Sei sulla pagina 1di 19

Int. J. Engng ScL Vol. 35, No. 5, pp.

485-503, 1997

Pergamon

(~) 1997 Elsevier Science Ltd. All rights reserved


Printed in Great Britain
0020-7225197 $17.00 + 0.00

PII: S0020-7225(96)00126-7

LIQUID-METAL MHD FLOW IN RECTANGULAR DUCTS WITH THIN


CONDUCTING OR INSULATING WALLS: LAMINAR A N D
TURBULENT SOLUTIONS
S. CUEVAS t and B. E PICOLOGLOU
Argonne National Laboratory, Argonne, IL 60439, U.S.A.

J. S. W A L K E R
Department of Mechanical and Industrial Engineering, University of Illinois, Urbana, IL 61801, U.S.A.

G. TALMAGE
Department of Mechanical Engineering, Pennsylvania State University, University Park, PA 16802, U.S.A.

Abstract This paper treats the steady, fully-developed flow of a liquid metal in a rectangular
duct of constant cross-section with a uniform, transverse magnetic field. Thin conducting wall
boundary conditions at the top/bottom walls (perpendicular to the magnetic field) are extended
to allow electrical currents to return through either the wall or the Hartmann layers. Hence,
a unified analysis of flows in ducts with wall conductance ratios in the range of interest of
fusion blanket applications, namely, from thin conducting to insulating wall ducts, is conducted.
The flow in laminar and turbulent regimes is investigated through a composite core-side-layer
spectral collocation solution which explicitly resolves the flow in the side layers (parallel to the
magnetic field) even for very large Hartmann numbers. Turbulent profiles are obtained through
an iterative scheme in which turbulence is introduced through an eddy viscosity model from the
renormalization group theory of turbulence [Yakhot, V. and Orsag, S. A., J. Sci. Comput., 1986,
1(1), 3]. The transition from a flow in a duct with thin conducting walls to one with insulating
walls is clearly displayed by varying the wall conductance ratio from 0.05 to 0 for Hartmann
numbers in the range 103-105. In turbulent regime, Reynolds numbers vary in the range 5 x 1045 x 105. For thin conducting wall duct flows, turbulence is concentrated in the increased side layers
while the core remains unperturbed. In insulating wall ducts, the flow remains in the laminar
regime within the considered range of Reynolds numbers. 1997 Elsevier Science Ltd.

1. I N T R O D U C T I O N

Self-cooled liquid-metal blankets have been favored as a feasible alternative for fusion reactors
due to their performance characteristics and simplicity of design [1]. The absortion of energy
from the neutron flux, the breeding of tritium and the transfer of heat to the external energy
conversion system can be properly carried out by liquid-lithium blankets. The main constraints in
the design of self-cooled blankets occur as a result of the fact that, being located inside the windings,
the liquid metal flows in a very strong magnetic field (5-10 Tesla) which originates very intense
magnetohydrodynamic (MHD) effects. For instance, the strong MHD pressure losses increase the
power required to pump the liquid through the ducts forming the blanket. Large MHD pressure
drops may also originate unacceptable mechanical stresses in the first wall. As the flow patterns are
greatly modified by the MHD interaction, the heat transfer characteristics are also affected.
The dimensionless parameters that characterize the liquid-metal flows in self-cooled blankets
are the interaction parameter, the Hartmann number, the magnetic Reynolds number and the wall
conductance ratio, defined as

N=

2
o'B,,L,
pO

M=

oL

cr

Rm=lJcrOL,

O-wtw

c= o'L'

t To whom all correspondence should be addressed. Present address: Centro de Investigaci6n en Energia--UNAM,
A.P. 34, Temixco, Mor. 62580, Mexico.
485

