Sunteți pe pagina 1din 24

MMAE 514 PROJECT REPORT

Stability Analysis of Taylor-Couette Flow

Katherine Asztalos & Jorge Pulpeiro González


May 5t h , 2017

A BSTRACT
The flow field contained between two concentric rotating cylinders is a classi-
cal fluid stability problem. In this project, we studied the stability of the full 3D
case as well as the axisymmetric 2D case, and obtained accurate results when
compared with G.I. Taylor’s experimental results from his classical paper pub-
lished in 1923. Also, we studied the flow characteristics and determined the cor-
responding critical parameters governing the stability of different geometrical
configurations of the two concentric rotating cylinders.

Taylor-Couette flow.
N OMENCLATURE
T Taylor number
Tcr i t Critical Taylor number
Re Reynolds number
Re cr i t Critical Reynolds number
µ Ratio of outer to inner angular velocity
α −(1 − µ)
η Ratio of inner to outer radius
g
� Gravitational constant
ur Radial velocity
uθ Azimuthal velocity
uz Height velocity
P Pressure
� P /ρ
ur � Perturbed radial velocity
uθ � Perturbed azimuthal velocity
uz � Perturbed height velocity
P� Perturbed pressure
�� Perturbed P /ρ
uˆr Eigenfunction for radial velocity
uˆθ Eigenfunction for azimuthal velocity
uˆz Eigenfunction for height velocity
P̂ Eigenfunction for rressure
�ˆ Eigenfunction for P /ρ
uˆr∗ Non-dimensional scaled radial velocity
ξ Non-dimensional coordinate in radial direction
k Dimensional wavenumber in "z"
a Non-dimensional wavenumber in "z"
m Dimensional wavenumber in "θ"
c Dimensional eigenvalue
σ Non-dimensional eigenvalue
1 I NTRODUCTION

1.1 B ACKGROUND AND M OTIVATION

The study of fluid between two concentric cylinders can be considered a classical stability

problem. G.I. Taylor’s results, published in 1923, are still being examined due to an interest

in the interactions between axisymmetric and non-axisymmetric modes, and due to the va-

riety of flow configurations based on different geometric constraints and motion constraints

that still provide opportunities to learn something new about this flow configuration.

Figure 1.1: Coordinate system of definition given by Taylor[6].

The purpose of this project was to solve the temporal stability problem for fluid in-between

two concentric cylinders. Therefore, we selected real wavenumbers in the height and az-

imuthal direction and searched for imaginary eigenvalues such that the disturbances grew

exponentially with time; that is, σI m < 0.

The software Mathematica was used along with a package written specifically for non-

modal stability analysis through use of a spectral expansion using the tau method with

Chebyshev polynomials as the basis functions [5].

1
Analysis of the full 3D, non-axisymmetric problem was performed. Upon solution of the

problem, we found that the most unstable mode always occurred when the azimuthal wavenum-

ber m was zero; this led us to focus on the 2D axisymmetric problem. Comparison with G.I.

Taylor’s experimental results was completed, which showed very good agreement between

our results and Taylor’s.

1.2 O BJECTIVE

In addition to solving the temporal stability problem for both the non-axisymmetric and

axisymmetric case, we also looked to recover aspects of the flow field due to the most un-

stable mode and compared this to what we expected based on the existing literature on the

subject as well as our own results from computational fluid dynamics (CFD).

2
2 M ETHODOLOGY

2.1 S TABILITY A NALYSIS USING Mathematica

The software Mathematica was chosen to study the temporal stability problem with Taylor-

Couette flow in order to test the use of the package NMStools.m (also called Eigen.m), which

was written specifically to solve differential eigenproblems which arise in stability calcula-

tions [5]. The differential equation supplied to the function EigenNDSolve is discretized with

a spectral expansion with Chebyshev polynomials using the tau method. At the most basic

level, a spectral tau method looks to represent a function as an infinite sum of some basic

functions φn (x) such that




u(x, t ) = a n (t )φn (x), (2.1)
n=1

where the coefficients a n are sought for a finite sum [4]. Of course, the function u can be

described by some spatial operator L and corresponding boundary operator B such that

