See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/35964244

Mathematical models of solidification : forced flow in circular pipes.

Thesis · November 1981

Source: OAI

CITATIONS READS

0 62

1 author:

Paul Sampson
Northumbria University
10 PUBLICATIONS 63 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Diamond VoD View project

All content following this page was uploaded by Paul Sampson on 04 August 2015.

The user has requested enhancement of the downloaded file.

Mathematical Models of Solidification Forced Flow in Circular Pipes

ABSTRACT

In the flow of liquid metals through pipes, the relatively cold

walls of the pipe induce heat losses and consequent freezing of the
metal. In seme ca3es, the pipe may freeze shut completely and thus
prevent further fluid from being transmitted. A mathematical model is
developed to describe this process and a novel method is used to
predict when blockage might occur. If the fluid flow is laminar then
two dimensionless parameters, and a certain relationship between them,
provide such a blockage criterion. If turbulent, then four parameters
are necessary to describe the flow. Blockage in this case is predicted
by a numerical model and a FORTRAN program is included. An attempt at
the modelling of fluids with a melting range, rather than a single
melting point, is outlined and some results presented which again may
be used as the basis for predicting blockage in this case.

Paul Sampson
School of Maxhematics,
Statistics and Computing,
Newcastle upon Tyne Polytechnic.

- i - Abstract
Mathematical Models of Solidification Forced Flow in Circular Pipes

PREFACE

I would like first to make a few remarks about some of the notation
used in this thesis. The reader may notice that occasionally, the use
of bold face characters may be found in the text. There are three main
uses for this. The primary use is in the two letter dimensionless
groups such as St, Pr etc., which serves to mark them as single
entities and not as an implied algebraic multiplication (Pr might be
interpreted as pressure drop x radius, for example). The second use is
for special emphasis of concepts and the third, to denote vector
quantities. There should be no conflict between their usages.

References to published work will be enclosed in brackets, and will

additionally be preceded by a chapter number and a full stop if cited
in chapters other than that chapter for which the reference was
originally intended. The references themselves are to be found at the
end of the text and before the appendices. Equation numbers are
enclosed in parentheses, again indicated by a preceding chapter number
where necessary. I have chosen to number equations, references and
figures consecutively within each chapter, rather than globally, to
avoid the appearance of large numbers. There will also be found
referrals to sections. These will be enclosed in parentheses preceded
by the chapter number in which the section shall be found, and a colon.
A few well known elementary functions will be found in the text. These
are ln(x) signifying the natural logarithm function, log(x) the
logarithm base 10, exp(x) is the exponential function, T(x) the Gamma
function, erf(x) and erfc(x) the error function and complementary error
function (erfc(x)=1-erf(x)) respectively. Bessel functions will be
denoted by Jv(x).

Acknowledgements

I would like to thank Professor R D Gibson and Dr R W Young for their

invaluable assistance over the past three years, Roy Bradley for his
reading of the manuscript and helpful comments, Cindy for the typing,
Peter Blachford of this school and the staff of the Polytechnic
Computer Unit for their help and the Polytechnic in general for funding
my research.

Paul Sampson BSc, MSc

School of Mathematics,
Statistics and Computing,
Newcastle upon Tyne Polytechnic.
October 1981

Preface
- ii -
Mathematical Models of Solidification Forced Flow in Circular Pipes

CONTENTS

CHAPTER PAGE

1:INTRODUCTION 1

2: INPUTS AND OUTPUTS 8

3: SOLIDIFICATION 12

3:1:The inward solidification problem

in cylindrical coordinates 17

4:HEAT TRANSFER 20

4:1'.Dimensional analysis 21

4:2:The interface 22

5: FLUID FLOW 24

5:1:Standard solutions 24

5:2:Turbulent flow. . . . . . . . 28

6:REVIEW OF PREVIOUS MODELS 31

6:1:Steady state laminar flow 31

6:2:Transient laminar flow

and solidification .33

6:3'• Turbulent flow 34

6:4:General conclusions and hiatus 36

7: FORMULATION OF THE PROBLEM 38

8: THE GRAETZ PROBLEM 42

8:1:Laminar flow - Sellars, Tribus

and Klein 42

8:2:Turbulent flow - Sleicher, Notter

and Crippen 44

- iii - Contents
Mathematical Models of Solidification Forced Flow in Circular Pipes

9: LAMINAR FLOW 48

9:1 integration 50

9:2:Results 52

9:3:Phase plane analysis 55

9:4:Stability and steady states 57

9:5:Asymptotic analysis 63

9:6:Experimental validation 66

10: TURBULENT FLOW 68

10:1:The specifications .68

10:2 thermodynamics 72

10:3:Remarks on c and G 74

10:4:Solution examples 77

10:5:Stability revisited 81

11:MUSHY REGIONS AND VARIABLE VISCOSITY 87

I1:
1 :Modelling ...88

11:2:Fluid temperature 92

11:3:Fluid velocity 93

11:4:Non-dimensionalization 95

11:5:Numerical treatment. . . . . . . . . . . . 9 7

II :6 :Results 101

11:7:Conclusions and directions 103

References R1-3

A:Appendix - Program listing A1-21

B:Appendix - Program example . B1-6

C: Appendix - Laminar paper C1-11

D:Appendix - Software paper D1-9

E:Appendix - Conference paper . . . . E1-11

- iv - Contents
 DECLARATION

While registered as a candidate for the degree of Ph.D. the author has

not been a registered candidate for another award of the CNAA or of any-

other body. None of the material contained in this thesis has been used

in any other submission for an academic award. Some of the work which

appears in the thesis has however been published in the literature and

reference is made to these publications in the body of the thesis.

During the research program the author was required to attend, and

occasionally present, departmental colloauia and to attend such seminars

at the University of Newcastle and Newcastle upon Tyne Polytechnic as

were relevant to the research. The author has also attended national and

international conferences and meetings of the British Theoretical

Mechanics Colloquium, the Institute of Mathematics and its Applications,

The Numerical Methods in Thermal Problems Conference and the North East

Polytechnics Mathematical Modelling and Computer Simulation (POLYMODEL)

Group.

P.Sampson B.Sc., M.Sc.

October 1981

- v -
1: INTRODUCTION

In many metals—processing operations, molten metals are poured

through nozzles, the walls of which are maintained at a temperature

below that of the freezing point of the liquid. As a consequence, a

solidified crust can form on the inside of the walls and this may lead

ultimately to nozzle blockage. Whilst guidelines have been evolved for

blockage avoidance, largely through practice and experience on the

works floor, there has been little analytic treatment of this problem

although some work has been done on the formation of crusts.

The work in this thesis deals with the range of mathematical

theories which are necessary to describe the process of blockage and

the flow of liquids through pipes. The physical processes which need to

be discussed and modelled are:

(1) Solidification

(2) Heat transfer in cylindrical pipes

(3) Fluid flow in cylindrical pipes

There is no apparent hierarchy involved in these broad divisions

and it is proposed to describe the work done in terms of these three

processes. The main contribution of the thesis lies in the modelling of

the flow of liquids in pipes with solidification. This involves a study

of the interplay of the three processes and the numerical and

computational solution of an underlying set of equations.

We ' discuss three applications of this work which relate directly

1 1 introduction
the metals processing industries.

(1) Simple transport. In this application it is necessary to

transfer molten metal from one place to another. There are

various methods used to effect this, transport through pipes

being only one of them. (Another might be direct pouring

from a casting ladle.) Clearly, it would be undesirable for

the pipe to block solid and so the inputs have to be

adjusted accordingly. In practice, this is done continuously

by the human or computer controlling the process.

Figure (1) _ ^ Tundish Nozzle Joint

- 2 - 1:Introduction
(2) Continuous casting. In this case the reverse situation is

the required option since blockage, here used in the sense

of total internal solidification, is the desired result. If

the interior were not totally solid there could be a

spectacular - and highly dangerous - release of liquid metal

as the measured lengths are cut from the continuously moving

tube. It is necessary to know where and when it is safe to

cut.

Figure (2) - A Typical Vertical Continuous Casting Set-up [l]

Solid
coifinq

Dummy

3 1:Introduction
Static solidification. This is a less direct application but

possibly one of the more important. Suppose that molten

metal has been transported through a nozzle, as in the first

example, and has been poured into a mould. The metal will

then solidify from the boundaries of the mould to form the

product, a billet. Liquid metals are rarely pure and contain

many components or contaminants. In general these impurities

will be pushed towards the centre by diffusion at the

advancing interface before complete solidification occurs.

As a result, the centre can be of a markedly different

material to the rest of the casting. This is usually

undesirable but the effect will be reduced if the metal can

be made to freeze more quickly.

Figure (3) ~ Cross Section of Billet

- .
4 - 1:Introduction
 The first two of the above cases have obvious practical solutions

since to prevent or force blockage it is easy to see what steps are

required. For example, if the temperature of the nozzle wall is

controlled then it should be as high as possible if blockage is to be

prevented. There are no specially imposed minimizing or maximizing

criteria. The third example is more interesting. To achieve rapid

solidification in the mould, as much heat as possible should be removed

from the liquid while it is still in the pipe. In this context, the

nozzle wall temperature should be as low as possible and this is in

opposition to the 'solution' outlined in the first example. In other

words, example three is modelled in exactly the same way as example one

but requires solutions with low wall temperatures and a high throughput

of liquid metal. This is in order to achieve rapid freezing in the

billet and consequent improvement in the quality of the casting.

The model is also applicable to other fluids and so should not be

regarded as being in any way restricted to the flow of liquid metals.

In fact, the experimental verifications for the laminar flow model was

performed with data for ice and water [6.2], as will be shown in

chapter 9«

Fine details of blockage will not be considered since we do not

aim to discuss the cases where blockage develops but rather those cases

where fluid is still successfully being output. Nevertheless, it of

interest to know if the pipe is going to block and so criteria will be

presented for predictions of blockage.

- 5 - 1:Introduction
 We propose to deal with the work in sections. The first, chapter

2, will outline the physical parameters which govern the type of flow

and serve to familiarize the reader with some of the notation used in

this work. There follows a section, chapter 3, on solidification, which

will discuss the conditions at solid/liquid interfaces and some of the

approximations to be used in the model. We will use simple cartesian

models to demonstrate the validity of these approximations and also to

introduce some dimensionless parameters which will govern the rate of

freezing. The next section, chapter 4, will deal with the diffusion

equation, in cylindrical coordinates, and will introduce the standard

scalings and non-dimensionalizations which are most often used in this

type of modelling. The fluid flow is then discussed in chapter 5 and

the two types of flow, laminar and turbulent, which will be used in the

model are described along with the fundamental boundary conditions.

Before describing the models proper, a review section, chapter 6,

serves to place the work in context. The problem is then formulated in

mathematical terms in chapter 7. Chapter 8 on the Graetz problem will

describe earlier work on a purely heat-transfer problem which will be

of use in determining the position of the solid/liquid interface.

Having introduced the basic ideas behind the model, the remainder of

the thesis will then apply these to the problem of nozzle blockage.

These models deal with laminar flow (chapter 9), turbulent flow

(chapter 10) and fluids with temperature dependent viscosity and a

solidification range of temperatures rather than a single melting point

(chapter 11). Much of this work has been published in the scientific

press !_2]>l3]»[4] and is included, as appendices, at the end of this

work. Hiere is also a paper [lO.l] which is currently in press.

- 6 - 1:Introduction
 The method developed to find steady state crusts leads, by

corollary, to the prediction of blockage by an original method. The

author believes that there is much of value in this work since it is

essentially a complete theory to the extent that it is possible to

forecast whether or not the nozzle will block, given the inputs. Furth­

ermore, the model predicts the throughput of liquid metal and in the

important case of a melting range, the throughput of partially frozen

mush.

- 7 - 1:Introduction
2: INPUTS ANH OUTPUTS

The wall temperature is only one of the inputs which may be used

to control the throughput of metal. In practice it is a parameter over

which there is little direct control. In this section, we will

introduce some of the main factors which determine the flow of fluid

and heat, and which will also ultimately determine whether or not the

nozzle will block. We can divide the inputs as follows:

(1) Those relating to the material being transported. These are

not generally alterable, except by changing the fluid or the operating

temperature range.

(a) Density is denoted in this work by p. In general this will

be taken as a constant throughout all phases of the

material. Units are kg m-3.

(b) Thermal conductivity is denoted by k and is constant

throughout the solid phase. In turbulent flow, it is not

constant in the liquid phase and variations will be

incorporated into an eddy diffusivity (see below). In

laminar flow the thermal conductivity is constant and the

appropriate phase subscripts will be used to distinguish

between them. Units are W m ^ K ^.

(c) Specific Heat will be denoted by C and is constant in each

phase although only the specific heat of the liquid will

have any direct effects. Units are J kg ^ K ^.

- 8 - 2:Inputs and Outputs

 The above three properties are commonly combined into k =

k/pC, known as the thermal diffusivity. Units are m s .

(d) Latent Heat of fusion, denoted by L, is constant. Units are

J kg"1 .

(e) Melting Point needs some discussion. For pure substances, we

will assume that the latent heat of fusion will be liberated

at the melting temperature, T . If the fluid is not pure,

latent heat will be released over a temperature range

bounded by Tg - the solidus temperature - below which no

liquid metal is present and T^ - the liquidus temperature -

above which no solid metal will be found. Between these two

temperatures we have a 'mushy' region consisting of a

mixture of liquid and small solid particles. Mushy regions

will be discussed in greater detail later. Units are K.

(f) Viscosity, dynamic denoted by y and kinetic by v = u/p»

1 1 1
units being kg m s and m^ s respectively.

Note also the Prandtl Number, Pr = V/K, of the fluid. This

plays a part in turbulent flow, as will be seen. It is

typically much less than unity for liquid metals.

All of the fluid properties are assumed to be independent of

temperature in the simpler models. In more complex models, they can

vary with temperature and this will be taken into account in chapter

11. Other factors which affect the fluid properties, particularly the

- 9 - 2:Inputs and Outputs

viscosity, are the local shear rates of the flow and the impurities in

the fluid. It should also be pointed out that fluid properties of

liquid metals are notoriously difficult to measure accurately and that

there are many variations even if the investigation is restricted to

iron and steel. Figures for these metals in both solid and molten

states may be found in Angus [l].

(2) The second group of inputs concerns the physical set up:

(a) Pipe Length and Pipe Radius, denoted by 1 and rQ

respectively. The mathematical model will refer to an

idealized pipe with these dimensions and the cylindrical

wall will be of zero thickness. Units are m.

(b) Pressure Drop, P, over the length of the pipe will be

-1 - 2
constant. Units are kg m s

(c) Wall Temperature and Inlet Temperature denoted by T^ and Tq

S
respectively will be assumed constant . The model will thus

assume isothermal boundary conditions. Units are K.

8
A possible extension to the model would include heat transfer at the
wall/air interface, with a finite thickness pipe wall. This will not be
dealt with in this thesis, however, since it will not affect the
blockage criteria significantly.

- 10 - 2:Inputs and Outputs

 One of the outputs required is the amount of liquid metal which

leaves the exit of the pipe. This is a flux integral and can be

calculated as a proportion of the input flow rate before the

solidification process starts. In mushy type flows, it is also of

interest to know the proportion of the fluid which is mushy. The

maximization of the mushy throughput ensures both that the impure fluid

is well mixed and that a significant amount of specific and latent heat

has been removed. (This is relevant to application (1.3)).

The most useful output from the point of view of the model alone,

independent of any particular practical applications, is the position

of the solid/liquid interface measured from the axis of symmetry of the

cylinder. We denote this quantity by 5, and it is in general a function

of time, t, and axial distance down the pipe, z, measured from the

entrance at z=0. Most of the effort, analytical and numerical, in this

work is directed towards finding this function and how it is affected

by the inputs.

It is not generally useful to know the temperature distributions

in the fluid or solid. The heat equations will be used primarily to

determine the heat fluxes at interfaces and not the temperatures. It is

the interface heat flux which governs the growth of the solid crust.

- 11 - 2:Inputs and Outputs

3: SOLIDIFICATION

Solidification processes have been of interest for some years now

and there are many models available which have been designed to cope

with the multitude of physical situations which arise. Applications of

such descriptions are many and varied, ranging over the fields of

medicine, construction and engineering [l]. It is this last area for

which the work presented is intended, but the mathematics may find uses

in other places, with or without modification. To be more specific, the

immediate concern is to develop analytical and numerical models with a

view to describing the freezing of liquid metals in pipes. Before

embarking upon the description of the model however, it would be

appropriate at this point to briefly outline the historical context of

solidification models.

Firstly, we are not concerned with a detailed description of the

fine structure of liquid metals near their fusion temperatures. A

description at crystallographic levels would be of little value here

since blockage of pipes is a global phenomenon and would moreover

render such a model inappropriate to even the closely related

description of the freezing of water in pipes. Instead, we will make

use of the diffusion equation in order to calculate the temperature

distributions, heat fluxes and related quantities which will allow us

to determine conditions under which pipes block.

Possibly the simplest of all models is the well known 'freezing

of ice on a pond' model characterized by the error function solutions

with argument proportional to the inverse square root of time. This is

- 12 - 3:Solidification
the classical analytic solution to the one-dimensional freezing

problem, known after Neumann, and can be found in Carslaw and Jaeger

[2]. The position of the solid/liquid interface is easily tracked as a

function of time. It will be of some interest to restate the Neumann

solution here in terms of our notation since it allows us to make a few

definitions which will be used hereinafter.