486

S. CUEVAS et al.

respectively, where o-, p, v,, and ~ are the electrical conductivity, mass density, kinematic viscosity
and magnetic permeability of the liquid, respectively, and Crw and tw are the electrical conductivity
and thickness of the wall, respectively. In addition, 0 is the average flow velocity, L is a characteristic
length of the flow and B,, is the applied magnetic field strength. Notice that N = M 2 / R e , where
R e = ( J L / v o is the ordinary Reynolds number. Due to the large magnetic field strengths involved, N
and M are typically of the order 103-10 s. Under these conditions, electromagnetic forces dominate
over inertial and viscous forces in most of the flow. Hence, most of the theoretical studies in
this field have been performed within the inertialess approximation which assumes a laminar flow
where the electromagnetic force is the dominant interaction that determines the flow and pressure
distributions in the liquid metal, except for thin boundary layers [2]. In some approaches, the
velocity profile near the walls is not obtained in detail but only the volumetric flow rate is given.
Fully numerical solutions allow the exploration of the detailed flow in the near-wall region, however,
the high requirements of storage and computing time strongly limit these solutions to rather small
Hartmann numbers (M < 103) [3,4]. On the other hand, for fusion applications, the condition
R m << 1 holds, which indicates that the induced magnetic field is negligible in comparison to the
applied field. For electrically conducting wall ducts in fusion reactors, c ranges form 0.01 to 0.1 and
most of the theoretical analyses are based on the thin conducting wall approximation, expressed
by the condition M - l << c << 1 [5]. Since insulating wall ducts may eventually become a feasible
option, the range of interest of c for fusion applications can be extended from 0 to 0.1.
Although, in general, a very good agreement between experimental results and theoretical laminar
predictions has been found [6], blanket related experiments have demonstrated the existence of
persisting velocity fluctuations which are hitherto not satisfactorily explained. These fluctuations
appear in the region near the side walls of a rectangular thin conducting wall duct near blanket
relevant operation conditions [7]. It is speculated that fluctuations are originated as a result of a flow
instability in the high-velocity side layers which are characteristic of MHD flows in thin conducting
wall ducts. It is expected that the existence of high-energy, periodic fluctuations will produce an
excellent mixing in the side layer and bring about an enhancement of the heat transfer rates from the
first wall. However, a deeper insight into the mechanisms producing and controlling the instability is
required for practical applications. A preliminary linear stability analysis of the side-layer instability
flow indicates that the steady flow in the core is unconditionally stable and is never perturbed
by any fluctuations in the side layers [8]. Independent experimental studies have corroborated the
existence of near-the-wall velocity fluctuations in MHD flows in ducts with conducting walls [9],
though these experiments were carried out far away from reactor relevant conditions.
The experimental evidence of persisting side-layer velocity fluctuations makes it necessary to
extend previous studies in order to consider flows beyond the laminar regime. At present, the lack of
a thorough understanding of the side-wall instability prevents attempts at performing a theoretical
modeling of this phenomenon. Instead, the present study is aimed at the theoretical investigation
of turbulent liquid-metal duct flows in strong, uniform magnetic fields. We are mainly interested
in how the structure of the side layers is influenced by the electrical conductivity of the walls, the
magnetic field strength and the inertial effects. Instead of performing a separate analysis for thin
conducting and insulating wall ducts, in this investigation extended boundary conditions that allow
duct walls to vary from thin conducting to electrically insulating are formulated. A composite coreside-layer spectral collocation solution is obtained for the laminar regime and extended to include
turbulent effects through an eddy viscosity model from the renormalization group (RNG) theory of
turbulence [10]. This model has been used previously in the analysis of liquid-metal M H D turbulent
flows [11,12].
In Section 2, the governing equations and boundary conditions are formulated. In Section 3, a
composite core-side-layer solution is established and results obtained with the spectral collocation
method are shown in Sections 3.1 and 3.2 for laminar and turbulent flows, respectively. Finally, a
discussion of the results and concluding remarks are presented in Section 4.

Liquid-metal M H D flow

487

2. G O V E R N I N G E Q U A T I O N S
The system of equations governing the steady flow of a liquid metal immersed in a magnetic field
is
p ( U V ) I ! = - - V p q- rloV211 --I- j X B,

(1)

V u = 0,

(2a)

V l = 0,

(2b)

j = o-(-V4)

+ u x B),

(3)

V B = 0,

(4a)

V x B =/Jj.

(4b)

where r/o = pro is the dynamic viscosity of the fluid. In turn, u, p, j, qb and B are the velocity,
pressure, electric current density, electric potential and magnetic induction fields, respectively. All
transport coefficients are assumed to be constant. The Navier-Stokes equations with the Lorentz
force are given by (1) while conservation of mass and electric charge are expressed through equations
(2a) and (2b), respectively. Equation (3) represents Ohm's law in a moving media and equations
(4) govern the magnetic induction field where, in principle, B represents the superposition of the
applied magnetic field produced by the external magnets and the induced magnetic field generated
by the electric currents circulating in the liquid.
In order to get the governing Reynolds equations for the turbulent regime, we consider the flow
variables u, p and qb to consist of a mean part, which will be denoted by an overbar, and a fluctuating
part with a zero average, denoted by a prime, namely
u = fi + u',

(5a)

p =/~ + p',

(5b)

q6 -- 4] + q~',

(5c)

where, by definition, u' =/~' = 4]' = 0. Because of the approximation R m << 1, pulsations of the
magnetic induction field can be neglected in comparison with its averaged value, therefore, B = 13.
The mean flow equations are obtained by substituting equations (5) in (1)-(4) and averaging the
resultant equations. In dimensionless terms we get
N -1 (fi V ) u =

--XTp + M-2V (Vfi - Reu'u') + j B,

(6)

V fi = 0,

(7a)

V . j = 0,

(7b)

= -v~

+ ~ ~,

(8)

S. CUEVAS et al.

488

~y=a
Z--1

Z---1

~Z

T ~--

y ~--a 7

Fig. 1. Dimensionless duct cross-section.

V g = 0,

(9a)

V x B = Rm],

(9b)

where mean variables fi, /~, j, ~ and 13 are normilized by O, crOB2L, (rOB,,, OB,,L and B,,
respectively. We consider a turbulent liquid-metal flow in a constant area rectangular duct whose
axis coincides with the x-coordinate axis, in the presence of a uniform, transverse magnetic field,
B = ~ (see Fig. 1). All length dimensions are normalized by L, half the distance between the side
walls. Under these conditions the flow is fully developed and the problem becomes two-dimensional.
This is actually the case in most of the blanket regions. Laminar flows in regions where the magnetic
field is non-uniform and the flow is not fully developed are treated elsewhere [13,14]. Hence, fi =
(~(y, z), O, 0), ~ = (b(Y, z) and Vfi = ( - K , 0, 0), where K is a constant. Therefore, equation (6)
reduces to

M -2

-~y -~y - Reu'v'

~t7

a___~_~= 0.