∂u
= Lu + f (2.2)
∂t

and

Bu = 0. (2.3)

The differential equations supplied to Mathematica were obtained from the cylindrical Navier-

Stokes equations after we assumed small perturbations about the base flow solution for

the azimuthal velocity u θ (r ) and implemented our normal mode ansatz. The axisymmet-

ric equations were simplified such that we obtained a system of two coupled equations for

the radial and azimuthal velocity; this was accomplished through use of continuity and the

z-momentum equation in the elimination of the pressure term.

3
2.2 P ROBLEM F ORMULATION

The incompressible cylindrical Navier-Stokes equations are described by the following set

of differential equations:
∂u r u r 1 ∂u θ ∂u z
+ + + =0 (2.4)
∂r r r ∂θ ∂z

2 � �
∂u r ∂u r u θ ∂u r u θ ∂u z ∂ 1 ∂� ��
+ ur + − + uz =ν r ur
∂t ∂r r ∂θ r ∂z ∂r r ∂r
2 2 �
1 ∂ ur ∂ ur 2 ∂u θ 1 ∂P
+ 2 2
+ 2
− 2 − + gr (2.5)
r ∂θ ∂z r ∂θ ρ ∂r

� �
∂u θ ∂u θ u θ ∂u θ u r u θ ∂u θ ∂ 1 ∂� ��
+ ur + + + uz =ν r uθ
∂t ∂r r ∂θ r ∂z ∂r r ∂r
2 2 �
1 ∂ uθ ∂ uθ 2 ∂u r 1 1 ∂P
+ 2 2
+ 2
+ 2 − + gθ (2.6)
r ∂θ ∂z r ∂θ r ρ ∂θ

� �
∂u z ∂u z u θ ∂u z ∂u z 1 ∂ � ∂u z � 1 ∂2 u z ∂2 u z
+ ur + + uz =ν r + 2 +
∂t ∂r r ∂θ ∂z r ∂r ∂r r ∂θ 2 ∂z 2
1 ∂P
− + gz (2.7)
ρ ∂z

with continuity given by equation 2.4 and equations 2.5 - 2.7 describing momentum in the

r, θ, & z direction, respectively.

Figure 1.1 shows the configuration of the two concentric cylinders confining the fluid that

we are interested in studying. We are interested in a steady, axisymmetric solution fully de-

veloped in "z" such that we only have one velocity component, u θ (r ). With these assump-

tions and the assumption that we can neglect gravitational forces, equation 2.5 simplifies

to
u θ2 1 ∂P
− =− (2.8)
r ρ ∂r

4
which effectively describes a balance between centrifugal forces and a radial pressure gra-

dient. From this, the pressure can be solved using


dr 2
P − Po = ρ u (r ). (2.9)
r θ

Equation 2.6 ultimately provides us with our steady solution for u θ (r )

� � �
∂ 1 ∂� ��
0=ν r uθ , (2.10)
∂r r ∂r

which can be solved using the following boundary conditions for u θ

u θ (r = R 1 ) = Ω1 R 1 (2.11)

u θ (r = R 2 ) = Ω2 R 2 . (2.12)

Of course, the corresponding boundary conditions for angular velocity are described by:

ω(r = R 1 ) = Ω1 (2.13)

ω(r = R 2 ) = Ω2 . (2.14)

It can be easily verified that the solution to 2.10 is simply

B
u θ = Ar + , (2.15)
r

5
where the constants A & B are determined by the boundary conditions to be 2.13 and 2.14

Ω2 R 22 − Ω1 R 12
A =
R 22 − R 12
Ω1 − Ω 2
B = . (2.16)
1/R 12 − 1/R 22

Finally, the angular velocity can be described by

B
Ω(r ) = A + . (2.17)
r2

2.3 3D E IGENPROBLEM

We first consider the non-axisymmetric problem formulated from the cylindrical Navier-

