λ− √ − 2λ − √
α λ(1 − 2α ) 1 − 2iα 8λ(2 − α2 ) 1 − 2iα

Finally we also get:



2 " √ #
A λ − 1 λ(1 − 2α) 3
γ = 384 Re + −
α2 λ3 4α4 2
 2 " √ #
A 3 − λ λ( 2α − 3) 1
 = 48 Re + +
α2 λ3 4α4 12

If we now solve for (3.20) using the values we have obtained so far we have an equation for V2 :
1
∇2 V2 − α2 V2t = (ψ1θ V0R − ψ1R V0θ )
R
1
V2RR − α2 V2t = ψ1θ V0R
R 
ψ1 R A h i cos θ
= − + 2 Re λeλ(R−1) ieit
R 2 α sin θ

This is quite straight forward to calculate using the methods we have seen before in this section but it is
lengthy and as we are more interested in the stream function which doesn’t make use of V2 we can ignore
this equation. If we look at our final vorticity equation (3.22) we see that due to the fact that V1 = 0
and V0θ = 0 we end up with:

∇2 Ω2 − α2 Ω2t = 0

25
∂2
which if we solve for ∇2 ≡ ∂R2 and by the fact that we cannot have exponential growth towards the
centre of the flow, we get :

Ω2RR − α2 Ω2t = 0
h √ i
Ω2 = F1 (θ) Re C14 e iα(R−1) ieit
  
C14 √iα(R−1) it
ψ2 = F1 (θ) Re e + C 16 R + C 17 ie
iα2

where F1 (θ) is function of θ that can be decided on depending on any other constraints the system may
have and can be expected to be proportional to a combination of cos θ or sin θ. If we look at our boundary
conditions we are simply solving:
  
C14 it
ψ2 (1, θ.t) ≡ 0 = F1 (θ) Re + C16 + C17 ie
iα2
  
C14 it
ψ2R (1, θ.t) ≡ 0 = F1 (θ) Re √ + C16 ie

√ 
C14 iα−1
what we end up with is that C16 = − √ iα
and that C 17 = C 14 iα 2 leaving us with a general solution
that satisfies the boundary conditions and is quite flexible.

3.2 Analysis of Flow


Using the derived equations, much like in the first section, we wish to see how time affects the flow and
so we wish to calculate the time averages for V0 and ψ1 :
   3  
Z 2π Z 2π J i 2 αR Z 2π
1 1 0 A 0 0 A 0 0 0
hV0 i = (1 − R2 ) dt + 2 sin t dt + 2 Re   3  ieit dt 
2π 4 0 α 0 α J i2 α 0
0

1
= (1 − R2 )
4
and has flow rate:  
π A 2A  
{V0 } = 1+ 2
sin(t) + 2 Re (e−λ − 1 + λ)eit
8 α α
With the calculated time average expected and is the same as the previous section. Now for the more
interesting case of ψ1 where rather than calculating the time average using the lengthy equations we
found, we calculate the time average through our equation (3.21):

2
∇ Ω1 − αΩ1t = ∇2 hΩ1 i − α hΩ1t i = 2 hV0 V0R i sin θ

26
 
and as V0 = g(R) + Re f (R)eit and the time average of Ω1t = 0 we can write this as

  
dg df ∗
∇2 hΩ1 i = 2 g + Re f sin θ
dR dR
  
1 3 A2 2α(1 − i)  √2α(R−1) α(1−i)
√ (R−1)
= (R − R) + 4 Re √ e −e 2 sin θ
4 α 2
 

1 1 5 1 3
∇2 ψ1 = hΩ1 i = R − R + L1 R sin θ + ...
4 24 8
2
  
A 2(1 − i) 1 √2α(R−1) α(1−i)
√ (R−1)
... + 4 Re √ e − ie 2 sin θ
α α 2 2
1
hψ1f i = (4R − 9R3 + 6R5 − R7 ) sin θ + ...
4608   
A2 2(1 − i) 1 √2α(R−1) α(1−i)
√ (R−1)
... + 4 Re √ e +e 2 sin θ
α α3 2 4
   
A2 (1 − i) 3α − 2i 5
... − 4 Re √ √ − R3 sin θ + ...
α α3 2 2 2 4
   
A2 (1 − i) 3α − 2i 15
... + 4 Re √ √ − R sin θ
α α3 2 2 2 4

We can then compare this with the time average for ψ1 using our equations and we get something quite
similar with differences arising due to the complex conjugate of λ:
 2 " √ ! √ ! #
sin θ A λ − 1 λ(1 − 2α) 3 3 − λ λ( 2α − 3)
hψ1s i = Re + R + + R + ...
2 α2 λ3 4α4 λ3 4α4
1 
... + sin θ R(4 − 9R2 + 6R4 − R6
4608
 2  
A 1 λ(R−1) λ √2α(R−1)
... − sin θ Re e − e
α2 λ3 4α4

The flow rate of ψ1 = 0 due to sin θ.