(10)

In order to determine the turbulent Reynolds stresses, we use Boussinesq's approach which assumes
that the mean turbulent stresses can be expressed in the same form as the laminar stresses with a
turbulent viscosity, namely
3~7
Vey~-fy,

(1 la)

0t7
- R e u ' w ' = Ve~ 3z'

(1 lb)

- R e u ' v'

where vej (j = y, z) is known as the eddy viscosity (normalized by the molecular kinematic viscosity
Vo). In this approach, only mean values are calculated while fluctuations are modeled through
an eddy viscosity. Due to the anisotropic effect of the magnetic field, we allow for different eddy

Liquid-metal MHD flow

489

viscosities in each direction. Once we accept the Boussinesq postulate expressed by equations (11),
the closure problem is reduced to determining the scalars vej. Introducing equations (11) in equation
(10) and defining the total viscous diffusivities as
vty = 1 + Vey,

Vtz = 1 + Vez,

which account for the molecular and turbulent contributions, we finally reduce the set of equations
(6)-(9) to the form

M_ 2

VtYOU + "~ZO VU-~Z

-- fz . .3X
. .

K,

(12)

]y= a~,
ay'

(13a)

]..= - a~
a--z + 5,

(13b)

ai.

(14)

The mean variables also have to satisfy Ohm's law and current conservation at the walls

JYw = -

ay '

jzw = -

3z '

__

3]yw + a ~ w = 0,

(16)

ay

where the subindex w indicates a property or variable of the wall. At a wall-liquid interface, both
qb and j fi are continuous, where fi stands for the normal to the surface. By setting vt = 1 and
dropping the overbars, equations (12)-(16) reduce to the laminar set of equations.
2.1 S y m m e t r y a n d boundary conditions

Because of symmetry conditions in both the y- and z-directions, we need to consider only one
quarter of the duct's cross-section, for instance, y ___0 and z >__0. Equations (14) and (16) allow the
elimination of the electric current density by introducing electric current stream functions h and hw
for the fluid and the walls, respectively, defined as

]Y=
jyw =

3h
az'
3hw

~Z'

] _ = _ ah
ay'
f~w

3hw
= ay"

In terms of h, the Navier-Stokes equation (12) and Ohm's law (13) can be rewritten as

M -2

-~y Vty-~y

+ ~Z Vtz-~z

-- -~y = - g ,

(17)

S. CUEVAS et al.

490

Oh

aTb

Oz

Oy'

Oh

(18)

O4,

ay + 7 z =

19)

for 0 < y _< a a n d 0 _< z _< 1, where a is the duct's aspect ratio. In the wall, equations (15) b e c o m e

az

ay'

a---y-

= 0,

(20a)

(20b)

for a _< y < a + T a n d 0 _< z _< 1 + T, as well as for 1 _< z < 1 + T a n d 0 _< y < a, where T = tw/L
is the dimensionless wall thickness. In a d d i t i o n , we have:
A t y = 0:

0~
--

= 0,

0y

(21a)

0_~_~ = 0,

(21b)

h = 0;

(21 c)

Oy

at z = 0:
~t7

Oz

aty=aandatz=

O,

(22a)

qb = 0,

(22b)

Oh
Oz

(22c)

0;

1:

= 0,

and aty=a+Tandatz=

(23a)

qb = q~w,

(23b)

h = hw;

(23c)

I+T:

hw =

0.

(24)

Since ah/Oz = aqb/Oy, a suitable p o t e n t i a l function F(y, z) can be i n t r o d u c e d for h a n d q~, defined
in the following way:

~F
h=Ty,
and, similarly, in the walls we have

OF

Liquid-metal MHD flow

hw = Oy'

491

Oz"

In order to assure the uniqueness of F and Fw, the following condition must be satisfied:
Fw(0, 1 + T) = O,

F(O, 1) = Fw(0, 1).

In terms of the new potential, the equations and boundary conditions (17)-(24) take the form

M-2

0 Vty~yy -'1- ~zz V~ ~Z


~yy

= - ay
2

--

32F
32F
= ~
+ az2 ,
forO<_ y < _ a a n d O <

K,

(25)

(26)

z<_ 1.

a2Fw a2Fw
ay-----5-+ 3--ff~ = 0,
fora<_y<_a+TandO<z<

l+T,

or 1 < z <

(27)

l+Tand0<-y<-a.

At y = 0:
3/i

ay

(28a)

~0~

OF

(28b)

aFw -

0;

(28c)

atz=O:
3t~
~0~
Oz

(29a)

OF
Oz

(29b)

aFw

- -

3z

0~

= 0;

(29c)

a t y = a + T:

aFw =
ay

atz=

0;

1 +T:

Fw = 0 ;
atz=
]3ES 35:5-C

(30)

1:

(31)

492

S. CUEVAS et al.

fi = 0,

(32a)

F = F~,

(32b)

(?) az'

aTz

(32c)

and at y = a:
t7 = 0,
OF

(33a)

OFw

(33b)

Oy"

0Fw
( ? ) O=z

OF
Oz"

(33c)

2.2 Thin conducting and insulating wall conditions


In general, the electric current density and electric potential function o f the fluid, walls and
surrounding m e d i u m are coupled. However, the problem can be decoupled by assuming that the
walls are thin, T << 1, and the surrounding medium is an electrical insulator. With these assumptions
a b o u n d a r y condition on the fluid variables at the inside surface o f each wall can be derived [5]. In
order to do that, let us consider first the region 1 _< z _< l + T and define the stretched coordinate
I a Then, from equation (27) we get
Z = y1( z - l), so that, O/Oz - raz"
02Fw OZ 2

T 2 02Fw
ay2 '

0Fw
OZ - -

OF
c-~z (Y" 1).