Suppose the semi-infinite region x > 0 holds pure liquid at a

temperature T q at time t = 0 and that for t > 0 the boundary x = 0 is

held at a temperature T , which is less than the melting point of the

liquid, T . Solidification will occur and the region will be divided

into two phases. The first region will contain the material in its

solid phase at some local temperature T^(x,t) lying in the range

Tw _
< T^ _
< T . This region will be defined by 0 _
< x <
_ X(t) where X(t)

will be the position of the advancing solid/liquid interface. The

second region, x >

_ X(t), will contain liquid at some temperature

T_(x,t) such that T < T« < T .

2X m — 2 — o

Clearly, at x = X(t) the temperatures of the two regions will be

equal. This is the first fundamental boundary condition and simply

states that no discontinuities in temperature occur anywhere. Although

the temperature is continuous, the heat flux, q = k grad T, is not

continuous at phase change boundaries. The operative word here is

change since two states of a m a t e r i a l i . e . solid and liquid in this

case - can coexist with continuity of heat flux provided that no phase

change is actually taking place; in other words when the two states are

in equilibrium and have reached a steady state. Discontinuities of flux

are however the most common case. Quite apart from discontinuities in

- 13 - 3:Solidification
the properties (density, thermal conductivity etc.) of the materials,

there is an enthalpy jump, AH, equivalent to L, at the melting point.

This can he thought of as a difference in the heat contents of the two

materials although this is an oversimplification of the concept of

enthalpy. The enthalpy difference is related to the difference in heat

flux since conservation of energy demands that the heat supplied in the

volume element of material dxdA by freezing, viz LpdxdA, in time dt,

must equal the net amount of heat, -( k^lgrad |— k^|grad T.J )dtdA,

removed in that time by conduction. This gives the second fundamental

boundary condition at the interface, (3) below.

We now return to the Neumann problem, which is solved using the

diffusion equation in each region separately:

/j\ 9 t
, 3T.\ 9 (pC.T.) /. _ . s
(1) 1 1 o r
" st

where the conductivities and specific heats are piecewise constant and

we are introducing the notation i=1,2 to denote solid and fluid regions

respectively, the densities of each phase being constant and identical.

These equations are subject to

(2) T2(X(t),t) = T1(X(t),t)

k
lf' ' ^
Other boundary conditions, specific to this problem, are that T^

approaches T q as x tends to infinity and that T^ = T^ at x = 0, both

for all time t > 0. The solution is well known, and in terms of our

notation is

(4) T = T + (T - T )
1 w m w erf(X)

- 14 - 3:Solidification
(5) T2 = T M + (TQ - TM)[ erf(x-p/x) - erf(\y) ]/erfc(ty )

(6) X(t) = 2 \ ( k t)Va

s
2
where y = k C„/k„C = K / k?* It is equation (6) which defines .the
SI IS S I
interface. To determine X» we use equation (3) to recover the following

transcendental equation for X;

2 2 2
fn\ , . St r „ exp(-A ) exp(-Y X ) i -I
T J
erf(x) " erfc(xy) "

The following important dimensionless parameters have been used in (7):

(8) St = C^(Tq - Tm)/L (the Stefan Number)

(9) B = k (T - T )/k (T - T )
s m w f o m

The Stefan number, being the ratio of sensible heat to latent

heat, will in the cases studied be a small number. This means that the

solidification rate parameter, X, will be correspondingly small. In

cylindrical solidification processes, the relationship between X and St

will be similar in that a small Stefan number will give rise to slow

crust growth rates. Since this is the case, we will be using

quasi-steady models in the diffusion equation. To demonstrate the

validity of this approximation, we non-dimensionalize the above

Cartesian problem, temporarily signifying dimensional variables with an

*
asterisk superscript, by choosing a length xq for x - a purely

arbitrary choice since in the one dimensional cartesian freezing

problem there is no typical length scale. Thus the following

transformed coordinates

(10) x = x x ( X = x X )
0 0

- 15 - 3:Solidification
(11) t* = ( rp A / k J T - T ))t
o f o m

(12) T* = T - (T - T ) T. (i=1,2)
l m o m l
reduce equation (3) to the convenient dimensionless form:

(13) fe3k —ax1 - -jr2

3x " 7dt
T at X=x

and the diffusion equations to

(14) St — 1 = y2-^-j1 for 0 < x < X(t)

3t 3x

(,5) st = 2
fS
subject to

(16) T = T 2 =0 at x=X

(17) -
*
• 1 as x for t > 0

(18) T = - (T - T )/(T - T ) at z = 0 for t > 0

1 m w o m

For small Stefan number, we can expand (4) and (5) as a power

series in X and these equations become in dimensionless form,

(19> T 1 + + a + 0
i f -rf «">
o m

(2°) T 2 = I Xy [ (F - 1) • I Xy (F - 1) ] • O U 3
)
rT
I /tt

near x = X(t), whence (13) becomes

- 16 - 3: Solidification
 x B A
(2D H - - ^ y - <fB +
|y2'*2 <»3>
rT
7

Now the solution (6), expressed in dimensionless terms is given by

2 2 2
X = 4y X t/St and so for consistency with (21), X must be given by the

root of the equation

3
(22) 4y2X2/St = 2B - 4 \ y / v /2
- i[B + 6y2/7r]x2 + 0 ( x )

from which X can easily be expanded as a power series in St 1 / 2:

(23) X = (BSt/2 y 2 ) l/2 [ 1 - (St/2TTB)1/2 -

(St/2TrB)(2B - .5 + 2ttB2/y2) ] + 0(St2)

As an example, with B = 1, y = 1 and St = 0.05 then expansion (23)

yields a value of 0.1342 for X (cf X = 0.1581 for zero order terms only

on the right hand side of (23) and X = 0.1440 for terms up to the first

order) and for figures typical for solid steel near its melting point

(k = 250, C = 700, p = 7000) this produces a solid layer about ?mm

s s
thick after one second has elapsed. In the type of problem that we will

be considering, a constant supply of quickly flowing liquid metal

through a cylinder will reduce this rate even further, and so

quasi-steady models will be used.

3:1:The inward solidification problem in cylindrical coordinates

We now pose a similar problem to the above one-dimensional

Cartesian solution which is directly relevant to nozzle blockage.

Suppose we have an infinitely long cylinder of radius rQ, containing

pure fluid, at temperature Tq, which freezes at a temperature T^ and

* *
that at time t = 0 the cylinder wall, r = rQ, has a temperature

T (<T ) imposed upon it. The equations to be solved are

w m

- 17 - 3:Solidification
 1 9 , * 9Ti*x _ 3TJ *
(24) * * ^Ki * 9t ' ' °r '
r 9r 9r

where 'r' is the radial coordinate. This equation has no Neumann type

solution (see [2]). The temperature field is subject to the same

interface conditions as before, couched in terms of cylindrical

coordinates;

(25) T1(r,t) = T2(r,t) = T^ at r =6

9T * 9T 2 * *
fpfO > 1 1 t dS * *
(26)
s — * ~ kf ~ = pL
dt*
QC
at r = 6
9r 9r

Where again we use asterisks to denote dimensional variables. Note that

the 'external' boundary conditions are a little different,

* *
geometrically and physically. One condition is that T,(r ,t ) = T for
1 0 w
*
t >0, and the difference is fairly minor. However, there is no

equivalent condition at infinity. This is a serious drawback since in

the finite and closed region of the cylinder there are no means by
*
which the temperature at r = 0 could be kept at T , unless Tq is

exactly equal to T . We do not propose to investigate the heat transfer

in the fluid region at this stage however. Other problems arise as the

fluid solidifies completely since singularities arise at the centre.

This same problem arises with inward solidification of spheres and

again, will not be discussed here. Some progress has been made however

and the interested reader should consult Riley et al. [3], Stewartson &

Waechter [4] or Huang & Shih [5] and other relevant work. As far as

this work is concerned, we note that in cases of interest to us, the

Stefan number will be small and similar progress can be made.

- 18 - 3:Solidification
If we make the quasi-steady assumption for cylindrical coordinates, we

have to solve, for the solid region,

(27) -^1 -^1 = 0

3r 2
* r 3r
* * * *
subject
0 to T, = T on r = r and T. = T on r = 5< . This has the
1 w o 1 m
straightforward solution

(28) T* = T + (T_ - T ) l n ( r o / r ^
1 w m w ^—
ln(ro/fi )

with an associated interface flux

(29) k — - - k (T - T )/&* In (r /5*) at r* = 6*

s . * s m w o
3r

It is in fact equations (28) and (29) which will be used in our

modelling process. This completes the brief discussion of

solidification processes relevant to fluids with a single melting point

and isothermal boundary conditions.

- 19 - 3:Solidification
4:HEAT TRANSFER

After having outlined the mathematical modelling of

solidification, it is appropriate to discuss the underlying heat

transfer processes. The physical set-up under consideration consists

of:

(a) A cylindrical pipe of radius rQ and length 1 q .

(t>) Fluid enters the pipe at z = 0 (from the left in the diagram

below) with uniform temperature T q .

(c) The hot fluid loses heat through the outside walls, held at

constant temperature T^ and freezes inwardly from the cylinder

wall.

Figure (1) is a diagram of the circumstances and the expected

temperature profile
A

> 2'

Figure (1) - Diagram of system

The energy equation for the fluid is, neglecting viscous dissipation,

(this is indicated by the Brinkman Number, Br <<1, see (9.12-d))

/ \ 3 , * 3 , * * 3 , * *
(1) (pr CT ) + — ( p r u CT ) + — • (pr w CT )
3t 3r 3z
* *
3T 3T
a kr
* ? + kr
a 9
* < — ) — * < — >
3r 3r 3z 3z

20 - 4:Heat Transfer
 *
where u is the radial component of the velocity vector, being of an
*
order of magnitude smaller than the axial component, w . The axial

diffusion term and the time derivative are also small since the flow is

comparatively rapid in the axial direction. Thus radial conduction

dominates and the crust growth rate is small - hence the temperature

field can adjust itself to the new interface much more quickly than

this interface can change. To quantify this, and also to make the

mathematics non-dimensional, we perform some scalings.

4:1:Dimensional analysis

We will obtain solutions initially with constant fluid

properties. We scale the radial and axial coordinates as follows,

*
(2) r = r r
0
*
(3) z = r Pe z
o 0
,
where PeQ is the Peclet number of the fluid flow at time
*= 0
t just

before the crust starts to develop. The Peclet number, akin to the

Reynolds number (Re), for momentum, is the ratio of convected to

diffused heat transport. In fact, Pe = PrRe. We base the Peclet number

on the tube diameter (as is usual in cylindrical coordinates) and an

initial measure of axial velocity, w q ,

(4) Pe = 2r w / K (K = k /pC )
0 0 0 i t

where p„ = p = p. For consistency, we have

x s
(5) w = w w
0
(6) u* = w u/Pe
0 o
The temperature is scaled, as in (3.12),
*
(7) T = T + (T - T )T_
2 m 0 m 2
and the timescale, similarly from (3«11)> mutatis mutandis,

- 21 - 4:Heat Transfer
(8) t* = (pA/2k.(T - T )) t
o r o m
(The factor of 2 is an adjustment for cylindrical coordinates, and is

introduced for convenience). We then recover

(9) 2St |j- (rT2) *| ^ (ruT2) * ± fj- (rtfy . | - + i _ i £ .

Pe 3z
o

And so the equation to be solved in the fluid region will be

9T
0°) "§7 (ruT2) + — (rwT2) = 2 (r )

having neglected terms of the order of St and l/Pe . This is

o
subject to

(11) T2(r,0) = 1

(12) T2(5,Z)=0

Also, by symmetry, is an even function of r and so

(1 3) "^2
3 7 (o,2) - o

4:2:The interface

For small Stefan number, the quasi-steady solution (see (3.28))

to the diffusion equation applies in the solid crust. In dimensionless

form this is

<H>
1115
(T - T )
o m
*
(where we have scaled 5< with r ) and so the equivalent dimensionless

form of the heat balance equation at the interface is

(15) — = 2 — at r = 6
5 In 6 3r 3t

(where B has been defined in (3«9))»

Clearly, to solve equation (15) for 6 requires the solution of

- 22 - 4:Heat Transfer
(10), to obtain the interface temperature gradient. By integrating (10)

radially from r=0 to r=6 we obtain

3T?
(16) = (RUT
2>S - (RUT
2>O - (RVT
2»S H " +
I ? rwT2dr
o

In practice, equation (16) is not used but it does illustrate the

salient features. Thus by the no-slip condition (where u=w=0 at r = fi)

and u=0 at r=0 we obtain

3T
2 1 9
rwT2dr at r= 6
o

At time
t=0, we can assume that the temperature of the fluid is
*
uniformly at T2 = 1 (T2 = T ) and so, when the wall temperature
* *
condition (T. = T at r = r ) is imposed
r for t > 0, the crust
1 w o
equation is

(18) -22- = H 2
In 62 9t
2
for small t, subject to 6 (z,0) = 1 for z _
> 0. Integrating (18) gives

(19) <S2(1 - in S2) = 1 - 2Bt

i.e.

(20) - 1 - (Bt)1"'2
6 v for Bt << 1 and z > 0

which represents an initially thin uniform distribution of solid. As

soon as this happens however, the temperature distribution in the fluid

begins to change due to convection along the pipe axis and conduction

through the walls and so the dimensionless heat flux, 3T2/3r, begins to

have an effect. To be able to calculate T2, knowledge of u and w is

required and so the hydrodynamics needs to be investigated. It will be

found that the fluid velocity depends upon <5, and so will present us

with a coupled problem which can however be simplified by a change in

the coordinate system.

- 23 - 4".Heat Transfer
5:FLUID FLOW

In equation (4.1) we have already made an implicit asssumption

that the flow is radially symmetric and so no azimuthal components or

dependencies will be included. This is a reasonable assumption if

buoyancy effects and natural convection are negligible.

5:1:Standard solutions

The classical formula for the velocity of a fluid in a constant

radius cylinder under fully developed laminar flow is (see eg Kays [l])

(1) w* = 2V (1 - r* 2 /r 2 )

(2) u* = 0

where V is the mean axial velocity given by

1
r
/" ° * * *
(3) V = — 2r w dr
r J
o °

f 2 - 1
and the total mass flow rate is dM/dt = M = irr^pV kg s . This can be

directly related to the constant pressure gradient, -P/l , over the

o
length of the pipe from the solution of
. , 1 3 / r *3v*\ 3P1
(4) — — (^ — * ) = *
r 3r 3r 3z
giving

(5) V = Pr2/8ul0

and hence M, in terms of the supplied pressure drop, P.

We would like to be able to preserve this simple Poiseuille type

- 24 - 5:Fluid Flow
flow as far as possible and so we make the assumption that the flow is

fully developed. This allows us to neglect the non-linear convective

terms in the Navier-Stokes equations and regard the entire flow inside

the cylinder as a boundary layer despite the large Reynolds number. We

have already seen, in the last chapter that we can safely neglect the

time derivative if the Stefan number is small - i.e. the hydrodynamics

adjusts itself much more quickly to the changing crust than does the

crust change. We can also neglect axial diffusion since the Peclet

< r* _
number is large. Hence (4) is applicable in the region 0 _ < 5 * and
*
0 <
_ z _ < 1q. The radial momentum equation simply gives us that
* *
3p /3r = 0, as in Poiseuille flow, since everything in this equation

is at least a factor of 1/Pe0 smaller than the axial pressure gradient

(momentum) equation. Integration of (4), assuming constant viscosity,

gives
1 2
C6) J*
w _ _- - ip!.
_j>_ (fx*
6 _ r *2^)
3z

* * *
where we have used the no-slip condition, w = 0 at r = 6 , and
* * *
symmetry 3w /3r = 0 at r = 0. We can now apply the bulk version of

the mass conservation equation

6
* 2_irr* pw*dr* = dM = •
/ *v)
M(t
(7)
o
which gives

(8) = - 8VM/it<5*4
3Z

It is fundamental to our model that equations (7) and (8) be

applied as they normally would be in constant diameter cylinders. Many

earlier models have been based on the assumption that the inlet
*
velocity (at z = 0) is constant rather than the pressure drop over the

pipe length. Conservation of mas3 would then require, no matter how

- 25 - 5:Fluid Flow
thick the crust, that the axial velocity increase indefinitely to

maintain the constant mass flow rate as the pipe blocks. In practise

this does not happen since viscous resistance and crust thickening

tends to slow down the flow as a whole. The axial velocity must still

rise as the fluid travels into narrower regions but the inlet velocity

decreases and less mass will be transmitted per second as time passes.

This will tend to result in a faster solidification process since the

amount of heat supplied will drop with the velocity. One of two things

will happen. Either the flow is sufficiently fast, and the fluid can

therefore supply sufficient heat, so that crust growth ceases

altogether; or, if the flow is sufficiently slow, total solidification

will occur and other more detailed, models must be used. We do not

intend to model remelting.

* *
Thus, integrating (8) from z = 0 tc z = 1 , and rearranging,
*_k * * *
(9) M = irP/{8v 6 (z ,t )dz }

*
which, since 6 is expected to decrease with time, demonstrates that

the mass flow rate decreases monotonically with time.