Stokes equations given by equations 2.4 - 2.7. We are interested in the temporal evolution of

small perturbations to our base state; therefore, we introduce the following perturbations

u = ur �

v = uθ + uθ �

w = uz �
p P P�
= + = � + � �. (2.18)
ρ ρ ρ

Linearizing about the base flow, we obtain the following set of perturbed, linearized equa-

tions for the 3D case:


∂u r� u r�
1 ∂u θ ∂u z�
+ + + =0 (2.19)
∂r r r ∂θ ∂z

� �
∂u r� u θ ∂u r� 2u θ u θ ∂� � 2 � u r� �
+ − =− + ν ∇ ur − 2 (2.20)
∂t r ∂θ r ∂r r

6
∂u θ� �
u θ � � u θ ∂u θ
� ∂u θ 1 ∂� � �
2 �

2 � uθ �
+ + ur + =− + ν ∇ uθ + ur − 2 (2.21)
∂t ∂r r r ∂θ r ∂θ r r

∂u z� u θ ∂u z� 1 ∂� � 2 �
+ =− ν∇ u z (2.22)
∂t r ∂θ r ∂z

where the operator ∇2 can be described by

∂2 1 ∂ 1 ∂2 ∂2
∇2 = + + + . (2.23)
∂r 2 r ∂r r 2 ∂θ 2 ∂z 2

We assume a form of the solution such that

ˆ ) e i (c t +mθ+kz) ,
(u r � , u θ � , u z � , � � )(r, θ, z, t ) = (uˆr , uˆθ , uˆz , �)(r (2.24)

where c is a parameter that can be complex and m and z are the wavenumbers in the az-

imuthal and "height" direction, respectively.

We obtain the eigenproblem describing the full 3D problem upon substitution of the nor-

mal model ansatz given by equation 2.24 into our linearized equations 2.19 - 2.22. The eigen-

problem can be described by the following set of equations

r uˆr � + uˆr + i m uˆθ + i kr uˆz = 0 (2.25)

� �
i r c + i (Ar + B )m + νm + νr k + ν uˆr − νr uˆr � − νr 2 uˆr ��
2 2 2 2 2

� �
+2 i m − (Ar 2 + B ) uˆθ + r 2 �
ˆ� = 0 (2.26)

7
� �
i r 2 c + i (Ar 2 + B )m + νm 2 + νr 2 k 2 + ν uˆθ − νr uˆθ � − νr 2 uˆθ ��

+2(Ar 2 − νr )uˆr + i mr �
ˆ =0 (2.27)

� �
i r 2 c + i (Ar 2 + B )m + νm 2 + νr 2 k 2 uˆz − νr uˆz � − νr 2 uˆz �� + i kr 2 �
ˆ = 0, (2.28)

with the corresponding boundary conditions

uˆr (R 1 ) = 0 uˆr (R 2 ) = 0

uˆθ (R 1 ) = 0 uˆθ (R 2 ) = 0

uˆz (R 1 ) = 0 uˆz (R 2 ) = 0

ˆ 1 ) = 0.
�(R (2.29)

Thus, equations 2.26 - 2.29 describe the full 3D flow field for the normal mode ansatz

described by equation 2.24 for the linearized set of equations about the base flow 2.19 - 2.22.

2.4 2D E IGENPROBLEM

The 2D cylindrical Navier-Stokes equations are formulated by an additional assumption

that the fluid flow is fully developed azimuthally. We are still interested in the temporal

evolution of small perturbations to our base state; therefore, we introduce the same pertur-

bations as equation 2.18 into equations 2.4 - 2.7 with the assumption that the flow field is

fully developed azimuthally. Linearizing about our base state, we obtain the following set of

perturbed, linearized equations for the 2D case:

∂u r � u r � ∂u z �
+ + =0 (2.30)
∂r r ∂z

8
∂u r� 2u θ u θ� � u � � 1 ∂P �
− = ν ∇2 u r� − 2r − (2.31)
∂t r r ρ ∂r

∂u θ� � ∂u θ uθ � � u� �
+ + u r = ν ∇2 u θ� − 2θ (2.32)
∂t ∂r r r

∂u z� 1 ∂P �
= ν∇2 u z� − , (2.33)
∂t ρ ∂z

where the operator ∇2 can be described by

∂2 1 ∂ ∂2
∇2 = + + . (2.34)
∂r 2 r ∂r ∂z 2

We assume a form of the solution such that

(u r � , u θ � , u z � , P � )(r, z, t ) = (uˆr , uˆθ , uˆz , P̂ )(r ) e i (c t +kz) , (2.35)

and upon substitution of our normal mode ansatz 2.35 into equations 2.30 - 2.33, we obtain

the eigenproblem describing our fluid flow. The system of equations can be manipulated

to eliminate pressure using continuity and the z-momentum equation such that we are left

with a system of equations for u r and u θ such that

� �� � �1 �
DD ∗ − a 2 − i σ DD ∗ − a 2 uˆr∗ = −T a 2 2 − κ uˆθ (2.36)
r

� �
DD ∗ − a 2 − i σ uˆθ = uˆr∗ , (2.37)

9
where the operators D & D ∗ are given by

d2 1 d 1
DD ∗ = 2
+ − 2 (2.38)
dr r dr r
2
d 1 d
D ∗D = + (2.39)
dr 2 r dr

for the general case. The corresponding boundary conditions are no-slip at the cylinder

walls such that

u r = u θ = 0 at r = R1 & r = R2
d ur
= 0 at r = R1 & r = R2 (2.40)
dr

We have non-dimensionalized by the characteristic length scale important to this problem;

namely, the radius of the outer cylinder, R 2 . The wavenumber in z and the complex fre-

quency are therefore non-dimensionalized by

a 2 = k 2 R 22 (2.41)
cR 22
σ= . (2.42)
ν

We can also normalize the radial velocity such that

2AR 22
uˆr∗ = uˆr (2.43)
ν

The remaining non-dimensional parameters describe the geometry of the problem as well

as the properties of the fluid. The Taylor number T can be described as

4AB 2 (1 − µ)(1 − µ/η2 )


T =− R = , (2.44)
ν2 2 (1 − η2 )2

10
where κ is another constant described as

µ
1−
AR 22
η2
κ=− = . (2.45)
B 1−µ

Finally the important parameters µ & η are

Ω2
µ= (2.46)
Ω1
R1
η= . (2.47)
R2

If we assume that the gap between the concentric cylinders is small such that d = R 2 −

R 1 << R 2 , the operators D and D ∗ collapse into the same operator and we are left with the

simplified equations

� 2 �� � � �
D − a 2 − i σ D 2 − a 2 uˆr = 1 + α ξ uˆθ (2.48)

� 2 �
D − a 2 − i σ uˆθ = −T a 2 uˆr , (2.49)

where again we have non-dimensionalized the wavenumber and frequency by equations

2.41 and 2.42, respectively, and normalized the radial velocity by

2Ω1 d 2 a 2
uˆr∗ = uˆr . (2.50)
ν

The corresponding boundary conditions are again no-slip at the cylinder walls such that

u r = u θ = 0 at ξ=0&ξ=1 (2.51)
d ur
= 0 at ξ=0&ξ=1 (2.52)
dr

11
where ξ is the non-dimensional radial distance such that

r − R1
ξ= , d = R2 − R1 . (2.53)
d

The remaining non-dimensional variables describe the fluid flow; α can be written as

α = −(1 − µ), (2.54)

and the Taylor number appropriate to the dimensionalization of the axisymmetric problem

using the thin-gap assumption can be described as

4AΩ1 4
T =− d . (2.55)
ν2

µ and η are described by equations 2.46 and 2.47, respectively.