(34)

and from equation (32c) we have

(35)

Integrating equation (34), using equations (35) and (31) and neglecting terms O ( T 2) we get

Fw =

c OF (
~z y, 1 ) ( Z - 1 ) ,

but at Z = 0, Fw = F(y, 1), so we finally have


OF
F(y, 1) = -c--~z(y, 1).

(36)

Equation (36) is an integral of Shercliff's b o u n d a r y condition [15] for the current stream function
h. We now consider the region a _< y _< a + T and define the stretched coordinate Y = (y - a ) / T .
F r o m (27), we have
02Fw _ _ T 2 02Fw
Oy 2
Oz2

(37)

In addition, from equation (30), at Y = 1

aFw
OY
and, from equations (33b,c), at Y = 0

-- 0,

(38)

Liquid-metal MHD flow

493

OFw
OF
= T-g-y(a, z),
OY
oy

(39a)

3z =

~z"

(39b)

Expressing Fw as a series in the parameter T, Fw = Fw0(Z) + T2FwI(Y, z)


becomes

02Fwl

02Fw0

3Y 2

3z 2 '

O(T4), equation

(37)

(40)

while equation (38) gives


aFwl

- -

3Y

= 0,

at Y = 1,

(41)

and, neglecting terms O(T2), equation (39b) reads

(42)

"-~-z"

az =

Integrating equation (40) and using boundary condition (41) we get

OFwl
3-7 =

O2Fw0, ,
~z 2 tz) ( Y -

(43)

1).

Then, from equations (39a), (42) and (43), we finally obtain


r (a, z) =
_

/.
1

3Fw (0, z)
3

TOY

= T

(rwo(Z)
+ TZVw~<O,z) +
\
(0, z) + O(T 3) =

OtT4)]
/

T d2Fw (z) + O(T 3)


dz 2

32F
= c--~z2 (a, z) + O(cT2),
that is

~yy(a,z)=

02F
C~z 2 (a, z).

(44)

Boundary condition (44) can also be expressed in terms of the electric current stream function as
Shercliff's condition. Condition (44) is valid within the thin wall approximation, M - l < < c < < 1,
which ignores the jump in jy across the Hartmann layer. In other words, the Hartmann layer is
completely disregarded. It is clear that boundary condition (44) does not give the correct limit when
c - 0, that is, when the walls become insulators. In such a case, all the return currents should flow
through the Hartmann layers, but equation (44) predicts no current at all when c = 0. Since we
are interested in exploring the transition of the flow patterns as c - 0, we must modify boundary
condition (44) to take into account the Hartmann layers. In order to do this, we need to estimate the
amount of current flowing through this layer. The component of the current density in the z-direction
is j_, = u - 3ch/Oz, where u is the core-layer velocity profile given by u = Uc( 1 - e-M~). Here Uc is the
external or core velocity and ~ is the coordinate perpendicular to the top wall, measured from the
wall. Since qb is continuous across the Hartmann layer within an error of O(M-2), Oqb/Oz can be

494

s. CUEVAS et al.

considered constant. On the other hand, the current density outside the layer is jzc = u - 3ch/Oz,
and then, the total current through the Hartmann layer is
(jz - jzc)d~ = uc
that

e-M~d~ = ~ ,

is

Ah = M-luc(a,z)

= M -1 [ 0 2 F
02F "
)
~ ~y2 (a,z) + -~z2 (a,z) ,

where equation (26) has been used. Finally, adding this contribution to equation (44), we get
__OF
O2F
= M _ 1 O2F
Oy (a, z) - (c + M - l ) - ~ z 2 (a, z)
0y2"

(45)

Hence, in the limit c = 0, boundary condition (45) states that all return currents will close through
the Hartmann layer. Strictly, this modification holds only for a laminar Hartmann layer and we
should be able to devise a proper modification for a turbulent layer. However, there are some
physical reasons for expecting that an instability in the Hartmann layer is not very likely to grow
into a turbulent boundary layer. The argument is that the shear layers in the Hartmann layers
have vorticity that is perpendicular to the magnetic field. Therefore, if the Hartmann layers become
unstable, the resultant instability will have vorticity predominantly perpendicular to the magnetic
field and will be quickly damped. In this way, the Hartmann layers serve mostly as return paths for
the currents. For these reasons, we consider (45) as a suitable boundary condition for both laminar
and turbulent cases. Hence, the boundary value problem can be stated as

M- 2

/)]

O Vty Oil -I- -~zO Vtz -~z

02 F
02 F
z7 = ~y2 + Oz2'
forO<_ y < a a n d O < _ z
At y = 0:

= - Oy
- 2

K,

(46)

(47)

<_ 1.

0IT

0y =

--

OF
- -

0y

0,

(48a)

= O;

(48b)

a t z = 0:

--

317
= 0,
3z

(49a)

3F
= 0;
3z

(49b)

ff = 0,

(50a)

3F
F + C-~z = 0;

(50b)

- -

atz=

1:

and at y = a:

Liquid-metal MHD flow

495

~=0,
OF

a--y- -

(51 a)

(c + M -l 02F = M -l 02F
) - ~ z2
3y2"

(51b)

In addition, the volume flux conservation condition has to be satisfied, namely

~0

' f o tTdz dy = a.