*
The radial component of velocity, u , is not zero since the

developing crust tends to push fluid slowly towards the centre. The

differential form of the continuity equation,

(10) 9
i_ , * *s
# (r V) 3 C
r r* V ) = C
9r 9z

- 26 - 5:Fluid Flow
maintains this form where scaled and it is straightforward enough to

show that

(11) u* = (r*/6*)(96*/3z*) w*

If we define the quantity D(t) as the ratio of initial to curent mass

flow rate, viz:

'
1 r i
(12) D = i- I—) dz
o o 5<

where D is a monotonically increasing function of time, and D(0) = 1,

then we can write (6) and (11) as

(13) w* = ( p / A p v l o ) D _ 1 ( % 2 [ 1 - % ]
6 6
* 2
_1
(14) u* = CPr^Apvl o )D ^ 2 - [ 1 - % ]
6 3z 6

£
• ^
The bulk mean axial velocity, w, is given by M/piriS :

(15) w* = (Pr^/8pvlo)D_1 (-J)2

and it is at this point that we can formally identify

w = wfz ,0) = Pr2/8pvl . Hence the initial Feclet number can be

o o o
expressed as

(16) Pe = Pr^/4pv<l .

Vhen variations in density are included in the model, the effect of

density needs to be incorporated as an additional radially convective
term in the heat diffusion equation. However, the time dependent term
in the continuity equation would be neglected anyway due to the small
Stefan number.

- 27 - 5:Fluid Flow
5:2:Turbulent flow

Equation (4) was used to derive a laminar axial velocity from a

pressure gradient which was known to be independent of the radial

location. An equation often used in pipe flow is

* * 1 —*?
(17) 3p /3z = - — f' w /d

where d is the pipe diameter and f' is a friction factor dependent upon

the nature of the flow. In the laminar flow case, d is simply equal to
*
25 and the friction factor is given by the Hagen [2] - Poiseuille [3]

law

(18) f' = 64/Re

where Re is, in our case, defined by

(19) Re = = (Pr3/4Pv2l )r /D(t*)<5*(z*,t*)

0 0 0

and ReQ is consequently given by

(20) Re = Pr3/4pv2l
0 0 0

In turbulent flow, we use the well known Blasius [4] law

(21) f* = 0.3164 R e " 1 ^

from which we obtain upon substitution into (17),

/ \ *, * 1/&.—*!/L *-15/4
(22) 9p /az = -0.06651pv w " 6 57

- 28 - 5:Fluid Flow
In this case, w is a turbulent average axial velocity and is still

subject to the definition of the mass flow rate;

(23) w* = M/Ptt6*2

Thus

(24) 9p*/3** - -0.06651p-3/4v1/i[k/T]T/4s»-1-9/4

* * *
Integration of this over z from z = 0 to z =1 then gives

(25) P = 0.06651 p " 3 / V / 4 l r'19/4[M/Tr]7/4 ° (-§) 9 /

dz* ]
o 6

in order to satisfy the constant pressure drop condition. Thus the mass

flow rate can be given in terms of the imposed pressure drop and the

current crust configuration: >

PkI 2vr 3/7 Pr U/7
(26) M = T.T20U .1 -j-a ( - ^ ) ( j ^ )
o o

where D is again the ratio of initial to current turbulent mass flow

rates,
n tlr 19/k „ k/j
(27) D = [I I ° (_|) d2 ]
o o 6

s _ * *
The initial Peclet number (definition (4.4)) where w = w (z ,0) =
o
* 2
M(0)/pirr as in the laminar case, is given by
0 U
(28) Pe - 15.V.O8 ,fZo,3/T 1'
t/7
0
r ICL UpVKI 2
0 0 o

and the initial Reynolds number, similarly by

KI b/j Pr4 k/t

(29) „ = 30.8816 ( ^ ) )
o o

The radial velocity is again related to the axial by (11) in order to

satisfy the turbulent averaged mass conversation equation.

- 29 - 5:Fluid Flow
 It should be borne in mind that equations (22), and (24) through

(29) are a direct consequence of the Blasius friction factor law (21),

the mean bulk axial velocity, (23) and the boundary condition that the

pressure drop over the pipe length is held constant. In fact (21) is

suitable for Reynolds numbers not exceeding approximately 100 000. Note

also that nowhere do we have an expression for the turbulent axial

*
velocity w as we have for the laminar case (equation (13)). This does

not turn out to be a serious drawback as might be expected, since only

the behaviour in certain regions is of any importance. There are many

other friction factor laws, mostly derived from experimental results

and curve fitting (see for example [5]) but these suffer from the

disadvantage that they are either implicit or consist of a sum of at

least two terms in fractional powers of the Reynolds number. This

presents difficulties for analytical methods and it is not possible to

obtain the necessary relationship between pressure drop and mass flow

rate in closed form. This is not to say that numerical methods could

not be used but the effort involved in this must be weighed up against

their accuracy - or even applicability - to the flow of liquid metals.

- 30 - 5:Fluid Flow
6:REVIEW OF PREVIOUS MODELS

Most of the work in this area has been carried out over the last

two decades and many of the elements of this present work can be found,

in some form or another, in the literature. There has not been,

however, a coherent attack on the problem such as can be found in this

thesis.

6;1:Steady state laminar flow

One of the earlier systematic models was developed by Zerkle and

Sunderland [1 J. They examined a steady state model, by which we mean

that the crust has stopped growing, and compared the results with

experimental data for water and ice. The basic equation used for the

energy equation and associated boundary conditions are the same as in

this thesis, (4.10) - (4.12), with the logarithmic solution in the

solid region, (4.14). The hydrodynamics differed in that whilst they

assumed the parabolic axial velocity profile, which is almost

universally accepted in models of laminar flow, they did so without

neglecting the non-linear convective terms in the axial momentum

equation. They gave arguments why the parabolic velocity profile is

retained, despite the narrowing of the pipe and the presence of these

non-linear terms, but at the same time stated that the flow is fully

developed. This leads to a rather different expression for the fluid

pressure gradient to (5.8). One item that did emerge from this analysis
1/B
was a single curve, effectively 6 as a function of z in our

notation, which it was claimed would suffice for all pipe lengths. In

other words, given B, it is possible to determine 5 at any place along

- 31 - 6'.Review
the pipe axis. Additionally, they derived an equation for the pressure

drop which depends upon the Prandtl number of the fluid, and the

analysis was carried out with Pr = 5, 10 and B = 0, 0.5, 1 , 2 as

examples. A semi-empirical solution was also developed which provided a

1/B
set of curves for 6 , variable with the Reynolds number, which were

flatter than the theoretical curve nearer z = 0 but steeper further

along the axis. The main drawback in this model is that there was an

indefinite pipe length.

In a later paper [2], a fixed pressure drop over a specified

length was used in conjunction with a bulk temperature (ie radially

averaged) in the fluid with heat transport by fluid flow rather than

conduction at the interface. This is the Nusselt Number approach, used

in much of this type of work and depends upon heat transfer

coefficients (h) rather than thermal conductivity (k). The Nusselt

number, Nu = hd/k where d is the cylinder diameter, is usually used to

calculate h rather than as a parameter, or as a comparison between

convection and conduction.

1/B
This provided a model which led to a series of curves, $

against z, dependent upon the Reynolds number, and a crucial

observation that two solutions existed which satisfied the constant

pressure drop condition. A valuable graph is provided which plotted

experimental results - those where blockage occurred and those where it

did not - against axes with B and what is effectively ReQ, or 1/a, in

our notation. We use this data in the validation of our laminar model

(Chapter 9)• The double solution was noted independently at about the

same time by Stephan [3_| who dealt with the solidification problem with

- 32 - 6:Review
fluid transport and the heat transfer coefficient approach in two

coordinate systems (Cartesian and cylindrical polar).

Whilst the heat transfer coefficient approach seems to be

favoured by some workers, it is important to stress that this is only

one way to provide a counterbalance term for solid heat conduction in

the crust development equation (4.15) as a replacement for aT^/Sr. It

is not the case that this is the only way by which double solutions for

steady state crusts can arise.

6:2:Transient laminar flow and solidification

By transient, we mean to take into account the development of the

crust in time as well as in space. Many of the transient models do not

assume that the time derivative of temperature is negligible in the

heat equation. Such an example, Cheung and Baker [4], is the basis for

a model of liquid metal flow in a fastbreeder nuclear reactor. The

convective terms in both the axial and radial momentum equations are

retained but the radial pressure gradient and axial momentum transport

by viscous diffusion are neglected. This model is used for the purpose

of obtaining the relevant dimensionless parameters and attempting to

correlate these with experiments using Stainless Steel, Oxide Fuel,

Naphthalene, Water, Paraffin Wax and Olive Oil, in order of increasing

Prandtl number. An earlier paper by Martinez and Beaubouef [5J had used

a similar model and included the pressure drop equation used in [1],

12J to provide a model for the development of the crust configurations

with time by analysis and computaticn.

- 33 - 6:Review
 The main point to note about these models is the retention of the

non-linear convective terms in the axial, and sometimes the radial,

momentum equations. This can be a drawback from the point of view of

computation and it necessitates a justification of the parabolic

velocity profile. It would be more appropriate if this level of

refinement were applied to flow of liquid and heat in the entrance

region of the pipe where the flow may not be fully developed. Hwang and

Sheu ]_6J performed some theoretical analysis in this region, with time

derivatives in energy and momentum neglected (the small Stefan number

approximation). In this case, the interface position, interface heat

flux from the liquid side and also pressure drop were compared with [2J

and showed good agreement. Again, these quantities were noted to be

dependent, by analysis, on Reynolds and Prandtl numbers.

6:3tTurbulent fi.ow

The earliest work we will consider in this section is a paper by

Oziqjik and Mulligan [7] which is the first in a series of

investigations by Ozi^ik and his school. In this work they assumed a

slug flow axial velocity profile. This is simply a uniform velocity and

is a reasonable approximation to the mean-flow axial velocity in

turbulent flow (it does however necessitate ignoring the no-slip

condition). In order to satisfy the continuity of mass equation, this

velocity varies slowly with z and is proportional to 1/5 . Thus as the

pipe blocks then 6 -

*
• 0 and consequently the model breaks down. Again,

there is no definite pipe length included in the analysis. Transient

terms, ie 3T/3t, are retained in the heat equation for the fluid but

quasi-steady heat flow is assumed in the solid. Most of the analysis is

- 34 - 6:Review
performed with integral transform (Hankel) methods. Mulligan and Jones

followed this with a paper [8j documenting some experimental work on

•water.

Shibani and Oziqjik then introduced the work done on the turbulent

Graetz problem (see chapter 8) to the calculation of the crust position

from the liquid interface heat flux, first for flow between parallel

plates [9] and then for cylinders [10]. There is however an indefinite

pipe length and no conclusions can be reached about blockage.

An important paper by Thomason, Mulligan and Everhart [11]

considered a steady state turbulent model. They retained convective

terms in the axial momentum equation and averaged radially to obtain

the equation for the axial pressure gradient used in this thesis. By

employing the Blasius friction factor, they derive an expression

equivalent to our equation (5.24). This is, however, without the

crucial step of using a finite pipe length to produce the control over

the axial velocity in order to satisfy global mass conservation (5.25).

Finally, Cho and Ozi^ik [ 123 applied the Laplace transform technique to

the explicitly time dependent form cf the heat equation for the fluid.

Curiously, they neglected the radial convection of heat in this

equation. Again, no finite pipe length is included in the analysis.

Another series of slug flow models in which the transient fully

convective axial momentum equation was used appears in work by Epstein

et al. [13]> and soon after by Epstein and Hauser [14]» These use the

Blasius friction factor approach to recover equations similar to those

of (5.22) and its derivations. They assume that 6 is known, and

- 35 - 6:Review
postulate simple functional forms for it and derive expressions for the

leading edge of 6 in the axial direction; that is to say, how far along

the pipe solidification has occurred. Used in conjunction with the

small t solution (U. 19), this model has been quite successful in

predicting the penetration depth.

Finally, there is the direct approach of the Szekely and DiNovo

model [15] which employs diffusion equations in the liquid, solid and

pipe wall, with heat transfer coefficients between liquid/solid,

liquid/wall and wall/air. Some analysis is performed, notably the

recovery of many dimensionless parameters, but the work is geared

mainly to solution by computation. Various graphs are presented which

show the percentage closure of the nozzle at various positions along

the pipe length. It is interesting to note that the furthest that the

crust develops is no more than about thirty percent of the pipe radius

although it is clear from the information displayed that the crust

thickness would have increased had the computer time been available.

Also, the possibility of remelting is a feature of this model.

6:4:General conclusions and hiatus

A number of different models of the solidification problem with

convected fluids in pipes have been developed. This is to be expected

since on the one hand, variations may be needed for different types of

fluid and on the other, the work is not invariably directed toward the

same end. For example, some of the models are constructed to calculate

correlations between bulk temperature and Nusselt numbers since this is

useful in certain engineering applications. Others are developed in

- 36 - 6:Review
order to calculate pressure drops. Only a few are developed explicitly

to predict blockage.

Since blockage, or more properly non-blockage - as should become

clear in the following chapters - is what we are interested in, we are

able to select from this past work those theories which are directly

relevant to this end in order to construct a reasonably eclectic model

from them. In the final analysis, we can claim that it is possible to

find steady-state solutions, i.e. crust configurations, for both

laminar and turbulent fluid flow where, in the case of the latter, it

is in the regime adequately described by the Blasius friction factor,

and that where no such solutions are possible then blockage is the

predicted outcome. We can approach steady state solutions directly or

via transient modelling. We will adopt the transient approach since the

mechanisms are much clearer, the possibility of timing is not lost,

and, with the experience gained in this way we can approach the more

complex variable viscosity and mushy-region flows, to be found in the

last chapter, directly as steady state models with greater confidence.

- 31 - 6:Review
7:FORMULATION OF THE PROBLEM

On combining the material of the preceding chapters, the

situation is as follows. We wish to calculate the position of the

solid/liquid interface, ^ , as a function of the axial distance down the

pipe, z, and also as a function of time, t. The governing equation is

(4.15) i.e.
(1 \ o _ B
^ ' at 5 In 5~

where the freezing parameter, B, is defined as (3«9)

(2) B = k (T - T )/k_(T - T )
s m w i o m

and the interface heat flux from the liquid side, q, must be obtained

from the solution to (4.10):

/_n 3T 3T 2 3 / 3T \
(3) u +
17 " U " 7 1 ? (r ®37 '

with boundary conditions (4.1l) and (4.12) where q is equal to 3T/3r at

r = & Equation (3) above, as written, is slightly different to (4.10)

since we have used the alternative form of the energy equation which

can be obtained by employing the mass continuity equation in the

left hand side of (4.10). We have also divided throughout by pC on the

assumption that this is constant, and denoted the dimensionless thermal

diffusivity by g. We have the relationship between u and w, (5• 11)

(4) u = w
t H
If we now make the following coordinate transformation, a standard

device often employed in this type of work, to remove the moving

boundary from the equations

- 38 - 7:Formulation
(5) R = r/ 6
, z = z, t = t

then equation (3) in conjunction with (4) yields

(r\ ,2 3T _ 2 3 / _OT\
W
^ ^ 3z R 3R

subject to

(7) T = 0 on R = 1

(8) T = 1 at z = 0

where T, w and g are now to be regarded as functions of R,z and t. This

transformation has removed the necessity for tracking the crust

position from the heat equation and has effectively transformed a

variable pipe radius system into one with a constant radius. This does,

however, modify equation (1) which is now

(9) 2 6 36/3t = B/lnfi + Q

where Q = -3T/3R, evaluated at R = 1. How we proceed to find Q depends

upon the nature of the flow. If fully developed laminar flow, then the

dimensionless thermal diffusivity, g, is simply equal to unity and

plays no further part in the analysis. The dimensionless axial velocity

is, from (5.13) onward,

(10) w = W*/Wq = 26~2D~1(1 - R 2 )

where

(11)
o

(12) a= 1 /r Pe = ApvkI^/'Pv^
0 0 0 0 0
Thus, we have to solve

2
= D
[ |3p -T . 1 3T I
3R2 R 3R J

- 39 - 7:Formulation
for 0 <R< 1 and 0 < z < a, subject to (7) and (8).

If on the other hand the flow is turbulent, then the diffusivity

is radially dependent and will be denoted by g(R). The actual turbulent

thermal diffusivity will be equal to the laminar value near the wall

but will be, in general, greater than the laminar diffusivity in the

central regions of the pipe. The turbulent axial velocity is similarly

difficult to formulate precisely. If we assume Blasius type flow, as

developed in (5:2), then we can write

(14) w = 6"2D_1f(R)

where f(1) = 0, and

n
(15) 2Rf(R)dR = 1
o * _ * *
which ensures that w has been correctly scaled with w(z ,0) = w q ,

for which the expression is given by (5.23) ff. The dimensionless mass

flow rate parameter is given by (5.27), and so

(16) D -[ i V9/4ds ]4/7

where again, the dimensionless pipe length, a, is given by

(17) a = 1 /r Pe
0 0 0

and so the heat equation is

(18)
J

again subject to (7), (8) and (9).

Note that the initial Peclet number for laminar and turbulent

flow is in both cases the ratio of convective to thermal dissipation

but in each case is given by a different formula in terms of the

- 40 - 7 formulation
inputs, laminar "by equation (5*16) and turbulent by (5.28). This has

corresponding effects on the definitions of a for laminar and turbulent

flow in terms of inputs.

- 41 - 7:Formulation
8:THE GRAETZ PROBLEM

This is concerned with the solution of

to v , .1,1.,
( w
3z r 3r 3r
subject to the boundary conditions

(2) T*(r*,0) = Tq
* *
(3)
v T (r ,z ) = T
o m

*
where k is the thermal diffusivity of the fluid and w is the axial
* *
velocity of the fluid. Both w and < are function of r only. This set

of equations is a special case of (4-.1) in that it governs a

steady-state situation with constant density and specific heat,

negligible axial conduction and constant radius pipe (studied

originally by Graetz [l]).

