Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Finite Element Simulation of Water Cooling Process of Steel Strips On Runout Table

Download as pdf or txt
Download as pdf or txt
You are on page 1of 240

Finite Element Simulation of Water Cooling Process

of Steel Strips on Runout Table

By

Fuchang X u
B . S c , University of Science and Technology Beijing, China, 1984
M . S c , University of Science and Technology Beijing, China, 1987

A THESIS IN PARTIAL F U L F I L L M E N T OF
THE REQUIREMENT FOR THE D E G R E E OF

DOCTOR OF PHILOSOPHY

in

THE F A C U L T Y OF G R A D U A T E STUDIES
(Mechanical Engineering)

THE UNIVERSITY OF BRITISH C O L U M B I A


March 2006
Fuchang Xu, 2006
Abstract

This study engages in acquiring reliable heat transfer data from experimental tests using a
test facility of industrial scale for an effective F E simulation of the water cooling
processes of steel strips on run-out table (ROT).

General 2D F E programs are developed for the direct and inverse heat transfer analyses.
Both the Flux Zoning Method (FZM) and Flux Marching Method (FMM) are proposed to
specify the heat fluxes in the inverse analysis. Parametric studies have been conducted
and show that the direct analysis progran may produce results with a higher accuracy,
and that the F Z M and F Z M are suitable for obtaining an accurate relationship between
the heat flux and the surface temperature in the impingement zone and in the parallel
zone, respectively.

The direct numerical investigations show that the isothermal condition does not exist for
thermocouples (TC) with separate measuring junction in the water cooling process, and
that the temperature difference between the two separate TC wires is affected strongly by
the conduction of the surface TC wires in the impingement zone and by the progressing
speed of black zone front in the parallel zone, respectively. The numerical analyses also
show that the horizontal distance between TCs should be less than 8-10 mm to warrant a
real 2D inverse calculation, and that there is a minimum depth for the embedded TCs for
an undisturbed surface temperature field.

The inverse analysis results of experimental tests show that the heat transfer behaviour at
the stagnation for stationary plates is mainly and greatly affected by the water
temperature, slightly by the steel grade and hardly by water flow rate. By contrast, the
cooling behaviour in the parallel zone is obviously affected by the flow rate and hardly
by the water temperature. It is also found that the plate's motion evidently reduces the
heat flux magnitude.

Extensive ID simulations to the cooling processes of the steel strips on an industry ROT
show that the final coiling temperatures of steel strips may be predicted with a reasonable
accuracy. Also, initial studies show that 2D modellings may have modest impact on the
accuracy of the coiling temperature prediction.

ii
Table of Contents

Abstract n

Table of Contents iii

List of Tables viii

List of Figures ix

List of Symbols and Abbreviations xv

Acknowledgements xvn

Chapter 1 Introduction 1

1.1 Controlled Cooling and Its Benefits 1


1.2 Controlled Cooling Systems 2
1.3 Heat Transfer during Water Cooling Process on Runout Table (ROT) 3
1.4 Experimental and Numerical Investigations 5
1.5 Objectives of This Study 6

Chapter 2 Backgrounds and Literature Review 8

2.1 Water Jet Impingement Cooling 8


2.1.1 Hydrodynamics of water jet impingement 8
2.1.2 Representative investigations . 12
2.1.3 Research work at U B C ; 17
2.2 Temperature Measurement Errors 20
2.2.1 Thermocouple (TC) thermometry .....20
2.2.2 Measurement errors using TC 22
2.2.3 Section summary and discussion 25
2.3 Numerical Methods for Inverse Heat Conduction Problem (IHCP) 27
2.3.1 Common techniques 28
2.3.2 Typical previous studies 30
2.3.3 Specification of heat flux 32

iii
2.4 TC Installation 33
2.4.1 Horizontal distance 34
2.4.2 Vertical depth 34
2.4.3 Effective TC depth 35

Chapter 3 Program for Direct and Inverse Heat Conduction Analyses 37

3.1 Program for Direct Analysis 37


3.1.1 Formulation 37
3.1.2 Implementation and validation .40
3.2 Formulation for Inverse Analysis 40
3.2.1 Objective function 40
3.2.2 General calculation form of heat flux 42
3.2.3 Iterative and sequential algorithms 44
3.2.4 Convergence norm 47
3.2.5 Flux zoning method (FZM) 48
3.2.6 Flowchart 49
3.3 Parametric Study to F Z M 49
3.3.1 Test procedure and data 51
3.3.2 Triangular heat flux inputs ..' 52
3.3.3 Impulse-like heat flux inputs 56
3.4 Summary 61

Chapter 4 Flux Marching Method ( F M M ) for IHCP in Water Jet Cooling 62

4.1 Progression of Water Jet Cooling Zones 62


4.1.1 Stationary plate case ...62
4.1.2 Moving plate case 64
4.1.3 Moving speed of waterfront in stationary cases 64
4.2 Drawback of F Z M 66
4.3 Main Features of F M M 69
4.4 Numerical Tests 73
4.4.1 Test procedure and setup 73
4.4.2 F E model 75

iv
4.4.3 Realization of waterfront movement 76
4.5 Results and Discussion 77
4.5.1 Reduction of peak value using F Z M 78
4.5.2 Shape change of heat flux using F Z M 79
4.5.3 Mis-prediction of surface temperature using F Z M 81
4.5.4 Effect of speed mismatch using F M M 81
4.5.5 Effect of flux jump using F M M 87
4.6 Summary and Prospects 89
4.7 Flowchart for the IHCP Solution Procedure 90

Chapter 5 Appropriate Installations of Thermocouples in Water Jet Cooling 93

5.1 Error in the Measured Temperatures 93


5.2 Stationary Case 96
5.2.1 Effect of TC wire conduction 96
5.2.2 Effect of drilling a hole 100
5.2.3 Effect of plate material 104
5.2.4 Effect of contact with water 105
5.2.5 Section summary 106
5.3 Moving Case 107
5.3.1 Motivation of study 107
5.3.2 F E Model 107
5.3.3 Results and discussion... Ill
5.3.4 Section summary 121
5.4 Appropriate TC Position 122
5.4.1 General modelling aspects 122
5.4.2 Appropriate horizontal distance 123
5.4.3 Appropriate vertical depth 125
5.4.4 Effective TC depth 127
5.4.5 Section summary 129
5.5 Summary 130

v
Chapter 6 Heat Transfer Behavior under a Circular Water Jet 131

6.1 Experimental Tests 131


6.1.1 Tests using stationary plates 131
6.1.2 Tests using moving plates 134
6.2 Validation and Application of the Effective Depth Approach (EDA) 137
6.3 Heat Transfer in the impingement Zone of Stationary Plates 143
6.3.1 Initial investigations 143
6.3.2 Heat transfer behaviour 145
6.3.3 Section summary 155
6.4 Heat Transfer in the Parallel Zone of Stationary Plates 156
6.4.1 Case study 157
6.4.2 Parameter effects 159
6.5 Heat Transfer for the Moving Plates 162
6.6 Summary 164

Chapter 7 Simulation of Run-out Table Cooling Process 165

7.1 Simplication of Governing Equation 165


7.2 Modeling and Implementation 169
7.2.1 Numerical method 169
7.2.2 Initial condition 169
7.2.3 Boundary condition 170
7.2.4 Temperature dependences of thermophysical properties 174
7.2.4 Latent heat 175
7.2.6 Microstructure evolution and property predication 175
7.2.7 Program structure 176
7.3 Studies on ROT Cooling 179
7.3.1 Case study 179
7.3.2 Parametric study 184
7.4 Investigation of Necessarity of 2D Simulation of ROT Cooling. 191
7.4.1 2D transverse modelling 191
7.4.2 2D longitudianl modelling 195

vi
7.5 Summary 201

Chapter 8 Conclusions and Future Work 202

8.1 Conclusions 202

8.2 Suggestions and Future Work 208

Bibliography 212

Appendix A Finite Element Matrix Equations for Conduction Heat Transfer 220

vii
List of Tables

Table 4.1 Test scheme and result using F Z M 74


Table 4.2 Node in calculation zone 76
Table 4.3 Calculated heat flux accuracy at TC3 with speed mismatch in FMM....85

Table 5.1 Data for F E M model 97


Table 5.2 Heat flux type notation 110
Table 5.3 Minimum depth for practical water cooling 126

Table 6.1 Experimental tests using stationary plates 132


Table 6.2 Hydrodynamic parameters at the stagnation point 132
Table 6.3 Comparison of results using different approaches 137

Table 7.1 Bank state code 180


Table 7.2 Examples of cooling parameters 180
Table 7.3 Examples of calculation results 181
Table 7.4 Effect of time step on the coiling temperature for Case 192 185
Table 7.5 Longitudinal modelling parameters and results 196

viii
List of Figures

Figure 1.1 A typical hot rolling strip mill 1


Figure 1.2 Controlled cooling systems 2
Figure 1.3 Schematic of jet configurations 4

Figure 2.1 Schematic of the boiling curve .9


Figure 2.2 Variation of HTC with temperature for saturated water 9
Figure 2.3 Velocity and pressure distribution 10
Figure 2.4 Heat transfer regions for steady state water jet 11
Figure 2.5 ROT test facility at U B C 17
Figure 2.6 Schematic arrangements of TCs 19
Figure 2.7 Seebeck voltage 21
Figure 2.8 Measuring junctions 21
Figure 2.9 Parameters for TC installation 33

Figure 3.1 Boundary conditions for a general heat conduction problem 38


Figure 3.2 F Z M F E Model for inverse calculation 48
Figure 3.3 Flowchart for F Z M IHCP solution procedure 50
Figure 3.4 Triangular heat flux inputs 52
Figure 3.5 Effect of measurement error level 53
Figure 3.6 Effect of number of future steps 54
Figure 3.7 Effect of regularization parameter 55
Figure 3.8 Impulse-like heat flux inputs 56
Figure 3.9 Effect of regularization parameter 57
Figure 3.10 Effect of future step number 58
Figure 3.11 Calculated heat flux 60

Figure 4.1 Progression of cooling zones 63


Figure 4.2 Example of measured temperatures 65

ix
Figure 4.3 F Z M FE model for verifications 67
Figure 4.4 Heat flux required for a same cooling curve 68
Figure 4.5 Effect of distribution of heat flux components
on the required heat flux 69
Figure 4.6 F M M FE Model for inverse calculation 70
Figure 4.7 Overstepping of heat flux 73
Figure 4.8 Crucial part of typical heat fluxes 74
Figure 4.9 FE-model for numerical test 75
Figure 4.10 Comparison of heat fluxes 77
Figure 4.11 Internal temperature directly calculated in case 2 78
Figure 4.12 Input q vs. ratio
p 79
Figure 4.13 V vs. ratio
w 79
Figure 4.14 Shape change of heat flux using F Z M 80
Figure 4.15 Heat flux space distribution at time 12 s 80
Figure 4.16 Surface temperature differences using F Z M 82
Figure 4.17 Comparison of Ts Vs. heat flux 82
Figure 4.18 Effect of higher speed mismatch using F M M 83
Figure 4.19 Intensification of higher speed mismatch 84
Figure 4.20 Calculated heat fluxes with higher speed mismatch 85
Figure 4.21 Calculated heat fluxes with lower speed mismatch 86
Figure 4.22 Calculated heat flux accuracy with speed mismatch 87
Figure 4.23 Assigned heat fluxes along the second zone 88
Figure 4.24 Effect of heat flux jump 88
Figure 4.25 Flowchart for IHCP solution procedure 92

Figure 5.1 Typical temperature profiles during water cooling 94


Figure 5.2 Inversely calculated heat flux 95
Figure 5.3 F E M model for analysis of effect of wire 96
Figure 5.4 Disturbed field of surface temperature around the wires 98
Figure 5.5 Effects of wire conduction on surface temperatures 98
Figure 5.6 Calculated temperature gradient along the wires 99

x
Figure 5.7 Example of F E M model for the effect of hole or cave 100
Figure 5.8-a Effect of 6 mm hole on temperature filed 101
Figure 5.8-b Effect of cave at 6 mm on temperature field 101
Figure 5.9 Disturbed temperature fields at 6 mm 102
Figure 5.10 Effect of 4 mm hole on temperature filed 102
Figure 5.11 Effect of 6 mm hole and K-plus wire 103
Figure 5.12 The effect of the inserted third metal 104
Figure 5.13 Cooling waterfront and TCs 108
Figure 5.14 Typical mesh and sampling points 109
Figure 5.15 Types of heat flux profiles 109
Figure 5.16 Space distribution of heat flux 111
Figure 5.17 Temperature profile results of 3D and 2D analyses 112
Figure 5.18 Comparison of temperature difference in 2D and 3D 113
Figure 5.19 Effect of the progressing speed of water cooling zones 114
Figure 5.20 Effect of stopping distance 116
Figure 5.21 Effect of heat flux magnitude 117
Figure 5.22 Effect of heat flux magnitude and progressing speed 117
Figure 5.23 Effect of the depth of hole and heat flux type 118
Figure 5.24 Time instant vs. the depth of hole and heat flux type 119
Figure 5.25 Effect of initial temperature 120
Figure 5.26 Effect of specific heat 121
Figure 5.27 Input of practical heat flux 123
Figure 5.28 Temperature vs. TC distance 124
Figure 5.29 Minimum depth of TC for ideal cooling 125
Figure 5.30 Minimum depth of TC for practical water cooling 126
Figure 5.31 Null-calorimeter techniques for carbon steel 127
Figure 5.32 Coincidence of temperature for practical cooling 128
Figure 5.33 Effective and actual depth of TC 129

Figure 6.1 Example A of measured temperatures 133


Figure 6.2 Example B of measured temperatures 133

xi
Figure 6.3 Example of TC arrangements for moving plates 135
Figure 6.4 Example of measured temperatures in moving plate tests 136
Figure 6.5 A n enlarged portion of Figure 6.4 136
Figure 6.6 Boiling curves using different modellings for test #6 138
Figure 6.7 Boiling curves using different modellings for test #9 138
Figure 6.8 Sensitivity of results to the effective depth for #24 139
Figure 6.9 Relationship between the actual depth and effective depth 140
Figure 6.10 Overall view of heat flux on the top surface for Test #25 142
Figure 6.11 Comparison of results with and without E D A for Test #25 142
Figure 6.12 Heat flux in the impingement zone -
Two components of heat flux assumed 144
Figure 6.13 Heat flux in the impingement zone - One heat flux assumed 145
Figure 6.14 Water temperature vs. surface temperature at Q = 15 1/min
w 146
Figure 6.15 Water temperature vs. surface temperature at Q = 30 1/min
w 147
Figure 6.16 Effect of water temperature at Q = 15 1/min
w 147
Figure 6.17 Effect of water temperature at Q = 30 1/min
w 149
Figure 6.18 Effect of water temperature at Q = 45 1/min
w 149
Figure 6.19 Effect of water flow rate at T = 30 C
w 150
Figure 6.20 Effect of water flow rate at T = 40 C
w 151

Figure 6.21 Effect of water flow rate at T = 70 C


w 151

Figure 6.22 Effect of water flow rate at T = 80 C


w 152

Figure 6.23 Effect of water flow rate at T = 50 C


w 152

Figure 6.24 Effect of water flow rate at T = 50 C for SS316


w 153
Figure 6.25 Effect of steel grade at T = 30 C
w 154
Figure 6.26 Temperature-dependent conductivity 155
Figure 6.27 Effect of steel grade at T = 50 C
w 156
Figure 6.28 Comparison of cooling curves at TC3
using different approaches for #25 157
Figure 6.29 Comparison of peak heat fluxes at the first four locations for #25 158
Figure 6.30 Comparison of cooling curves by F M M and E D A for #25 158
Figure 6.31 Comparison of cooling curves for #3 159

xii
Figure 6.32 Boiling curves at TC3 by F M M
for #24 (T = 30C) and #27 ( T = 50C)
w w 160
Figure 6.33 Boiling curves at TC3 by F M M
for #22 ( Q = 30) and #24 ( Q = 45)
w w 161
Figure 6.34 Boiling curves at TC3 by F M M for #3 (DQSK) and #22 (SS316) 162
Figure 6.35 Effects of the plate's speed and initial temperature
on the peak heat flux 163

Figure 7.1 Coordinate system used for modeling 165


Figure 7.2 Cooling zones under one active jetline 171
Figure 7.3 Design of the cooling zone lengths ...172
Figure 7.4 H T C in air cooling zone 173
Figure 7.5 H T C in impingement cooling zone 174
Figure 7.6 Schematic illustration of a typical ROT 179
Figure 7.7 Prediction accuracy 182
Figure 7.8 Temperature profiles for Case 192 ....183
Figure 7.9 H T C profiles around the first bank for Case 192 183
Figure 7.10 Temperature profiles around the first bank for Case 192 184
Figure 7.11 Effect of impingement width
on the coiling temperature for Case 192 185
Figure 7.12 Effect of the starting temperature
on the coiling temperature for Case 192 186
Figure 7.13 Effect of the strip speed on the coiling temperature for Case 192 187
Figure 7.14 Effect of the strip thickness on the coiling temperature for Case 192.... 187
Figure 7.15 Effect of the strip conductivity
on the coiling temperature for Case 192 188
Figure 7.16 Effect of the strip specific heat
on the coiling temperature for Case 192 189
Figure 7.17 Effect of the distance b l on the coiling temperature for Case 192 190
Figure 7.18 Effect of the distance L2 on the coiling temperature for Case 192 191
Figure 7.19 2D transverse modelling 192

xiii
Figure 7.20 Effect of transverse HTC difference
on the coiling temperature for Case 192 193
Figure 7.21 Effect of transverse HTC difference
on the coiling temperature for Case 192 194
Figure 7.22 Temperature distributions on cross section 195
Figure 7.23 Movement of model piece 197
Figure 7.24 HTC values for the two-jetlines cases 197
Figure 7.25 2D-temperature and lD-2D-temperature difference for Case 752-1.... 198
Figure 7.26 2D-temperatures and gradient for Case 752-4 200

Figure 8.1 Extra TCs and nozzles for future tests 209

xiv
List of Symbols and Abbreviations

c p Specific heat, J /kg-C

C a Seebeck coefficient of thermocouple, liV/K

C a s Seebeck coefficient of third metal, uV/K


C Heat capacity matrix
h Heat transfer coefficient (HTC), W / m C 2 o

i i time step
t h

J Number of heat flux components


k Conductivity, W/m-C
k,k x y Conductivities in the x- and v-directions, W/m-C
K Heat conduction matrix
L Number of measurement points
N Number of total time steps
n Iteration step
n,n x y Components of outward normal vector
Q Flux load vector, J
q h
Heat generation per unit volume, W/kg
q Heat flux, W/m 2

q j
Heat flux at the i time step, W/m
t h 2

qj Component of heat flux vector, W/m


q Heat flux vector, W/m
q 1
Heat flux vector at the i time step, W/m
t h 2

S Total boundary domain


Si to S 5 Parts of the boundary domain
T Temperature, C
T s Prescribed temperature or surface temperature, C

Tf Fluid (ambient) temperature,C


T Temperature vector,C

T Vector of temperature derivate to time,C/s

xv
V m Measured temperature vector at the i* time step,C

Calculated temperature vector at the i time step,C


t h

t Time, s
x, y Lagrangian coordinates, m
X Total sensitivity matrix for multi-dimensional problem
X I
Sensitivity matrix at the i time step
t h

p Density, kg/m

At Time step, s
AT Temperature difference, C
AT Temperature difference vector,C
a Regularization parameter
a Stefan Boltzmann constant, W/m -K 2 4

s Emissivity

FE Finite element
FTS Future time step
HTC Heat transfer coefficient
IHCP Inverse heat conduction problem
TC Thermocouple

ID One dimensional
2D Two dimensional
3D Three dimensional

xvi
Acknowledgements

A Ph.D. thesis is never the result of the author but an achievement of many persons.