(52)

3. C O M P O S I T E C O R E - S I D E - L A Y E R

SOLUTION

The heat transfer characteristics and the eventual appearance of instabilities are mainly determined by the flow in the side layers, while the Hartmann layers play a rather passive role. For the
very high Hartmann numbers involved in fusion applications, the complete numerical resolution of
all the flow regions, especially the Hartmann layers, results in such an expensive task in terms of
computing time and storage that it becomes practically impossible [4]. Here, we address the problem
in a simplified way by considering a joint core-side-layer flow which occupies the region 0 < y _ a,
0 _< z _< 1. Assuming that M is sufficiently large, the Hartmann layers and corner regions are ignored in the sense that they are not resolved numerically. This implies that the non-slip condition
(51a) at the top (bottom) wall is not satisfied. Nevertheless, the Hartmann layers are taken into account as a return path for the electric currents through boundary condition (5 lb). The composite
core-side-layer flow does not distinguish between the core and the side layer and is governed by a
boundary value problem with M and c as O(1) parameters. This approach coincides more closely
with realistic flow situations and correctly predicts the division of vertical electric currents between
the side layer and the side wall [13]. We now simplify the system of equations (46)-(51) at O ( M - I ) ,
that is, we retain terms up to O ( M -1) and neglect all the other smaller terms in both the core and
the side layer. Hence, assuming that Vty = vt~ = O(1), equation (46) becomes

M -2

~ z ( 0Vtz-~z
tT)
32Ff - K,
= Oy---

(53)

and using equation (47) we have


= K + m -2

Vtz

+ Og2

However, M - 2 a \{v t,_~}


aa'~ < O ( M - l ) iT; then, for this order, we get

fi=K+--

32F
022

(54)

Substituting (54), equation (53) becomes


32F M-2 F
3y----~ -

34F

avt.a3F I
+ - az- -Oz3 J = K,

(55)

for 0 _< y _< a and 0 __<z _< 1. In addition, at y = 0:

03F

ayaz 2

-- O,

(56a)

S. CUEVAS et al.

496

~F

ay

- 0;

(56b)

= 0,

(57a)

0;

(57b)

-K,

(58a)

a t z = 0:
03F
0z 3

OF
Oz

atz=

1:

32F
az 2

OF

F + c-~z = 0;

(58b)

OF (c + M _ t ) a2F
-~y ~
= M-1K.

(59)

and at y = a:

Notice that, at this order, there is no dependence on Vty. Actually, in the core-side-layer flow the
main viscous effects come from the side wall, for H a r t m a n n layers are neglected. Therefore, the
relevant viscosity is vt:. At higher orders, Vty may be assumed to be 1. In order to close the system
of governing equations (55)-(59), the explicit form of the total viscosity Vtz = vt must be given.
In laminar regime, vt = 1. For turbulent calculations, an algebraic expression for vt provided by
the R N G theory of turbulence [10] was used. The boundary value problem given by the set of
equations (55)-(59) and the corresponding laminar or turbulent expression for vt, was solved with
the spectral collocation method [13,16], expressing the unknown function F as a finite series in
terms of even Chebyshev polynomials in y and z and using appropriate G a u s s - L o b a t t o collocation
points. The system of simultaneous, linear algebraic equations for the coefficients of the series was
solved with the G a u s s - J o r d a n elimination method. All the solutions were normalized by applying
condition (52) and correspond to a square cross-section duct (a = 1).

3.1 Laminar flow results


The relevant parameters governing the behavior of the flow in the laminar regime are M and c.
We explored three high values of M, namely, 103, 104 and 105, and four values of c, namely, 0.05,
0.01, 0.001 and 0. Numerical solutions were validated by comparison against available analytical
solutions, finding a very good agreement [17]. Figures 2(a) and (b) show the core-side-layer velocity
profiles in one-eighth of the duct at the mid-plane y = 0 for the three different M, for the cases c =
0.05 and c = 0, respectively. In both cases, the O ( M -1/2) side-layer thickness is clearly displayed.
On the other hand, as for c = 0.05 the side-layer velocity is O(Ml/2), its increase with the H a r t m a n n
number is dramatic, the peak velocity for M = 103 being about 10% the maximum velocity for
the case M = 105. For c = 0, the electric currents close completely within the fluid (through
the H a r t m a n n layers) reducing the electromagnetic drag. In this limit the velocity overshoots are
completely suppressed and the side-layer velocity is O( 1). The non-slip condition at the side wall is
matched smoothly through the side layer with the uniform flow at the core and the higher M, the
flatter the profile. Figure 3 shows the transition of the velocity profiles at y = 0, as c decreases from
0.05 to 0 for M = l 0 4. The thin conducting wall approximation holds for c = 0.05, 0.01 and 0.001.
For c = 0.001, c lies in the range M -I << c << M -1/2, so that, the side-layer velocity is O(cM) but
the layer flow does not convey any of the O(1) volume flux [5]. For c = 0.01, we have c = M -1/2,

Liquid-metal M H D flow

497

(a)

60.0 -

50.0 M=;O0,O00

.M..=.~O..,.o.o.o.

.....

. k t ~ Lo.o.o.

40.0

. . . .