8:1;Laminar flow - Sellars, Tribus and Klein

The axial velocity has a parabolic profile and the thermal

conductivity is constant. Equations (l) - (3) can then be written in a

dimensionless form, using the standard non-dimensionalization factors

described in the last two chapters.

(4) (1 - r2) f = i | - (r |£)

3z r 3r 3r

(5) T(r,0) = 1

(6) T(1,Z) = 0

It is then possible to write a superposition solution for T(r,z) in the

form

- 42 - 8:The Graetz Problem

(7) T(r,z) = I CnRn(r)exp(-X2z)
n=o

where the C are constants and the eigenfunctions R (r) must each
n n
satisfy the eigenvalue problem:

(8) rR" + R' + X2r(l-r2) R = 0 for 0 < r < 1

n n n n — —
(9) Rn(0) = 1

(10) RJ1) = 0

The first few eigenfunctions and eigenvalues (X ) can be calculated

numerically but for higher values of n, and hence large X , asymptotic

n
formulae are needed. For large n (n >> 1) Sellars et al [2] split the

domain into three regions, viz: r ^ 0, 0 < r < 1, and r ^ 1. The region

near the centre of the cylinder has the solution, satisfying (9),

(11) R(r) = J (Xr) as r + 0

0

where we are temporarily omitting the subscript n. Away from the centre

and the wall, the WKB approximation was used to provide an equation

expressed as a power series in 1/X. This can be matched with the large

Xr asymptotic expression for (11) and gives

1
1/2 2 l/4
(12) R(r) = (2/ l rXr) (l-r )- cos(|^Cl-r ) 5 +| a rcsinr - £ )
2

= 2/3
For r ^ 1, a change of variable n X (l-r) yields another power

series in l/X from (8) with a first order solution which can be matched

with (12) to provide eventually

3
l/2
(13) Rn(r) = [2(1 - r)/3j ( - D n J l / 3 ( | \ J 2 (l-r) )

and also the eigenvalues

(14) X ^ 4n + 8/3 for n -

»
• »
n
which are chosen to satisfy (10). It is then a simple matter to find the

- 43 - 8:The Graetz Problem

values of C n> by standard fourier analysis for example, in order that (5)

be satisfied.

The dimensionless temperature gradient at the pipe wall is then

given by

( 1 5) = I CnR^(l)exp(-^z) at r = 1
n=o

and the asymptotic form for the coefficient is easily shown to be

(using (13)),
rC2/3)
(16) C R'(1) = - 7 ^~ 1 / 3 = 1-276 (n + f ) ~ l / 3
11 n 3
3 r(U/3>

8:2:Turbulent flow -Sleicher, Notter and Crippen

In turbulent flow, a mean axial velocity is used to describe the

flow of the fluid along the cylinder. There is, unfortunately, no

simple form for this velocity, although it is a function of radial

position only. Another complication arises from the fact that the

thermal conductivity is no longer constant, and we must use a radial

dependent turbulent thermal diffusivity. Enough is known about the

behaviour of these quantities near the centre and also near the edge of

the pipe to enable much progress to be made. Work by Notter et al [_3J,

L4-J, [5 J is directed to this end. They consider equations (1) - (3) in

the same way as (7), which in their notation leads to the eigenvalue

problem

( 1 7) ^-(rg(r)-^n) + -5-A^rf(r)R (r) = 0

dr dr 2 n n

subject to the usual conditions (9) and (10). Here, f(r)/2 and g(r) are

the dimensionless forms of the axial velocity and turbulent thermal

- 44 - 8:The Graetz Problem

diffusivity respectively. By using the following transformations;

(18) y = (r 2 fg/2) l/4 R

n

i rr *
(19) 5 = Q (f/2g) dr
o
which maps r = 0 to £ = 0 and r = 1 to ? = f

(20) w(5) = (r 2 fg/2)" l/4 — 0 (r 2 fg/2) 1/4

d
£~
(21) k = GX

where k is a redefined eigenvalue and

1 f1 2
(22) dr(f/2g)
•'o
they are able to reduce (17) to

(23) + (k2 - w(g))y = 0

a r

Now near the centre of the pipe, f and g are approximately constant and
2
so w(g) - 1/45 as £-»• 0. Away from the centre and the wall, w is
2
small and so k >>w. Near the wall, the turbulent diffusivity is very

nearly the same as the laminar diffusivity and so g = 1 and f = 2H(l-r)

is a good approximation to the turbulent velocity where

(24) H = Ref/32

and f' is the Moody friction factor. At this point, some difficulty

arises from the lack of knowledge concerning the turbulent velocity

gradient near the wall and they propose that in this region, w can be

approximated by

(25) w(^) = c / U - 0 - 5/36(n - $) 2

where c is undetermined and depends upon the derivatives of the

- 45 - 8:The Graetz Problem

velocity and diffusivity at the wall. After a complicated series of
—
1 /2 "3/2
manipulations involving asymptotic power series in k and k~ ln(k)

like terms, with matched asymptotic techniques used to patch the three

regions together, they finally obtain for the heat flux at r = 1;

00 T

(26) " F = IL C R (l)exp(-X2z)

3r n=o n n n

where

(27) GX = (n +|-) + — - — 5 * [ c In (nit + -fir) + -ri- ]

n 3 3 l8lT
2ir(n-t|)

(28) C H'(1) = ~ "

[ ± _ 1 { |_ ( ^ J }
n n
irGXnl/3 3 r(U/3) (
n 2
* n

+
>]
which, given H, G and c, is the information that we will require. It

should be noted that the laminar flow result can be recovered from this

by the substitution of the laminar flow friction factor definition

(equation (5.18)) into (24) and the fact that g = 1 and f = 2(l-r2)

imply that H = 2 and G = 1/4 respectively. In general, however, H, G

and c will depend upon the Reynolds number of the flow and the Prandtl

number of the fluid. Reference [5] gives tables of these numbers, with

non-asymptotic (small n) numerical calculations of X and C R'(1), for

n n n
a wide range of Reynolds and Prandtl numbers.

The quantity c, above, was obtained by applying the asymptotic

analysis outlined above and using the results to calculate the low n

eigenvalues. This calculation is then compared with the numerical

computation which will give these eigenvalues directly since it is not

too difficult to calculate the first few eigenfunctions numerically.

The value of c can then be found by equating the two results. This

attack is clearly suspect but the authors acknowledge this and further

- 46 - 8:The Graetz Problem

point out that although the value of c obtained in this way will be

incorrect for the higher eigenvalues, the effect will not be as

noticeable since c appears only in the second order approximation to

the eigenvalues. They also point out that this value of c has the

advantage that it then enables the asymptotic analysis to apply to the

low eigenvalues.

This author, however, suspects that (25) may be misleading since

the expression for w(^), if laminar functions are used in (20), is

(29) w(£) = -(1 + 4r4)/64r2(l - r 2 ) 3

(30) £ = 2r(l - r )2 + 2arcsin(r;

and an expansion of w near G = I

T leads to the result that
2 2/3
(31) w(5) = -5/36(tt - 5) + 0.102VU - s) + 0(1)

which seems to suggest that c is a fiction and that an analysis based

—2/*5 —1
on an order (ir-G)~ term rather than the (TT-^) approach might be

more fruitful. This is borne out by the fact that the published values

for c appear to be rather erratic.

- 47 - 8:The Graetz Problem

9:LAMINAR FLOW

To recapitulate, we wish to calculate the position of the solid

liquid interface, 5, either as a function of z and t or as a steady-

state function of z. We use the equation

(1) 25 — = _ IT fort>0,0<z<a
- -
3t In 6
5 3R _ _
-L
subject to

(2) 6(z,0) = 6(0,t) = 1

using

(3) ' 1-(Bt)^ 2

6v for t ^ 0 and z > 0
00
to provide an initial guess since 36/3t -
*
• - as t 0. The

dimensionless number B is the freezing parameter and depends upon the

material and the uniform entry and wall temperatures:

(4) B = k (T - T )/k (T - T )
s m w f 0 m
The dimensionless pipe length, 10 , is given by

(5) a = 1 /r Pe = 4pVKl2/Pr4
0 0 0 0 0
To calculate 3T /3R at R = 1, we are required to solve
2
/r \
(6)
/
.
(1
t
-,2^ 3T
R
_
D
f 3 T + 3T -]
- = L-9R2 M R J

(7) T (R,0) = 1, T (1 ,z) = 0

where, in order to satisfy the constant pressure drop condition,

fa -k
(8) 6 (z,t)dt
a

Equations (6) and (7) have the solution for the flux,
OO
A
(9) l i = - LI expC- X2Dz) at R = 1
3R n=o n * n

where, for n small, X and A are tabulated and for n large,

n n
(10) X * 4n + 8/3
n

- 48 - 9:Laminar Flow
 2Jl//30 r(2/3) "I ,~l/3
I
T A
n
r(U/3)

The assumptions we have made are:

(12) (a) The liquid is pure, incompressible and Newtonian.

(b) Physical properties of both solid and liquid phases are

constant.

(c) Axial conduction is negligible (Pe >> 1).

£
(d) Viscous dissipation is negligible (Br << 1)
(e) The densities of each phase are identical, or the difference

between them does not affect the liquid velocity.

(f) The system has a uniform temperature T at t = 0, and at

0
t = 0, the nozzle wall temperature is changed to T and held
w
constant

(g) New fluid enters the system at uniform temperature T .

0
(h) The axial velocity is parabolic, with no-slip at the

solid/liquid interface, fully developed laminar flow.

(i) Pressure drop over the length of the freezing section is

constant for all time.

(j) Fluid flow and heat transfer adjust to the changing interface

much more rapidly than the interface changes (St << 1).

The non-dimensionalization is

(13) (a) r = [rQ]r radial coordinate

*
(b)z = [L r P e J z axial coordinate
0 0
(c) t* = [ p A / k f ( T o - T m ) ] t time
* r n
(d) T = T + [T -T IT temperature
m 0 m
(e) 6 = [rj 5 interface location

(f) p = [P]p fluid pressure

'The Brinkman number is the ^ratio of heat generated to heat

transferred, defined by Br = [yV /k„(T -T )].
f o m

- 49 - 9:Laminar Flow
and the moving boundary, 6 , has been explicitly eliminated from the

heat equation via the transformation.

(14) r = R6

with the consequences to differentiation;

9 1 3

9r* 8 3R
(1f>) 1_ 4 L J M L
U D ;
3Z 3z 6 3z 3R

9:1:Integration

Due to the occurrence of the time dependent parameter D in (9),

numerical integration of (1) is not quite as straightforward as the

equation would indicate since D depends upon the entire crust at the

time of integration. It is still easy to perform, however, and a four

step Runge Kutta integration procedure is used. The scheme adopted is

as follows, where superscript 'i' refers to an axial grid point and

subscript 'j' to a time step. Initially, we provide, from (3);

(17) (a) = 1

(b) rj » 1 - 3/2(BAt) l/2

(c) = 1 - 2(BAt)1^2 for i = 2,3...Nz

i 2
where r. = S (iAz,jAt) is used ra~her than 6 since it is easier to
J
2
integrate (1) for 6 . Initialization 17(b) is present purely to make

the transition from the uniform crust to zero crust at the entrance

slightly less abrupt. The number Nz is the number of axial grid points

used, and was typically 20-30. The initial estimate of the mass flow

rate parameter D can be calculated from the formula, valid for all j,

(,8)
V w T 2 [ trj)"2 + C r f ]
1—O ,e:,

for j = 1 in this case. This is Simpsons rule, and necessitates

- 50 - 9:Laminar Flow
choosing Nz to be even. From this initialization, we can proceed with

the familiar Runge Kutta process;

N
(19) k 1 = At [ 2B/ln t] + I A11 exp(-X121D.iAz) ] i=l,2,...Nz
« J

where the summation of the Graetz series is performed with the N+1

values of A q and necessary for convergence of the series within an

imposed criterion. (The axial step length is z = a/Nz).

At this point, we have a choice. This is only the first step in

the fourth order scheme and we can either proceed with the following

steps or we can recalculate D. We choose the latter course since D is

developing in time and as such can be regarded as a dependent variable

on much the same footing as The 'old' value of D is hence replaced

J
by the new value:
Nz-2
(20) D (I + ki
o - 3 k .1 I 1 I >"2 +
^ + 1 +
I ki+1
)~2 +
<rj+2 + \ * i + 2
K
i—o ,2,

and the process can proceed in like fashion:

(21) l 1 = At [2B/ln(r^ k1) + ]

J ^ J

C(r +
(23) - - 3Ni r j 2 ^ * I li+1)"2 - ( f 2
• I l i + 2
n
X—O ,

(23) m 1 = At [ 2 B / l n ( j l1) + ]
J ^ J

(24) D =
i1 3 ^
JJNZ
£ L ( ^ + m 1 )" 2 + l+(rj+1 + m 1 + 1 ) " 2 + (l4+2 + m i + 2 ) " 2 ]
J
i=o,2, J J J

(25) n 1 = At [ 2B/ln (r* + m1) + ]

J J
where at each stage,

(26) - V A exp(-A2D.iAz)
Jl L
n=o
*•*
n n °jn '
n

- 51 - 9:Laminar Flow
the D. in (26) being the most recently calculated value at each step,
J

viz (20), (22) and (24). Finally we calculate;

(27) rj +1 = rj + | (k1 + 211 + 2m1 + n 1 )

( r ) 1 / 2
* k

and D. . is computed from (18) with the appropriate subscript

J

modification. Subscript j is then incremented, and we return to (19)

for the calculation of the crust at the next step, and so on.

9:2:Results

It is not immediately obvious from the foregoing what will

happen. There appear to be three possibilities:

(1) Swill decrease indefinitely as time passes, eventually turning

negative.

(2) 6 will eventually stabilize and remain constant for further time

steps.

(3) 5 will reach a minimum configuration, then increase again (i.e.

re-melting.)

The corresponding behaviour of D would be; an indefinite increase, an

increase followed by a stabilisation at some constant value, or an

increase up to a maximum value followed by a decrease; respectively.

The possibility that the crust would turn negative is of course purely

mathematical/numerical and corresponds to blockage of the cylinder

where the interface reaches the centre at a finite, non-zero, velocity.

As mentioned near the end of chapter 3, other more detailed analysis

must be applied to modelling of total solidification. As far as this

work is concerned, we merely note that blockage will or has occurred

and stop the calculation. Figure (1) shows just such a case.

- 52 - 9:Laminar Flow
 Figure (1)

Blockage

z/a —
Q-.qq 0.10 120 a30 0-40 0.50 0.60 070 0 8Q
090 1.00

a * 1CT5
B =9-5

Figure (2)

Steady State

z/cc-»

00° 0.10 020 0.30 0.40 050 0.60 0.70 0.90 0.90 1.00

- 53 - 9:Laminar Flow
 The second possibility is that of a steady state solution. This

occurs where each Q 1 in the aforementioned numerical scheme balances

J

the contribution from the 2B/lnr^ term to within some small limit.
J

Figure (2) shows an example.

Table (1) shows some results obtained by computation. Rather than

present graphs for each numerical experiment, we will note merely the

value of 6 at the pipe exit and the time taken to reach that

position, where steady states result,

Figure (3) Table (1)

B <5,(°0
0.20 0.9935 1.020 0.00297
1.20 0.9602 1.129 0.0225
2.20 0.9256 1.261 0.0450
3.20 0.8895 1.423 0.0675
4.20 0.8513 1.627 0.0975
5.20 0.8107 1.892 0.125
6.20 0.7669 2.250 0.160
7.20 0.7184 2.762 0.2025
8.20 0,6628 3.568 0.280
9.20 0.5936 5.085 0.410
10.20 0.4718 10.790 1.280
11.20 <0.25t
12.20 <0.15t
13.20 <0.10+

time taken t o reach steady

state.

t.
blockage times

The table presents results for a fixed length pipe with the parameter B

controlling the solidification depth. As the wall temperature decreases

below the melting point (or alternatively, as the inlet temperature

approaches the melting temperature from above), the parameter B

- 54 - 9:Laminar Flow
increases and so does the crust thickness. Figure (3) plots the values

found in table (1) for the exit crust and time taken versus B.

The third possibility, that of remelting, never occurs. To

explain this, and the behaviour of this model in general, we will

consider the crust velocity, otherwise known as the blockage rate.

9;?:Phase plane analysis

Figure (3) seems to indicate that at or near B=10.5, the crust

position at the pipe exit must drop rapidly to zero. In fact this is

not the case and we have indicated the doubt on this graph by the wavy

line. What actually happens is that the curve has reached a turning

point at that particular value of B and is about to reverse direction,

thus providing, for a given value of B less than about 10.5, two values

for the outlet crust position. One is near the pipe wall and the other

is nearer the axis. The break point occurs when is near 0.4. If we

plot, for fixed a, the position of the crust at the exit versus its

(negative) velocity we obtain the series of curves shown in figure (4).

Each curve, for a different value of B, starts on the right hand side

at t = 0 with an initially thin crust, 6 ^ 1, developing at a

relatively rapid rate. As time elapses the crust develops but more

slowly. Either the blockage rate eventually slows to zero (shown for

BO0.2), in which case a steady state has been reached, or the blockage

rate reaches a minimum value, indicated by the bunching together of

subsequent crust profiles as time passes, in figure (1), and then

increases again as the pipe blocks. At no time does the crust velocity

become positive. The points of intersection of the curves with the

- 55 - 9:Laminar Flow
36/3t axis appear to be points of stable equilibrium.