M y sincerest gratitude and deep appreciation go to my supervisor Prof. Dr. Mohamed S.


Gadala for his judicious academic advice, for his financial and personal support. His
invaluable guidance and encouragement had been the foundation and source for me to
effectively proceed and finish this work. His careful revision enormously contributed to
the production of this thesis.

Many thanks also go to Prof. Dr. M . Militzer and Dr. V . Prodanovic for their valuable
suggestions and insightful discussions at various stages of this work.

Thanks to my colleagues in FE-lab and in the Strategic Group for their help and
suggestions. They had also shared their knowledge, experience and culture with me and
thus my view had been broadened.

Special thanks to my family and my mother who offered continuous moral support and
encouragement. To them and my past-away father, I dedicate this thesis.

Fuchang X u

xvii
Chapter 1

INTRODUCTION

1.1 Controlled Cooling and Its Benefits

Steel strip is one of the most versatile hot rolled products with a wide variety of applications
ranging from automobile bodies to soft drink cans. For the production line shown in Figure
1.1 [1], steel strips are produced from slabs approximately 250 mm thick. The slabs are
reheated in a reheating furnace to hot rolling temperatures close to 1250 C, rolled
sequentially by the roughing and finishing mills, and turned into strips of about 1 mm to 20
mm in thickness. After being rolled, the strips are cooled down to 550-650 C while they
pass through a so-called run-out table (ROT).

/Reheat ing\ /"RoughingN /FinishingY fRun-Out\ f~ .. "\


Vfurnacc J V Mi I la J \ Mills J \ Table J V J

Figure 1.1 A typical hot rolling strip mill [1]

While the geometry and surface quality of steel strips are mainly influenced by the rolling
deformation procedures, the microstructure and mechanical properties are highly dependent
on the controlled cooling procedure on a ROT. The controlled cooling process of steel refers
to the cooling process very specially designed and applied immediately after the finishing
stand. In a controlled cooling process, the cooling start and finish temperatures, and the
cooling rates are exercised accurately according to the expected properties in terms of phase
composition, size, and distribution.
One of the benefits due to this process is grain refinement. A finer grain size leads to an
increase in strength, an improved notch toughness and resistance to brittle fracture; and
contributes to an appreciable reduction in carbon content, while still maintaining the same
strength level. The lower carbon content improves formability and contributes to superior
notch toughness and weldability. In short, a controlled cooling process will assure a balanced

1
combination of mechanical properties, and narrows the gap in the properties between the as-
hot rolled high strength, low alloy steels and those requiring a special heat treatment step.

1.2 C o n t r o l l e d C o o l i n g Systems

Typically 50-100 m in length, a ROT is made up of a set of cooling banks placed above and
under the strips, and an array of motorized rollers. Cooling systems are required to have a
large cooling efficiency, and to be capable of producing the desired microstructures and
mechanical properties. There are three main types of cooling systems used for the controlled
cooling process of steel strips on a ROT: water spray, laminar flow (circular jet) and water
curtain, as shown in Figure 1.2 [2].

Laminar Water curtain Spray

Figure 1.2 Controlled cooling systems [2]

In a water spray system, the water as coolant impinges from a row of specially designed
nozzles onto the strip. The water spray is often used to cool the bottom surface and to sweep
the top surface to ensure the accuracy of pyrometer measurements. This kind of cooling
system is not included in this study and will not be discussed any more.
The laminar flow is preferred because it can penetrate the vapour film on the cooled surface
and increase the coolant residence time. While each of the laminar cooling apparatus consists
of two or four rows of U-pipes, this system creates low-pressure, laminar flow streams, and
therefore has a higher cooling efficiency than the water spray system.
In the water curtain system, the strip is cooled by a planar jet, which spans the entire width of
the strip. In addition to the advantages of a laminar jet system, it also improves the
uniformity of cooling in the strip's transverse direction.

2
Typical cooling conditions on a ROT for the strip hot rolling [3] are:

Cooling rate: 5-150 C/s


Cooling efficiency: 10-400 kJ/kg
Heat flux: 10 -10 W/m
5 7 2

Heat transfer coefficient (HTC): 1 0 - l 0 W/m -C


3 5 2

The basic requirements for the cooling system are summarized as follows. First, cooling must
be uniform in every location of the top and bottom surfaces, the edge and centre, and the
leading and trailing ends. Also, there must be no hard spots at any location. Second, any
distortion and residual stress due to water-cooling must be minimized. Third, the cooling rate
and the cooling start and finish temperatures must be accurately controlled. These
requirements are not simple ones because, for example, there are already temperature
variations between the centre and the periphery of a strip, even immediately after the finish
rolling.

1.3 Heat Transfer during Water Cooling Process on Run-out Table (ROT)

Although this cooling process has been used in the steel industry for decades, it is still mainly
a trial and error procedure when new products are developed. This is because the thermal
events during the cooling process are very complex, and a fundamental understanding of this
process is not yet complete.
In a typical hot strip rolling mill, for example, a steel strip leaves the last finishing stand at
temperatures ranging from 750 to 1000 C, and is rapidly transported along a ROT; the strip
is subsequently or alternatively subjected to air-cooling and water-cooling. Due to the high
temperatures of a steel strip during the water-cooling periods, the nucleate boiling is typically
confined to a small region beneath the jet, while the film boiling exists over most of the
surface at locations upstream and downstream of the stagnation point. It is clear that there are
conductive heat transfers in three directions, i.e., width, thickness and length directions
within the strip. It is also obvious that there appears heat loss due to the conduction of the
strip to the water and the rolls and due to the radiation of the strip to the air.
As for the water-cooling mechanism, both the laminar flow and water curtain systems belong
to jet impingement cooling with continuous cross sections. They may be either free-surface

3
jet or plunging jet, as shown in Figure 1.3 [4]. The free-surface jet is injected into an
immiscible atmosphere, and the liquid travels relatively unimpeded to the impingement
surface. The bottom surface cooling as well as the top surface cooling (when there is no
residual water) can be classified in this category. The plunging jet impinges onto a pool of
liquid covering the surface, where the depth of the pool is less than the nozzle-surface
spacing. If there is a residual water layer on the top surface from the other jets, this kind of
jet appears.

Nozzle

Gas

H
< Liquid
\\\\\\\\\\\\\\\\\\\\\\\\\ \\\\\\\\\\\\\\\\\\\\\\\\\

a. Free-surface b. Plunging

Figure 1.3 Schematic of jet configurations [4]

For stationary cases [5-12], i.e., when the cooled plate does not move, the cooled area on the
top surface is symmetrical to the water jetline for the water curtain or a row of circular
nozzles, or axisymmetrical to the stagnation point for a single circular jet. When water
impinges onto the cooling surface, it first vertically hits the plate and then moves transversely
or radially. With the spreading out of the cooling water the plate cools down. This stationary
cooling process exists only in laboratory conditions.
For moving cases [13-22], i.e., when the workpiece moves, the heat transfer is no longer
symmetrical to the water jetline for the water curtain or a row of circular nozzles; and it is
only symmetrical to the moving direction of workpiece for a single circular jet. On the side
where the water flow and workpiece move in the same direction, the thickness of the vapour
layers beneath the pools and in the film boiling regime decrease because vapour flow may be
promoted by the workpiece motion. By contrast, the vapour layer thickness may increase due

4
to stretching toward the jet by such a motion on the other side where the water flow and
workpiece travel in the opposite direction.
Making the cooling process on a ROT even more complicated is the variation of the thermal
conductivity and the specific heat of the strip with temperature, and the phase transformation
as the strip is cooled typically from about 800-900 C to 550-650 C. The latent heat of
phase transformation has an effect on the temperature profile and the final temperature. It is
worth noting that the latent heat may vary with the cooling rate and the temperature of the
strip.

1.4 Experimental and Numerical Investigations

There is considerable interest in controlling the local temperature of the strip as a function of
time to achieve the desired metallurgical and mechanical properties. Thus, it becomes a
necessity to understand, predict and simulate successfully the controlled cooling process on a
ROT. To realize these objectives, it is essential to obtain the accurate HTC or heat flux value
and its distribution along length and width. Therefore, numerous investigations have been
done that provide insight into the fundamentals of the complex cooling process.
Extensive work has been done at the University of British Columbia on the novel properties
of steel strips using an industry-scale cooling facility [23-27]. They include the following:
1) Studying and modelling the evaluation of austenite microstructure and the optimum
"conditioning" of austenite before transformation for optimum grain refinement.
2) Studying and modelling the thermal behaviour of a stationary or moving steel plate,
cooled by the circular jet under complex cooling conditions similar to the industrial
ones.
3) Studying and modelling the austenite decomposition of.steel under complex cooling
conditions.
4) Setup of the quantitative relations between chemistry, microstructure, and property as a
guide for steel grade (alloy) design and steel processing.

5
1.5 Objectives of T h i s Study

The aim of this research is to establish procedures and modelling strategies for an effective
and robust finite element simulation of the controlled cooling process of steel strips on a
ROT. To successfully realize this objective, several aspects of subtasks ought to be
considered as follows:
1) Development of a general two-dimensional finite element heat transfer analysis program
The program should be capable of analyzing both the state-steady and transient heat
conduction problems. Different heat transfer boundary conditions should be adopted,
including time and temperature dependent convection and radiation boundary
conditions. The program should take into account the heat generation option due to
phase transformation and the temperature dependence of thermophysical properties. The
program should be capable of using a mixture of different types of elements.
The program can be used either for the transient heat transfer to simulate the cooling
process as a function of time or for the steady-state heat transfer to simulate the cooling
process as a function of location.
2) Development of an algorithm for the inverse heat conduction problem (IHCP)
Based on the literature review, an iterative and sequential inverse heat transfer analysis
procedure will be developed and implemented into the proposed 2D FE program. In this
method, the least-squares technique, sequential function specification, and regularization
are used; simplifications in the sensitivity matrix calculation and the iterative technique
are to be adopted to reduce the CPU time while maintaining accuracy; both absolute and
relative norms are used to obtain a reasonable convergent solution.
3) Development of new techniques for the IHCP
A new method will be developed to specify the heat fluxes on the cooled surface in
IHCP. This new method should take the effect of water movement into account. The
profiles of heat fluxes simulating those that occur in ROT applications will be used to
investigate the accuracy and stability of the proposed inverse algorithm with the new
technique.
A method to compensate for the effect of the holes for installing thermocouples (TC)
will also be developed and discussed for simplifying the inverse calculation procedure.
4) Investigations of TC locations and temperature measurement errors

6
The developed F E program will be used to investigate the appropriate locations of TCs
and the probable temperature measurement errors. In U B C experimental tests, both
imbedded and surface TCs with a separation measuring junction are originally used to
measure the temperatures at different locations. It is generally known that both the holes
for installing the TCs and the attachment of TCs to the surface will disturb the
temperature field. The magnitude and pattern of their influence under water jet cooling
conditions are not reported. Also a detailed study on the appropriate locations of TCs for
inverse calculation problems has not been found, and an initial probe will be presented
in this study.
5) Determination of surface heat flux and heat transfer behaviour
The developed inverse analysis program and techniques will be used to determine the
heat flux during a water jet cooling process. Special procedures for specifying the heat
flux distribution on the target cooling surface will be checked and compared. A n
analysis of data and comparisons with other resources will be performed.
6) Simulation of ROT cooling process
The final stage of this study includes two tasks:
(i) Development of a special program for ROT simulations
This work focuses on the frame work of the F E program. Models and correlations
for boundary conditions, microstructure evolutions and mechanical properties from
other researchers can be implemented. The program will be used to predict the final
coiling temperature of the steel strip.
(ii) Investigation of 2D modelling for ROT applications
The effect of transverse and longitudinal heat transfer on the accuracy of coiling
temperature predictions will be investigated.

7
Chapter 2

BACKGROUND AND LITERATURE REVIEW

This chapter presents various aspects regarding water jet impingement cooling. They
include representative research studies, temperature measurement, and IHCP as well as
TC installation. The literature review of the simulation of a real ROT cooling process is
presented in Chapter 7.

2.1 Water Jet Impingement Cooling

2.1.1 Hydrodynamics of water jet impingement

Before focused on water jet impingement cooling, the heat transfer modes between the
water and hot surfaces are first discussed. Figure 2.1 schematically shows the boiling
curve for a saturated liquid pool cooling at a steady-state condition [4]. The single-phase
forced convection regime represents heat transfer in the absence of boiling, i.e. at a lower
surface temperature. When the surface temperature increases and reaches the onset of
nucleate boiling, discrete bubbles begin to detach from the surface and the convection
coefficient increases. The transition boiling regime refers to the formation of unstable
vapour blankets, while the film boiling regime represents heat transfer from the hot
surface to water across a vapour film (layer). If the initial surface temperature is high
enough, there is consequently convective heat transfer, nucleate boiling, transition
boiling, and finally film boiling.
Figure 2.2 depicts H T C variations with the surface temperature for a transient pool
cooling process [7]. When a hot surface is cooled down by water, the first cooling stage is
the stable film boiling; during this stage there is an uninterrupted film of steam between
the hot surface and the impinging water, and H T C is low. Reaching the so-called
Leidenfrost temperature, this steam film collapses, causing the transferable heat flow
density to increase rapidly and H T C to rise sharply. H T C is continuously increasing,
passing through the transition regime and reaching a maximum at the burnout point. After
passing the burnout point and entering the nucleate boiling and then the free convection

8
regime, HTC decreases rapidly. It is clear that surface temperature has a major influence
on HTC.

Single-Phase Nucleate Transition Film

Nucleate Boiling He>t Flux

Wall Superheat log AT,


Figure 2.1 Schematic of the boiling curve [4]

0 200 400 600 800 1000


Surface temperature,C

Figure 2.2 Variation of HTC with temperature for saturated water [7]

In addition to the surface temperature, there are a variety of factors affecting water jet
impingement cooling. Among them are jet velocity, jet diameter, pressure, and the
saturation temperature of water.

9
The jet velocity (also called impact velocity) with which water vertically hits the plate
can be calculated by:

V^4v 2gH n
2
(2.1)

where Vj is the jet velocity, m/s; V is the water velocity at nozzle exit, m/s; g is the
n

gravitational acceleration, m/s ; H is the vertical distance from nozzle exit to plate
surface, m. Positive and negative signs refer to the cooling on the top surface and bottom
surface, respectively.
The water jet diameter Dj is the diameter of the area between the plate's surface and
water at the instant when water hits the surface. It can be calculated by:

D
j= -^JVj D
n (-)
2 2

where D is the nozzle diameter.

Nozzle

Free surface

Stagnation Point
ujx)

A. Stagnation Region
B. Acceleration Region
C. Parallel-Flow Region

Figure 2.3 Velocity and pressure distribution [4]

10
The pressure at the stagnation point is given by:

P.=P +jpVja
2
(2.3)

where P is the atmospheric pressure and p is the density of water.


a

The saturation temperature T sat can be obtained from the saturation table of water
according to the pressure.
Figure 2.3 illustrates the representative distributions of the streamwise velocity and
pressure for a planar, free-surface jet (water curtain) [4]. When water vertically impinges
with the jet velocity onto the cooling surface, its streamwise velocity accelerates from
zero at the stagnation point (directly beneath the jet) to the jet velocity with increasing
streamwise distance. It is accompanied by the monotonic decrease of pressure from a
maximum at the stagnation point to the ambient value. The stagnation region A and
acceleration region B roughly form the impingement region. In the parallel flow zone the
wall jet flows parallel to the surface; the water velocity is assumed constant and equal to
the jet velocity, and the pressure is equal to the ambient pressure.

Coolant
/

I: Single Phase Forced Convection


II: Nucleate/Transition Boiling
III: Forced Convection Rim Boiling
IV: Agglomerated Pools
V: Radiation A Convection

Stagnation P t
7 7'/' / / / /
II J, I IV J, V

Figure 2.4 Heat transfer regions for steady state water jet [5]

For a single steady state free water jet with a stationary plate (in the following only
"plate" is used), the cooled area on the top surface can be divided into five different heat
transfer regions, as shown in Figure 2.4 [5]. The first region is called the impingement

11
zone and is located just beneath the jet. In this region, heat transfer is performed by
single-phase forced convection, and the cooling effectiveness is very high; the surface
temperature drops sharply to a low temperature such that boiling is not possible. Different
research shows that this zone may extend about 2 to 4 times the nozzle diameter or the
curtain width [5-9].

The water is heated as water spreads outwardly, the surface temperature is kept at a
higher level, and then the onset of the boiling is eventually reached, forming a
comparatively narrow nucleate-transition boiling region. In the third region the forced
convection film boiling occurs, the cooling water reaches the spherical state, and the heat
transfer gradually decreases. Thereafter, due to surface tension effects, the water may
agglomerate into pools, overriding the vapour layer. In this region, the water insulates
with the plate surface through a vapour film that runs away irregularly from the plate, and
the heat loses by convection between the vapor layer and the plate surface, and
particularly by radiation from the unwetted surface beneath the pools to the ambient. For
the last region, the plate is still dry and not covered by water; heat transfer occurs by the
radiation and convection from the dry plate to the surroundings.
For the cooling process on a ROT, the heat transfer may be different since the plate is
moving at high speed. This will be discussed in Chapter 4.

2.1.2 Representative investigations

The above example of water jet impingement cooling clearly shows that water jet cooling
involves a large number of subprocesses, and each of them is a complicated heat transfer
process taking place on stationary or moving surfaces. Much research, both experimental
and/or numerical, has been performed for a fundamental understanding of the cooling
process of hot strips/plates. The experimental tests should determine the cooling capacity
of the used facility by measuring the temperature profiles to provide important
information for the process design and development. In addition, the measured
temperatures in experiments are usually utilized for estimating the surface heat flux or
HTC, which is used for recognizing the heat transfer mode and for further numerical
simulations of the cooling process.

12
Surveys show that most of the experimental investigations have worked on the heat
transfer phenomenon with a single jet on stationary plate [5-12] or moving at a speed
much lower than the operational speed of a strip on a ROT [13-22].
Ochi et al. [8] experimentally investigated the transient boiling heat transfer to a circular
water jet impinging on a hot plate. The test plate was made of stainless steel of 210x50x2
mm in dimension and was heated up to 900-1100 C. The distance from the nozzle exit
to the cooling surface was 25 mm; nozzle diameters were 5, 10, and 20 mm; the water
subcooling ranged from 5 to 80 C and the impinging velocity changed from 2 to 7 m/s.
The temperature at the bottom surface was measured, and the temperature and flux at the
top cooling surface was inversely calculated using the finite difference method (FDM).
In the experiments, it was observed that a stable film boiling was maintained over the
whole cooling surface of the heated plate, and the temperature fell linearly with time. At
the end of the film boiling, the water contacted the plate surface on the impingement zone
and the heat flux increased. Then the cooling surface of the impingement zone got
completely wet and the wet zone advanced outwards. It was found that the heat flux at
the stagnation point was higher with a decreasing nozzle diameter. It was also observed
that the heat flux was highest at the stagnation point, and that the impingement zone was
in the range r/D <l .28 for the experimental cases.
n

The following empirical correlation obtained shows the effect of the water subcooling
AT b, jet velocity Vj, and nozzle diameter D on the minimum heat flux q :
su min

qmin =3.18xl0 . ( l - r 0 . 3 8 3 A r J ( ^ A , )
5
v
0 8 2 8
(2.4)

It is readily observed that the minimum heat flux is linear with the subcooling. A similar
correlation was observed by Ishigai et al. [9] whose experimental tests had a jet velocity
and subcooling in the range of 0.65-3.5 m/s and 5-55 C.
Chen et al. [13-15] studied the heat transfer phenomena of a circular water jet impinging
upward onto both a stationary and moving metal plate. In the work, the bottom surface of
a low carbon steel plate of 355 mm in width by 254 mm in length by 6.35 mm in
thickness was electrically heated slowly and uniformly to a temperature of 240 C and
88 C, while other surfaces were thermally insulated to minimize heat loss to the
ambient. The plate was cooled by water from a circular nozzle of 4.76 mm in diameter.