12
3 R ESULTS AND D ISCUSSION

3.1 3D C ASE

First, the 3D differential eigenproblem was solved for a range of k’s and m’s and the most

unstable eigenvalue for each k and m was saved. This study of the parametric space allowed

us to confirm that the most unstable eigenvalue would always occur for m = 0. A plot of this

study can be seen in Figure 3.1. Also, a slice of this plot for constant k can be seen in Figure

3.2.

Figure 3.1: Variation of the most unstable eigenvalue with the k & m wavenumbers.

13
{ cIm vs. m }
c Im

m
-10 -5 5 10

-0.5

-1.0

-1.5

Figure 3.2: Variation of the most unstable eigenvalue with the m wavenumber.

This finding allowed us to simplify the 3D differential eigenproblem to a 2D one simply

by setting m = 0. Once this was done, another parametric study for the wavenumber k was

performed. The result can be seen in Figure 3.3. The effect of the Taylor number in this

plot was also evaluated. Increasing the Taylor number would make the eigenvalues move

upwards and would make the minimum eigenvalues move towards k’s with a bigger absolute

value.
{ cIm vs. k }
c Im

k
-30 -20 -10 10 20 30

-1

Figure 3.3: Variation of the most unstable eigenvalue with the k wavenumber.

14
The Taylor number was varied in the same way as Taylor’s experiment [6]. This was done

because it was very desirable to compare the results obtained with our model to Taylor’s ex-

perimental results. In his results, Taylor has a fixed geometry with a fixed outer cylinder ro-

tational velocity, and varies the inner cylinder rotational velocity until observing instability

in the flow. This is how he would define the limit of stability. We followed the same approach

with our model. We fixed the outer cylinder velocity to the same value Taylor used, and then

changed the inner cylinder velocity until we found instability. In our problem this was rep-

resented by the zero crossing from positive to negative of the minimum eigenvalue. This is,

effectively, the definition of marginal stability for this problem.

A wide range of angular velocities of the outer cylinder was selected for analysis so that

we had a meaningful representation of the range that Taylor had used in his experiment. In

the following table we can see Taylor’s experimental results and our numerical results. Note

that there is good agreement between the two. Also note that if we consider the experimental

results as a reference, the error is less than 10% for nearly the entire range studied.

A plot of the results can be seen in Figure 3.4. Some numerical results from Taylor are

plotted, as well as the Rayleigh criterion. The Rayleigh criterion comes from the inviscid

stability analysis such that in the absence of viscostiy, the necessary and sufficient condition

for stability is µ < η2 everywhere in the flow field. The flow field is unstable if the the square

of the angular momentum r 2 Ω decreases anywhere in the interval. In this plot it becomes

even more evident that our numerical results show good agreement with the reference data

we considered.

15
Ω2 /ν Ω1 /ν (Taylor) Ω1 /ν (Our) Error
790.0 914.0 911.8 -0.24%
530.0 626.0 628.6 0.42%
340.0 431.5 430.0 -0.34%
196.3 294.0 294.8 0.29%
99.1 220.1 224.1 1.84%
54.9 200.1 203.4 1.67%
0.0 193.1 192.4 -0.34%
-41.5 192.0 195.8 2.00%
-101.5 219.0 215.7 -1.49%
-141.0 227.0 235.0 3.54%
-320.0 320.0 329.4 2.95%
-462.0 367.0 389.8 6.22%
-876.0 496.0 525.8 6.02%
-1104.0 553.0 589.7 6.64%
-1362.0 608.0 655.6 7.84%
-1672.0 666.0 728.4 9.36%
-2120.0 733.0 824.1 12.42%

Table 3.1: Table with Taylor experimental results and our numerical results.

16
Figure 3.4: Limit of stability according to various sources and methods.

3.2 2D C ASE

The results from the 3D non-axisymmetric case have shown that the most unstable mode

occurs for which the azimuthal wavenumber m is set to zero. Therefore, the most unstable

mode occurs for the axisymmetric 2D case, and we can use either equations 2.36 - 2.37 to

study the general case, or equations 2.48 - 2.49 to study the case for which d << R 2 .