30.0
"d
<

20.0

10.0

0
0.75

o:so

0.85
0.90
Transversal length Z

0.95

1.00

(h)
1.25

1.00
%

.~. 0.75

?,

',

>

~
<

\
,

:..

0.50

M=100,000

0.25

.,

......M.~.I.O..,Q.o..o....
...~-LqO0....

0
0.75

0.'80

0.85
0.;0
Transversal length Z

,,,!

0.95

1.00

Fig. 2. L a m i n a r velocity vs z at y = 0 for different M. (a) c = 0.05. (b) c = 0.

the velocity in the side layer is O ( M l/2) and it carries a fraction of the O(1) volume flux. The case
c = 0.05 fits in the range M -1/2 << c << 1 where the side-layer velocity is still O ( M 1/2) and carries
a constant fraction of the O(1) volume flux [5].
3.2 Turbulence model and numerical results

In dimensionless terms, the total viscous diffusivity (molecular + turbulent) given by the R N G
theory [10] is
Vt

[1

+ H (CiRe2vt(Vu)2l~ 4 - C2)\..11/3
J ,

(60)

where the Heaviside function H ( x ) is defined as H ( x ) = x when x > 0 and H ( x ) = 0 when x < 0;
Cl = 0.0256 and (?2 = 110 are constants of the model used by Talmage et al. [11] for the M H D flow
in homopolar devices and lo is the integral length scale, normalized by L. Since the steepest velocity
gradients are in the z-direction (perpendicular to the side wall), we can assume [V~[ ~ d~/dz. In
order to avoid oscillations of vt given by model (60) at the point of the laminar-turbulent transition,
an iteration scheme devised by Talmage et al. [18] was implemented. The integral length scale, l,,,

S. C U E V A S et al.

498
20.0 -

C=0.05

..c...~.0..:01.........
..c-9:9.0.1..._
c=0

15,0"

10.0-

<
5.0-

0.75

0.80

0.85

0.90

0.95

.00

Transversal length Z
Fig. 3. L a m i n a r velocity vs z at y = 0 lbr different c; M = 104.

is a free parameter and must be specified outside the theory. Here, it was determined through the
model proposed by Yakhot et al. [19], namely
l, = min(z, y 6 , ) ,

(61)

where z is the distance to the wall and = yo( 1-9-/-1 ) - 1 . ]/o is a constant which value was estimated
as 0.3 in two limiting n o n - M H D cases of fully-turbulent and transitional boundary layer over a flat
plate [19]. For the present M H D flow ,, was determined to be 0.4 [17]. In addition, 9ac-1 = 0 / 6 , is
a shape factor, where 6 , and 0 are the displacement and momentum-loss thicknesses, respectively,
defined as

6. =

1U

0=

1-

(62a)

dz,
U

dz.

(62b)

For a duct flow, using definitions (61)-(62), the outer integral length scale can be written as

y,(1

l:,,,t,r)
,

9/--1)-1
--

" o [ S ( U " - u)dz]2


= y

u)Zdz

(63)

The introduction of boundary layer characteristics in the definition of the integral length provides
a better assessment of the factors influencing the turbulence. Under this approach, the magnetic
effects are considered by the introduction of the particular structure of the M H D boundary layer
in the integral length model. The total viscosity model defined by equations (60)-(63) closes the
system of governing equations (55)-(59). Turbulent flow solutions involve an iterative procedure in
which initial Reynolds number and pressure gradient values are input, being properly rescaled once
convergence is reached [17].
Turbulent solutions were analyzed for three Re, namely, 5 10 4, 105, and 5 105 while M and
c took the same values as in the laminar case. Hence, N varied from 2 to 2 105. In order to
contrast the turbulent effects, Fig. 4(a) compares the laminar velocity profile at y = 0 as a function
of the transverse z-coordinate for the case M = l0 s and c = 0.05, against the corresponding mean
velocity profiles for the three different Re. Two main effects are clearly noticed in the turbulent cases
with respect to the laminar one: a dramatic decrease of the side-layer velocity and the thickening

Liquid-metal MHD flow

499

(a)
60.0
Laminar ,,
.R.e.-.5.O.,.O.O.O.
...R..e-. L0.0.,.0.Q.0..
Re.--50.0.000_
....

50.0

.....

",~ 40.0
o

.~ 30.0'

20.0

10.0

0
0.75

o~8o

i
0.'85
0.90
Transversal length Z

q-~t.......0.95

N,J

1.00

(b)
3000.0

2500.0

2000.0

8o
1500.0
o
["