13
The distance from the nozzle exit to the plate surface was 90 mm. The water temperature
ranged from 25 to 27 C. A nozzle exit velocity of 2.66 m/s and a jet velocity of 2.30 m/s
were maintained. The plate moving speed was 0.5 m/s.
Results showed that the HTC was larger at the stagnation point than at other locations,
and was about 210 kW/m -C at a plate surface temperature of 240 C. The HTC values
2

at larger r/D locations were reduced because the temperature gradient decreased as the
n

distance from the stagnation point increased. It was suggested that the transient
phenomenon, rather than boiling, dominated the heat transfer mechanism when the plate
surface temperature was 88 C.

Moving plate tests were conducted using a longer plate of 609 mm in length to achieve
steady state heat transfer on the moving plate with 0.14-1.44 m/s. The tests revealed that
the surface motion had no appreciable effect on local H T C near the stagnation point,
while the overall heat transfer rate and cooling effectiveness increased with the increase
of the moving speed.
Stevens et al. [16] investigated the effects of the jet Reynolds number, nozzle-plate
spacing, and jet diameter on local HTC for round, single-phase free liquid jets impinging
onto a flat uniform constant heat flux surface. In the experiment the nozzle diameters
were 2.2, 4.1, 5.8, and 8.9 mm with 39 mm in length. Test Reynolds numbers ranged
from 4000 to 52000 such that turbulent flow prevailed in all tests. It was found that in the
region of r/Dj <0.75 the H T C was nearly constant; beyond that the H T C decreased
sharply. It was concluded that the impingement zone was around 1.5 times the nozzle
diameter.
Zumbrunnen et al. [17-18] measured the heat transfer distributions on moving and
stationary plates cooled by a planar liquid jet. The experiments were transient because of
the short length of the plates. In the experiment, the plate surface temperatures were
measured and utilized with a numerical solution to the one-dimensional heat transfer
energy equation for constant thermophysical properties to determine the local HTC. Two
nozzle widths of 10.2 mm and 20.3 mm were used. The plate moved at less than one half
of the impingement velocity, with an actual maximum moving speed of 0.8 m/s.
It was found that for t<0.6 s, the local Nusselt numbers were generally larger than those
at larger times. For a given time, Nusselt numbers had two local maxima: one beneath the

14
jet due to the enhanced heat transfer by the high shear flow associated with fluid
acceleration away from the stagnation point; the other adjacent to the jet caused by the
transition to turbulence and subsequent growth of the turbulent boundary layer.
It was verified that plate motion had different effects on the different cooling regimes.
First, the maximum Nusselt number did not shift as a consequence of plate motion and
remained fixed at the stagnation points. That is, the H T C in the single-phase forced
convection zone were not appreciably influenced by the surface motion.
Second, the critical Reynolds number of the transition of turbulence depended on the
relative velocity between the plate and flow, and the pressure gradient in the vicinity of
the jet. The data showed that the transition was delayed in the flow direction.
Third, when nucleate boiling occurred over the entire surface of the moving plate (plate
surface temperature T =644 C, jet velocity Vj=2.60 m/s, and plate velocity V =0.6 m/s),
s p

the position of peak heat transfer shifted from the jet centerline in the direction opposite
to the plate motion because nucleate boiling was more effective at the higher plate
temperatures.
The following local heat flux correlation was obtained by Filipovic [19-20] for the
impingement zone on stationary plates:

q =C(AT )V;- D-ATj"


im sub
5
(2.5)

where q im was the heat flux in the impingement zone in W/m , C was a constant
decreasing with decreasing subcooling, with C=1.3xl0 at AT b=S0C
7
su and C=0.81xl0 7

at AT b=65 C. Because the coefficient C related to the subcooling, the heat flux was not
su

a linear relationship to the subcooling.

Hatta et al. [21-22] examined the cooling process of both a stationary and moving plate
cooled by a water curtain. The plate was made from 18Cr-8Ni stainless steel of 240 mm
in length, 100 in width, and 10 mm in thickness. The plate was heated up to 900 C and
then cooled by water with a temperature of 8-18 C. Three TCs were used and arranged
with a space of 60 mm along the centerline in width. The vertical distance of the nozzle
exit to the impinging surface was fixed at 100 mm. For moving cases, the travelling
velocity of the plate ranged from 0.48 to 2.4 m/min.
It has been observed that the water formed a thin film as soon as it impinged onto the
plate surface. The initial red hot plate surface became darkened within a so-called black

15
zone near the impinging point. The black zone was assumed to correspond to the wetting
zone; i.e., the water-cooling distance is equivalent to the distance from the impinging
point to the boundary between the wetting zone and the non-wetting zone. In the black
zone the water temperature was below the boiling point. They postulated that the plate
surface was unwetted if either of the following two conditions was satisfied:
For stationary plates:

T > 1100-87;
s
(2.6)
T > 68 C
s

For moving plates:

T > 1100 -8.57;


(2.7)
T > 710 C
s

where both the plate surface T and water temperatures T have units of C.
s w

Prior research has provided insight into water jet cooling on a ROT. The main findings
can be summarized as follows:
1. The heat transfer greatly depends on the subcooling, and is also affected by jet
velocity and jet diameter;
2. The heat transfer is much stronger in the impingement zone than in the parallel
zone;
3. The effect of plate movement on the heat transfer is to some extent conflicting.

There are still many important points to be examined. Most of research results in the
literature address the jet impingement problem under steady-state conditions; i.e., the test
specimens are constantly heated while being cooled by the water impinging jet. As for
the tests using transient conditions, the dimensions of the nozzles, the distances of the
nozzle to the plate are quite small, and the plate temperatures are low; all of these do not
reflect the conditions used in practice on a ROT. The results of such research efforts have
been important in providing insight into the problem but, generally, are difficult to apply
to practical situations on a ROT.
With respect to temperature measurements and analysis methods, one of the most
common procedures used in the literature is to use TCs on the opposite side of the plate

16
or impeded below the top surface, and to obtain the heat flux distribution on the surface
through an inverse analysis. More discussions are presented in Sections 2.3 and 2.4.

2.1.3 Research w o r k at U B C

By recognizing those defects in the literature, a ROT test facility of industrial scale has
been constructed at the University of British Colombia, as schematically shown in Figure
2.5 [23-27]. This facility consists of a top water bank and a bottom water bank, a top
header and a bottom header (not shown), a water pump and pipe circuits and flow control
valves, a furnace and apparatuses (not shown) for heating water and measuring water
temperatures.
Three nozzles can be set up onto the top header and can be easily changed i f necessary.
Nozzle dimensions are compatable with those used in the steel industry. The distance
between the top nozzles is adjustable from 50 to 90 mm. The vertical distance of the top
nozzle exit to the test plate can also be adjusted from 0.6 to 2 m. The top nozzles are used
for top surface cooling.

Top water bank

Header

Nozzle
Furnace

Bottom water bank

Pump

Figure 2.5 ROT test facility at U B C

17
Only one nozzle is available on the bottom header whose distance to the plate is fixed.
The alignment angle of the bottom nozzle to the plate may be changed. This bottom
nozzle is used for bottom surface cooling.
Total water flow capacity has a maximum of 138 1/min. The water can be heated to 90
C. A l l the abovementioned parameters are very close to those used in the steel industry.
The test plate dimensions are up to 280x280x10 mm for stationary tests. Most of the test
plates have a nominal thickness of 7 mm. Although the furnace can heat test plates up to
1200 C, the initial temperatures of the test plates are roughly 700-900 C.
The materials of the test plates are carbon steel D Q S K (Drawing Quality Special Killed)
and stainless steel SS316. The density and specific heat of both steels are assumed to be
constant and are 7800 kg/m and 470 J /kgC. For the conductivity of D Q S K there is the
3

following equation with a correlation coefficient of 0.977 and a standard error of 1.0:

k = 60.571 -0.03849xT (C) in Wlm-C (2.8)

The conductivity of SS316 is also temperature dependent and the following equation is
assumed:

= 11.141 + 0 . 0 1 4 x r ( C ) in Wlm-C (2.9)

These equations are valid for temperatures ranging from zero to 1000 C.
TCs of Type K are used in all experiments for measuring the temperatures. TCs of Type
K would normally have an error of approximately 0.75% of the target temperatures when
used at a temperature higher than 277 C. Each of TC wires is around 0.051 mm (0.002")
in diameter and is insulated each other and sealed by a metallic protective coat that has an
out diamter of about 1.6 mm (1/16").
As shown in Figure 2.6, a total of 16 TCs are used in each test for stationary tests, and are
installed at 8 locations in the circumferential direction starting in the center of the plate
with an increment of 15.9 mm (5/8") in the radial direction, numbered from 1 at the
centre to 8 at the farthest. At each location, an internal TC is installed in a blind hole with
a diameter of 1.6 mm (1/16") that is drilled from the plate's bottom surface. The
measuring junction is fixed onto the end surface of the hole that is about 1 mm (0.04")
below the plate's top surface. A thickness-through hole of the same diameter is drilled

18
with a centre distance of 3.2 mm (1/8") from the blind hole. The surface TC is inserted in
the through-hole and its wires are spot welded on the top surface above the blind hole.
It should be noted that the TC wires are welded separately to the designed locations; i.e.,
the separation junction is adopted (see next section). The distance of wire leads is
approximately 1 mm. It should also be noted that an amount of length of wires is bare and
is exposed to ambient.

Nozzle

Air

Water
Plate 0 O 0 0 / 0 \ 0 0 0

TI T2

Insulation

Figure 2.6 Schematic arrangement of TCs

The approach of using a TC on the cooled surface and an imbedded TC to measure the
temperatures and then to obtain the heat flux has not been reported in the literature. In the
following, this approach will be called "Two-TCs". The rational behind using this Two-
TCs approach is to estimate the temperature gradient directly from the two measured
temperatures at that location and then evaluate the surface heat flux from the equation:

19
where q is heat flux in W/m , k is temperature dependent conductivity in W/mC, and
2

dT/dy is the temperature gradient in C/m at the top surface point.

This approach had been used to obtain the heat fluxes and boiling curves [26-27]. It can
be seen from those analyses that the heat flux values are generally much higher than
those reported in the literature for the same or equivalent cooling conditions. This
suggests that this approach may hold its pertinent problems, which is briefly analyzed in
Section 2.2.3, after the temperature measurement errors associated with TCs are
discussed.

2.2 Temperature Measurement Errors

TCs have been used widely in temperature measurement during water-cooling


experiments. It is crucial to obtain accurate and reliable temperature measurements in the
experiments under the designed cooling conditions. In this section, the measurement error
sources associated with TCs are discussed.

2.2.1 T C thermometry

A TC consists of two dissimilar metallic wires. The measuring junction is the point where
the wires are attached or soldered to the surface of the specimen or the plate. The other
ends of the wires are referred to as the reference junction. When the two junctions are at
different temperatures, e.g., when the measuring end is heated, an emf (electromotive
force) is intrinsically developed, and a continuous electric current will be generated and
flow in this thermoelectric circuit. This phenomenon is called the Seebeck effect [28-29].
For a given length of homogenous materials, the Seebeck voltage depends on two factors:
one being the temperature difference between the two junctions, the other being the
Seebeck coefficient. If the circuit is somewhere broken, as shown in Figure 2.7, the net
open circuit voltage (e^g) can be expressed as

^=C -AT
a (2.11)

20
where C is the Seebeck coefficient and AT is the temperature difference between the two
a

junctions. In this study, it is assumed that the Seebeck coefficient Ca is a material


constant and does not change with temperature.

T, T 2

heat metal B

Figure 2.7 Seebeck voltage

There are two common methods to connect the TC wires to the target surface at the
measuring junction, as shown in Figure 2.8. The bead-type junction is practically realized
by twisting the wires together and then soldering the joint on the specimen, forming a
small bead. It is obvious that this kind of junction has a finite volume of bead and hence a
large thermal inertia. Therefore, the junction temperature may not be the same as that
undisturbed one although the generated e m f is uniquely related to the temperature at the
bead junction.

Bead-type Separation -type

Figure 2.8 Measuring junctions

21
Another type of TC junction is called separation junction. In this type of junction, the
individual wires are separately soldered to the specimen, and the material of the specimen
whose temperature is to be measured forms a part of the thermoelectric circuit. The
spacing between the two wires ranges from one to two wire diameters. The presence of
specimen metal between the two soldered points is assumed not to influence the resulting
emf as long as the whole measuring junction has a uniform temperature [29]. This kind of
measuring junction is also called intrinsic junction because this approach will greatly
reduce the thermal inertia of the junction and hence decrease the response time of TC
close to zero.
Any other contact between the wires, except the reference junction, and any other contact
between the wires and the specimen, except the measuring junction, may result in a
premature connection and hence a wrong output of emf [29-30]. In addition, the contact
resistance of TCs to the parent materials, the change of chemistry compositions of TCs
and oxidation of the parent material surface etc may also affect the measurement
accuracy. The following discussion is confined to the effects due to the installations of
TC to specimen.

2.2.2 Measurement errors using TC


The installation of TCs either on the specimen surface or inside will generally deform the
original temperature field and will usually cause measurement errors. Most of the
following discussions focus on the intrinsic TCs.
Hereby, the concept of "combined wire" should be introduced. The combined wire is a
single cylinder of a radius that may be given by

r = 4lr v (2.12)

and a thermal conductivity


k = (k +k )/2
l 2 (2.13)

where r is the radius of a single physical TC wire, and A:, and & are the individual
w 2

thermal conductivities of the wires.


Analytical methods are widely used to investigate the errors associated with TCs on the
surface or inside [29-31]. In these studies, the dimensions of the parent solid are taken as
a semi-infinite body, and generally a steady-state condition is assumed. Also, a single

22
cylinder to represent the combined wire is used. Analyses show that the wire conduction
is the crucial factor among the factors influencing the accuracy of temperature
measurement using TCs. Therefore, a reduction of such errors can be obtained by
decreasing the heat flux conducted from the measuring point along the wires.
Park et al. numerically examined the error in the measured surface temperature using an
intrinsic TC [32]. The simulated process was the quenching of a hot nickel cylinder of
500 C into cold water of 20 C. A 2D axisymmetrical F E model with a single combined
wire was used first. In this model, the wire of 0.0635 mm in radius was laid along the Y -
axis. The surface of the wire and the top surface of the parent plate of 1 mm thick and
4.36 mm wide were specified temperature dependent heat flux as the boundary condition.
This analysis indicated an error of about 40 C that occurred within 0.1 s of immersion
and persisted for almost 1.7 s.
A 3D F E model was then developed. In this model, two square columns, each of which
had the same cross-area as the wire in the 2D model, were apart of two wire diameters.
Different thermal properties were assigned to each of the equivalent wires. The same
boundary conditions were specified. This simulation predicted an error of similar
magnitude. Unfortunately, the fact that the temperatures at the two wire roots were found
being different because of the difference in thermal properties had not been given
attention.

The accuracy of intrinsic TCs in the transient radiant heating process was experimentally
investigated in [33]. Two intrinsic TCs were installed on each specimen. One was placed
on the top surface and the other, directly opposite, on the bottom surface. A l l TC wires
were 0.01" in diameter, with a separation of approximately 0.05" between wires. It was
assumed that the heat transfer was one-dimensional and that the TC on the bottom surface
was subject to a lower order of errors than that on the top surface exposed to the radiant
energy. Consequently, the accuracy of the top surface TC could be determined by
comparing the temperature difference between the two measured temperatures with the
theoretical one that was obtained from the solution of the heat transfer equation. It was
found that all tests produced consistent results, the temperature errors could reach 35 F,
and that the temperature errors increased with increases in heating rates or in specimen
thickness. The author attributed the errors to an amount of bare TC wire that lead to more

23
loss of heat to ambient, and precluded the effect of temperature gradient between the two
wires because the included area of the third metal, i.e., the specimen itself, was very
small. This postulation may not be justified, since the ratio of separation distance to wire
diameter is about 5 and the heating process is highly transient.

Attia et al. studied the distortion in a 2D thermal field around inserted TCs that were
presented by combined wires [3436]. The 2D model was a square with a void or two
voids with the x-y plane perpendicular to the axis of the hole (void). The heat was
assumed to flow in one direction from the positive y side to the negative y side. It was
found that the distortion was symmetrical about both the x-axis and v-axis; and that the
maximum existed at the top and bottom points of the void when one void was located in
the centre of the square; and that the temperature disturbance was inversely proportional
to the temperature gradient, the void diameter, and the distance from the boundary.
In reference [37], Attia et al. extended their analyses to the general case of 3D TC
installation and focused on the end effect representing the disturbance of temperature
field surrounding the bottom of the hole for installing TC. It has been concluded that the
eccentric position of TC was the most important factor and caused a significant error,
especially when the temperature gradient was large.

The effect of TC wire was also analyzed using a regular 2D F E axisymmetrical model
[38]. The TC wire was 0.04 mm in diameter and 10 mm in length that was exposed to the
ambient. A constant H T C of 2 kW/m -C was applied to the top surface, and the same
2

material properties of 15 W/m-C for the conductivity and 5x106 J/m -C for the specific
3

heat were assumed for the wire and parent object. It was found that an error could be of
40 C or 4.6% due to wire fin effect.

A n approach was then developed to investigate the interaction between the two wires of
TC. In this approach, the F E model consisted of two portions: one was a normalized
regular 2D F E axisymmetrical submodel representing the TC; the other was the parent
object that did not include TC in the F E mesh. The submodel was superimposed at the
actual location of the surface TC on the parent material. Its properties were assigned
according to its particular location. The top surface of the submodel was exposed to a
convective medium, and other surfaces were specified with temperatures as thermal

24
boundary conditions, also according to its particular location. Unfortunately, no
information was given about how to get the specified temperatures. And from the point of
view of the author of this thesis, the results shown do not represent the real interactions
but only the temperatures at locations with a distance of one or two wire diameters.

The effect of hole for imbedded intrinsic TC was also numerically studied by L i [39]. In
this research, the distance of imbedded TC to the surface was fixed at 1.5 mm. It was
found that the difference between the temperature at TC tip and the undisturbed one
could be up to 100 C as the cooling time was about 2 s under the conditions that the
applied maximum heat flux was 20 MW/m and the plate material was AISI 316. This
kind of temperature difference was not significant when the plate material had higher
conductivity and the maximum was around 5 M W / m . It was generally concluded that the
2

hole should be included in inverse calculations.


The measurement error due to the installation of surface intrinsic TCs was also
investigated by using analytical and numerical methods in L i ' s work [39]. It was pointed
out that the heat flux at the measuring junction was about 10 times as that on the plate
surface. The temperature error may reach 300 C at around 2 s after water cooling.
Li's work did not address the effect of the hole on the surface temperature and did not
pay attention to the temperature differences between the two wires for the intrinsic TC.