Figure (4) - Pipe Exit Crust Trajectories

If we use the right hand side of equation (1) and plot it as a function

of <$(z,t), including the effect of D(t) from the upstream crust

positions, one of the curves shown in figure (5) would result. Which

one would depend upon whether B were high, critical or low,

respectively the curves annotated 1,2 and 3. This critical value

(B .,) will in general depend upon a, unlike the case in certain

cri u
1/B
models reviewed in chapter 6 which present a single curve for 5 (z),

valid for any z.

If a is fixed, (say the pipe length is constant), then it can be

seen that where B < B there are two steady state solutions for
crit
5< (a) (where we are henceforth using the notation 6 (z) to signify
s s

- 56 - 9:Laminar Flow
steady state solutions). This was noted by Stephan [6.3] and

independently by DesRuisseaux and Zerkle [6.2] as pointed out in

section 6:1.

Figure (5)

It was not discussed in great detail

however and an investigation into

this situation would be in order

since while it appears obvious that

the two solutions are stable (marked

s) and unstable (marked *) merely by

inspection of figure (5), it is in

reality not quite so simple due to the presence of D in the right hand

side of (1) through 3T/3R and D depends upon the entire crust

configuration and not just upon the crust at the particular point which

we are examining (currently at z = a).

9:4:Stability and steady-states

We can approach this problem from another direction, namely from

the definition of D, equation (8). If we denote a steady state value of

D (where extant) by D , then it is soon realised that it is not a

s
number which can be determined in advance of the numerical solution by

inspection. The steady state version of (1) is:

36
(29) 26 - a T s = 0 " B/ln(fi ) + Q(zD )
sot s s
where
00

(30) Q(x) = J
L A exp(-X2x)
n=o n * n

and so S< (z) cannot be immediately deduced. We may instead plot a

s

- 57 - 9:Laminar Flow
function D as a function of D where
3 3

(3D D* = -1 S'^DZ = - exp 0*B/Q( ZD )]dz

s a . s a S

and then, either graphically or by numerical means, find a value of Dg

which when inserted into the right hand side of (31) would yield a
*
value for D = D , if possible. Figure (6) is a sketch of the expected
S 3

functional forms of equation (31). Curves 1,2 and 3 in figure (6) are

the three possible outcomes of plotting D against D for three

3 3

different values of B, respectively B > B = and B < B

*
The diagonal is the line D = D and where this line cuts curve 3 are
3 3

to be found the two steady state solutions for D , and hence for 5 (z)
3 3

by employing (29) as a prescription for 5 . The smaller value, ^,

3 3

corresponds to the thinner crust - point 's' in fig. (5) - and the
C2 }
larger value, , to the thicker (point in the figure). We have

to prove that the former, thinner, crust represents the stable

steady state solution. In the following, we shall replace the function

Q(zD(t)) by the function f(z,t), and 0(zD ) by f (z).

3 3
*
Figure ( 6 ) -D versus D
S S

9:Laminar Flow
 two
For a given a with values of D return two possible

steady state configurations for the crust. The larger value of D

would result from a crust which had encroached further into the liquid.

It is maintained that the crust would never actually reach this state

and would stop development at the lower value of D . This may seem

clear from figure (5)» however, figure (5) shows only the

phase plane of the value of the crust at the end of the pipe. In fact,

we need to show that the point marked S is stable in the entire p h a s e —

c o n t i n u u m for all z since (1 ) is an integro-differential equation.

This is a fairly straightforward but lengthy process.

Equation (1) can be written as

+
(32) =
H r for 0 < z <_ a

and 0 < t

where
r 2 d z
(33) = 6 and D ( t ) = ^
2
o ? (ztt)

It is known that as z-*0

(3U) f = 0(z and T(Zjt) •

+ 1
and as -
*
• >
*
<
(35) fOCe'b*)

and that f is montonically decreasing with z . If we assume a steady

state solution given by T = T C z ) and D =

•» D then
8 3

(3S)
v r r + i'2' m 0
s

i.e. r = exp[-2B// ( z ) \

- 59 - 9:Laminar Flow
 i <•
and s o D = — exp [4B/f3 (z)~]dz , or
s a 3
0

oD
S
<3T> » S " 3 C exp [ 4 B / f s ( z / D s ) ~ \ d z
S

Equation (37) has two, one or no solutions for D , as depicted in

s

figure (6), since f is monotonically decreasing, and so the right

©

hand side is monotonically increasing from the value one. We now

perturb this solution and set

(38) D(t) = D + d ( t ) ' T ( Z j t ) = T ( z ) + y(Zjt) . :

S 3

and this gives, from (32)-(36-) as y and d are small

<39) H = +
d t
T In T ~ **
S 3

y(zat)dz
(UO) d = —
A 3
o r (z)

Introducing normal modes by letting

a i
(Ul) y(z 3 t) = e g(z)

where a is the growth rate, then

g(z)dz
(1,2) (A + ^ , "J
r ln r z
a >o r (z)
s s s

This is a homogeneous integral equation for the eigenvalue a with

separable kernel and can be solved in a standard manner.

The integral on the right hand side of (1+2) is a constant, albeit

unknown, which we denote by G and so (42.) yields

(1.3) slsj—
r * J z ) aD r 3 ( o + 2 B / T l n z T )
3 3 3 S S

We then integrate (1*3) from z*0 to z=a and. tbe condition for a non-

trivial solution is
ra (z/D ) ( d f /dz)dz
s s
(UU) 1 -
a
o r 3 f0 + 2 B / T s l n z rs )
•s

which determines a.

- 60 - 9:Laminar Flow
 Substituting for T (z) from (36) in the integral then'gives
S

a z ( d f /dz)exp(6B/f (z))dz
(U5) 1 = - 4 5 s s

aD- o 2Ba+fl( z ) e x p ( 2 B / f (z))

s s

If a = 0 we recover

d f ,
-45

n(x)
(U6) 1 " x exp[4B/fs(x)] _
aD
s

which can be integrated by parts, in conjunction with (37)» to

give

e x
(1+7) P [ ? B / f ( a f ] when a - 0

Note that this equation can be obtained directly from (37) by

differentiation with respect to D and so (1*7) corresponds to

O

the critical case of the single (repeated) root solution to (37).

When (37) has two distinct roots, D and D ^ 2 ^ (> D

s s s

it is fairly clear that for the same a and B there exists a unique

solution D+ to (iTf) for which D < D+< D ^ 2

\ although D + does not
o s s s s

of course satisfy (37). It is also apparent that

( 1 ) ( 1 )
(U8) D s > \exp(4B/f s (*))

and
2
(U9) dJ ^ < 4 exp(4B/fJ2^(a))
S u s

We now assume that a is positive, then equation (^5) gives

a z ( d f /dz)exp(6B/f )dz
^4B S o
(50) 1 =
(1) ( 1 } 2 ( 1 )
aD o 2Bo+f e x p( 2 B / f )
s s s
•a z ( d f ^/dz)exp(4B/f ^ )dz
-45 S 8
(1) * (1)2
aD
s

— n ] exp(4B/f (a)) - 1
U J
D
s

- 61 - 9:Laminar Flow
i.e. D ^ <4 exp(4B/f ^ (ct))
S u s

This contradicts (V® and so cannot satisfy O+5) for positive

cr, hence D represents a stable crust. The above and following

s
2
arguments depend upon the quantity a + 2 B / Y I n T being positive,
s s

otherwise the inequalities would be reversed. If this were not so,

equation (1*1*) would admit no real solutions for D since d f / d z < 0

s s

for 3>0. Equation (UU) could conceivably have real solutions for

D if the quantity a + 2 B / T I n T
2 changed sign but then g ( z ) would
O S S
be unbounded when this happened, as indicated by (^9). Next, assume
(2)
that D is a stable solution with a < 0\ then 0+5) gives
O

2
a z(dfJ ^ / d z )e x p ( 6 B / f ^ ) d z
o S
( 5 1 ) ( 2 ) 2 ( 2 )
aD s o 2Ba + f s ezp(2B/f s )

2
a z ( d fJ ^ / d z ) e x p ( 4 B / f ^2^ ) d z
-4 B s s
( 2 ) f (2)2
" aD
s * s

— ( 2 ) ( * ) ) ~ !

D
s

i.e. D J 2 ) > | e x p [ 4 B / f s ( 2 ) (a)~]

(2)
which contradicts (U9) and so D cannot represent a stable solution.
s

- 62 - 9:Laminar Plow
9:5:Asymptotic analysis

For liquid metals the dimensionless pipe length, a, (equal to

1 /r Pe ) is often a small number, as illustrated by the previous

O O C

examples where = 10~^. If so, then the function Q(x) in (30) is a

slowly convergent series and thus is difficult to evaluate numerically.

Figure (7) Q(x) versus x

Fortunately, there is available a solution to the Graetz problem valid

for small z. This is known as the Leveque solution [l] which we can

employ to some considerable advantage since it will also give us a

blockage/non-blockage criterion which will be valid for any value of a •

- 63 - 9:Laminar Flow
 s v 1/3
The Leveque solution is a power series in z which solves (8.4)-(8.6)

for T in terms of Kummer functions [2]. This provides a straightforward

evaluation of the temperature gradient at R=1 and we can write this,

taking advantage of work by Richardson [3] in extending the Leveque

solution,

(52) Q(x) - (6/z)1 /3

/ (1/3) -.6 - 2 2 (x,l/3 _ ^2/3,
35 6
r (1/3)

Figure (7) shows a log-log plot of Q(x), versus x, taken from the

full series (30) and for comparison are also plotted the curves using

the first terms of the series (30) and series (52), indicated by the

curved dashed line and the straight dashed line respectively. It is

clear that for all x > 0,

(53) Q(x) < A x ~ 1 ^ ( where A = 61 ^/r( 1/3) )

thus, if a steady state solution exists, we then have the following:

(54) B/ln & > -A (D Z ) ~ 1 / 3

3 —

i.e.

(55) fi(z) < exp(-B(D Z) 1 / 3 /A)

S — S

and so from the definition of D,

(56) > - faexp[4B(D z)1 ^/Ajdz

D
s — a s
o
which can be integrated:

(57) D > 6\ (1 - h + h2/2)exp(h) - 1 ]/h3

S —

where

(58) h = 4B(D c O ^ / A
S

We may then invert (58) to recover the necessary condition for the

existence of a steady state solution:

(59) Ea l / 3 = A h D s " 1 / 3 / 4 £ h 2 [ 0 - h + h 2 /2)e h - 1]~1 /3

/4r( 1/3)

from (57). Now h is ln(<s"4(a)) and so the function of h on the right

s

- 64 - 9:Laminar Flow
hand side of (59) can be plotted as a function of 5 (a). The graph in
S

figure (8) results.

Figure (8)

This function has a maximum value of 0.232403 at 5 (a) = 0.412721

s
1/3
(by Newton-Raphson iteration). Clearly, if Ba exceeds this value,

then (59) is violated and therefore there are no steady state

solutions, for any a. If it is less than this value then it is not

possible to say for certain that there are a pair of solutions for the

crust, although the closer t

c is to zero, the more likely it is that

there are. The two values for the exit crust (one stable, the other

unstable) would be given approximately by the points of intersection of

1J'T
Ba = constant with the curve in figure (8). We can go some way

towards checking the numerical solutions in 9:1 against this since

a = 10"5 is reasonably close to asymptotic, (if figure (3) is correctly

oriented, we effectively recover half of figure (8)). Table (2)

- 65 - 9:Laminar Flow
represents a series of calculations of B from h using the values of

63(a) given in table (1). The critical value of B, for a=10~^, using

the above asymptotic analysis, is B = 10.787.

Table (2) Figure (9)

5 B (predicted) Limiting Crust Configuration
S(C0
0.9935 0.204
0.9602 1 .228
0.9256 2.252
0.8895 3.275
0.8513 4.304
0.8107 5.334
0.7669 6.366
0.7184 7.405
0.6628 8.451
0.5936 9.515
0.4718 10.6407

Figure (9) shows the limiting critical crust. There are no

steady state solutions for which 5 (ex) is less than 0.412721 or for
s
which D is greater than 17.2324. We can then use these numbers to
s
formulate the blockage criterion, valid for any ,
00
3 a =
p
(60) cr it^ ^
-1II(0.412721) J A exp(-17.232.4X^a)
n=o
which is indicated on figure (7) by the superimposed a,B axes. With

B > Bcr^^., no steady state solutions exist, or in other words

blockage results.

9:6:Experimental validation

Using the experimental data for water, provided in [6.2], we can

- 66 - 9:Laminar Flow
construct figure (10). The solid circles represent experiments where

the nozzle froze up completely and the open circles where it did not.

The superimposed black line represents our numerically derived blockage

criterion. Whilst the values of B are correct, since DesRuisseaux and

Zerkle used the same definition as we do, there is some doubt about the

exact location of the a gradations (and hence the actual location of

the data points may be inaccurate). This is because in the temperature

ranges near the freezing point of water, quite significant changes in

fluid properties occur and consequently there is some difficulty in

translating their vertical axis (which was a pressure drop) into our a,

for which they have no equivalent. However, the numerical evidence is a

reasonably close fit to the experimental data.

Figure (10) - Numerical Blockage Criterion versus experimental data

10:TURBULENT FLOW

The flow of the liquid metal through a nozzle is likely to be

turbulent and consequently it is difficult to analyse the flow and heat

transfer. Turbulence affects these variables through viscosity and

thermal conductivity and so it is apparent from the outset that

solutions will depend upon the Reynolds number of the flow and the

Prandtl number of the fluid. Because of this, it is necessary to

restrict numerical analysis of the problem to certain ranges of

Reynolds and Prandtl numbers although the mathematical analysis should

remain valid for any turbulent flows within the confines of the model.

Most of the equations used in this chapter have been developed in

chapter 5» on fluid flow. The heat transfer problem is essentially the

same as that for laminar flow, save for the analysis of the Graetz

problem, which was covered in chapter 8. The following is essentially

the text of a paper [l] with modifications appropriate to its inclusion

in this thesis, and also with some expansion.

10;1:The specifications

A hot fluid of constant density and specific heat is forced

through a cold pipe of finite length and radius by a pressure drop

maintained over the length of the pipe. The pipe wall is held at a

constant temperature, below the melting point of the fluid, providing

an isothermal boundary for the fluid whilst it is inside the pipe. The

fluid temperature on entry to the pipe is constant and is greater than

the melting temperature of the fluid. The flow is considered to be

turbulent, fully developed and radially symmetric, allowing us to

- 68 - 10:Turbulent flow
dispense with angular components and dependencies in the cylindrical

coordinate system used in the analysis. This last assumption is valid

if buoyancy effects and free convection are neglected. Since the flow

is turbulent, conduction of both momentum and heat in the axial

direction will be ignored as being negligible in comparison with radial

conduction, allowing us to drop the axial double derivatives, an

assumption valid for a large Peclet number.

Initially there is no solidified layer but as time passes, fluid

loses heat to the pipe wall and begins to freeze inwardly. This solid

layer will have constant specific heat, thermal conductivity and a

density equal to that of the fluid. If the Stefan number of the system

is sufficiently small, then the flow of the fluid can be described as

quasi-steady allowing us to neglect the time rate of change of

temperature and momentum in the fluid equations. This effectively means

that the fluid flow and temperature fields will change more quickly

than the position of the solid/liquid interface. The mathematical model

is developed in the following manner. Conservation of mass, in

cylindrical coordinates, is expressed by the equation:

a . * *. a . * *.
(1) —*(r u ) + —*(r w ) = 0
3r 3z

* *
where mean turbulent radial and axial velocities, u and w

respectively, are used. This equation holds in the fluid region defined
# # #
by 0 <
_ r _ < 5 and 0 <
_ z _< 1q. TO satisfy equation (1), we write

expressions for the axial and radial velocities:

, % * * # * * * • , * ? * *
(2) w (r ,z ,t ) = F(t )f(r /$ )/<5 (z ,t )
* * * * * * * » *
(3) u (r ,z ,t ) = (r /$ )w (35 /a z )

- 69 - 10:Turbulent flow
Radial symmetry and the no-slip condition on the solid/liquid interface

together imply that f(R) is an even function of R and that f(1) = 0.

Without loss of generality, we may impose the further condition on f,

namely: #
r<5
# v / " * , * » * *?
(4) 2r f(r /$ )dr /$ = 1
Jn
'O
or alternatively;

(5) f12Rf(R)dR = 1
'o
This ensures that the hulk mean axial velocity, defined by
(c\
(6) ~*r 1
w (z * ,t±*\) =
— f5 2r*v *dr*
6 •'o
is given by the expression
/ \ -* / *s / *2/ *
(7) w = F(t ) / S (z ,t )

Compared with (5«23)» F = M/pir is the monotonically decreasing (with

time) volumetric flow rate. The local Reynolds number, based upon the

local diameter ( i.e. 26 ) is then expressed as:

(8) Re = 25 w /v = 2F(t )/v6

* *
Note that this is a function of z and t . Now many of the terms in the

axial momentum equation, being either of the order of the Stefan number

or inversely proportional to the square of the initial Peclet number,

are neglected. There remains the standard formula relating the pressure

gradient to the bulk axial velocity:

,» * * — *
(9) 3p /3z = - f'w /45

where f' is the friction factor, the well known Blasius formula:

(10) f = 0.3164 R e " 1 ^

which will be used in equation (9). More accurate expressions for

friction laws do exist, valid over larger ranges of Reynolds number,

but these are usually implicit in nature, making analysis difficult.