1000.0

500.0

0
0.75

0:80

0.85

0.90

0.95

1.00

Transversal length Z

Fig. 4. Laminar and turbulent profiles as a function of z at y = 0 for different Re; M = 105,
c = 0.05. (a) Velocity profiles. (b) Total viscosity profiles.
o f the b o u n d a r y layer. Both effects are more p r o n o u n c e d the higher Re. For Re = 5 105, the
m a x i m u m side-layer velocity is about one-sixth o f the m a x i m u m laminar velocity, while the turbulent
b o u n d a r y layer thickness is about five times the thickness o f the laminar layer. Figure 4(b) shows
the corresponding total viscosity profiles at y = 0. Turbulence b r o u g h t about a marked increase in
total viscosity in the extended b o u n d a r y layer region, the higher Re, the higher the total viscosity.
While in the core the viscosity holds its laminar value, vt = 1, in the b o u n d a r y layer it goes f r o m
O(102) for Re = 5 104 to O(103) for Re = 5 105. Figures 5(a) and (b) show similar mean velocity
a n d total viscosity profiles for the same M but a lower c, namely, c = 0.01. The overall behavior
o f b o t h profiles is the same as in the previous case although turbulent effects are less notorious.
This tendency persisted as c decreased to 0.001 [17]. Finally, Fig. 6 presents the velocity profiles for
c = 0. In this case, the model predicted a laminar behavior for the three Re (vt = 1 everywhere),
a n d laminar a n d Re = 5 x 104 curves coincide. However, convergence o f the iterative scheme was
difficult to reach when c = 0. As a consequence, mean velocity profiles are not very s m o o t h and,
for Re -- 5 x 105 a n d Re = 105, small oscillations appeared near the wall. Essentially, by decreasing
M to 104 a n d 103, the same tendencies with respect to variations o f Re and c found for M = 105
were observed, but turbulent effects were less pronounced. Figures 7(a) and (b) show the velocity

S. C U E V A S et al.

500

(a)
60.0

Laminar
50.0

.... R~=..5.0,.0. 0..0.......


.- .R.e.-.1.0.9,99.0...
Re=50Q~000

"~ 40.0
o
~

30.0

o~ 20.0
10.0

o
0.75

0.ks

o.'8o

0.'90

079~

....

"

1.00

Transversal length Z

(b)
3000.0

2500.0
Re=500,000
Re=100,000
2000.0

~
[-o

1500.0

1000.0

500.0

0
0.75

0.'80

0.85
0.90
Transversal length Z

0.95

1.00

Fig. 5. L a m i n a r and turbulent profiles as a function of z at y = 0 tbr different Re; M = 105,


c = 0.01. (a) Velocity profiles. (b) Total viscosity profiles.

profiles for c = 0.05, for M = 104 and M = 103, respectively. Total viscosity profiles presented lower
values in the b o u n d a r y layer than for M = 105, for corresponding Re [17]. Convergence problems
appeared as c decreased from 0.01 to 0.001 and were even stronger for c = 0 where, for M = 104,
laminar behavior was again predicted. However, for M = 103 it was not possible to find a converged
solution for the cases c = 0.001 and c = 0 which, incidentally, lie outside the thin conducting wall
approximation.

4. D I S C U S S I O N

AND

CONCLUDING

REMARKS

The steady, fully-developed flow of a liquid metal in a constant area rectangular duct with a
uniform, transverse magnetic field was investigated in laminar as well as in turbulent regimes. Thin
conducting wall boundary conditions at the top/bottom walls were extended to include the insulating
wall condition in which Hartmann layers are the return path for electrical currents. Therefore, the
range of interest of wall conductance ratios and Hartmann numbers for fusion blanket technology
was explored. A composite core-side-layer spectral collocation solution was obtained. This solution

Liquid-metal M H D flow

501

1.25

1.00

~.

"-*,i 0.75

\.

~ o.5o
Laminar

......a..e.
. -..5..0.,o..o...o......

0.25

...R..e- !99.,O.99..
Re=500,00.0_

0.75

0.~0

0.~5

o.bo

o.b5

i oo

Transversal length Z
Fig. 6. L a m i n a r and turbulent velocity vs z at y = 0 tbr different Re; M = 105, c = 0.

successfully resolves the side layers in a very efficient computing way, even for Hartmann numbers as
high as 105, contrasting with finite difference approaches which prove inefficient even for M _< 103.
The transition from a thin conducting to an insulating wall duct flow dearly reveals the dramatic
change in the laminar flow structure, from a flow with high-velocity side-wall-jets to a slug-like flow.
Turbulent effects were introduced through an eddy viscosity model from the R N G theory of
turbulence, finding physically reasonable trends. The lack of experimental data on the cases considered here prevented a comparison with numerical results. In all well converged cases, the thin
conducting wall approximation holds. In these cases, the presence of turbulence was manifested by
a marked decrease of the mean side-layer velocity, a thickening of the boundary layer and a notorious increase of the total viscosity. Former effects were amplified as Re increased but turbulence
only affected the boundary layer region, the core remaining unperturbed. The general trend shown
for fixed M and Re indicates that the smaller the wall conductance ratio the less pronounced the
turbulent effects. In this sense, a reduction in c tends to laminarize the flow. Evidently, this behavior has to do with the fact that, for a given M, a reduction in c implies a decrease in the velocity
overshoots in the side layer which are the main source of turbulence. In the insulating wall limit,
the turbulence model predicted a laminar flow behavior whatever the Reynolds number, within the
studied range. The slug-like flow in an insulating duct at very high M is hardly destabilized even
with a strong perturbation. This may be the cause of the difficulty of reaching convergence in the
insulating wall duct case. As c is lowered (= O ( M -l)), oscillations in the laminar-turbulent transition region may appear as the velocity overshoots in the boundary layer disappear and the flow
becomes more stable. As Hunt pointed out [20], since the velocity profile in an insulated duct at
M >> 1 has no points of inflexion in the side-wall boundary layers, the magnetic field presents a
stabilizing effect on the flow. Hence, a uniform decrease of the conductivity of the walls will tend
to stabilize the flow in these layers.
Numerical results also show that, provided the thin conducting wall approximation holds, an
increase in M leads to stronger turbulent effects for a given Re. This implies that the higher and
thinner the side-wall-jets the more suitable the boundary layer flow to promote a high level of
turbulence. So that, for thin conducting wall ducts at M >> 1, the magnetic field may play a
turbulence promoting role. Hunt [20] observed that for flow in a rectangular duct with conducting
walls, the value o f the Re at which the side-wall boundary layers become unstable decreases as M
increases, contrasting with the case of flow in a plane channel where the stabilizing effect of the
magnetic field has been theoretically and experimentally demonstrated. Nevertheless, it is incorrect
to assume that the side-wall boundary layers are always unstable as M - co. For small enough