2.2.3 Section s u m m a r y and discussion

i) Summary

The above discussion clearly shows that the fin effect of TC conduction wires may cause
significant errors. The indicated temperature would be lower or higher than the
undisturbed one for the cooling and heating process, respectively.
The hole for installing TCs will also disturb the temperature field. If there are two
adjacent holes a more complicated temperature pattern would occur, and the magnitude
of error would be higher.
The isothermal condition for two wires of intrinsic TCs has not been, however,
extensively studied and discussed. This is an important for a rapid cooling/heating

25
process, as in the jet impingement cooling of hot steel plates. It is reasonable to say that
these errors may be more significant for a highly transient intensive cooling process.
Moreover, the work in the literature has not addressed the possible effect of water
movement on the temperature measurement errors using intrinsic TCs. That would be a
critical aspect of intrinsic TC applications.

ii) Discussion
This discussion is oriented to the appropriateness of the Two-TCs approach. From the
point of view of this author, such an approach may not produce credible results. Several
reasons support this postulation.
This approach is based on the two measured temperatures, one from the surface TC,
noted as T , the other one from the imbedded TC, as Tj . Here, take the notation T and
sm m st

Tu for the "true" surface temperature and "true" internal temperature at the corresponding
locations. Only when

T ~T =T -T
sm im st lt or T - T = T - T
sm sl im (2.14)

is valid at any time instant during the cooling process, the true temperature gradient may
be estimated correctly. However, this condition is fairly difficult to satisfy. It may be
easily concluded that the respective errors in the measured surface temperature and the
interior one is different (Chapter 5) because the temperatures are different but have the
same nominal error level, even assuming the temperature field has not been disturbed. In
fact, the temperature field is greatly affected by both the attachment of TCs and the holes,
and the measured temperatures may be far from the true values.
Second, the actual distance between two measurement points is in question because of
few reasons. The first one is the methods used to calculate the distance between the two
TCs. This is usually obtained by taking the difference between the measured values of the
plate thickness and the hole depth. The defect is that the thickness is an average value for
the whole plate, and no individual thickness for each location is determined.
The second reason is that the exact locations of the TCs are unknown. It is clear that the
two points should be aligned along a vertical line, and the distance should be their
vertical space. However, it is very difficult to know the exact location of the interior TC.

26
Therefore, the top surface point may have a small shift from the interior one. This may
lead to a larger error especially when large temperature gradient exists.
Another reason is that the heat transfer is highly transient with large temperature gradient
in the transverse direction. It is reasonable to assume that the transverse gradient is not
uniform through the thickness. Thus, there will definitely be a temperature difference
between the two target points even i f the top surface is thermally insulated. The obtained
heat flux value from equation (2.10) would be in error because a thermal insulation is
already assumed and no vertical heat flux should exist.
Therefore, other options to determine surface boundary condition (heat flux or HTC) are
required.

2.3 N u m e r i c a l M e t h o d s for I H C P

The measured temperatures in experiments are usually used to estimate the surface
boundary condition, which is used for recognizing the heat transfer mode and for further
numerical simulations of the cooling process. Generally, the analysis methods used are
associated with temperature measurement approaches.
A variety of numerical methods and computational algorithms have been developed in
the literature to obtain the surface boundary condition. By skipping all of the analytical
methods and the direct methods, the following discussion focuses on the numerical
inverse methods.
A n IHCP means that the boundary conditions or the thermophysical properties of
material are not fully specified, and they are determined from measured internal
temperature profiles. The effects of changes in boundary condition are usually damped
and lagged; i.e., the varying magnitude of the interior temperature profile lags behind the
changes in boundary conditions and is generally of lesser magnitude. Therefore, an IHCP
would be a typically ill-posed problem and would normally be sensitive to the
measurement errors. Thus, in general, the uniqueness and stability of an IHCP solution
are not guaranteed [40-41].

Among the methods for IHCP are the least-square regularization method [40-41], the
sequential function specification method [40], the space marching method [42], the

27
conjugate gradient method [43], the maximum entropy method [44], and the model
reduction method [45]. Important applications of these methods and their numerous
modifications have been performed in various branches of thermal engineering [23-25,
39, 46-52]. It is not an appropriate approach to review all of them. Instead, the emphasis
in this section is on the computational methods and techniques used in the water jet
impingement and similar cooling processes such as quenching.

2.3.1 Common techniques

The techniques that stabilize an IHCP solution are highlighted in the following
subsections. For simplicity, the following discussion is based on one single heat flux q'
(for more than one heat flux component, just change the notation to vector form q ) and 1

multiple measurement points.

1. Least squares

A n IHCP may generally be converted into an optimization problem. The sum of squares
of the differences between the calculated temperatures and the measured ones are usually
targeted as the objective function:

This objective function is minimized with respect to the unknown heat flux component q'
by insuring:

^ =0 (2.16)
dq

The heat flux q' can be obtained by solving the above equation.

2. Future time step and function specification

Since IHCP is ill-posed, the inverse solution may not be unique and may be sensitive to
measurement random error. To reduce such a sensitivity and to recover the effect of
boundary condition changes, in addition to the measured temperatures at the current time
step T , the measured temperatures at future time steps (FTS) T '
1 + /
, T' , ...,T (where N-
+2 N

1 is the number of future steps) are also used to estimate the heat flux q'. In this case, the
objective function becomes:

28
5
= t i ( 7
i - 7
i ) 2
( -
2 1 7
)
j=\ 1=1

or in vector form

s^iX-xfiX-X) (2-18)
1=1

It is clear that the heat flux q' (/ = 1-A ) at different time steps may be different. One way
7

to treat the problem is to make a temporary assumption for the value of q' , q ,..., q . +I l+2 N

This is called function specification. Thus in this stage, a function specification is always
related to the future time step, and is about the function form of the surface heat flux
variation with time. The simplest and most widely used assumption is a sequence of
constant segments; i.e., q' =q' for 1 < k < N-l. Other assumptions are possible, such as
+k

linear function, parabolas, cubics, and exponentials. For the constant and linear
assumptions, the objective function has only one variable q'. Therefore, the equations and
procedure are still valid.

3 . Regularization

To damp the fluctuation of a solution from the measurement errors, more items may add
to the objective function. This procedure modifies the objective function and is called the
regularization method. The simplest one is the variable q itself.

5 = f(T:-TJ) (T:-Ti) + - q - q
7 r
(2.19)
/=i

This term a q q is called the zeroth-order regularization term. The fraction value a is a
T

weighting factor and is called the regularization parameter.


The higher regularization items such as the difference between the consecutive heat
fluxes in time, the difference between the adjacent heat fluxes in space (these two are
called the first-order regularization term), the difference between the derivates of heat
fluxes with respect to time or the spatial derivates of heat fluxes (these two are called the
second-order regularization term) are possible to add into the objective function.
It is interesting to note that the first order term is occasionally used and the others
infrequently used [39]. It is also important to mention that the zeroth order term reduces

29
the magnitude of heat flux; the first-order term reduces the magnitude of change in heat
flux, and the second-order term decreases the rapid oscillations in heat flux.
It is worth noting that the function specification technique also works as a regularization
procedure and stabilizes the solution process. And the function specification and
regularization techniques can be used for both a linear and nonlinear IHCP.

4. Whole time domain and sequential approach

If a method can estimate simultaneously all the parameters for the total time interval, it is
called a whole time domain method or simply a whole domain. In this case, all the data
from the measurements are input into the objective function, and the function
specification is also made for the total number of time steps, if used.
The transient heat conduction is a kind of diffusive process. Therefore, the whole domain
method is not an efficient procedure. The most computationally efficient estimation
approach is the sequential one. In the sequential procedure, the heat flux q' is estimated
after q ' ; after q' is obtained, /' is increased by one and the estimation procedure is
1 1

repeated.

Improvements to and adaptations of the numerical algorithms to complex applications


such as the water-cooling of hot plates by jet impingement on a ROT are still an active
area of research for obtaining stable and reliable results. In this study, an FEM-based
inverse heat conduction method will be developed and used to determine the surface heat
flux and surface temperature from the internal temperature. A l l the abovementioned
techniques will be used in the proposed algorithm in this study (Chapter 3).

2.3.2 Typical previous studies

Vader et al. [10-11] investigated the convective nucleate boiling on a heated surface
cooled by an impinging planar jet of water under a steady-state condition. The
temperatures were measured by TCs at the opposite dry face to the cooled top surface.
The temperature and heat flux at the cooled surface were found by solving the steady-
state equation. In the calculations, the boundary conditions were assigned from the least
squares cubic spline fit of the measured temperatures. It was reported that there were
uncertainties of 10% and +12% in the surface temperature and HTC, respectively.

30
Wolf et al. [12] also studied the local jet impingement boiling heat transfer. A similar
procedure as Vader's was used, except that only the top surface temperatures were
calculated by solving the heat transfer equation. The heat flux on the top surface was
inferred from the measurement of the voltage difference across a definite plate's length.
This length was changed to obtain the heat fluxes at different locations from the
stagnation line. Overall uncertainties were the surface heat flux, 2%, the surface
temperature, 2C, and the resulting HTC, 7%.

Liu et al. [23-25] investigated the boiling heat transfer on steel plates cooled by a circular
jet. In their experimental studies under transient conditions, the temperatures were
measured with TCs inserted inside the plates and close to the top surface. The top surface
temperatures and heat fluxes were simultaneously calculated out from an FEM-based
inverse algorithm based on the zeroth-order regularization. The calculations were only
based on current temperatures. The function specification and iteration techniques are not
used.

Kumagai et al. [49] studied the transient cooling of a hot plate with an impinging water
jet. Temperatures were measured at four points at different depths along the plate
thickness (7 direction) and at nine locations in the width direction (X direction). A n
exponential equation with three coefficients was used to describe the temperature profile
in the Y direction, and then the coefficients were determined from the measured
temperatures and the least squares method. The approximate temperature curve was
extrapolated to the surface to obtain the surface temperature. The gradient at the surface
produced the heat flux value.
The error level was also reported by Kumagai et al. [49]. At the stagnation point, this
method had an error of 1-3.3 % in the surface temperature estimation and of 10-35 % in
the heat flux; at other locations the errors were 1-1.5 % in the surface temperature and
10-15 % in the heat flux after the solid-water contact.

The iterative regularization and the conjugate gradient method were used in the inverse
determination of the local H T C for the nucleate boiling on a horizontal pipe cylinder
under a steady-state condition [50]. In this work, the Lagrangian functional was defined
using the sum of the temperature differences between the measured and calculated ones,

31
and the governing equation for steady-state heat conduction and all boundary conditions
as multiplier terms. The conjugate gradient method was used to find the descent
parameter. The iterative technique was adopted to get a more accurate solution of heat
flux from the measured temperatures at locations inside the cylinder body. However, the
claim of using the regularization technique should be viewed cautiously since no other
terms were added into the objective functional.

K i m used an FEM-based inverse model to evaluate H T C during heat transfer of a


cylinder cooled by a fan or water [51]. Multiple embedded TCs at the same radius were
used to measure the temperatures. In the method, a spatial regularizer composed of the
zeroth- and first-order regularization terms was added to the objective function together
with the future step technique and the iterative technique. This approach is quite similar
to our development.

2.3.3 Specification of heat flux

No matter which theory or principle is used to develop the inverse algorithm, only a
limited number of unknown heat fluxes may be obtained mathematically. The number of
unknowns may not be greater than the number of temperature measurements for an
IHCP. When the geometry of a 2D or 3D heat transfer problem is discretized into the
F D M grid or F E M mesh, there are a number of nodes (elements) on the surfaces where
the heat flux should be determined. As the algorithm developed is implemented in a
numerical program to determine the heat flux during the cooling processes such as water
jet cooling, special procedures for specifying the heat flux distribution on the target
surface should be set up. This is a challenge for a 2D or 3D IHCP.

No detailed studies have been found in the literature regarding the above point. Most of
the published research didn't identify appropriate methods for specifying the heat flux on
the cooled surface in a 2D or 3D problem. In [23] the number of heat fluxes is implicitly
assumed equal to the temperature measurements, and a local uniform heat flux
assumption is used. This will be discussed in detail in Chapter 3 and Chapter 4.

32
2.4 T C Installation

As discussed above, the temperature measurements in water cooling experiments have


two purposes. The first one is to determine the cooling capacity of the cooling facility to
provide important information for process design and development. The other is to obtain
accurate and reliable surface heat fluxes or HTCs under the designed cooling conditions
by interpreting the temperature profiles.
Those studies, briefly reviewed in Section 2.2, are important for obtaining a real
temperature field but they have put little emphasis on the measurement and application of
temperature for an inverse heat conduction analysis. The crucial aspects for an inverse
analysis would lie in two categories: to keep the cooling condition undisturbed by TC
attachments and to make the temperatures relevant to benefit in reducing the effect of
random temperature noise on inverse calculations. Therefore, the installations of TC
should be designed accordingly.
Since a surface TC on the cooled surface may change the surface cooling condition as
well as cause large errors (Chapter 5), it will not be discussed here and only imbedded
TC of the bead type is considered in this discussion.
As shown in Figure 2.9, there are several parameters in installing TCs: the diameter D or
radius R of a hole for installing TC, the vertical distance h of TC to the cooled surface,
the horizontal distance / between two adjacent TCs, and the plate's thickness H.

Cooled Surface
s \
<i i

Figure 2.9 Parameters for TC installation

33
2.4:1 Horizontal distance

Among the measures to reduce sensitivity and to improve the accuracy of inverse
calculation results, the correlation of the temperatures at different locations should be
considered. As shown in Figure 2.9, i f the temperature at location TC2 is strongly
affected by the heat source at location T C I and vice versa, the influence from the errors
in temperatures at these two locations may be diminished. Thus, the inverse calculation
becomes more stable and reliable. This discussion makes it clear that the distance /
between two TCs should be kept to a reasonable range. Unfortunately, little work has
been done on this critical issue.
The importance of determining the critical distance between two TCs becomes more
crucial for the water jet impingement cooling process in which the cooling conditions are
highly transient and any material point outside of the impingement zone is not cooled
simultaneously. Analysis to experimental results shows that the progressing speed of the
water cooling zones is fairly slow and is 1 to 6 mm/s (Chapter 4). The temperature at
TC2 may hardly relate to the heat flux at location TCI i f the TC distance is around 16
mm, which was used in the literature [23-27].

2.4.2 Vertical depth

As TCs are implanted below the cooled surface through holes drilled from the plate's
backside, it is inevitable to have a distance between TCs and the surface. When
determining an appropriate vertical depth of TC, two conflicting factors should be
discussed.
The first factor is the suppression effect. The higher frequency fluctuations of boundary
conditions is increasingly damped and suppressed with the increasing of the depth. From
heat transfer conduction theory there is an exponential correlation in the form:

(2.20)

where Ar and A r a r e the amplitude of the temperature changes at the surface and at
o

depth x, respectively; a is the thermal diffusivity of the specimen; and/is the frequency
of fluctuations. If the detection limit is defined at 5% of the surface amplitude and the
depth is set at 2 mm, a sensor can capture only a frequency up to 10 Hz for steel plates.

34
To reduce the lagging effect in inverse calculation, TC should be as close to the cooled
surface as possible; i.e., the depth h in Figure 2.9 should be a small value. In reference
[23-27] h was set up at around 1 mm. For most of steels the thermal diffusivity is about
9.55x10" m /s, so the dimensionless time step can be 1.0 with a time step of 0.01s. This
5 2

number implies that the inverse calculation may not meet a difficult problem.
On the other side, as the depth is designed at a very small value, the surface temperature
above the void would be much lower than the undisturbed one, and a cold spot is formed,
which has a major influence on the cooling condition. The inversely obtained heat flux
may not reflect the real heat transfer mode even i f the measured temperature is without
noise, and is interpreted with a higher accuracy. In this respect the distance h should be a
value with which the surface temperature will be very close to the undisturbed, and as a
result the real heat transfer may be attained.

Therefore, it is desirable that a TC is implanted at a depth at which the surface


temperature is almost not affected, and at the same time the detection of a higher
frequency of boundary condition can be realized. Again, little work has been done on this
issue.

2.4.3 Effective T C depth

Although the surface temperature may be kept undisturbed with an appropriate depth, the
internal temperature field is still deformed when an imbedded TC is used. The
temperature at the internal sampling point is lower than the undisturbed. If the hole effect
should be taken into account for a general modelling, the heat transfer analysis becomes
three-dimensional. This is a challenge for an inverse calculation. Therefore, this hole
effect is neglected in most inverse calculations. This approach definitely causes an even
greater deviation of the calculated heat flux from the real one (Chapter 6).
Developed by Beck et al. [53] and frequently used in aerospace research [54], the "null-
calorimeter" technique postulates that the measured temperature at the tip of TC, which is
installed in a hole, will closely follow the undisturbed surface temperature i f the ratio of
the hole diameter to the TC depth is kept at 2.2.
Literature survey shows that the "null-calorimeter" concept has not been introduced into
the cooling process of steel. So it is of interest to investigate whether this idea is

35
applicable to steel cooling. Moreover, it would be useful to develop an approach in which
the hole may not be included in the geometry and its effect can be compensated by
adjusting the TC distance in the modelling of inverse analysis. This approach is called the
"effective depth approach (EDA)" in this study (Chapter 5).

36
Chapter 3

P R O G R A M FOR DIRECT AND INVERSE H E A T CONDUCTION ANALYSES

To study the water cooling of hot steel plates by jet impingement, a 2D F E heat transfer
analysis program is first developed; and then an algorithm for inverse heat transfer
analysis is formulated and implemented into the 2D F E heat transfer analysis program.
The Flux Zoning Method (FZM) is discussed and used for specifying the heat fluxes on
the nodes at the target surface. Parametric studies are performed with abrupt heat flux
inputs. Validation results show the proposed method and procedures are robust and
reliable and have an acceptable level of accuracy [55-56].

3.1 Program for Direct Analysis

3.1.1 Formulation

The modeling and experimental procedures are based on a 2D planar and axisymmetric
assumption. For completeness of the treatment, a brief outline of the equations for a
direct F E formulation for 2D planar transient conduction heat transfer problem is given in
the following section. A detailed account of this formulation and the formulation for
axisymmetric transient heat conduction problem may be found in reference [57-58].
The general governing equation for 2D conduction heat transfer problems, shown in
Figure 1, is written in the form:

d 57\ d S7\ b 8T

where T is the temperature, C; q is the heat generation rate per unit volume, W/m ; k
b 3
x

and k are the conductivities in the x- and v-directions, respectively, W/m-C; p is the
y

density, kg/m ; c is the specific heat, J /kg-C;


p t is the time, s ; and x, y are the

Lagrangian coordinates of the point.


The boundary condition (BC) may be one or a combination of the followings cases:
Prescribed temperature: This is an example of the Dirichlet B C . The prescribed
temperature T may be a function of time and boundary coordinate (spatial function):
s

37
T = T (x,y,t)
s onS, (3.2.a)

Prescribed heat flux: The specified heat flux (q ) may be a spatial function or a function
s

of time:

-k = q (x,y,t) onS (3.2.b)


on
a 2

where q is the specified heat flow per unit area, W/m , n is the normal of the point
s
2

considered. The prescribed heat flux is an example of the Cauchy's or Neumann BC. If
q is zero, it will represent a natural B C .
s

Temperature, T s eatflux,q
H s

Figure 3.1 Boundary conditions for a general conduction problem

Convection heat exchange: When due to contact with a fluid medium, there is a
convective heat transfer on part of the body surface, S4.