27
3.3 Graphical Interpretation
We are interested in the contour plots of the different time average for the ψ1 term of our flow and how
they behave at different values of amplitude A as we vary the Womersley parameter α. What determines
whether or not α is large is how it affects the Bessel function we have and how it approximates it to an
exponential function. Therefore in our requirement of α  1 we can set α to around 5 and above. The
ratio of the amplitude A to α is important due to the arrangement of parameters in the time average
equation and A needs to be larger than α, otherwise the purely radial term dominates and we end up
with a flow much similar to that of ψ1 in the first section:

Figure 8: The contour plots on the left show the time average for ψ1f at fixed α = 8 and changing the amplitude from
A = 15 in the top graphs to A = 70 in the bottom ones. Whilst the graphs for ψ1f don’t appear to change much we notice
it starts to head from the centre towards the top and bottom of the pipe. The graph on the right shows the contour plots
for the time average equations of ψ1s . We look at these two cases as we notice an additional aspect of the flow forming in
the upper and lower regions of the pipe as the amplitude increases in the contour plots for ψ1s .

The calculated time average for ψf suggests a deformation in the flow at a lower amplitude and as
this increases the average effect of time leads to an influence across fluid that is focused very much at
the centre of the flow. The contour plots for ψ1s acts very much like the ψ1 seen in the first section but
as the amplitude increase, the effect of time seems to be felt close to the boundary of the pipe. What is

28
also interesting is that if we lower the value of α by half we end up with a flow that differs greatly from
our initial ψ1 and the variation occurs at slight changes in amplitude:

Figure 9: The contour plots on the left show the time average for ψ1s at fixed α = 7 at A = 19 and the graph on the right
is A = 20. Unlike ψ1f which develops into one kind of flow and remains as such, ψ1s seems to transition between three
forms as the amplitude changes. The core also appears to shrink as the additional layers of flow form.

Figure 10: The contour plot on the left shows the time average for ψ1s at fixed α = 5 and an amplitude of A = 8 and the
graph on the right is of ψ1f at amplitude A = 103 just so we can understand how the flow behaves at different extremes of
amplitude. If we alter the Womersley parameter to 10, we cannot obtain the above graph for values of A up to and
probably beyond 10150

What we see here is a case of steady streaming, even though there appears to be a problem at the
centre of the flow unlike ψ1f , but it does show how the flow has settled down over time, for further work
see Lyne [6]. Unlike the previous section where the chosen scaling restrained the time dependence, which
was seen in the contour plots, the scaling for this section has allowed for a system which depending on the
parameters α and A we see a flow whose dependence on time is greater. Furthermore, the time averages
for the last section did not alter with the change in value for amplitude but in this section, the increase
in amplitude manages to have a noticeable effect on the flow.

29
3.4 Summary
Looking over this section, we have managed to obtain the following important results:
  3  
1 A A J 0 i 2 αR

V0 = (1 − R2 ) + 2 sin t + 2 Re   3  ieit 
4 α α J i2 α
0

1 A A h i
≡ (1 − R2 ) + 2 sin t + 2 Re eλ(R−1) ieit (in the case of high frequency limit)
4 α α
V1 = 0

1
hV0 i = (1 − R2 )
4
 
π A 2A  −λ it

{V0 } = 1 + 2 sin(t) + 2 Re (e − 1 + λ)e
8 α α
{V1 } = hV1 i = 0

ψ1 = see (3.28)
" √ ! #
1 √iα(R−1) 1 iα − 1 it
ψ2 = F1 (θ)Re C14 e − √ R+ ie
iα2 iα iα2

1
hψ1f i = (4R − 9R3 + 6R5 − R7 ) sin θ + ...
4608   
A2 2(1 − i) 1 √2α(R−1) α(1−i)
√ (R−1)
... + 4 Re √ e +e 2 sin θ
α α3 2 4
   
A2 (1 − i) 3α − 2i 5
... − 4 Re √ √ − R3 sin θ + ...
α α3 2 2 2 4
   
A2 (1 − i) 3α − 2i 15
... + 4 Re √ √ − R sin θ
α α3 2 2 2 4
 2 " √ ! √ ! #
sin θ A λ − 1 λ(1 − 2α) 3 3 − λ λ( 2α − 3)
hψ1s i = Re + R + + R + ...
2 α2 λ3 4α4 λ3 4α4
1 
... + sin θ R(4 − 9R2 + 6R4 − R6
4608
 2  
A 1 λ(R−1) λ √2α(R−1)
... − sin θ Re e − e
α2 λ3 4α4
hψ2 i = 0
{ψ1 } = 0
{ψ2 } = 0