To gain a better understanding of the relationship between the most unstable mode for

which σI m < 0 and the wavenumber a as well as the Taylor number T , we varied both of

these to see for which combination we obtained a maximum for σI m . Figure 3.5 shows the

results of σI m < 0 for different wavenumbers and Taylor numbers. It is clear to see how

increasing the Taylor number subsequently increases the magnitude of the most unstable

mode.

17
Figure 3.5: Variation of the most unstable eigenvalue with the wavenumber and the Taylor number.

It is of interest to separately study how changing the wavenumber and the Taylor num-

ber affect the most unstable mode. Figure 3.6 shows the results for a fixed Taylor number

with the wavenumber varied between two extremes. Figure 3.7 shows the results for a fixed

wavenumber with the Taylor number continually increased.

18
Figure 3.6: Variation of the most unstable eigenvalue with the wavenumber.

Figure 3.7: Variation of the most unstable eigenvalue with the Taylor number.

19
3.3 F LOW F IELDS & CFD

3.4 F LOW F IELDS

The 3D and 2D flow fields were reconstructed using the normal mode ansatz given by

equations 2.24 and 2.35, respectively. We set boundary conditions such that the only coef-

ficients left were the ones corresponding to the most unstable wavenumbers (positive and

negative k). This left us with a cosine profile in z for the perturbations in u r � , u θ � , and � � .

Also, we had a sinusoidal profile for u z � . With this and the eigenfunctions, we were able to

reconstruct all the perturbation functions. Figures 3.8 and 3.9 shows the results for the base

flow plus the perturbation, u r and u θ .

Figure 3.8: Superposition of the base flow and perturbation for a time of t = 0.18 seconds with z = 0 for ur .

20
Figure 3.9: Superposition of the base flow and perturbation for a time of t = 0.75 seconds with z = 0 for uθ .

3.4.1 O PEN F OAM

The steady, laminar fluid flow was simulated for the inner cylinder rotating at an angular

velocity of 4 [r ad /s] while the inner cylinder was stationary using the open-source software

OpenFoam [3]. A coarse 3D mesh was created using OpenFoam that contained 276, 268

mesh points. The simulation time step was Δt = 1 second; a large time step was cho-

sen for the steady problem. The steady solver simpleFoam was used with no turbulence

model. Pressure boundary conditions were zero pressure gradient for all boundaries. Ve-

locity boundary conditions were Dirichlet boundary conditions for angular velocity at the

cylinder walls and slip boundary conditions at the top and bottom of the cylinder. The

inner angular velocity was set to 4 [r ad /s] and the outer angular velocity was 0 [r ad /s],

corresponding to a case where µ = 0. The inner and outer radius were R 1 = 0.038[m] and

R 2 = 0.04035[m] as they are for Taylor’ s problem. For the results shown for the magnitude

of the velocity, the Reynolds was calculated to be Re = 1786 and the Taylor number was cal-

21
culated to be T = 382, 694 using Equation 2.55. This is far above the critical Taylor number

and Reynolds number, which is why we suspect that we see wavy rolls in our solution. How-

ever, it should be noted that at this high of a Reynolds number we have turbulent flow, and

therefore the absence of a turbulence model is a simplification that cannot really be justified

for anything further than just to get an idea of the velocity field. A more robust solution of

this problem using CFD would need to make the following changes:

• Use of a finer mesh

• Use of an unsteady solver (such as pimpleFoam, a merged piso - simple model. PISO =

pressure implicit with splitting of operators, where SIMPLE = semi - implicit pressure

linked equations) with smaller time steps

• Use of a turbulence model (start with a simple algebraic model, such as Spalart-Allmaras,

and go from there).

Please see the included video entitled "umag t a yl or couet t est ead y.avi " for the solution

to the velocity magnitude for a simulation run-time of 861 seconds.

22

S-ar putea să vă placă și