-k = h(T -T ) s f onS 4 (3.2.c)


on

where h is the convection heat transfer or film coefficient, W/m .C, which may be 2

temperature dependent; T is the surface temperature (C) on S , and 7} is the fluid


s 4

temperature (C), which may be a spatial or time function.

Radiation: Assuming a grey body, the B C is given by:

38
- 9L
k = e Jg* T*\ o n S 5 (3.2.d)
on
where e is the emissivity of the body's surface, a is the Stefan-Boltzmann constant
(W/m -K ), T
2 4
sr is the absolute temperature of surface Ss (K), and T er is the known
absolute temperature of the external radiative source (K). The radiation boundary
condition may be dealt with as a nonlinear convective boundary condition with an
equivalent temperature dependent film coefficient, K ; where:

K = a(T +T lT +T )

2
s r
2
e r sr er (3.2.e)

With the use of a weighted residual Galerkin procedure, the final F E equations may be
written as:

C 1 + KT = Q (3.3)

where C is the equivalent heat capacity matrix; K is the equivalent heat conduction
matrix; T and T are vectors of the nodal temperature and its derivatives, respectively; Q
is the equivalent load vector. Detailed expressions of the matrices in equation (3.3) are
given in Appendix A .
A general family of solution algorithms for equation (3.3) may be obtained by
introducing a parameter a where (0.0 < a < 1.0) such that

I+dAt rj. _ 1 /t+Atrj. J \ _ 1 ^-t+aAt rp I rp\


tr
^ 4a)
~ At aAt

t+aAt T=a t + A t T + (l-ayT (3.4b)

If a = 0, an explicit Euler forward method is obtained; i f a= Vi, an implicit trapezoidal


rule is obtained; and i f a = 1, an implicit Euler backward method is obtained.
Substituting equation (3.4) into equation (3.3) yields:

t+oAtr / i n "1 t+aAt/ , N t+aAtr^u - V+oAr/ \

where a * 0. The definition of all terms is given in Appendix A .

Depending on the value of a, the solution procedure may be either conditionally stable

(a<0.5) or unconditionally stable a > 0.5.

39
3.1.2 Implementation and validation

In the developed program, the geometry domain can be discretized into an assembly of
triangular and quadrilateral isoparametric 2D finite elements. The triangular elements
may have three or six nodes whereas the quadrilateral elements may have four, eight, or
nine nodes. The program is capable of dealing with a mixture of element types.
Various types of boundary conditions discussed above are adopted in the program. The
values of HTC or heat flux may be a function of both time and space. The heat flux may
be specified as surface or nodal input. It is worth noting that numerical experiments
showed that using the nodal heat flux is better than using the surface heat flux when an
inverse calculation with flux marching method (FMM) (Chapter 4) is involved.
Nonlinearities may arise from the dependence of thermophysical properties on
temperature, as in the case of a radiation boundary condition. In the developed program,
nonlinearities are handled in a step-wise staggered approach; i.e., the values of the
parameters at the current step are calculated based on the temperature at the previous step
and are assumed to be constant during the current step. The heat generation due to phase
transformation can be treated in a similar way, i f considered. In fact, heat generation is, if
not specially stated, not considered in all calculations in this study because the austenite
is transformed to martensite under the cooling conditions in experiments. The developed
program is capable of handling both steady-state and transient heat transfer cases.
Various verification cases have been performed to assess the accuracy, reliability and
stability of the 2D F E program for direct heat transfer calculations. Verification cases
were solved using both the commercial program A N S Y S [59] and the home program
developed; the differences between the results from the developed program and A N S Y S
were shown within less than 0.5% in all cases. Details of these verification cases and
results comparisons are omitted for space consideration.

3.2 Formulation for Inverse Analysis

3.2.1 Objective function

The key points discussed in Section 2.3 are summarized in the following for the
introduction of the proposed objective function. A n IHCP may be generally converted

40
into an optimization problem. The objective function of the optimization problem may be
considered as the sum of the squares of the differences between calculated and measured
temperatures. Because inverse problems are generally ill posed, the solution to the IHCP
may not be unique and would normally be sensitive to measurement errors. To reduce
such sensitivity, information at a number of future time steps is employed to estimate the
heat flux q' at the current step. And a temporary assumption, normally called a function
specification, would be normally considered for the values of heat fluxes at all time steps
considered. The function specification technique works as a regularization procedure and
stabilizes the solution process. To damp the solution fluctuation due to the measurement
errors, the objective function may be made more extensive by including more terms in
the expression. A commonly used scalar quantity in this regard is the zeroth-order term
based on the heat flux vector q with a weighting factor, a, which is normally called the
regularization parameter.

Based on the above discussion, an objective function in the least-squares method, with
the future time steps technique and regularization, may be expressed as follows:

F(q)^f (T -T ) (T -X)
d m c
T
m + af j q' q'
T
(3.6)
1=1 1=1

where X > X a r e
the experimentally measured and the theoretically calculated

temperature vectors at the / time step in a computation window of size N (Section 3.2.3
th

B), respectively; q'is the heat flux vector at the / time step, a is the regularization
th

factor and N is the number of total steps considered in the computation window.

The heat flux and temperature vectors are

q = [q' q 2
... q T (3.7a)

q'=fo' Qi - q'A T
(3-7b)

t = l X T lm ... T[J (3.8)

X=[T' C T\
2 ... TJ (3.9)

where L is the number of measurements, J the number of heat flux components that can
be determined for flux space distribution on surface. It is obvious that J must be less than
or equal to L, the number of measurements.

41
It should be noted that the dimensions of the heat flux vector q ' at each step is lxJ while
the total heat flux vector q is JxN as it includes the data in TV steps; and the temperature
vector T ' at each step is IxL.
Note that higher order regularization parameters involving the spatial derivates of q are
not adopted in this study. The rationale is due to the fact that heat fluxes in the water jet
impingement cooling may dramatically change in space, and therefore it is not
appropriate to confine the changing rate of the heat flux in space.

By defining AT = (T.J, - T' T - T


2 2
... T ^ - T J where A T has dimensions of Ix
C
N

(LxN), equation (3.6) may be rewritten as

F(q) = AT AT + aq q T T
(3.10)

3.2.2 General calculation form of heat flux

It should be noted that the temperature T* would be determined or affected only by the
heat fluxes q where m < k. Mathematically, T* may be expressed as an implicit function
m

of the heat flux:

t =f(q\q\...,q ) k
(3.11a)

or in a successive form as:

T* =/(T* ,q*) _1

T?-'=/C- ,q*->) 2

i (3.11b)
T =/(T ',q )
c
2
c
2

The following linear equation is valid by neglecting the terms of higher order:

t = T r ^ ( q - q V f t ( q - q * ) + -.- + ^ ( q - q * )
1 , 2 2 i i
(3.12)
oq oq dq
+

The values with an ' * ' superscript in equation (3.12) may be considered as initial guess

values that would ultimately lead the temperature T**.

42
Here, the first derivative of temperature T' with respect to heat flux q' is defined as the
c

sensitivity matrix:

dT'
X =
c

dq'
a (0
n a (i) n

(3.13)
a (i)
2X a (i)
22

dV
a (i) = ^

where i=l, 2...N, r =1, 2...L, and 5 =1, 2...J. The sensitivity matrix X ' is an LxJmatrix.

The optimality of the objective function may be obtained by letting 8F/dq = 0 (note

thatdF/dq should be done with respect to each component q', with i=l, 2...N), and as a

result the following set of equations is obtained:

5t;
a
5T a
N
f dT' ^ T

+ al (T -X')-aq
m
jt

(3.14)
(=1
7=1,2 TV

where q is the initial guess of heat fluxes, and T ' is the calculated temperature vector
J
c

with the initial guess values.


Recalling equations (3.12) and (3.13), equation (3.14) may be rearranged and rewritten in
the following form:

(X^ .Xq = q .+I)(q-q) = X AT-aq /


(3.15)

where X is labeled as the total sensitivity matrix for a multi-dimensional problem and has
the following form:

X 1
0 0 0
X 2
X 1
0 0
X (3.16)
: : 0
X N
... X 2
X 1

43
It should be noted that the dimension of matrix X is (LxN)x(JxN). Also, it is worth noting

that performing the calculation in equation (3.15) may be easily done in the entire time

domain, and no function specification for q' is needed. If the total sensitivity is known

and accurate, iteration may not be required to obtain a final solution.

3.2.3 Iterative and sequential algorithms

1. Function specification and iteration technique

The form of the governing differential equation for both the temperature T (x, t) and the
sensitivity coefficient X (x, t) is the same [39] and, therefore, the same F E program may
be used to calculate them. While this is efficient from the programming point of view, it
may not be practical, especially when the temperature time history is somewhat long, the
density of mesh is high, and the coefficients at all points are not all needed to get the heat
flux q.
A n alternative way is to calculate the sensitivity matrices at each time step for the target
points. Such a procedure would still require an extensive calculation time and a higher
cost when the number of time steps considered and/or the number of heat flux
components are reasonably large.

A perturbation algorithm [23] was used to obtain the sensitivity matrices X (only the
1

information at the current step was included in the work presented in [23]). First, a given

value q * is assumed for all components of the heat flux vector q ; the direct heat
1 1

transfer calculation is conducted to get the temperature distribution, say TrJ, for the given

future steps at each TC location. Then, one component of heat flux q , say the J
1
h

component, is increased a reasonable amount, such as 10%, to obtain new

temperatures, T ] . The ratios of temperature difference at each TC location to the

difference of the J h
heat flux component are the sensitivity coefficients. Such a

perturbation is repeated for each component of the heat flux q * until all sensitivity
1

matrix components of X are obtained.


1

The above method would be adopted in this study. By applying this approach, several

issues should be resolved. First, the heat fluxes q'*for /' = 2... TV in the consecutive steps

should be assigned. As mentioned in Section 3.2.1, the function specification would

44
stabilize the solution process. Moreover, it would simplify the calculation of the total

sensitivity matrix X . A constant assumption is used in this study, i.e., q * = q * = q* for 1


1+i 1

< k < n Ts- Therefore, all sensitivity matrices X ' for / = 1,2,...,TV may be obtained from
F

the above method by one assignment when the direct calculation is performed for N

steps.
The second issue is the nonlinearity. The whole sensitivity matrix X is independent of the
heat flux q only i f the conductivity k and specific heat c p are not functions of the
temperature or i f the average values for these quantities are used when the dependencies
on temperature exist. The thermophysical properties of most steels are temperature
dependent. If this kind of dependency is considered, all properties should be updated at
the beginning of each time step, which is time-consuming, especially for large size
models. Moreover, such changes in properties would not be very large and would not
significantly change the magnitude of X . Also, updating the material properties at the
beginning of each time step would be based on the temperatures T** obtained from the
initially given values of heat flux q*, which is essentially an approximation. The
assumption of a constant value is, therefore, justified. As a slight modification to the
above assumption, another option is to update the sensitivity matrix X every M steps (in
our numerical experiments, M=10). Both methods led to similar results of inverse
calculation, so the use of constant sensitivity matrix is justified.

Other factors affecting the accuracy of the calculated heat fluxes are the future
information and the addition of regularization parameters. To improve the accuracy due
to these factors, the iterative technique with convergence limits is adopted in this study.
Some modifications to equation (3.15) are to be considered. The term a q* (eventually a
"q in the iterative process) would lead to an increment A'q and a better estimation for the
first value of 'q. This term makes the calculation more cumbersome with very little
benefit in convergence and, therefore, will be neglected, and the equations may be written
as:

( X X + a I ) A q = X 'AT*
r 1 7

(3.17a)

q = q* +A'q (3.17b)

(X 'X + I)A' q = X A " - T


7 , r 1
(3.17c)

45
" q="q+A"q
+1
(3.17d)

where n is the number of iterations.

2. Sequential technique and function specification

Starting with either equation (3.15) or equation (3.17), a number of N flux vectors
q' for i = \...N, corresponding to each time step, can be estimated simultaneously.
When N is equal to the whole time step in the measurement, the developed method may
theoretically be used to obtain the entire time history of heat fluxes.
The focus of this study is, however, more on the application of the method than on the
method itself. During the water jet cooling process, the temperature at each measurement
point will drop sharply within only a few limited time steps. Such a sharp drop means
that a large load or heat flux vector would occur during such a small fraction of the time
domain. Other measuring points will have the same phenomena but at a different time
window. This means that i f the whole domain approach is used there will be a large
fluctuation in the load vector, and as a result the convergence of the solution may be
significantly affected. Therefore, the number of time steps N is normally less than 10.
A "computation window" of size N may be used sequentially to determinate the heat
fluxes in the span of time considered. To clarify this procedure, an example is illustrated
in the following where N 3 is assumed at the beginning of the analysis (the same i f
assuming at any time step). In the first sequence, the heat fluxes at the first three steps
may be obtained by iteration as:

q 2
= q !
+ Aq 2

[Aq J
{<
3

The subsequent three heat flux vectors q to q may be estimated i f the temperatures T
4 6 3

are considered as the initial temperature for the next sequence, and the computation
window moves 3 steps; and so on for each subsequent sequence.
The above procedure presents a hybrid approach between a whole domain one and a true
sequential one that will be addressed later in the following paragraphs. This hybrid
method implies that the heat fluxes at the previous iteration at each time step will be used

46
in the next iteration for the corresponding time step; i.e., at the (n+1) iteration "q will
1 1

be used for the first time step, "q for the second one and "q for the third one. There
2 3

will be no need for function specification.


In this thesis, a completely sequential approach with function specification is used. First,
the newly calculated "q is used for all time steps within the computation window after
1

the first iteration; i.e., a constant function specification is used for this computation
window. Second, the computation window moves one time step at the next sequence after
obtaining a convergent solution in the current sequence. Once again, using the previous
example, the first computation window consists of time step 1 to time step 3; and the
second one includes time step 2 to time step 4.

3.2.4 Convergence n o r m

In each time step, the iterative procedure is used until the inversely predicted temperature
T converges to the measured temperature T . The convergence criteria used to define
c m

the acceptance of the predicted temperature are based on an error norm for the
iteration defined by:

Error-norm" =|| AT || n
(3.19)
Two convergence criteria for ending the iteration process at each time step are used:
Error-norm " < 5T (3.20a)
or

lError - norm n+1


- Error - norm"
[
<e (3.20b)
Error-norm"
The values of ST and e depend on the measurement error level. The rationale behind
using absolute criteria is that while the norm at a given previous iteration is already very
small, the relative norm criterion is still not satisfied at the last iteration, and therefore, a
convergent solution can not be obtained. For example, i f assume ST=0.5, =0.05, Error-
norm " = 0.49 and Error-norm n + 1
= 0.46 (here n+l is the last iteration), then there is:

10.46-0.491
J ^ = 0.06 > 0.05
0.49 (3.21)

47
3.2.5 Flux zoning method (FZM)

The developed inverse analysis algorithm is a general approach that was implemented in
a F E program to determine the heat flux during the water jet cooling process. Special
procedures for specifying the heat flux distribution on target surface are considered and
implemented.
A common approach is to divide the target surface into several subregions; each
subregion may correspond to one temperature measurement. The same value of heat flux
is assumed for all nodes in a given subregion. The heat fluxes for various subregions are
generally not equal, i.e., qi^qi+i. The heat flux vector composed of all of heat fluxes on
each subregion may be determined using the above inverse algorithm. This approach is
named Flux Zoning Method (FZM). The model is schematically illustrated in Figure 3.2.
In this figure, two adjacent subregions are ploted.

There are eventually many nodes in each subregion, and the sampling (virtual
measurement) point should be located in the middle. Therefore, even number of elements
should be used. In fact, the number of nodes in each individual subregion may have little
effect on both the calculation procedure and accuracy, and may have a limit only from the
consideration of the element aspect ratio.

i zone q, (t)
t h

> (i+l) " zone q i (t)


1
i+

Figure 3.2 F Z M F E Model for inverse calculation

This approach is fairly simple and easy to implement. It may also be a good practice for
the pool cooling process or cooling surface that is covered simutanously by water, for
example, the impingement zone under the jet water cooling. In fact, the heat flux
determined using F Z M would be the average one for each subregion, there is

48
q (t) = - ^ f q ( x , t ) d x
i i (3.22)
b-a *

where a and b are the starting and end points of the subregion.
When the heat flux is nearly constant or does not dramatically change spatially in a given
subregion; the obtained heat flux would be a good approximation. However, this
approach may not be an appropraite method for the cooled surface outside of
impingement because it does not consider the progression of water cooling zones, which
will be discussed in detail in Chapter 4.
It is possible that the number of subregions is less than that of the measurements when
several TCs line up vertically (in our test configuration).

3.2.6 Flowchart

The developed algorithms and the F Z M are successfully implemented in the F E program,
which is capable of solving both direct and inverse heat transfer problem. Figure 3.3
shows a simplified flowchart for the proposed F Z M IHCP solution procedure. In the
procedure, the initial guess q* of heat flux q may be taken as zero or any other value.
However, the heat flux q obtained at the previous sequence may be used as the initial
guess of heat flux at the current step to accelerate and enhance convergence. The
numerical results show that such a setup of the initial guess of the heat flux is better than
a random guess.

3.3 Parametric Study to F Z M

The following section provides only highlights of the test cases performed for the inverse
analysis using F Z M and the results obtained.

3.3.1 Test procedure and data

The inverse analysis procedures are tested with verification cases that are designed to
simulate the cooling of steel plates on a ROT. Typical values of heat fluxes in these
applications are provided as an input for a direct analysis, and the direct calculation data
are used to verify the inverse analysis.

49
Setting of number of total time step

Open a computation window of size N

Calculation of
sensitivity matrix

Initial guess of q

Next
Direct Calculation of T
sequence

Next
iteration

Calculation
of new q

End

Figure 3.3 Flowchart for F Z M IHCP solution procedure

Two cases are tested to investigate the effects of number of total (future) time steps on
the elimination of damping and lagging behavior, and to identify the appropriate value of
regularization parameters. Triangular and impulse heat flux inputs, with the values that
simulate actual conditions in the ROT cooling, are used.
The first stage of the verification involves specifying input heat fluxes and solving a
direct heat transfer analysis problem to obtain the corresponding temperature field. The
second stage involves an inverse analysis in which the internal temperatures at the target
points (here, two locations) calculated from the first stage are used as virtually-measured
ones, called as the virtual internal temperature or directly the measured temperature later
on. From the virtal internal temperatures the heat fluxes and surface temperatures are
inversely calculated. A comparison of the inversely calculated surface temperatures to the
directly calculated ones is performed to further verify the accuracy and stability of the
inverse analysis algorithm.