30
4 Conclusion
The desire to understand how the flow behaves under a pulsatile driving force lies in the understanding
of blood related diseases, such as high blood pressure or those involving the wall shear stress, much like
what Siggers et al investigated. The importance of curvature should also be recognised as even with slight
curvature, the centrifugal forces introduce cross flow, which is a contributing factor to health concerns
linked with blood flow. For example, Peripheral arterial disease is another case which would benefit from
the incorporation of pulsatile flow through a curved flow with a narrowing one but for further compre-
hension of this disease, we need to fully understand unsteady flow through a curved pipe.

We have seen how the pulsatile factor within a slightly curved pipe affects the flow and it is interesting
to compare the effect of time in differently scaled models. The next stage would be to look more thor-
oughly into steady streaming, see how the flow settles down over time and how it varies with curvature.
This would involve further expansion in powers of the Dean number K so as to include powers of K 3
and note the effects that time has on other components of the velocity and the stream function. We also
assumed that the Dean number was small and adjusted our model accordingly but an interesting case
to consider would be that of a high Dean number as it exhibits a recognisable boundary layer and the
flow interaction in that layer is also of importance. A variation in curvature will make the initial model
harder to simplify but it is also an aspect of the blood flow that needs to be fully understood.. The
constraints in the model limit how much we can analyse the flow and apply towards real life scenarios
involving blood vessels as veins and arteries tend not to have such limited curvature. Another area worth
investigating is the elasticity associated with the walls of veins and how the deformation in shape effects
the oscillation of the flow.

The first half of this project dealt with a quasi steady flow where we effectively let α −→ 0 whereas
the second half of the project dealt with a high frequency limit where α −→ ∞ and the difference in
the cross sectional pipe flows is due to the steady streaming present. The assumptions made helped to
simplify the model of the pipe flow yet the calculations were still lengthy and complicated. By looking
into pulsatile flow through a curved pipe I feel that this project should be able to contribute to an aspect
of understanding blood flow. Whilst there are still several areas that still need to be understood when
talking about blood flow, hopefully this project has helped to highlight the importance of both combing
time with pipe curvature. Finally, I would like to thank my project supervisor Doctor Johnathan Mestel
for all the invaluable help he has provided in the completion of this project.

31
References
[1] BELL, W. W. 1968 Special Functions for Scientists and Engineers. D. Van Nostrad Company Ltd.
London. Princeton. New Jersey Toronto Melbourne. Pg 110
[2] BERGER, S.A., TALBOT, L. & Yao, L.S. 1983 Flow in curved pipes. Ann Rev. Fluid Mech. Annual
Reviews. Pg 461-512
[3] BOWMAN, F. 1938 Introduction to Bessel Functions, Longmans. Green And Co. Ltd. Pg 41

[4] DAVID, N. Ku. 1977 Blood Flow In Arteries. Ann Rev. Fluid Mech. Pg 399-434

[5] GOLDSMITH, H. L. & SKALAK, R. 1975 Hemodynamics. Annual Reviews. Pg 213-242


[6] LYNE, W. H. 1971 Unsteady viscous flow in a curved pipe. J. Fluid Mech. Pg 45, 1331.

[7] MAZMUNDAR, J. N. 1992 Biofluid Mechanics. World Scientific Publishing Co.


[8] MESTEL, J. BioFluids Lecture 15: Steady flow in slightly curved pipes: the Dean equations.
http://www2.imperial.ac.uk/ ajm8/BioFluids/lec1015.pdf. Imperial College of London

[9] MESTEL, J. BioFluids Lecture 16: Steady flows for various Dean numbers.
http://www2.imperial.ac.uk/ ajm8/BioFluids/lec1016.pdf . Imperial College of London

[10] PEDLEY, T. J. 1977 Pulmonary Fluid Dynamics. Ann Rev. Fluid Mech. Annual Reviews. Pg 229-274

[11] PEDLEY, T. J. 1980 Fluid Mechanics of Large Blood Vessels. Cambridge University Press

[12] SIGGERS, J. H. & WATERS, S. L. 2008 Unsteady flow in pipes with finite curvature. Cambridge
University Press. J. Fluid Mechanics vol.600. Pg 133-165

[13] SKALAK, R., OZKAYA, N. & SKALAK, T. C. 1997 Biofluid Mechanics. Ann Rev. Fluid Mech. Pg
167-204

32

Вам также может понравиться