Numerical analysis can be performed with some confidence on pipe

configurations where the initial Reynolds number lies within the range

- 70 -
10:Turbulent flow
of validity of (10). Above this range results may be suspect, but no

more so than the tacit assumption that the Blasius friction law holds

good for pipes of variable radius. Below this range, the laminar

analysis as presented in chapter 9 may be used. On using the above

definitions of Reynolds number and friction factor in equation (9), we

recover:

(11) = -0.3164(2~9/4) pv 1 / 4 F 7 / 4 (t*) 5 * ~ 1 9 A

3 z
* *
The pressure must satisfy the boundary conditions p (0,t ) = P and
* *
p (l0>"t ) = 0 appropriate to the assumption of a constant pressure drop

maintained over the pipe length. There are thus two boundary conditions

for a first order differential equation and so the solution is:

l 0
A*-1
# 6
9
A ,dx
z
*. * *
(12) p (z ,t ) = P
^ * 19/1;
<
5 - dx
o
*
where P(t ) is also determined by this equation and is given by

(13) F(t*) = (8v)3/7[P/(0.3l64pv lo

* - 1 9 / 4 dz*)] 4/7
•'o

In this way, we see that the velocity field is defined by the

fixed initial pressure drop. It is clear that as the crust thickens

»
(i.e. as 6 decreases) the axial inlet velocity will decrease although

the axial velocity as a whole will increase along the pipe. The

decrease of the inlet velocity is due to the increasing viscous

resistance whilst the narrowing of the effective radius along the pipe

accounts for the increase in speed. This approach to inlet conditions

is fundamental and is the main reason why it is possible to obtain

blockage criteria. Other models have proposed that it is the inlet

axial velocity which, as opposed to the pressure drop, is held

- 71 -
10:Turbulent flow
constant, an untenable position we believe since the axial velocity-

must then increase indefinitely as the pipe blocks.

10;2:Thermodynamics

For a description of the heat transport we use the convective

diffusion equation, ignoring axial conduction and the time derivative

since they are of the order of the inverse square of the initial P^clet

number and the order of the Stefan number respectively. Thus:

(U) ^
3r dz r 3r 3r

*
where mean turbulent quantities are again used and g is the radially

dependent turbulent thermal diffusivity. This diffusivity is typically

the same as the laminar diffusivity near boundaries but is higher near

the centre of the flow, Unfortunately, from the point of view of

*
simplicity, g also depends upon the Prandtl and Reynolds numbers.

* * *
In the solid region, defined by 5 r _
< rQ the velocities u
* *
and w are zero and g is constant, hence the temperature in the crust

is given by:

(15) T = T + (T - T ) ln(r /r )/ln(r /<$ )

w m w o o
in order to satisfy the isothermal boundary conditions on the pipe wall

and the solid/liquid interface. The fluid region needs more detailed

consideration and to this end it is convenient to non-dimensionalize

the equations:

(16) T* = Tffl + [Tq - Tn]T(r,z,t)

(17) 6* = [ro]6(z,t)

(18) r = [r Q ]r

- 72 - 10:Turbulent flow
(19) z =[rPe]z
o o
(20) t* = [r2/2icSt]t
o
(21) g = [ icjg(r/ 6)

as in the laminar case, except that the initial Peclet number is

defined differently (see (25) below). This gives:

r r H !I , !I T _ K r o P e o 1 3 , 3T,
1 ( / ) +
' 2 S 2 - m * r 7 a F ( r g
9? 1
where T is subject to the boundary conditions

(23) T(r,0,t) = 1

(24) T(5,z,t) = 0

Certain conditions on f(r/fi) have already been noted and those on

g(r/<S) are that g(0) > 1 and g( 1) = 1 with dg(x)/dx < 0. Note that Re
0
(the Reynolds number at t = 0) is given by: ^

(25) Reo - 2P(0)/vrQ - [ ^ ^

w e
and the initial Peclet number is given by PeQ = ^ define the

quantity, which again is the ratio of initial to final mass flow rates,

(26) D(t) = F(0)/F(t)

we have that
19/k
(27) D(t) = r
u
k p S~ (Z>t)dxlU/7
a 4 ^
where ,
-r fill ~k/l Kl
3/7
(28) a = l 0 /r Q Pe o = [ 0.316U 1+pvicl 2 ^ E vr^ ^
o o
There is the further interface condition to (14), viz:

(2q) pL
pij = k I * *+ - k I * *-
^ ' 3t* s 3?* >r =<5 f 3r* 'r =6
which in dimensionless form now becomes, using (15);

(30) 2 — = — I
* 3t 5 In 6 3r 'r=6
where the dimensionless freezing parameter is the usual:

(31)
v 3 = k (T - T )/k,(T - T )
sv m w" f 0 m'
To remove the difficulty of having to numerically track a moving

- 73 - 10:Turbulent flow
boundary, we transform the independent variables by writing R = r/<5

(and keeping z as it is) to obtain the 'crust free' version of (22):

(32) f ( H ) f - B(t) f |^(H g (H) | | )

subject to

(33) T(R,0) = 1

(34) T(1,Z) = 0

the time t appearing only parametrically. Equation (30) becomes

(35) 25 — = — — I
3t In 6 9R 'R=l
In practice, we look for a solution to (35) subject to (32), (33) and

(34) rather than vice-versa. The boundary and initial conditions on

(35) are

(36) s(z,0) = 6(0,t) = 1

With these equations, we are now able to take advantage of the work

done on the turbulent Graetz problem, described in 8:2.

10:3:Remarks on c and G

We can write, for use in (35):

an
.2
(37) Q( ZD( t)) = -3T(1,Z)/3R = £ AQexp(-A^zD( t))
_ n ^ O rj p
(38) GAn = n + | + [cln(n +2 /3)+^-]/2Tr(n+| )

(39) A 1 +
n " 6
C - 7 7 ^ 2 < «U»Cav)-H lfc> ]
TTGX1/3
3 rO/3) 2TT(GX )'
n n

where H is defined as

(40) H - (0.3164)Se ^ 4 / 3 2

The quantities c and G which are to be found in the above

asymptotic formulae have been listed for certain discrete values of Re

and Pr in L6.5 J• It is emphasised that the expression for the interface

- 74 - 10:Turbulent flow
heat flux will be calculated as a function of z and t and, since the

flux depends upon the flow, the coefficients A q and eigenvalues X q must

be calculated as functions of the local Reynolds number at each axial

and temporal step in any numerical discretization process. There is no

single function Q(x) as for laminar flow. Similarly, there is no single

set of A and X which will suffice to describe the entire problem

n n
ma thema ti cally.

Figure (1) Figure (2)

Figures (1) and (2) represent the data, marked with open circles,

from j_6.5J over the range of interest. (They in fact provide data for

Pr >> 1). Figure (1) presents values of G as a function of Re (shown on

a log He axis), parametrized by a series of curves for different

Prandtl numbers. We present them in this way since in our model the

fluid properties are constant and hence Pr will be constant. Figure (2)

- 75 - 10:Turbulent flow
presents the data in a similar fashion for c. Once Pr is known, a

polynomial function can be interpolated from these data and henceforth

only their dependence upon Re is required. Unfortunately the data for c

are not particularly well behaved. There are a few points (in the
4 5\
region Re ^ 10 -10 ) which appear to be shared, in a non-systematic

way, by fluids with different Prandtl numbers. Interpolation by

polynomial is therefore somewhat unreliable in this region. We have

already commented upon the erratic nature of c at the end of chapter 8

and the possible reasons for it. We have some idea of the possible

effects that c may have upon the accuracy of the solutions since we

know that they were originally undetermined and were calculated to give

the best match with the lower eigenvalues which can be determined

numerically rather than asymptotically. Consequently, since we will be

calculating all of the 'constants' A^ and XQ which are necessary to the

convergence of (37) from these data, the first two or three terms in

(37) may be less accurate than is desirable. The alternative is to use

the numerically computed values for the early Aq and XQ as

interpolation points for polynomial fits. These are also provided for a

wide range of Re and Pr in [6.5]• The disadvantage here is that there

2 2
would be ten (A - A,, X - X, plus c and G for the later
o 3 0 ?
eigenvalues) polynomial interpolations to perform at each numerical

integration rather than two, requiring the use of an additional 480

data points. It may be worth noting, since it is not very clear from

(38) and (39), that the A q and XQ are increasing functions of both Pr

and Re and neither are they simply functions of Pe.

Another point which is relevant concerns the overall drop in the

Reynolds number as time passes and the crust thickens. It may be the

- 76 - 10:Turbulent flow
case that the Reynolds number will drop below the range of validity of

the polynomial approximants calculated for c and G. It i3 not clear

that the flow will actually become laminar, if originally turbulent, as

He drops below what is usually regarded as the transitional value of

about 2100. Hysteresis effects may play a part, but this is not the

place to discuss this aspect of flow through pipes. In our model, if

this happens we limit G above by its laminar value of 0.25 and we will

limit c below by the value zero. The same remarks apply to H, (40), and

this will be limited below by the laminar value of two.

10:4:Solution examples

The first step is to calculate coefficients for a pair of fourth

order polynomial functions, dependent on Re, for the turbulent flow

parameters c and G. These are calculated from knowledge of the Prandtl

number. Subsequent values of these parameters may be calculated without

referring again to the Prandtl number. At the start of the experiment,

the temperature gradient is zero and so equation (35) can be integrated

exactly, for small t (and z > 0), to give

(41) 2Bt = 1 - 52 + <52ln(<52)

which can be inverted for small t:

(42) 6(z,t) 1 - (Bt)1/2

to give an initial evenly distributed 'thin-ice' crust, except at z = 0

where (36) always holds. The quantity D(t) can then be calculated by

Simpson's Rule. Following this, it is necessary to calculate He at

every axial grid point, from the relation:

(43) Re = 2F(t)/vr i(z,t) = He /6(z,t)D(t)

and hence it is then possible to calculate c and G from which we obtain

- 77 - 10:Turbulent flow
as many of the values of and that are necessary for the

calculation of the interface heat flux. A fourth order Runge-Kutta

process is then used to calculate new values for 5< and the procedure

continues with the next calculation for 3(t). In this respect, a

similar procedure to that used for laminar flow (9:1) is used. This is

done until either a steady state is reached or until the pipe blocks.

Table (1) - Inputs

run 1 run 2 run 3 run 4 unit

pipe length 20 20 20 20 cm
/
pipe radius 2 2 2 2' cm

pressure drop 452.4 452.4 452.4 452.4 kgm"1 s"1

density 7232 7232 7232 7232 kgtif3

I
6
kinematic viscosity 4.63 4.63 4.63 4.63 xlO m2 s"1

solid conductivity 198.7 198.7 198.7 198.7 Wm~l K"'

i J
i
fluid conductivity 231.2 231.2 231.2 j 231.2 Wm"' K"'
!

specific heat 72.4 72.4 72.4 i 72.4 Jkg"' K"'

latent heat 276 276 276 276 kJkg-1

j
! fusion temperature 1800 1800 1800 1800 K

inlet temperature 1900 19'00 1900 1900 K

wall temperature 1795 1790 1785 1780 K

| initial mass flow rate 7.783 7.783 7.783 , 7.783 kgs"1

1
j ( derived =7TpQ(0) ) >

For the purposes of numerical analysis, steady state will be said

to have been reached if the magnitude of the crust velocity at the pipe

exit becomes smaller than some appropriate value (typically about 10~^

in dimensionless units).

- 78 - 10:Turbulent flow
 Blockage can be defined when
Table (2) - Dimensionless parameters
the radial position of the
run 1 run 2 run 3 run 4
crust goes negative (as it
C

B 0.04297 0.08594 0.12891 0.17189

/ may in this model) but in
<X 0.1289 0.1289 0.1289 0.1289
most cases the calculations
Dp 77.6 77.6 77.6 77.6
-c© will stop since near
Pr 0.01049 0.01049 0.01049 0.01049
blockage the model will
i

Gz I30.598 30.598 30.598 30.598

become inaccurate and more
St 0.0262 0.0262 0.0262 0.0262
detailed analysis should be
Re 0 7399 7399 7399 7399
applied ([3.3] - [3-4]).

Instead, we note that the pipe will block and this is signalled either

by D(t) becoming too large or by the fact that the crust velocity at

the exit, having reached a non-zero minimum absolute value, is

beginning to increase again. In order to illustrate the behaviour of

the crust, we will provide four specific examples to demonstrate

behaviour with varying wall temperature. Table (1) lists the inputs for

these examples while Table (2) lists the associated dimensionless

parameters. Figures (3) and (4) show the growth of the crusts at the

pipe exit with time and the rates of crust growth respectively. From

these phase diagrams it is easy to see which will block well before the

same information can be deduced by looking at the crust alone.

The predicted blockage time of run (4), from figure (3), is about

61 seconds although it should be remembered that this is not to be

taken too seriously since the model would no longer be accurate. It

does, however, provide a reasonable minimum estimate. Run (2) seems to

be near critical since it has not succeeded in satisfying the criterion

for steady state in the time available although the exit crust has

- 75 - 10:Turbulent flow
remained at 16.65? blocked for over 10 seconds with a value for

D(t) = 1.288 for considerably longer.

Figure (3) - Exit Crust Positions

Run (1) has stabilised at 1% blockage after ju3t under 35 seconds with

D = 1.108. Another run, not shown in the figures since differentiation

between them would be difficult, was performed with exactly the same

inputs as run (l) but with double the pressure drop. In this case, a

steady state blockage value was 1 . 2 % with D = 1.1006 having taken just

over 45 seconds to reach steady state.

- SO - 10:Turbulent flow
 Figure (4) - Exit Crust Blockage Rates

10:5:Stability revisited

The phase plane plots for turbulent flow are similar to those of

laminar flow. Figure (5) shows this for the results presented in the

previous section. The same remarks apply concerning the two possible

steady-state solutions for each value of 3, provided that B < B .,.

cnt

We examine the existence of stable steady state solutions to

equation (35) by setting the right hand side equal to zero. It will be

- 81 - 10:Turbulent flow
shown that there are no, one or two solutions which satisfy (35) and

(27) dependent upon the values found for o, B, Reo and Pr.
#

Figure (5) - Trajectories

- 82 - 10:Turbulent flow
The analysis is similar to that described in the previous chapter but

since exponents of 4/7 appear in the turbulent model it is not

immediately clear that singularities will not arise.

Figure (6) - See text below

If we assume a steady state solution for 6, denoted by 5 (z) and

3

a corresponding steady state value of D for D(t) then from (35) and

(37) we can write:

0
5 (Z) = eX { B/ e x
(44) S l \ P ' P (- A n D < Z ) }
n=o " *
and we can define a function y:

(«) »<D«> • ° s 7 / 4 - i

By substituting (44) into (45) we can plot, for fixed a, Pr and Re , a

o
graph of y against D for parameter B. An accurate plot can be obtained
9
only by including the dependence of A and \ upon D but the general
n n s
shape of the curves will be unaffected. To obtain p(D ) accurately,

- 83 - 10:Turbulent flow
iteration would be performed with D and 5 (z) and it can be shown
3 3

that, provided that the initial estimate of D is sufficiently close to

unity, the iterative scheme would converge, if convergent, to a steady

state solution.

Figure (6) shows a sketch of y(D ) for variable B and fixed a, Pr

3

and HeQ showing the three different cases that might arise. In

particular we will examine the case appropriate to two steady state

solutions in the following analysis.