50
To study the effects of measurement errors in internal temperatures on the inversely
calculated heat fluxes, random errors are imposed onto the calculated exact internal
temperatures with the following equation:

T =T +a-r
m exacl (3.23)

where T is the virtual internal temperature, C; r ct is obtained from the direct heat
m exa

transfer analysis, C; r is a normally distributed random variable with zero mean and unit
standard deviation; and cr is the standard deviation, C. In this work, the maxiumum
additive random error a is 3 C, making a total difference of up to 6 C. In the water
jet cooling experiment, the plate temperature ranges from about 900 C to 100 C. This
means that the relative error has a range of about 0.33% (at higher plate temperatures)
to 3% (at lower temperatures). In the following discussion, the error is presented in its
absolute value.
The general FE model used for both the direct and inverse analyses is a 2D one with the
same mesh, as shown in Figure 3.2. The model is 7 mm high and 20 mm wide. The top
surface is evenly divided into two parts, which means that the horizontal distance
between sampling points in each subregion is 10 mm. Two components of heat fluxes are
applied, and each of them is specified on all nodes in the corresponding subregion. Other
surfaces are prescribed as thermally insulated. Four points, two on the surface and two 1
mm beneath the surface points, are selected. The temperatures at the two internal points
are used in the inverse analysis to estimate both the heat fluxes and the temperatures of
the points on the top surface.
The material is assumed to be carbon steel with a density of 7800 kg/m , a conductivity
3

of 32 W/mC, and a specific heat of 470 J /kg-C for the whole time period. The thermal
diffusivity for this steel is about 9.55x10" m /s.
The time step used for both the both direct and inverse analyses is the same and is equal
to 0.01 sec. As mentioned in Section 2.4, the dimensionless time step is 0.955, which
maintains stable calculations in the inverse scheme.
In the verification tests, the norms for convergence are assigned to be of the order of
magnitude of the imbedded measurement error level. The number of future time steps is
determined by successive trials and is kept constant throughout the analysis. More

51
discussion about the choice of the regularization parameter will be given in following

sections.

3.3.2 Triangular heat flux inputs

Figure 3.4 shows the two triangular heat flux inputs for the two subregions. They have

the same profile and peak value but with a time shift o f 0.5 seconds to produce a spatial

temperature gradient in the longitudinal direction.

0.0 " 1 1 "


0 0.5 1 1.5 2 2.5
Cooling time.s

Figure 3.4 Triangular heat flux inputs

Inverse analysis is first performed with the assumption that there is no measurement error

in the internal temperatures. For this case, the calculated heat fluxes match exactly with

the assumed input values. The regularization parameter has no effect on the results and

may be assumed to be zero, and the number o f future time steps has little effect on the

inversely calculated heat fluxes.

Other cases have been studied with various levels of measurement errors and different

regularization parameters and future time steps. Figures 3.5-3.7 show the results of these

cases and the corresponding parameters in each case.

Figure 3.5 shows the effect of the measurement error level on the inversely calculated

heat flux with a fixed regularization parameter and a fixed number of future steps. A s

expected, the heat flux fluctuation increases with the increase of error in the virtual

internal measured temperature. However, the profile of the calculated heat flux is still in

52
good agreement with that of the exact input, and the accuracy of the predicted peak flux
value is good.

Cooling time.s

Figure 3.5 Effect of measurement error level


a = 1 .Oe-11 3 future steps
Error level: upper = 0.5 C; bottom = 1.0 C

The effect of the number of future time steps is illustrated in Figure 3.6. In these cases the
value of the regularization parameter is kept constant at 1 .Oe-11, and the error level is
fixed at 1.0 C. Fewer fluctuations of the calculated heat flux are evident with the
increase of the number of future steps. A n obvious improvement occurs when the number
of future steps increases from 3 to 5, whereas there is slightly less impact for increasing
the number of future steps to 5 or 7. It may be concluded that using 4-6 future steps is
appropriate and reduces fluctuation levels.

53
4.0
3.5
JfiAh
E 3.0 #
N

I 2.5
2.0
1.5
re
X 1.0 Al '
0.5
0.0
0.5 1 1.5 2.5
Cooling time.s

0.5 1 1.5 2.5


Cooling time,s

0.5 1 1.5 2 2.5


c
Cooling time.s
Figure 3.6 Effect of number of future steps
oc=1.0e-ll error level = 1.0 C
a: 4 future steps b: 6 future steps c: 8 future steps

54
0.0 I 1 1 1 1 1
0 0.5 1 1.5 2 2.5
a
Cooling time.s

0 0.5 1 1.5 2 2.5

Cooling time,s

0 0.5 1 1.5 2 2.5

Cooling time.s

Figure 3.7 Effect o f regularization parameter


4 future steps error level = 1.0 C
a:a=1.0e-10; b:a=1.0e-ll; c:a=1.0e-12

55
Figure 3.7 shows the effect of varying the regularization parameter value a on the

calculated heat flux. Numerical experiments showed that this is a much more delicate

parameter to control. Stable and accurate results are obtained for a range o f values o f

ct=1.0e-10 to 1.0e-12, which is close to the diagonal value of the sensitivity matrix of

1.0e-10. That is quite an interesting discovery.

Lowering the value of a to less than 1.0e-14 w i l l increase fluctuations of the inversely

calculated heat fluxes, while increasing the value o f a to greater than 1.0e-9 greatly

increases the required number of iterations, and in many cases causes divergence. It

should be noted that for all the cases considered above, the maximum difference between

the exact value of the surface temperature and the inversely calculated one is only around

510 C; i.e., the inverse calculation error level is about 10%. This implies that the

proposed algorithm might successfully recover both the heat flux and surface

temperature.

3.3.3 Impulse-like heat flux inputs

To further examine the capacity of the proposed IHCP scheme and its appropriateness to
the simulation of impingement cooling, a second test with impulse-like heat fluxes, as
shown in Figure 3.8, is performed.

-i- q1
q2

n 2.0

0 0.5 1 1.5 2 2.5


Cooling time.s

Figure 3.8 Impulse-like heat flux inputs

56
The impulse test is known to be the most stringent one for the IHCP algorithms. To
simulate the real conditions on a ROT, a vary high value of heat flux (in the order of
those experienced practically) is applied in a very small time period, equivalent to only
five time steps in the solution scheme. The same FE model and procedures discussed in
the previous sections are used here.

6.0
q1-10
5.0 *
<

E 4.0 >

s
X
3
3.0 A

g 2.0

>vyv
1.0

0.0
0.5 1 1.5 2.5
Cooling time.s

0.5 1 1.5 2.5


Cooling time.s

Figure 3.9 Effect of regularization parameter


3 future steps, error level = 1.0 C
a: a=1.0e-10 (ql-10, q2-10) and b:1.0e-ll (ql-11, q2-l 1)

57
0.5 1 1.5 2 2
a
Cooling time.s
5.0
I
4.5
Aq1-6
4.0
I
N 3.5 IT
E
| 3.0

g 2.5

S 2.0
Q)
1
1.5

1.0

0.5

0.0
*t
0.5 1 1.5 2.5

Cooling time.s
5.0 I
4.5 n1 R
n
L) I O
LJ -

4.0 q2-8

3.5
5 3.0
5
2.5
2.0
o 1.5
1.0
0.5
0.0
0.5 1 1.5 2.5
Cooling time.s

Figure 3.10 Effect o f future step number


<x=1.0e-ll, error level = 1.0 C
a: future steps 3 (ql-4, q2-4); b: 5 (ql-6, q2-6); c: 7 (ql-8, q2-8)

58
The inversely calculated heat fluxes under different conditions are shown in Figure 3.9 to
Figure 3.11. As in the previous example, the inversely calculated heat fluxes are almost
identical to the input ones when no measurement errors are imposed onto the internal
temperatures.
Also in this case, the regularization parameter has no effect on the inverse results and
may be assumed to be zero, and the number of future steps has little effect on the
inversely calculated heat fluxes.
Similar to the outcome of the previous example, it is found that with the increase of the
measurement errors in the input temperatures, the value of both the regularization
parameter and the number of future time steps should be increased to obtain acceptable
levels of accuracy.
Figure 3.9 shows a comparison between the calculated heat fluxes using different values
of the regularization parameter and a level of 1 C error in the measured temperature.
Also, the inverse results show that the peak values of heat fluxes are quite accurate. Even
when fluctuations of the calculated values are evident, the average value of the
fluctuation represents a good approximation of the exact input. It is also apparent that the
regularization parameter has less impact on the fluctuation of the results.
The effect of the number of future steps on the calculated heat flux for the impulse-like
heat flux case is shown in Figure 3.10. It may be seen that there is a somewhat large
fluctuation when four future steps are used, and that the fluctuation from the
measurement errors is damped gradually with the increase the number of future steps.
When 7 future steps are used the calculated heat flux value is quite close to the input for
the period during which the input heat flux is constant.
However, the peak value is also smeared when the number of future steps increases,
although it may still be captured accurately. The damping effect of the future time steps
increases significantly when the number of future steps is larger than the number of time
steps of the input impulse. This is logical since more weight for the steps with a lower
heat flux input is assumed.
With the increase of the measurement error level, both the number of future steps and the
value of regularization parameter should be increased to get reasonable results. Figure
3.11 shows the calculated heat flux for error level 3.0 C. The result is obtained by

59
using 5 future steps and setting a=1.0e-10. It is obvious that the fluctuation of the
estimated heat flux is quite large. Although the peak value seems to be adequately
captured, it is difficult to say that this is a real peak because such a peak may come from
a larger measurement error. It should be noted that other choices of future time step
numbers and regularization parameter do not produce better results. This indicates that
the calculated heat fluxes (if convergence is achieved) would be quite inaccurate if the
random error in the actual measured temperature is larger than 3.0 C.

.1.0 I I 1 1 1 1

0 0.5 1 1.5 2 2.5


C o o l i n g time.s

Figure 3.11 Calculated heat flux


Error level = 3.0 C, 5 future steps, a = 1.0e-10

From the several examples discussed above, it may be concluded that the proposed
inverse calculation is generally stable and accurate but very sensitive to the measurement
error. The calculated peak heat fluxes are in reasonable agreements with the inputs only
when the error is not larger than 3 . 0 C. It is also apparent that the number of future
time steps should be around 3, while the regularization parameter value should be
equivalent to the diagonal value of the sensitivity matrix. In addition, the inverse
calculation error can be controlled within 10% for both heat flux and surface temperature.

60
3.4 Summary

The developed FE program has the following features.

1) It can solve steady-state or transient, direct or inverse, plane or axisymmetric heat


transfer problems;
2) Various types of boundary conditions such as specified surface temperature,
surface and nodal flux, convection and radiation as well as heat generation may be
considered, all of which may be functions of location and time;
3) Temperature dependence of thermal properties may be taken into account;
4) Implicit and explicit solution algorithms are available as well as consistent and
lumped capacitance matrix;
5) Domain discretization with a mixture of different element types is available.

In the proposed algorithm for IHCP, the sensitivity matrix is calculated only once at the
beginning of the iterative procedure and used for all steps. The iterative technique is
adopted for compensation of the use of regularization parameter and the simplification in
the sensitivity matrix calculation. Two criteria are used for convergence. Parametric
study results indicate that the use of the least-squares method, the sequential function
specification and regularization is appropriate for IHCP in water cooling applications;
and that in the inverse calculation the fluctuation of the inversely calculated heat flux due
to random errors in the measurement data may be damped by increasing the number of
future steps and the value of the regularization parameter.
A l l calculations show that the proposed algorithm and procedures are very simple, useful
and effective, and that they can recover the heat flux and surface temperature histories on
the cooling surface with 10% errors when the random error in the internal temperature is
within 1 C.

61
Chapter 4

F L U X M A R C H I N G M E T H O D F O R IHCP IN W A T E R J E T C O O L I N G

The cooling characteristics of water jet, especially the progression of cooling zones, are
first described. A new approach, "Flux Marching Method (FMM)", is then developed to
better suit the features of impingement water cooling. Numerical tests have been
designed to explore the effectiveness of the F M M and to compare its performance with
the more commonly utilized flux zoning method (FZM).

4.1 Progression of Water Jet Cooling Zones

4.1.1 Stationary plate case

The cooling water will spread out radically (circular jet) or longitudinally (planar jet)
after it impinges onto the surface of a stationary plate. In the following sections the
discussion is confined to a single circular jet, although such a discussion is generally
valid for a planar jet; and it is based on experimental results in the literature [4].
For stationary cases, the cooled area on the top surface is axisymmetrical about the
stagnation point for a single circular jet. For a steady-state jet cooling process, the cooled
area on the top surface can be divided into five different heat transfer regimes:
impingement zone with the single-phase forced convection, nucleate-transition boiling
regime, forced convection film boiling regime, agglomerated pool cooling regime and
dry zone with radiation and convection, as shown in Figure 2.4.
For a transient process using a single stationary jet, the latter four cooling zones should
progress outwardly until the full plate is cooled down to the ambient temperature or water
temperature.
According to the visual records [25], the cooled area on the top surface can be divided
into three zones, as schematically shown in Figure 4.1a. The plate immediately turns grey
around the impingement zone as the water impinges onto the plate surface. Outside the
darkened zone the plate is still red hot. Shortly after the start of cooling, the grey area
around the impingement zone begins to turn black and the cooling water progresses
outwards. In this black zone, the cooling effectiveness is very high, and the surface

62
temperature drops sharply to a low temperature such that the boiling may not be possible.
The radiation and air convection cooling are dominant in the red zone.
The progression to the black colour is very rapid. As the black zone grows within the
grey zone, the grey zone itself also grows outwards but at a considerably lower rate.
When the size of the black zone gets close to that of the grey zone, the two zones grow
approximately at the same speed. Hence, while the plate is being cooled by a circular jet
there is a black zone centered at stagnation point, followed by a relatively small grey
circular ring around the black zone, and finally by a bright red zone. It is clear that the
black zone grows until it finally covers the entire cooled surface. Such a progression of
cooling zones implies that: 1) only the material points in the impingement zone are
cooled simultaneously at the same rate; 2) any material point outside of the impingement
zone is cooled sequentially at different rates, and the further the point from the stagnation
point, the later it is cooled; 3) any material point outside of the impingement zone is first
cooled in the grey zone with a low cooling capacity, and its temperature should decrease
at a slower rate, and a sharp temperature drop should take place as the front of the black
zone approaches. Although the above observations seem obvious; they have not been
fully considered in an inverse heat transfer analysis.

Moving direction of plate

a: Stationary case b: Moving case

Figure 4.1 Progression of cooling zones

It can be assumed that the initial black zone is equivalent to the impingement zone while
in the red zone radiation is dominant.
The progressing speed of the black zone front for stationary cases will be discussed in
Section 4.1.3.

63
4.1.2 Moving plate case

The abovementioned three cooling zones: black zone, grey zone and red zone still exist
for a moving plate; with some important differences that will be discussed below. First,
the cooling zones are symmetric only about the moving direction of the plate, and take a
pearl shape or a comet shape whose long axis is along the moving direction, as shown in
Figure 4.1b. On the side where the water flow and the plate move in the same direction,
the water layer thickness may decrease because the flow may be promoted by the plate
motion, while the water layer thickness may increase due to stretching toward the jet by
such a motion on the other side where the water flow and plate travel in the opposite
direction.
Second, it can be assumed that all of the three zones simultaneously move with the same
speed, and that the pattern and size of the cooling zones do not change. This means that
any of the material points will pass sequentially through the red zone, the grey zone, the
black zone, the grey zone and finally back to the red zone. This final return to the red
zone occurs because the cooling time is very short and the material somewhat distant
from the black and the grey zones does not have enough time to cool down and transfers
its heat to the cooled part. The temperature recovery can be verified by simple inspection
of the cooling curves, one example of which is shown in Figure 6.5.
Third, the relative moving speed of the water to the plate may be much smaller, even in
the moving direction, whereas the water speed definitely decreases in the transverse
direction due to the decrease of water flow in this direction. This leads to the assumption
that the actual velocity of the water may be neglected, and the plate is cooled by a
moving pattern (here a pearl shape) of cooling water, and that the moving speed of the
cooling pattern is fairy close to the plate's moving speed.

4.1.3 Progressing speed of black zone front in stationary cases

It is of interest to investigate the progressing speed of black zone in stationary cases. As


discussed in Section 2.1.1, it is suggested that the cooling area outside the impingement
zone is classified as a parallel flow zone, and in the parallel zone the wall jet flows
parallel to the surface. The water horizontal velocity is also assumed to be constant and
equal to the jet velocity, and the pressure is equal to the ambient pressure. This claim

64
seems problematical. Assuming that the distance of the nozzle exit to the plate is 1.0 m
and the water exit velocity is zero, the jet velocity will be around 4.4 m/s. If the water
velocity is actually equal to the jet velocity the material points that are outside of the
impingement zone and 0.440 m away from the stagnation point will be covered by water
in less than 0.1 second; and the temperatures at those points should drop fast to lower
values. The experimental data, however, reveals another trend.
Figure 4.2 [27] clarifies this issue. This figure shows 8 temperatures numbered TI to T8
measured using the configuration shown in Figure 2.6. In the test, T C I is located at the
stagnation point, and TC 1 and TC2 are roughly inside of the impingement zone whereas
the others are outside of the stagnation zone. Because the distance between TCs is fixed
at 15.9 mm, the progressing speed of black zone front can be calculated by estimating the
time intervals taken by this front to travel from locations TC2 through TC8 from the
measured temperature-time profiles.

1000 , 1 1 1 , 1 1

Cooling time, s

Figure 4.2 Example of measured temperatures


From left to right: T i to Tg
DQSK; Q = 30 1/min; T = 70 C; D =19 mm
w w n

65
If the color of the steel is black, its temperature should be lower than 500 C. As
discussed in Section 4.1.1, the temperature at one point drops dramatically as the black
zone front reaches the point and the plate portion turns black. The temperature profiles at
locations TC3-TC8 have nearly the same trend: a sharp drop from 700 C to 200 C.
Therefore, it may be assumed that the time interval between two adjacent temperature
curves at 500 C is the same as that for the cooling water to move across 15.9 mm. Figure
4.2 shows that the time intervals are approximately 17, 12, 6, 4, 5 and 15 seconds for
locations TC3 through TC8, respectively. This indicates that the black zone front
movement is not uniform, and the average speed is changing from 1 mm/s to 6 mm/s. A l l
of the experimental data show the same range for the progressing speed of the black zone
front. In other words, the progressing speed of the black zone front is of 10 mm/s and is
much less than the jet velocity.

4.2 Drawback of F Z M

Besides its simplicity and easy implementation, the F Z M is proven in Section 3.3 to be
reliable and capable of recovering the applied heat flux with a higher accuracy. Because
of its averaging feature this approach may also represent a good practice, and the
obtained heat flux would be a good approximation wherever the heat flux is nearly
constant or does not change dramatically in each subregion, such as the case in pool
cooling process or where the cooling surface is covered simutanously by the water, for
example, the impingement zone under the jet water cooling. However, this approach may
not be an appropriate one for the cooled surface outside of the impingement zone
because it does not consider the water movement.
The following verification analyses clearly depict this drawback of F Z M . The procedure
described in Section 3.3.1 is also used here. The model is schematically illustrated in
Figure 4.3.
The F E model is set for a 16 mm (width) x7 mm (height) workpiece with 20 elements in
both the width and height directions. The sampling point is in the middle of the plate
width-wise and 1.05 mm beneath the surface. So, there are 10 elements on each side of

66
the sampling point. The material has a density of 7800 kg/m , a conductivity of 32
3

W/m-C and a specific heat of 470 J /kg-C for the whole time period.