502

S. CUEVAS et al.
(a)
20.0
Laminar
.....R.+..-.50..,9..o..9.......
Re=100~000.
15.0

A
II
] I

'_.~-360~_

"~ 1o.o

5.0

. .=.._..=..=--='

- : .........~ . . . . . ~

0.75

'

=..o.-

,, ........

. . . . . . . . . : ...................... ""

0.80

"

0.85
0.90
Transversal length Z

I,

"

0.95

1.00

(h)

20.0
Laminar
...8~-.5.o,.oo.o. ......

15.0

. .R..e--Jg.0.,0.9.0...
Re=50.O.0=9_00._

-~ io.o

= lo.o

5.0 ....................~ . , ~ . ~ . ~ . " ~

o.'2

. . . . .

-.-:- ' --':=- --- -..'.:..'.:':':.....

o.~

o;s

~ J ~ .

o.k

~.'1

1.o

Transversal length Z
Fig. 7. Laminar and turbulent velocity vs z at y = 0 for different Re; c = 0.05. (a) M = l0 a. (b)
M = 103.

Re (high enough N), the boundary layer flow should be stable whatever the value of M and c,

as actually predicts the R N G model. However, the laminar-turbulent transition was not explored
here. Duct's aspect ratio is an additional factor that influences the stability and turbulent behavior
of the flow and

should

be considered

in a future

analysis.

Acknowledgements--The first author wishes to acknowledge the hospitality of Argonne National Laboratory where most
of this research was completed. He also gratefully recognizes the financial support from the Consejo Nacional de Ciencia
y Tecnologia, Banco de M~xico and particularly, from the Instituto de Investigaciones El6ctricas.

REFERENCES
1. Cha, Y. S., Gohar, Y., Hassanein, A., Majundar, S., Pirologlou, B., Sze, D. K. and Smith, D. L., Fusion Technol.,
1985, 8, 90.
2. Hua, T. Q., Walker, J. S., Picologlou, B. F. and Reed, C. B., Fusion Technol., 1988, 14, 1389.
3. Ramos, J. I. and Winowich, N. S., h~t. J Num. Methods in Fluids, 1990, I1,907.
4. Sterl, A., J. Fluid Mech., 1990, 216, 161.
5. Walker, J. S., J MP.canique, 1981, 20, 79.
6. Reed, C. B., Picologlou, B. E, Hua, T. Q. and Walker, J. S., Proc. IEEE 12th Syrup. on Fusion Engng, Monterey, CA,
1987, p. 1267.

Liquid-metal MHD flow


7.
8.
9.
10.
11.

503

Reed, C. B. and Picologlou, B. E, Fusion TechnoL, 1989, 15, 705.


Ting, A., Walker, J. S., Moon, T. J., Reed, C. B. and Picologlou, B. E, Int. J. Engng Sci., 1991, 29, 939.
Sukoriansky, S., Zilberman, I. and Branover, H., Experiments in Fluids, 1986, 4, 11.
Yakhot, V. and Orszag, S. A., J. Sci. Comput., 1986, 1(1), 3.
Talmage, G., Walker, J. S., Brown, S. H., Sondergaard, N. A., Branover, H. and Sukoriansky, S., Phys. Fluids A, 1991,
3, 1657.
12. Sukoriansky, S., Staroselsky, I., Galperin, B., Roy, S. and Orszag, S. A., Progress in Astronautics and Aeronautics
AIAA, 1993, 149, 151.
13. Ting, A., Combined analytical and numerical solutions in liquid-metal flows in a rectangular duct with unitbrm or
non-uniform, strong magnetic fields. Ph.D. thesis, University of Illinois at Urbana-Champaign, U.S.A., 1991.
14. Ting, A., Hua, T. Q., Walker, J. S. and Picologlou, B. F., Int. J. Engng Sci., 1993, 31,357.
15. Shercliff, J. A., J Fluid Mecb., 1956, 1,644.
16. Canuto, C., Hussani, M. Y., Quarteroni, A. and Zang, T. A., Spectral Methods in Fluid Dynamics. Springer-Verlag,
New York, 1987.
17. Cuevas, S., Liquid-metal flow and heat transfer under strong magnetic fields. Doctoral Thesis, Facultad de Ciencias,
Universidad Nacional Aut6noma de M6xico, 1994 (in English).
18. Talmage, G., Walker, J. S., Brown, S. H., Sondergaard, N. A., Branover, H. and Sukoriansky, S., Progress in Astronautics
and Aeronautics AIAA, 1993, 149, 165.
19. Yakhot, V., Kedar, O. and Orszag, S. A., J. Sci. Comput., 1992, 7(3), 229.
20. Hunt, J. C. R., J. Fluid Mech., 21, 577.

(Received 23 February 1996; accepted 30 May 1996)

Potrebbero piacerti anche