On referring to the figure, we wish to show that D g g is a stable

solution and that D is unstable. First we will write:

su
(46) ||— = yH" + f U . t )

for 0 <
. z <
. a and for t > 0 where the function f behaves thus:
A q
(47a) f = 0(z~ ) as z • 0 for some 3^1/3
00
(47b) f = 0(exp(-kz)) as z •
+

(47c) 3f/3z < 0 for all z > 0

and S< is 0( 1) as z »
• 0. Equation (44) is then:

(48) <S2(Z) = exp(-2B/f (z))

5 3
where f (z) is the assumed steady state value of f(z,t). We introduce
3

linear perturbations about the steady state by writing

(49) D(t) = D + d(t) ; fi2(z,t) = 5?(z) + g(z)exp(at)

3 3

where a is a growth rate away from steady state. Substituting into (46)

and (27) gives, after linearisation,

(50) (CT+2B/5|1r>2i|)g(z) = ze"" t d(t)(df s /dz)/D s

(51) d(t) - 1 f g ( z ) s ' ^ (z)dz

o
where in (50) df /dz is an approximation since the small axial
3

variations in the A and A have been neglected as irrelevant to the

- 84 - 10:Turbulent flow
analysis. If we let

d
i ia - -27
(52) l 9(zKT(z)dz
5
*0

then (50) and (51) give

(53) (o+2B/521n2<S2)g(z) = " -J|. D$ ? d.jz(dfs/dz)

and so we obtain:
-27 . , q -7
(54) 9 ( 2 ) 5 ^ = - ||0 S * z(dfs/dz)/(a+2B/52ln252)

which can then be integrated, excluding the trivial solution where

d^ = 0 , to give the condition

(55) _ _ig _ "I l fa z

(dVdZ^dZ
1 =
T5" °S * a ~Z7
0
6 ^ (a+2B/6 2 1n 2 S 2 )

from which a can be, in principle, obtained. Now, on substituting (48)

into (55) we obtain

11
28z
(56) , , "19 D 1 f (dVd2'exP(27B''4f=)^
u
TT s * * i
0
(2gB + f?exp(2B/f ))

Clearly, if j
< = 0 then this reduces to

(57) 1 = DS"T^| f «xp(19B/4fs)zdfs/f|

which has the solution

4 -19 4

(58) D g = (yf)7 5 ^ (a) = (yp) exp(19B/7fs(a))

which is also obtainable directly from (45) and so (58) satisfies the

equation y = dy/dD = 0 which means that the case 0" = 0 represents the

critical case given by D in figure (6).

sc

When equation (45) has two distinct roots, D and D , it can be

ss su'
seen from the figure that there exists a value D such that
sz
D < D <
ss 3Z °su which will satisfy the condition dy/dDg = 0 but will

- 85 - 10:Turbulent flow
not satisfy y = 0. We know that since dy/dD > 0 at D = D :
^ 3 3 33

(59) D
ss » (tt'T exp(198 / 7f ss (o) )
and also that:
(60) < i 4
°su ( ^ e x p t ^ B ^ t c ) )
A A
where f (a) and f (a) are the corresponding appropriate values of
33 3U

f (a). If we assume that a is positive then (56) gives

3

^ -19B fa z
( d ^ ss / dz )exp(275/4f ss )dz
7aD
ss 2Bo+f2 exp(2B/f ) ^

-1 SB •
a
z
(^ ss /dz)exp(19B/4f ss )dz/f2 s

' y D s ; 7/4 exp(19B/4f ss (a)) - 4

since df /dz < 0, hence

ss
A 4/7
(62) Bss < % > exp(19B/7f ss (o))

which contradicts (59) and so <r must be negative if D is to satisfy

33

(56). This means that D represents a stable solution. If a is

S3

negative then the condition for D to satisfy (56) becomes, by exactly

3 VI

the same argument;

(63) 1 > y D s " u ?/4 exp(19B/4fsu(a)) - £

A2 A
where the inequality has been reversed since 2B<r + f gu exp(2B/f gu ) is
* 2 A
now less than fguexp(2B/fgu) due to the sign of a. Equation (63)

contradicts (60) and so D is an unstable solution.

su

- 86 - 10:Turbulent flow
11: MUSHY REGIONS AMD VARIABLE VISCOSITY

Most industrial fluids, especially those in metals processing,

are not in a pure form, nor is it necessarily desirable that they should

"be. For example, it is well known that there are many different types

of steel with varying quantities of trace elements such as chromium,

vanadium, titanium etc., quite apart from the usual carbon content of

such metals. These are introduced to give the steel certain properties

when solid but they also modify the behaviour of the metal as a fluid.

Due to the presence of different materials in the liquid metal, two main

effects are evident. Firstly, the fusion point of the fluid is no

longer at a fixed single temperature". ' " ' since

'' the components have different" melting-points We must

therefore consider the melting or solidification of such fluids to occur

over a range of temperatures bounded by the solidus temperature, T ,

s
below which we have metal in the solid state and the liquidus temperature,

T^, above which any metal is completely molten. Secondly, as individual

components solidify separately, at various temperatures within this range,

we can expect to find solid particles in the fluid. This is due to

straightforward freezing and also to some secondary effects because of

premature solidification of other components around those primary nuclei.

The latent heat of freezing, now dependent upon the components and their

relative distribution, is thus liberated over this temperature range

(Tg, T^) rather than at the single melting point (Tm) in pure fluids.

- 87 - 11: Mushy Regions

11:1: Modelling

The nature of the problem is inherently complex due to the

presence of a mushy region - part liquid, part solid - in the fluid.

Indeed, if the flow is turbulent the entire fluid may be mushy, usually

a desirable condition from the point of view of relatively rapid

solidification and evenly distributed material in castings (1.3). The

mush itself will not behave as a Newtonian fluid as viscosity will in

general be dependent upon the shear rate, solid fraction and temperature.

Some work has been done on shear-rate dependent viscosity and its

effects on the Graetz problem [9.3J, including contributions from viscous

dissipation. The modelling of mushy flow with solidification in a

cylinder is not a problem which has been discussed in the literature and

is a relatively new field of study.

The first step we take in describing blockage is based on the

experience gained earlier in this thesis. Instead of considering the time

development of solidified crusts, we will attack the problem directly by

attempting to find steady state solutions. We already have strong evidence

to support the claim that if no such solutions can be found for a given

physical arrangement then blockage is the outcome of the experiment.

The second step is to largely ignore the non-Newtonian nature of the

mush. We will assume that the viscosity of the mush varies not with shear

rate or solid fraction but only with temperature. There is some physical

justification for this since the solid fraction, and to some extent the

shear rate, will be temperature dependent and their effects could be

- 88 - 11: Mushy Regions

directly included in the viscosity. This approach leaves out many

important factors but as a first attempt it will serve its purpose.

Thirdly, we will assume that the fluid flow is laminar. Again,

the motivation behind this is in order to provide the basis for a first

attempt. It is also possible to compare results obtained from this

model directly with those for laminar flow.

Figure (l) - Diagram of system and notation

1

The mathematical model is, in most respects, identical with the

set up which was assumed for laminar flow in chapter 9. Isothermal

boundary conditions apply to the pipe inlet at z=0 and the pipe wall at

r=rQ where temperatures are T q and T^ respectively. As shown in figure

(1) three regions will be considered.

The first is the solid region, characterized by the fluid properties

,ss s' 13
' w
kich 1S .
foun<i m
.
the region defined by T^. _
< T
* _< Tg, or
*i
alternatively by the radial coordinate 'r lying in the range
* * * .
> r >
_ 6 Cz )• In this region, the temperature of the solid is given

by

-
- 89 11: Mushy Regions
(1) T = T + (T - T ) ln(r
v /r )/ln(r /5 )
w s v o o

as before. This is now the steady state temperature distribution and

the Stefan number no longer plays a part in the analysis. (Subscripts

•s', as used in previous discussions of steady state solutions will be

i
omitted throughout the remainder of this thesis).

*
The second is the mushy region, where T < T <
_ Tn , denoted by the
s X
*/ *N * */
radial coordinate in the range 5< (z ;_>r j
> A (z ) and the third the liquid
* *, * *
region, T^ _
< T <_ T q or A (z )_>r _>0.

For analytical purposes, it is convenient to refer these last

two as the fluid region, where the fluid properties are k^, and p.

Note that the density of the fluid is identical to the density of the solid.

This assumption has been made throughout the thesis since a change in

density would produce an additional radially convective term in the heat

equation. We might even allow the density to change between phases if

the density difference across phase boundaries were small enough to allow

this term to be ignored in comparison with the main radially convective

term.

The velocity field in the fluid region is given by

f~\ 3 / * *\ 3 , * *>
(2) (r u ) + — * (r w ) = 0

(3) = 1 (luW r* = o
dz r* 3r dP1 5
3?* ~ U

subject to

(U) u = w = 0 a t r = < 5 (no slip)

u* 3w* *
= gpe = 0 at r = 0 (symmetry)

- 90 - 11: Mushy Regions

 # « w
(6) p w) • = P ; p (1Q) = 0 (pressure drop maintained)

The only difference between -his formulation and that for laminar

flow is that the viscosity is temperature dependent, with the following

effects; we integrate (3) once in conjunction with (5) to obtain

* * *
(7) |2, = £_ IE,
3r 2u dz

# * * *
However since u is a function of r (through T (r ,z )) a second

integration, using (U), gives

*
* * * r<5
/o\ */ * 1 dp* f s ds
(8) w (r ,z ) = - J * —
2 dz" Jr'~ y(T(s
(mt *,z *))
n

A further radial integration, using ( 2 ) gives the mass flux as

r.*
* r
<5 . rfi * *
(9) M = * * * dra f * s ds
2 n r pw dr = - irp ; r dr
J

where 3M/3z = 0. Thus

6 * * -1
dt)
• - M_ r s ds , * i-
(10) dr J
dz TTp ^

axial integration of (10) then gives, using (6), the steady state mass

flux in terms of the applied pressure drop:

(ii) M .
.T *
o dz
r * ,* * #
5< * 6 s ds , *
r dr
o J
o r* U

Thus we obtain the complicated expression:# #

f5 2s ds
(12) w = , *
.
X #
o r ffi,4s *3^ds * 6 . *3 *
_ 1
* Us <is
dz *

6 * *
2s ds
J
2M
TTp 6 , *3, *
4s ds
- 91 - 11: Mushy Regions
for the axial velocity. This has been simplified a little since we

have integrated, by parts, the rightmost double integral in (9)- Note

that if the viscosity is constant, we obtain the laminar result;

? M
*2
tT3\
(13) w* = • (1 - — )
*2 *2
TTp 0 0

where

(U) „ =
°«"-V

At this stage, we transform the independent variables to remove the

. . * * *
undetermined solid/fluid interface boundary 6 by R = r /<5 and recover

2SdS/y(S,z)
m o R
(15) w* = • 2M
irp6 * 2
Us dS/y(S,z)

(l6) ii. iisS/Sl

0
5*"11 [ i432dS/u!s,z)] dz*

The continuity equation becomes

k • R
f £ k CB
"*> - H
fr* < s V
>

_ 1
_ 36 r 3ij)* _ j- 1 32ijj
6*2 3z* L 3 R R
3R2 J " 5* 3R3z™

where

(18) * = S6*2(z*)w*(S,z*)dS

thus
,.n, *2_ * _ 36 *2 * * aib"
(19) 6 Ru = R T—* 6 Rw - 6 A
dz dz

11:2: Fluid temperature

The heat equation for the fluid, in transformed coordinates is:

- 92 - 11: Mushy Regions

(20) R<S*V + R<5* [ u* - R ^ v*
w I ^ * = K — (RR
3z 3z*
3z " i ap
J 3R " 3R <- ~
3R )

for 0 < R < 1 and 0 < z _

< 1q where < = kf/pCf. In view of (18) and

(19), this reduces to

* * * *
(21) !i 3T, _ ^ II = K
<
1 _ ( R 3TP ,
3R 3z^ 3z 3R an 4" :3
3R >DR )

2
In constant viscosity laminar flow, 6 w is a function of R only and so
*/ *
3^ /3z is identically zero (demonstrating relationship (5.11)). However,

with temperature dependent viscosity, this is not the case and so the heat

equation is not expressible in a form suitable for analysis by the Graetz

problem methods outlined in chapter 8. Superposition cannot be used since

/ V .
equation (21) is non-linear in T* as jt* i
. .
s functionally dependent upon T*
*
through the viscosity integrals, (15;, needed to determine w .

Equation (21) is subject to the isothermal boundary conditions:

# £
(22) T = T at z = 0
o

(23) T = Ts at R =1

and the steady state heat balance at the fluid/solid interface is, using

(1),

3T , 3T
(2b) k K * *_
s 3p * *+ f aP r =6
r =<5

1•0 «

(25) k
s^ T s ~ /ro> = k
f 3R R=1

11:3: Fluid velocity

To solve (.21) in order to obtain 3T /3R, information about JI is

needed. We propose the following form for the viscosity, in terms of

- 9^? - 11: Mushy Regions
constant density, p, and constant kinematic viscosity, v, being fluid

properties measured at temperature T q (the initial global temperature

of the fluid before solidification commences):

*
=
(26) u pv for Tq >
_ T >
_ T^

T - T #
S
= PV T _ T for Ix > T > Ts
s

Assuming that the initial conditions of the system were the same

as the initial conditions of the previously discussed laminar and turbulent

models, we can recover the initial mass flux,

(27) M = irPr^/8vl
o 0 0

This is the mass flow rate of the fluid, of uniform temperature

T q and constant viscosity pv at t=0 before solidification starts. If a

steady state is reached, then the ratio of initial to final mass flow

rate is, as before, denoted by D and is given by

° f /x*^ 1 3 -1 *
(28) D = M /M = 7- (r /<S ) UpvR dfi dz
o 1 o

Previously, D was regarded as an unknown dimensionless number which reached

steady value after a certain period had elapsed and gave the steady state

mass flux. Alternatively, if the pipe blocked, D became unbounded and

the final mass flux dropped to zero. We have already seen, in section

9:5, that if D exceeds a value of about 17 then blockage is the predicted

outcome. Due to the presence of a viscosity dependent term in ( 2 8 ) , we do

not expect to recover the same critical value of D but we have a reasonable

idea what it will be. In the modelling of steady state mushy flows, we

will specify a value of D, greater than unity, as an input to the numerical

solution. (Note that y will in general be greater than pv and so D , >

mushy
^laminar * ^^ctively the same as specifying the final desired

- 9b - 11: Mushy Regions

output of fluid from the system and so tells the operator what the initial

mass flux should be to achieve this.

*
Thus at the pipe entrance (z =0) where the temperature is uniformly

at T we have
o

a R r W
(29) ff<R>0) o *(R»°) = £ CPr^/UpvlQ) R(l-R2)

(30) /(R,0) = ^ CFr^Apvlo) [l - (l-R2)2 ]

$
from (15)» (l8), (27) and (28) where y E pv and 6 (0) = r . Unfortunately

the radial velocity, u* (or alternatively, dip * /8z * \), is not known in
* *
advance of the solution since knowledge of 3T /3z is not as yet available.
* * * *
However, we do know that dip /3z = 0 at R=1 and at R=0 since (0,z )=0

and (l,z ) = M/2irp from (l8).

11:k: Non-dimensionalization

We proceed to non-dimensionalize the system using (9.13) (page U9)

where

(31) Pe = Pr3/Upv<l
o 0 0

is the Peclet number of the system before the freezing process starts, and

is the same definition as was used in laminar flow. (Additionally, instead

of T , we are using Tg to represent T=0). The dimensionless viscosity (we

have not used the asterisk superscript to denote the dimensional form of

y for consistency with previous chapters) is

(32) y = 1 for 1 > I > i

= T/T for x >T > 0

where the dimensionless liquidus temperature, x, is defined as

_
- 95 11: Mushy Regions
(33) x = (T. - T )/(T - T )
X s o s

and has a value between zero and unity. The dimensionless pipe length

is the same as the laminar definition:

(3U) a = Upvi<l2/Pr^
o o

and the freezing parameter is given by

(35) B = k s (T s - Tw)/kf(To - T s )

We have a three parameter system, viz a,B,x, whereas before we had

only two in the case of laminar flow of a pure fluid. We scale the

stream function:

(36) = [PrVUpvl 1 p\
0 o"

such that streamlines R=0 and R=1 are given by ip=0 and ip=l/kD.

Thus (21) becomes

(37) 12. 1 1 = r ill + 21 M 3<Ky

3R 3z 3R2 3R ^ 3z

for 0 _
< R _
< 1 and 0 _
< z <
_ tc subject to

(38) T(R,0) = 1 ; T(l,z) = 0

where, at z=0,

_ 1
(39) H=D R(1-R2)

(Uo) ^ r D ' 1
[l-(l-R2)2]

and in general; rR 1 ' '

2S dS
Us dS
(Ul) *j/(R,z) =
rl
3
US dS
o V I"1 Us(S2-R2)dS
u
JR
1
- F E E -
US3dS
o v
- 96 - 11: Mushy Regions
where
fa a
Us 3 ds -i"1 dz
*-h
02) D - i
a L U J

The solid fluid interface is then given by

B 3T
(U3)
ln<5(z) 3R R=1

11:5: Numerical treatment

Since (37) is not in a form suitable for treatment by Graetz

problem methods, we must solve it for T. We use the method of lines

(figure (2) is a diagram of the discretization) with Ur radial divisions

and Nz axial divisions. The step lengths are Ar = h = 1/Nr and Az = a/Nz.

We use Nr functions of z, T (z) ... T.T ,(Z), to enable us to calculate

o Nr-1 '
the temperature distribution. (The temperature at R=1 is given by T.T (Z)
Nr
and is identically zero).

There are two special cases. At i=0 (the cylinder axis), we have

that 3i|;/3z is zero since i|>(0,z) is a streamline. However, 3i|>/3R and

3T/3R are also zero (by symmetry) and so

( W = (r
° If r S lil+ = 2
M/fif at r = o

where we have used L'Hopitals rule (the functions are analytic at R=0).

The other special case is at R=l, or at i=Nr, where T^(Z)EO. Thus,

at R=l,

R +
("5' fsf § = °

since both 3tf>/3R and 3i^/3z are zero here (the no-slip condition). This

can be used to advantage, since the discrete form used to calculate the

- 97 ~ 11: Mushy Regions

dimertsionless heat flux at the solid fluid interface is:

(«» I 2h [ T
Nr+l^ ~ T
Nr-l^ ^
R=1

The discretization of (U-5) is:

2 T + T +
(1*7) (Nr«h) p L^r+i^) ~ Nr^ Nr-l^ ^ 2h ^ T N r + l ^ T
Nr-l^Z
= 0
which, since 3L. (z)=0 and Nr*h=l, gives us T w in terms of T„ , . This
» Nr Nr+1 Nr-1
is substituted into (U6) and gives
2
3T 2Nr m
(U8) (z)
9R
R=1 2Nr+l Nr-1

to second order accuracy.

.i+2

MUSHY REGION
•i+1
IR
filil
h
z
-i=l(z) LIQUID REGION
.i-1
Az

-•i-2

Ji"3

To calculate each T^ we also need Nr-1 functions I|K(Z), i=l,2,3, •. .Nr-1,

since ;
J
.
' (z)=0 and z)Hl/UD. At a given z, each of these IJK must be

calculated from a discretization of (Ul). The numerical solution procedure

is as follows. We will denote the discretized forms of dij^/dz by F^.