4C4C

1 1
1
|
1 I

Symmetric specification

4C2C

1
1
1
1
Un-symmetric specification

Figure 4.3 F Z M F E model for verifications

In the direct analysis heat transfer calculation, the full width of the top surface is assigned
a uniform heat flux of 2.5MW/m . The calculation time period is 2 seconds. These
2

conditions create a thermal history of the cooling temperature with a starting temerature
of 900 C and an ending temperature of 733.6 C. The temperature history, without the
imposing of random errors, is used as the virtual measured internal temperature for the
inverse analysis.
In the first round of inverse calculations, the heat flux is always assigned symmetrically
with the sampling point, and the element number on each side changes from 1 to 10.
For the identical input, the required heat fluxes in the inverse calculations are different
for the different heat flux assignments, as shown in Figure 4.4. In this graph, the x-axis
indicates the number of elements on each side, while the y-axis indicates the required
heat flux. It is clear that the heat flux is as high as 9.8 M W / m as the heat transfer is
2

localized within one element size at each side (totally 1.6 mm). When the cooling surface
is doubled the required heat flux dramatically decreases to 5.3 MW/m . This declining
trend becomes mild when continuing to increase the cooling surface from 2 elements at
each side. As 5 elements on each side are specified with the heat flux, the required heat

67
flux is 3.1 M W / m . The required heat flux decreases slightly from 3.1 to 2.5 MW/m as
the cooling area is doubled from 5 elements to 10 elements on each side.
The actual cooling process may be quite a local behaviour. However, the natural choice is
to apply the heat flux on the whole subregion if the F Z M is used. Thus, quite a large error
may exist.
The above results also imply that the cooling capacity 8 mm away from the target
location has less effect on the temperature at the target location. It may be inferred that
the actual 2D problem may be incorrectly handled by the F Z M as several individual ID
problems when the distance between two TCs is larger than 8 mm. In other words, the
distance between two TCs should be less than 8 mm if F Z M is to be used correctly.

o i i i i i i i i i IJ
0 1 2 3 4 5 6 7 8 9 10 11
Number of elements on each side

Figure 4.4 Heat flux required for the same cooling curve

Another round of analyses has been performed. In these analyses the total number of
elements is fixed at 6 on which the heat flux is specified. However, the heat flux
distribution is unsymmetric with respect to the sampling point. The scheme and the
results are shown in Figure 4.5. It can be seen that the symmetrically cooling case has the
highest cooling capacity, and that the more the cooling area is eccentric from the target
location, the more the heat flux is needed. The value of heat flux in the 1C5C case is 4.7
M W / m and is 16% higher than that in the 3C3C case (4.05 MW/m ) and is 88% higher
2 2

than that in the 5C5C case (2.5 M W / m , Figure 4.4).


2

68
It should be noted that the cooling pattern in the water jet cooling is fairly close to the
1C5C or the 2C5C during the period of sharp temperature drop. Assuming the
progressing speed of black zone front is 3.2 mm/s and 10 elements are used to discretize
the 16 mm length, only two-element lengths beyond the target location are covered by
water within 1 s that is needed for water to cover the two-element lengths. During this
time of 1 s the temperature at the target location would see a sharp drop. This cooling
situation may be very close to the 5C2C. When the F Z M is used, i.e., the 5C5C scheme,
there may be a large error in the estimation of heat flux from the same profile of
measured temperature.

4.8
2C4C
4.6 3C3C
J I L i i i ' i
4.4

s 1C5C
x" 4.2
3

ns
ai 4
X

3.8

3.6
3C3C 2C4C 1C5C

Distribution of heat flux components

Figure 4.5 Effect of distribution of heat flux components on the required heat flux

As discussed above, the F Z M does not correctly consider the cooling stage when the
cooling water is progressively covering the surface and a sharp temperature drop takes
place at short time interval.

4.3 F M M and Its Features

The above discussion suggests that it is crucial to find an appropriate approach that
accounts for the water movement. The flux marching method (FMM) is developed based
on this requirement. The key points of F M M are highlighted as follows:

69
1. As in the FZM approach, the target surface is divided into several subregions and
each subregion corresponds to one temperature measurement, as schematically
shown in Figure 4.6.

i zone
th
(i+l) zone
th

Figure 4.6 F M M FE Model for inverse calculation

2. There would generally be many nodes in each subregion, and the measurement
point corresponds to the first node in each subregion. The number of nodes in each
individual subregion generally may not be equal and may not relate to the
progressing speed of black zone front.
3. The heat flux at the first node is inversely calculated from the internal measured
temperature. Because the number of subregions is equal to the number of
measurements and only one heat flux in each subregion is considered as unknown,
the number of unknowns is equal to the number of measurements and the inverse
problem is solvable.
4. For each subregion, the following schemes may be applied:
a) The heat flux on a given node at the current time step is generally assumed to
be equal to the heat flux value on the neighbouring upstream node at the
previous time step. This is why the approach is named as flux marching
method.
The essence of F M M is like the wave propagation. The procedure is based on
the fact that the surface temperature in each subregion does not change much
before the instant at which water covers the surface; and on the assumption that
the small change of temperature field in each subregion would not affect the

70
heat transfer along the subregion surface; i.e., the heat transfer at the
downstream point that would be cooled at a subsequent time step would be the
same as that of an upstream neighbouring point at a prior time step. The heat
flux at the first node may be looked at as the impetus of a wave, and the flood
of heat flux would sweep the points in each individual subregion.
The progressing speed of black zone front may be different for each subregion,
as is the case shown in Figure 4.2. Therefore, to apply the approach discussed
above there should be different node numbers for each subregion because the
time step is fixed for the whole time domain.
b) The heat flux on the other nodes at the current time step may be generally
assumed to be the value on the prior node at the several previous time steps.
It is convenient and time saving to use the same mesh for different cases. And
from a practical point of view, it is sometimes necessary to solve the same
problem by using different time steps. These arguments postulate the necessity
to modify the approach stated in point 4-a to point 4-b as:

q r = q!. (4-i)
where the first subscript /' means z' subregion and the second subscript refers to
th

the node sequence number in the subregion. For example, qik is the heat flux on
the kth
node in the / subregion; the superscript means the time step. In this
th

equation, c is an integer.
c) The heat fluxes on a group of nodes at the current time step is equal to those at
the prior group at several time steps.
This situation is needed when a subregion must be discretized with a number of
elements that is larger than the number of time steps with which the water
moves from the first node to the last node in this subregion. Equation (4.1) can
be applied with c as a fraction.
d) General interpolation of heat flux
The procedures stated in points 4.a to 4.c require that the discretization of the
subregions be uniform along the water moving direction. Although the uniform
discretization is not difficult to realize, it may be necessary to consider a

71
condition with a general mesh. Thus, a linear interpolation procedure merits
attention and is discussed in the following:
i) Use the following equation to determine the time tsy that is needed for
water to move from the first node to the / h
node in the z subregion
th

ts = ^
s (4.2)
v
i

where dy is the distance between the first node and the / h


node in the z' th

subregion and VJ is the progressing speed of black zone front in the z' th

subregion.
ii) To find the time difference use

V=t -t k
S u (4.3)

where t is the total calculation time to the current /c time step and ty is the
k th

time difference to be used for interpolation.


iii) Use the following equations to find the heat flux at the / h
node in the z' th

subregion at the current /c time step:


,h

q^=q! when t ^ O
t- - t m

qj=qr+ U
A t m *q ' m + w h
en t <t <t
m
j J
m + 1
(4.4)

where m<k

This interpolation approach embodies the previous approaches and may be


used for all cases.

Two problems may appear when applying the F M M to the experimental cases: the step-
mismatch (overstepping or under-stepping) of heat flux and the jump of heat flux
(discussed in Section 4.4).
The overstepping of heat flux is illustrated in Figure 4.7. In the figure, #JL is the heat flux
on the last node of the i subregion and q
th
l+ \f is the heat flux on the first node of the
(z'+7) subregion. This figure shows that the heat flux on the last node of the z' subregion
th th

changes later than that on the first node of the (i+l) th


subregion. This implies that the last

72
node of the z' subregion is cooled down by water later than the first node of the (/+i)
th tn

subregion. This situation should not happen in practice i f the water jumping is not
considered. In fact, this situation is due to the inaccurate underestimation of the
progressing speed of black zone front at the z' subregion. A slight increase of the z' the
th th

flux marching speed should correct this disoperation. Similarly, the under-stepping
means that the heat flux on the first node of the (z'+7) subregion may greatly lag behind
,h

that on the last node of the z' subregion, and a small reduction of the z flux marching
th th

speed should be performed.

0 5 10 15 20
Time step
B qiL qi+1F

Figure 4.7 Overstepping of heat flux

4.4 Numerical Tests

In the following sections, the study focus is put on the relative effectiveness and capacity
of the F Z M and F M M in recovering input of heat fluxes that are designed to simulate
real-life situations of the water jet cooling process, from the input of virtual temperature
measurements for the zones outside of the impingement zone.

4.4.1 Test procedure and setup

The test procedure is the same as that described in Section 3.3.1. In the inverse estimation
stage, however, only the data at the current time step is used and no future time steps are
assigned; the regularization parameter and the convergence criteria are also fixed at the
same value for all cases; except for a special specification, no random error is imposed on

73
to the calculated temperature, and therefore the virtual measured temperature is without
measurement error.
Figure 4.8 shows the crucial part of typical heat fluxes in water impingement cooling that
is used as input for the direct heat transfer analyses. The nodes in the impingement zone
experiences the given heat flux change simultaneously, while the nodes outside the
impingement zone will experience this setup of heat fluxes with a time shift, which
simulates the water movement.
Ten cases are studied. The F Z M results of the first 7 cases are shown in Table 4.1. In
these cases, 4 peak values q and 4 moving speeds v are adopted. Case 2 will be discussed
p

in detail and the other three cases are presented in Section 4.4.

7
0 5 10 15 20 25 30 35 40 45 50
Step index

Figure 4.8 Crucial part of typical heat fluxes

Table 4.1 Test scheme and results using the F Z M

qp, Step V, Inversely calculated q , M W W


p

MW/m 2
shift mm/s
TCI TC2 TC3

Value Ratio Value Ratio Value Ratio


Case 1 10.4 8 1.25 10.4 1 6.86 0.660 4.68 0.450
Case 2 10.4 5 2 10.4 1 6.88 0.662 5.58 0.537
Case 3 10.4 2 5 10.4 1 7.24 0.696 7.28 0.700
Case 4 10.4 1 10 10.4 1 8.55 0.822 8.82 0.848
Case 5 7.8 5 2 7.8 1 5.16 0.662 4.25 0.545
Case 6 5.2 5 2 5.2 1 3.44 0.662 2.78 0.535
Case 7 2.6 5 2 2.6 1 1.72 0.662 1.38 0.531

74
4.4.2 F E model

The F E model used for both direct and inverse analyses in these numerical tests is a 2D
axisymmetrical one. The domain size is 7 mm in thickness and 80 mm in radial direction.
There are 80 elements in the radial direction and 9 elements in the thickness direction.
The elements are uniform in the radial direction to ensure an unbiased simulation of the
progressing black zone front, while they are variable in the thickness direction with a
relatively denser mesh close to the top surface. The 2D mesh is shown in Figure 4.9,
noting that the scale is not the same in the x- and v-directions.

Heat flux assigned


Impingement Parallel zone
7( in

TC
5

* g >
-9 3

a 1 2
>> B
* 1

0 20 40 60 80

X, mm thermally insulated

Figure 4.9 FE-model for numerical test

As illustrated in Figure 4.9, the domain is evenly divided into 5 subregions, and each
subregion has 16 elements. The first subregion is assumed to be the impingement zone
and the others are the parallel zone. Five sampling locations are set up, and the first one is
at the stagnation point and the other four sampling locations lie at the intersection
between the subregions. At each location, the temperatures at the two points, one on the
top surface and one 1 mm beneath the surface point, are sampled in the direct calculation.
The temperatures at the internal points are used in the inverse analysis to inversely
estimate the heat flux and the temperature of the points on the top surface.

75
There are 81 nodes in total on the top surface, and they are listed here as 1 to 81 and may
be in different calculation zones as stated in Table 4.2.
The sampling nodes are in the middle of each subregion for the F Z M (this is an
axisymmetrical problem, and node 1 is naturally the middle of the first zone) while they
are the first node for the F M M .
The material is assumed to have a density of 7800 kg/m , a specific heat of 480 J/kgC

and a thermal conductivity of 20W/mC.

Table 4.2 Node in calculation zone


Zone 1 Zone 2 Zone 3 Zone 4 Zone 5
Sampling Node 1 Node 17 Node 33 Node 49 Node 65
Direct Node 1 -16 Node 17-32 Node 33-48 Node 49 -64 Node 65 -81
FZM Node 1 -8 Node 9 -24 Node 25-40 Node 41 -56 Node 57-81
FMM Node 1 -16 Node 17-32 Node 33-48 Node 49 -64 Node 65 -81

4.4.3 Realization of water movement


The main objective of the current work is to investigate the capacity and accuracy of both
F Z M and F M M and their suitability for the water movement. The assumed boundary
conditions as shown in Figure 4.9 are such that all the sides except the top surface are
thermally insulated. This may not be practically accurate, but it should not influence the
coming conclusions.
The applied heat flux boundary conditions are calculated and applied on the nodes on the
plate's top surface basing on a time-space shift scheme. The scheme insures a space-time
marching approach that simulates actual cooling conditions and is briefly explained in the
following.
At the first stage (here 30 steps, 3s), the air-cooling is assumed for the top surface, and a
uniform heat flux (qi) is applied. At the second stage the water cooling starts and a heat
flux value (q2) is read in according to the curve input (see Figure 4.8) and is applied to all
the nodes in the impingement zone as well as the first node at the second subregion (later
these nodes are called impingement nodes). A l l other nodes would still be subjected to
air-cooling, i.e., the heat flux value of (qi). At the next time step, another heat flux value
(q3) is read in and is applied to the impingement nodes. The heat flux history on the

76
impingement nodes is shifted to the other nodes in the parallel zone according to the

assigned speed and element length. This space-time marching approach implies that the

cooling condition for all points is almost identical except for a time shift and a small

temperature drop due to the longer air-cooling period. A t any time instant, there is a

space distribution of heat fluxes, and one example will be shown in Figure 4.15.

4.5 Results and Discussion

For the condition stated in case 2, the time history profiles of input heat fluxes at the
locations T C 2 (TCI is the same as TC2) through T C 5 are presented by the dotted lines in
Figure 4.10. For simplicity, the same setting of heat flux is applied to all nodes with a
time shift.

12

10 I 1 1
TC2 JTC3 TC4 TC5
8

6
1
x
3
C 4
n
GJ
2

-2
' 10 20 30

Figure 4.10 Comparison of neat fluxes


Red lines: input; blue lines: calculated using FZM

The calculated internal temperatures from the direct simulation are shown in Figure 4.11.

The changes of the internal temperatures are fairly similar to those from the experimental

measurements in the stationary plates (see Figure 4.2) except for the large temperature

recovery, which happened in moving plate cases (see Figure 6.5). It may be stated that

the temperature profiles assemble the features for both stationary and moving plates.

7 7
The surface temperature profiles are almost identical to the internal ones except the
difference in values, and are not presented here.
The inversely calculated heat fluxes using the F Z M are also shown in Figure 4.10 with
the solid lines. It is clear that the calculated ones at locations TC3 through TC5 are nearly
identical, reflecting the fact that the input ones at these locations are the same. Therefore,
the following discussion is limited to locations TCI to TC3.

1000
900

| 600
| 500
H
.400
300
200
0 10 20 30 40
Cooling time.s

Figure 4.11 Internal temperature directly calculated in case 2

4.5.1 R e d u c t i o n of peak value using F Z M

There are two kinds of discrepancies between the input and the inversely calculated
values of heat flux profiles using the F Z M . First, the peak values may be underestimated.
For example, at location T C I , the calculated one is recovered exactly to the input because
the assigned heat fluxes on the nodes in this subregion are the same, and the average
value is equal to the value on each node. However, at locations TC2 and TC3, the peak
value decreases depending on the location and the progressing speed of black zone front.
As shown in Table 4.1 and Figure 4.12, the ratio of the calculated value to the input one
is around 0.78 and 0.48 at locations TC2 and TC3, respectively.
This ratio is not affected by the magnitude of input values and is fixed at 0.662 for
location TC2 and 0.53 for TC3 when the peak value decreases from 10.4 M W / m to 2.6
M W / m . This result shows that the F Z M has a limited capacity to capture the peak value.
2

78
0 2 4 6 8 10 12 0 2 4 6 8 10 12

Input q , MW/m
p
2
v^mm/s

Figure 4.12 Input q vs. ratio


p Figure 4.13 V vs. ratio
w

However, as shown in Figure 4.13, the accuracy of the prediction greatly increases when
the progressing speed of black zone front V increases. For location TC2, the ratio
w

slightly increases from 0.662 to 0.696 when the speed increases from 2 mm/s to 5 mm/s
while the ratio is 0.822 as the speed is 10 mm/s. The increasing intensity is around
24.5%. For location TC3, this ratio increases from 0.45 with a speed of 1.25 mm/s to 0.68
with a speed of 5 mm/s and then to 0.83 with a speed of 10 mm/s, and the increasing
intensity is as high as 88.4%.
It can be concluded that the accuracy of peak heat flux predictions using the F Z M
increases with the increase in the progressing speed of black zone front V . The rationale
w

behind this observation may be attributed to the evidence that a higher portion of the area
in each subregion is covered by water when the front moves faster, and that the cooling
condition would be similar between such areas. The reason why the increasing intensity
at location TC3 is higher than that at location TC2 may be because of the cooling
condition at the two sides of TC3 that would be affected by the water speed while only
one side of TC2 possesses a similar condition.

4.5.2 Shape change of heat flux using F Z M

The second kind of discrepancy between the input and inversely calculated values of heat
flux profiles using the F Z M is the change in shape of the heat flux profile and is
illustrated in Figure 4.14. The figure shows that the heat flux at location TC3 starts to
increase at time 9 s while the actual input increases at 12 s. This means that the calculated
heat flux increases much earlier than the input one.

79
Despite the inaccuracy of this prediction, it may be considered reasonable if compared to
the internal temperature profile and the averaging character of the F Z M . At time 9 s the
black zone front is 6 mm away from TC3 and will also affect the temperature at TC3, and
the internal temperature at TC3 starts a fast and sharp drop, as shown in Figure 4.13. The
temperature decrease at TC3 before time 12 s is due to the radial temperature gradient
and heat flow in the radial direction. This radial heat transfer is greatly underestimated
and is shifted to the effect of vertical cooling when using the F Z M . This may be one of
the reasons why the estimated heat flux using the F Z M increases at an earlier stage. On
the other side, there are two elements (2 mm long) in this region being cooled down at
time 9 s, so that the F Z M should capture this response.

1
12s-C

J I

' J 1
, / I

' / 1
10 20 30 40 50 60 70
Node index

Figure 4.15 Heat flux space distribution at time 12s

80
The averaging feature of F Z M may be further verified by the space distribution of the
heat flux, as shown in Figure 4.15, which shows the expected values (in solid line) and
the F Z M calculated ones (in dashed line). When the input heat flux changes from node to
node at each time instant, the calculated heat flux values using the F Z M for these nodes
are constant and do not reflect the actual conditions.

4.5.3 Mis-prediction of surface temperature using F Z M

Because the F Z M has an averaging character and may not be able to capture the actual
heat flux, the surface temperature predictions will also be inacurate. Figure 4.16 shows
the surface temperature differences when the calculated internal temperatures are
considered to be the virtual measured ones with an accuracy of 1 C. It can be seen that
the surface temperature difference may reach 150 C. Therefore, the F Z M may produce
an unrealistic relationship between surface temperature and heat flux, as shown in Figure
4.17.

200

150

100
o
jf 50
N
LL
V) 0
I-
o
-50

-100

-150

0 10 20 30 40
Cooling time.s
Figure 4.16 Surface temperature differences using F Z M

4.5.4 Effect of speed mismatch using F M M

It is to be first noted that i f the flux marching speed matches the progressing speed of
black zone front, the input heat flux may be exactly recovered for both the peak value and
the shape. A l l calcualtions relating to the input flux, including the suface temperature

81
predictions, would be produced with very good accuracy using the F M M . For the cases
considered, the F M M surface temperatures only differ by about 10 C with the same
convergent criterion of 1 C. This prediction accuracy is quite acceptable and will not
affect the relationship between the surface temperature and heat flux.