Initialize

1 Choose a > 0, 0 < T < 1, Hr even, Nz, D > 1, z = 0

2 Set T. = 1 for i = 0 to Nr - 1 and set T.t = 0

i Nr
3 Set ijK = (2i2h.2 - iSi^)AD for i = 0 to Nr

k Set F. = 0 for i = 0 to Nr.

l

-ps- il: Mushy Regions

Axial Step
A

1: Calculate T^ (at z + Az) for i = 0 to Nr - 1 by Runge-

Kutta methods using

dT.
( i tT 2 T
s r - 1 » e? i+i - i + Ti-i> • k lT
i,i - T i-i )(1 * V V

t 2h ^i+l " *i—l' 3

if i ^ 0, or if i = 0 use

dT
5 R = 2 T ,
R - "o1

since, by symmetry, T_^ = T^ and >

|
i ^ =>
|
i .

2: Calculate the grid position, denoted by i + p, where the

A
estimated temperature T is equal to T.

3: Calculate local viscosities on the basis of (32), for

each i. (In fact it is easier to use inverse viscosities

since l/u„T = 0 ) .
Nr

b: Perform a numerical integration by Simpsons rule to obtain

a number J
( where
f1 3
G= Usj(l/y)ds
o

5: For each i, (i.e. each R), perform a numerical integration

of
1
*+S(S2 - R2)(l/y)dS
R
A

ana hence calculate estimates from (bl) at z + dz.

A
6: Calculate estimates, for each i, F^ from
A A

F. = ('];.
r - Y .)/Az
1 l 1

- 99 - 11: Mushy Regions

Main LOOP

1: Repeat Axial Step until each has numerically settled

down to some stable value.

2: Save G and p.
A

3: Save new temperature T^ and stream functions from T^

A
and i»
..
l
U: Increment z pointer.

Program

1: Repeat main loop until z = a

*Up to this point we have made no reference to the function 6. The

*above is a once and for all calculation which is made given a,t and
*D. It produces the temperature distribution in the cylinder, the
*interface temperature gradient from (U8), the axially dependent
*function G from Axial Step k and also the position of the T = T
*isotherm in the transformed space R,z. To locate 6, we continue:

2: Choose an initial value for B

3: Calculate 6(z) from (.1+3)

U: Calculate the integral on the right hand side of (1+2)

and if this integral is close enough to D then the solution

is found. If not, then use Newton-Raphson on B as follows:

a
j
09) f(B) = D - i dz
exp(.-l+B/(3T/3R)) ^
o

(50) f'(B) = r exp(-l+B/(3T/3R))

where

(51) G(z) = US3Cl/y)dS

and it is understood that the 3T/3R in (1+9) and (50) is at R = 1. The

new value of B is calculated from

B = B ,, - f(B)/f'(B)
new old

as usual in Newton-Raphson iteration.

- 100 - 11: Mushy Regions

This completes the numerical procedure. It is possible to obtain from

this both the solid/mush interface, <5, and the mush/liquid interface, A.

We may also calculate the proportions of mush and solid inside the cylinder

from numerical integrations, thus

rl , *2 »2v . *, 2.
(52) tt(.<5 -A J dz /irr 1
mush o o

o / 2 *2, . *, 2.
C53) V, irCr -6 ) dz /Trr 1
solid o o o

11:6: Results

-U
We shall present some results for a = 10 . Figure 3 shows a

crust/mush profile for t = 0.5, D = 9 for which a value of B was found

to be 3.2096, and figure (U) shows a similar profile for D = 25 (B = 3.2066).

To obtain these results, a series of runs was performed with fixed a and

T (in this case a = 10 ^ and x = 0.5) varying D from 1.5 to 30. Table 1

shows the exit crust and exit mush locations, the value of B and the

volume fractions of mush and'solid in the cylinder.

Figure 3 - Sub Critical B

- 101 - 11: Mushy Regions

 Figure U - Super Critical B

The table shows that B reaches a maximum value of about '3.26 for

these particular values of a and. T and the mass flux has been reduced

by a factor of about Ik. Figure 5 shows plots of the solid crust

1/3
position at the pipe exit (i.e. 5(a)) versus the value of Ba (where
—U
a = 10 ) for three different values of the dimensionless liquidus

temperature, x. The graph is drawn with 5(a) as the horizontal axis in

order to allow comparison with figure (9.8). Also shown on the figure

are three regions relating to the curve for x = 0.9. The largest region
1/3
is that which is above the horizontal dashed line Ba = 0.118, in which

are to be found no steady state solutions for any value of B. However,

as can be seen, if the liquidus of the fluid is nearer to the solidus

temperature then steady state solutions are possible with larger values

of B. (We would rather effect changes in x by changing T^ since the

alternative method of lowering the value of x would be to raise the entry

temperature, Tq, which has the side effect of also decreasing the value of

B, all other parameters being constant). The second and third regions

are bounded by the aforementioned line and a vertical line at 5(a) = O.U-75.

To the right, we have stable steady state and to the left, unstable states

where D > lU.

- 102 - 11: Mushy Regions

 -1+
Table 1 - a = 10 t = 0.5

B 6(a) A(a) V V
solid mush

(1.0) (0.0) (1.0) (1.0) (0.0) (0.0)

1.5 0.82205 0.91297 0.8639U 0.17028 0.07U3U
2.0 1.1+9222 0.83727 0.78895 0.27896 0.06738
2.5 1.95296 0.78130 0.733U3 0.3521+8 0.06277
3.0 2.28335 0.73567 0.68720 0.1+0585 0.05957
1+.0 2.68710 0.66803 0.6181+9 0.1+7701+ 0.05595
5.0 2.90766 0.62211 0.57215 0.52310 0.05390
6.0 3.03835 O.5885U 0.53829 0.55610 0.05253
7.0 3.12070 0.56235 0.51182 0.5811+3 0.05151
8.0 3.17^35 0.5^097 0.1+9017 0.60178 0.05071
9.0 3.20962 0.52299 0.1+7192 0.61863 0 . 0 5 0 0 6 (Fig 3)
10.0 3.23271 0.50755 0.1+5623 0.63291+ 0.01+950
12.0 3.25631 0.1+8227 0.1+301+9 0.65629 0.01+860
ll+.O 3.2626U 0.1+6230 0.1+1011 0.671+83 0 . 0 1 + 7 9 0 (B ..?)
0.1+1+581 crit
16.0 3.26065 0.39325 0.69005 0.01+729
17.0 3.2573U 0.1+3858 0.38586 0.69673 0.01+702
18.0 3.25377 0.1+3191 0.37903 0.70293 0.01+676
20.0 3.21+322 0.1+2017 0.36691 0.71381 0.01+629
25.0 3.20659 0.39519 0.31+099 0.73527 O.OI+5I+O (Fig 1+)
30.0 3.16186 0.37529 0.32011 0.75123 0.01+1+82

It is just possible, therefore, by using the technique shown in

11:5) to generate the two steady state solutions for given a, B and T.

We say just possible, since if the value of D chosen to generate crusts

is too high (greater than about 30) then some numerical instabilities in

the temperature may develop. All of the data used to plot the curves

shown in figure (5) are 'reasonable' to the extent that no oscillations

in 6(z) have occurred. However, close inspection of figure (1+) will

show that at z = 3Az (where in this case, Az = a/32, there is a small

oscillation in 36/3z.

11:7: Conclusions and directions

The modelling of mushy regions presented in this chapter must be

regarded primarily as an attempt to demonstrate that the same arguments

- 103 - 11: Mushy Regions
used in the earlier laminar and turbulent models (chapters 9 and 10)

will apply to this more complex situation. We have shown that there are

again two possible steady state crust and mush, following the crust)

configurations for a given fluid (x) and pipe length (a), to use only

two of the physical analogues of the dimensionless system parameters.

This model can be categorized as an isothermal approach since to describe

the fluid (liquid plus mush) we merely draw a line at a certain temperature,
*
T = T^ or T = x. If it were not for the fact that we allow the viscosity

to vary with temperature in the mushy region there would be no difference

at all between this model and the laminar model. There has been no

discussion of or provision for • " heat and mass transport at the

liquid/mush interface. A discussion of mass transport would be especially

* * *
valuable since it is clear from the definition of p\ that the line r = A
*
(or T = T^) is not a streamline and so clearly some conversion process

must be occurring there.

There is still much that can be done with this laminar steady state

model. The formulation presented in this chapter could quite easily cope

with variable specific heat through a. function H, which is the enthalpy,

and another function, U, called the Kirchoff function:

T
. -1
(5*0 H = C(0)d© (units J kg )
T
s
T
. -1
(55) U = k(9)d0 (units W a
. )
T
s

We have used the solidus temperature as a reference point but there is

no reason why other temperatures cannot be used. A suitable energy

equation would then be, in physical (r,z) space

3H ^ 3H d2U x+ 1 3U
(56) 57 + (=» = IT? ? sF

- 10U - 11: Mushy Regions

(where we omit asterisk superscripts).

Also, other forms for the viscosity might be used. Temperature

dependent viscosity such as the Andrade equation

(57) u = A exp (B/T)

or power laws, usable over small (say 30K) temperature ranges:

(58) y = T~s

might be tried (see pp <?U ff. in jj?.5j). There are also shear-rate

dependent viscosity laws;

3w 1 _ 1
(59) v = m for m > 1 (see, for example, [9.3])
3r

(where m=l corresponds to a lewtonian fluid), which have the advantage

of being independent of z in our formulation in the (R,z) coordinate

space.
1/-3
Figure (5) - Ba versus 6(a)
(effectively, wall temperature versus exit crust location)

0.20 -

0.15

0.10

0.05

- 105 " 11: Mushy Regions

 Whichever model is used the essential point is that "blockage

criteria can be developed, as outlined in chapters 9 and 10, if constant

pressure drop is assumed. This can sometimes be done analytically, as

in chapter 9» but can always be done numerically for an individual

physical set up. Reference [l-3] gives a description of software

developed by this author specifically for the purpose of predicting

final steady state configurations for both laminar and turbulent flow.

- 106 - 11: Mushy Regions

Mathematical Models of Solidification Forced Flow in Circular Pipes

REFERENCES

Chapter 1

[ l] D.A.PEEL & A.E.PENGELLY, Heat transfer, solidification and

metallurgical structure in xhe continuous casting of aluminium in
Mathematical Models in Metallurgical Process Development,186-197,
THE IRON AND STEEL INSTITUTE (1970)

[ 2] P.SAMPSON & R.D.GIBSON, A mathematical model of nozzle blockage by

freezing, INT J HEAT AND MASS TRANSFER 24,231-241 (1981)

_| 3j P.SAMPSON & R.D.GIBSON, Solidification of a liquid metal flowing

through a circular pipe - a prediction of nozzle blockage, ADV IN
ENG SOFTWARE 3,17-25 (1981)

[ 4 ] P.SAMPSON, A mathematical model of nozzle blockage in Numerical

Methods in Thermal Problems 2,130-140 PINERIDGE PRESST1981)

Chapter 2

[ 1 ] H.T.ANGUS, Cast Iron: Physical and Engineering Properties, 2Ed,

BUTTERWORTHS (1976)

Chapter 3

[ 1] J.R.OCKENDON & W.R.HODGKINS (eds), Moving Boundary Value Problems

in Heat Flow and Diffusion, CUP (1975)

[ 2 ] H.S.CARSLAW & J.C.JAEGER, Conduction of Heat in Solids, 2Ed OUP

(1959)

[ 3 ] D.S.RILEY, F.T.SMITH & G.POCTS, The inward solidification of

spheres and circular cylinders, INT J HEAT & MASS TRANSFER
17,1507-1516 (1974)

[ 4 ] K.STEWARTSON & R.T.WAECHTER, On Stefan's problem for spheres, PROC

ROY SOC 348(A),415-426 (1976)

[ 5] C-L HUANG & Y-P SHIH, A perturbation method for spherical and
cylindrical solidification, CHEM ENG SCI 30,897-906 (1975)

Chapter 5

[ 1] W.M.KAYS, 'Heat Transfer: Turbulent flow inside smooth tubes' in

Convective Heat and Mass Transfer, Chapter 9, MCGRAW-HILL TMH
EDITION (1975)

[ 2] G.HAGEN, Uber die Bewegung des Wassers in Engen Zylindrischen

Rohren, POGG.ANN. 46,423ff (1839)

- R.1 - References
Mathematical Models of Solidification Forced Flow in Circular Pipes

3] J.POISEUILLE, Rfch^rches experimentelles sur le mouvement des

liquides dans les tubes de tres petits diambtres, COMPTE RENDUS
11,961ff (1840)

4] H.BLASIUS, Das Ahnlichkeitsgesetz bei Reibungsvorgangen in

Flussigkeiten, FORSCH.ARB.INGWES. 131 (1913)

5j G.W.GOVIER & K.AZIZ, The Flow of Complex Mixtures in Pipes, VAN

NOSTRAND REINHOLD (1972)

Chapter 6

[ l] R.D.ZERKLE & J.E.SUNDERLAND, The effect of liquid solidification

in a tube upon laminar-flow heat transfer and pressure drop, TRANS
ASME J HEAT TRANSFER 183-190 (May 1968)

[ 2] N.DesRUISSEAUX & R.D.ZERKLE, Freezing of hydraulic systems, CAN J

CHEM ENG 47,233-237 (1969)

[ 3] K.STEPHAN, Influence of heat transfer on melting and

solidification in forced flow, INT J HEAT & MASS TRANSFER
12,199-214 (1969)

[ 4] F.B.CHEUNG & L.BAKER, Transient freezing of liquids in tube flow,

NUC SCI & ENG 60,1-9 (1976)

[ 5] E.P.MARTINEZ & R.T.BEAUBOUEF, Transient freezing in laminar tube

flow, CAN J CHEM ENG 50,445-449 (1972)

[ 6] G.J.HWANG & J-P SHEU, Liquid solidification in combined

hydrodynamic and thermal entrance region of a circular tube, CAN J
CHEM ENG 54,66-71 (1976)

L 7] M.N.OZI^IK & J.C.MULLIGAN, Transient freezing of liquids in forced

flow inside circular tubes, TRANS ASME J HEAT TRANSFER 385-389
(Aug 1969)

_| 8j J.C.MULLIGAN & D.D.JONES, Experiments on heat transfer and

pressure drop in a horizontal tube with internal solidification,
INT J HEAT & MASS TRANSFER 19,213-219 (1976)

L 9 } A.A.SHIBANI & M.N.OZigiK,A solution of freezing of liquids of low

prandtl number in turbulent flow between parallel plates, TRANS
ASME J HEAT TRANSFER 20-24 (Feb 1977)

L10j A.A.SHIBANI & M.N.OZI^IK, Freezing of liquids in turbulent flow

inside tubes, CAN J CHEM ENGNG 55,672-677 (1977)

[11] S.B.THOMASON, J.C.MULLIGAN & J.EVERHART, The effect of internal

solidification on turbulent flow heat transfer and pressure drop
in a horizontal tube, TRAITS ASME J HEAT TRANSFER 100,387-394
(1978)

I12] C.CHO & M.N.OZI^IK, Transient freezing of liquids in turbulent

flow inside tubes, TRANS ASME J HEAT TRANSFER 101,465-468 (1979)

- R.2 - References
 Mathematical Models of Solidification Forced Flow in Circular Pipes

[13] M.EPSTEIN, A.YIM & F.B.CHEUNG, Freezing - controlled penetration

of a saturated liquid into a cold tube, TRANS ASME J HEAT TRANSFER
99,233-238 (1977)

[ 14 J M.EPSTEIN & G.M.HAUSER, Freezing of an advancing tube flow, TRANS

ASME J HEAT TRANSFER 99,687-689 (1977)

[15J J.SZEKELY & S.T.DINOVO, Thermal criteria for tundish nozzle or

taphole blockage, MET TRANS 5,747-754 (1974)

Chapter 8

[ 1] L.GRAETZ, Uber die Warmeleitungsfahigkeit von Flussigkeiten,

ANN.PHYS.CHEM. 25,337-357 (1885)

[ 2 ] J.R.SELLARS, M.TRIBUS & J.S.KLEIN, Heat transfer to laminar flow

in a round tube or flat conduit - the Graetz problem extended,
TRANS ASME 78C,441-481 (1956)

L 3j C.A.SLEICHER, R.H.NOTTER & M.D.CRIPPEN, A solution to the

turbulent Graetz problem by matched asymptotic expansions - I The
case of uniform wall temperature, CHEM ENG SCI 25,845-857 (1970)

[ 4 ] R.H.NOTTER & C.A.SLEICHER, A solution to the turbulent Graetz

problem by matched asymptotic expansions - II The case of uniform
wall heat flux, CHEM ENG SCI 26,559-565 (1971)

[ 5] R.H.NOTTER & C.A.SLEICHER, A solution to the turbulent Graetz

problem - III Fully developed and entry region heat transfer
rates, CHEM ENG SCI 27,2073-2093 (1972)

Chapter 9

L 1j M.A.LEVEQUE, Les lois de la transmission de la Chaleur par

Convection, ANN.MINES 13 (1928)

L 2j M.ABRAMOWITZ & I.A.STEGUN (Eds), Handbook of Mathematical

Functions, DOVER

I 3 ] S.M.RICHARDSON, Extended Leveque solutions for flows of power law

fluids in pipes and channels, INT J HEAT & MASS TRANSFER
22,1417-1423 (1979)

Chapter 10

[ 1] P.SAMPSON & R.D.GIBSON, A mathematical model of nozzle blockage by

freezing, part II - turbulent flow, to be published in INT J HEAT
AND MASS TRANSFER 25, (1982)

- R.3 - References

View publication stats