12

Surface temperature,c

Figure 4.17 Comparison of Ts vs. heat flux

A crucial point in using the F M M is to determine and to closely match the progressing
speed of black zone front. A mismatch may cause problems in recovering the required
heat fluxes. The influences of speed mismatch are numerically investigated.
Figure 4.18 shows the calculated heat fluxes at TC2 through TC5 with an increase of 5%
of the flux marching speed for the second zone only. This figure indicates that the heat
flux at TC2 has little change, while all the heat fluxes at TC3 through TC5 are severely
affected. At TC3, the heat flux first takes a negative value, indicating a heating process
before it starts to increase. Because the temperature at TC3 is reduced by water earlier

82
than it should be, an amount of heat should be input to maintain the temperature, and
therefore a negative heat flux occurs. This heating process would create the wrong
temperature field around the sampling point, and the peak value of the heat flux would be
mis-predicted.

It is expected that the temperature would be affected by the precedent heat fluxes. When
the inaccurate value of the predicted heat flux at TC3 marches to TC4 it will deform the
temperature field around TC4. Thus the heat flux at TC4 will be different from the input.
In this location the heat flux shows a small positive increase to balance the effect from
the negative one due to the marching scheme of the heat flux. The same phenomenon
repeats to some extent at locations TC4 and TC5.
If a mismatch of 5% is applied to all locations TC3 to TC5, the distortions of the
predicted heat fluxes would be intensified. Figure 4.19 shows one example of the impact
of this artificial mismatch on the calculated heat fluxes.
The influencing intensities of the mismatch level on the calculated heat flux accuracy are
depicted in Figure 4.20 to Figure 4.21 and Table 4.3. For all cases the peak value of input
heat flux is 10.4 M W / m in the direct calculation, and the flux marching speed at the
2

second subregion is artificially shifted with a deviation. The results at location TC3 are
presented.

83
20 22 24 26 28 30 32 34
Cooling time.s
TC5-ONE+5 -H-TC5-ALL+5

Figure 4.19 Intensification of higher speed mismatch

Figure 4.20 shows the results with a higher speed mismatch. With the increasing of speed
mismatch from +5% to +20% the unreal negative heat flux increases from -4.2 M W / m ' to
-8.6 M W / m , the absolute value of the latter being close to the input peak value. Because
2

of the negative portion the positive heat flux becomes 23% higher than the input value. It
is important to note that the positive portions of the heat fluxes are quite close to the input
heat flux if the negative portions are neglected.
The deformations of heat flux at TC3 are obtained i f the flux marching speed is less than
the progressing speed of black zone front. Figure 4.21 shows the deformations of heat
flux at TC3 with various values of imposed mismatches at the second subregion. When
the speed at the second subregion is slightly smaller than the input value, the marching
heat flux from TC2 increases at a later time than the setting and benefits the cooling
down of temperature at TC3. Thus less heat should be extracted away by the heat flux at
TC3 itself. Therefore, the heat flux at TC3 deviates widely from the input one. As the
slower speed mismatch increases, however, the deviation of the calculated heat flux to
the input one decreases, and the negative heat flux appears after the positive one. When
the progressing speed of black zone front is 20% less mismatched, the positive portion of
the heat flux turns closer to the input one.

84
15.0 15.0
12.5 12.5
10.0 I +5% 10.0 | +10%
7.5 7.5
\
V V
5.0 5.0
2.5 2.5
0.0 0.0
-2.5 -2.5
-5.0
V -5.0
-7.5 -7.5
V
-10.0 -10.0
10 15 20 10 15 20

Cooling time,s Cooling time.s

15.0 15.0
12.5 12.5
10.0
I +15% 10.0 Ii +20%

\V
7.5 7.5
5.0 I \ 5.0
2.5 2.5
0.0 0.0
n
-2.5 \ -2.5
X
-5.0 -5.0 \
-7.5 -7.5
-10.0 -10.0
10 15 20 10 '15 20

Cooling time.s Cooling time.s

Figure 4.20 Calculated heat fluxes with higher speed mismatch

Table 4.3 Calculated heat flux accuracy at TC3 with speed mismatch in F M M
Speed 5% 10% 15% 20%
Faster 11.6 111.5% 12.4 119.2% 12.7 122.1% 12.8 123.1%
Slower 5.59 53.8% 7.75 74.5% 8.69 83.6% 10.0 96.2%

The accuracy of the calculated heat flux for the above cases is presented in Table 4.3 and
Figure 4.22. Data shows that the calculated heat flux peak value at TC3 is about 11% to
23% higher than the input one with a higher speed mismatch. Compared to a 46.3%
deviation with the F Z M (Table 4.1 and Figure 4.12), this accuracy is acceptable. Note

85
that the heat flux at TC2 with the F M M is only slightly affected by the speed mismatch
while that with F Z M is also greatly influenced.

10
I
10

8 I 8

6 li " 1 5 %
6 Iv -20%

4 4 J
2 2

0 0

-2 -2

If
-4 -4

-6 -6

10 15 20 10 15 20

Cooling time.s Cooling time.s

Figure 4.21 Calculated heat fluxes with lower speed mismatch

It is clear from these calculations and analyses that the deformed profiles of the heat flux
at TC3 are different for an overestimation or an understimation of speed. This result is
extremely informative and is useful for figuring out whether the speed is overestimated or
understimated. The guideline for setting up the flux marching speed may be that the flux
marching speed should be slightly higher rather than lower than the progressing speed of
black zone front.

86
1.4

1.2

io 0.8
o
'3 0.6
on

0.4

0.2

0 5 10 15 20 25
S p e e d mismatch, %

Figure 4.22 Calculated heat flux accuracy with speed mismatch

4.5.5 Effect of flux j u m p using F M M

For this case, a new specification of heat flux on the top surface is performed in the direct
simulation. The heat flux profile shown in Figure 4.8, called here HP1, is simultanously
assigned to all nodes in the impingement zone bound by the TC2, while a profile HP2,
which is half of HP 1, is assigned to the nodes from the third subregion with a time shift.
While the time shift schedule is retained for the nodes in the second zone, the magnitude
of heat flux q is
q= ^ H P l l ^ H P 2
N2 N 2 (4.5)
i = l,2,-,N 2

where Ni is the number of nodes in the second zone (here A^=16).


This approach will make a smooth transition from a higher value of heat flux at TC2 to a
lower one at TC3. The resulting heat fluxes along the second zone are shown in Figure
4.23. In the inverse calculation, two methods are used to specify the heat flux on the
nodes in the second zone. One is just to assign the inversely obtained heat flux at TC2 to
all nodes, and the results are shown in Figure 4.24.

87
14

Nodel71 , _
12 i INOUCZI
TC2
10 | j Node25
E
\ 1 Node29
I
x"
\ \ \ \ Node33

\ \ \ \ I
6
*^
TC3

\A \
(0
O)

I 4

0 6 8 10 12 14 16

C o o l i n g time.s

Figure 4.23 Assigned heat fluxes along the second zone

As expected, the heat flux at TC2 is exactly recovered to the input. Note that this value is
also assigned to the node 32 in the inverse calculation, the left node to the node 33 where
TC3 is located. In the inverse calculation the temperature at TC3 will be lower than it
should be because the heat flux at node 32 almost doubles the one assigned in the direct
calculation. Thus, a heating flow is required to maintain the input temperature at TC3 and
then the negative flux appears. As discussed in the previous section, the wrong heat flux
at TC3 disturbs the temperature fields at the downstream nodes, so the heat fluxes must
be affected and then deviate from the input.

14
12 TC 9

10
8 TCI
E
6
E
4
\
x"
3

\ \ v^zuv p
2
ro
0
X

\
-2
-4
-6
10 20 30 40

Cooling time.s

Figure 4.24 Effect of heat flux jump

88
The solution to this problem is the second method. This method needs some iterations
and will be based on the results from the first method or from the F Z M . The inversely
calculated heat fluxes at TC2 and TC3 from the F Z M are also called as HP1 and HP2,
respectively. Then the heat fluxes along the second zone are determined according to
equation (4.5). If the heat flux at TC3 changes in the second iteration, a third iteration is
needed and so on. For this example, three iterations are needed to recover the distribution
of heat flux along the second zone and other values at TC4 and TC5.

4.6 S u m m a r y and Prospects

For a stationary water jet the cooled area on the top surface is axisymmetrical about the
stagnation point for a single circular jet. Only the material points in the impingement
zone are simultaneously cooled; any material points outside of the impingement zone are
sequentially cooled, and the further the point from the stagnation point, the later it is
cooled. The progressing speed of black zone front is much lower than the jet velocity.
For moving plate cases, the cooling zones are only symmetric about the moving direction
of the plate, and become a pearl or comet shape with long axis along the moving
direction. The velocity of water can be neglected compared to the plate velocity. The
moving speed of the cooling pattern is fairy close to the moving speed of the plate.
The F Z M is simple and easy to use and can be used in cases where the target surface is
simultaneously cooled by the same medium, such as the cooling in the impingement zone
or pool cooling, and in the air cooling stage.

A main feature of the F M M in comparison to the F Z M is the capture of the moving effect
as propagating wave. In this method, the cooling surface is still divided into several
subregions, and each subregion corresponds to one temperature measurement. The heat
flux at the first node in each subregion is inversely calculated from the internal measured
temperatures, and functions as the impetus of the wave; and the flood of the heat flux
would sweep the points in each individual subregion. Thus, the F M M takes the water
movement into account and may be more accurate in reflecting the actual cooling process

89
in jet cooling, and therefore may obtain a more accurate heat flux inversely calculated
from the measured temperature.
Using the F M M needs more attention when assigning the flux marching speed and
smoothing the jumping of the heat flux. Both overstepping and jumping of the heat flux
may cause disturbance in the heat flux profile with a virtual negative heat flux before the
real sharp increase of heat flux. Under -stepping (lower mismatch) may damp the real
sharp increase of heat flux, and casue a virtual negative heat flux after the real sharp
increase of the heat flux. A slight adjustment of flux marching speed may eliminate or
greatly reduce the effect from the speed mismatch. A factor should be used when
assigning the heat fluxes between two locations where the two inversely calculated heat
fluxes are jumping. It is to note that smoothing the distribution, rather than the speed
adjustment, should be performed first.
The results from the F Z M may benefit the smoothing operation when applying the F M M .
The F Z M and F M M may be adopted in one calculation such that the F Z M is used for the
air-cooling stage and the F M M for the water-cooling period.
In this study, although it is implemented in an F E program the F M M is a general
approach for assigning the heat fluxes along the cooling surfaces or subregions from the
design variables, and can be used in other computational methods. Moreover, it may be
possible to assume a heat flux profile with constants and to assign the heat fluxes from
the F M M , and then to determine these constants by an inverse method.

4.7 F l o w c h a r t for the I H C P Solution Procedure

Figure 4.25 shows the flowchart for the IHCP solution procedure combining the
developed F Z M and F M M into one inverse calculation program. The computation
window is composed of TV time steps including the current time step and all future steps;
and the computation window is shift only one time step after a convergent solution is
obtained. The parameter NSTEP is the total time step, and N F L O W is used to guide the
calculation method. The F Z M is always used first, and the calculation switches to the
F M M after the computation step reaches N F L O W . The F Z M is adopted through the
entire calculation as N F L O W is equal to NSTEP.

90
In this work, the sensitivity matrix is calculated only once at the beginning of the entire
inverse calculation, as the solid line after the operation "next sequence" expresses.
However, as the dotted line indicates, it can be newly obtained for each new step or every
M step to take into account the temperature dependence of thermophysical properties. In
this study, the latter approach is proved not to be necessary when an iterative technique is
adopted in this study. A constant function specification is used when calculating the
sensitivity matrix.
The initial guess q* of heat flux q may be taken as zero or any other value. However, the
heat flux q obtained at the previous sequence may be used as the initial guess of the heat
2

flux at the current sequence. Numerical results show that the choice of the initial guess of
heat flux has little effect on the iteration number. Either the initial heat fluxes or the
newly calculated ones should be assigned to the corresponding nodes according to the
adopted method, F Z M or F M M .
The iteration technique is used to get a more accurate solution. The first heat flux vector
at the previous iteration is used for all steps within the computation window; i.e., a
constant function specification is performed.

91
Open a computation
window of size N

Setup of N F L O W
Setup of NSTEP

Calculation of
sensitivity matrix

Using F Z M

Initial guess of q

Assign q on nodes

Direct Calculation of T Next


window
*
Next
iteration

Calculation of new q

Recording history of q

Figure 4.25 Flowchart for IHCP solution procedure

92
Chapter 5

APPROPRIATE INSTALLATIONS OF

T H E R M O C O U P L E S IN W A T E R J E T C O O L I N G

The reliability of TCs with a separation measuring junction in the temperature


measurement of water jet cooling, and the appropriate TC installations for inverse
calculation application cases have been extensively investigated using F E analyses [60-
61].

5.1 Error in the Measured Temperatures

While analyzing the experimental data of measured temperatures, significant


discrepancies are evident in rapidly decreasing temperature profiles and in the correlation
between the inversely calculated surface temperatures and the measured ones. One
example is discussed, which gives details of such discrepancies.
Experimental data [27] reveals that the heat transfer behaviour at points on the plate
within a 1.21.7*D around the stagnation point is identical, and may be considered as
n

one dimensional, where D is the nozzle diameter. Therefore, a small cylinder centered at
the stagnation point with the hole for TC is modelled for the inverse heat transfer
calculation (see Section 6.1 for the detailed geometry model). The temperature
dependence of thermophysical properties is considered.
A portion of typical temperature histories at the stagnation point is shown in Figure 5.1.
This figure shows the measured internal temperature, the measured surface temperature,
the inversely calculated surface temperature based on the internal temperature, and the
temperature difference between the measured surface temperature and the calculated one.
Both the measured surface temperatures and the internal ones have been filtered by a
simple but effective running filtration approach to reduce the noise level. In this
technique, the temperatures at a current time step plus 5 prior values and 5 subsequent
values are added together, and the average value is taken as the temperature at the current
time step. It should be noted that that the difference between the measured internal
temperatures and the calculated ones is less than 0.5 C.

93
It is apparent in Figure 5.1 that during air-cooling (/<11 sec) the measured surface

temperature and the internal temperature decrease simultaneously, and their difference is

negligible. This is acceptable since the two junctions are separated only by 1 mm, and the

plate has a uniform initial temperature. After cooling water impinges onto the plate

surface, the measured surface temperature drops very rapidly whereas the measured

internal temperature decreases at a much slower rate. The difference between these two

temperatures at this time point (t = \l sec) is up to 400 C, which means that the

average spatial temperature gradient is about 4E5 C/m. This sharp drop in surface

temperature is expected because of the high transient cooling condition.

900

800

700

600
af
3 500
s
|_ 400
E

300

200

100

10 11 12 13 14 15 16 17
Cooling,s
Figure 5.1 Typical temperature profiles during water-cooling
D =19 mm, Q =45 1/min, T =50 C, SS316
n w w

Ideally, the inversely calculated surface temperatures and the measured ones should be

approximately matched, even i f the internal temperature is lagged and damped, as proved

by the numerical tests in Section 3.4. This is not the case, however, as shown in the

figure; and there is a significant discrepancy in the magnitude and somewhat in the

decreasing profile of the surface temperature. The calculated surface temperature drops at

a much slower rate than the measured one. The difference is as high as 465 C at the

sharp dropping period of temperature, and it decreases slowly thereafter with the decrease

of absolute temperature. Even i f this large temperature difference can be partially

94
attributed to the lagging due to the highly transient condition, the nearly constant surface
temperature difference of around 100 C still suggests that there may be a significant
error in the measured surface temperature or internal one or both.
A direct heat transfer calculation using the same model has been performed in which the
measured surface temperature has been used as the boundary condition. In this case, the
calculated internal temperature is in good agreement with the measured one at the air-
cooling stage, while the difference between them jumps to about 250 C as soon as the
water hits the plate; the calculated temperature becomes lower than the measured one.
This significant difference is nearly kept unchanged during most of the water-cooling
period. This trend is slightly different from the trend obtained in the inverse analysis.
From the above data, both the direct and inverse calculations reveal that there is a
significant error either in surface temperature measurement (negative drift) or internal
one (positive drift) or in both cases.

14

12

1
E 10

I .
x" 1
B 6
1
2

0
10 11 12 13 14 15

C o o l i n g time.s

Figure 5.2 Inversely calculated heat flux

Figure 5.2 shows the changing mode of the inversely calculated heat flux with time.
Corresponding to the temperature change, the heat flux increases sharply to about 13.1
M W / m as soon as water impinges onto the surface, and then decreases to an average
2

value of around 2.1 M W / m that stays sort of constant during the time period of water-
2

cooling. Compared to the calculated surface temperature, the heat flux fluctuates more
widely because the latter is very sensitive to the errors in the measured temperatures. The
heat flux value of 2.1 M W / m is consistent with those reported in references [48, 63].
2

95
It may be concluded from the above discussion that the results obtained from inverse

calculations are reliable, and that most probably the surface temperature measurement is

not accurate.

5.2 Stationary Case [60]

5.2.1 Effect of T C wire conduction

Numerous numerical tests have been conducted where the length of wires and the value
of H T C or heat flux have been changed. For space consideration, only the typical case is
presented in the following sections. In fact, changing the length of wires and the value of
H T C or heat flux does hardly change the results presented.
The example of F E models used for the investigation of the effect of wire conduction is
shown in Figure 5.3. Axisymmetric conditions are assumed. A cylinder of 7 mm in height
and 7mm in radius is meshed. The choice of the model size is based on the value o f the
plate thickness (7 mm) and on the fact that the thermal insulation condition may be used
on the cylindrical outside surface, as mentioned above. The length o f bare portion of T C
wire is assumed to be 4 mm based on the installation configuration (Figure 2.6). In the
model, the wire is vertical to the plate although it is inclined in the installation. Such a
change w i l l not affect the simulation results. The material properties of the T C are taken
from reference [29]. Note that only one T C wire is included in one F E model; and two
separate models are used for two wires.

Figure 5.3 F E M model for analysis of effect of wire

9 6
The cooling process is divided into two stages: air-cooling and water-cooling. The heat
transfer coefficients for the TC wire are chosen to account for the convection and
radiation during air-cooling, and the convection during the water-cooling. The full length
of the TC wire is assumed immersed in water during the water-cooling stage. It should be
noted that a heat convection boundary condition, i.e., a specified HTC, is applied on the
outside surface of the TC wire rather than to specify a given heat flux. The rationale
behind this choice is to avoid having the TC wire temperature lower than the water
temperature. And the wire top is also assumed thermally insulated. The values of the
parameters are given in Table 5.1. A l l these values are typical for air-cooling and water
impingement cooling.

Table 5.1 Data for F E M model

Size Pi c, k, Ele- B.C.


(mm) kg/m 3
W/kgC W/mC ment Top Bottom &
Right side
Plate b=7 7800 470 31.5 1484 q=7.0e4 for air cooling Thermally
h=7 q=2.4e6 for water cooling insulated
Wire 1 d=0.1 8550 419 12.5 162 h=90 for air cooling
(K-Plus) h=4.05 h=4000 for water cooling
Wire 2 d=0.1 8700 544 58.6 162 h=90 for air cooling
(K-Minus) h=4.05 h=4000 for water cooling

Figure 5.4 shows the disturbed field of the surface temperature due to the attachment of
TC wires to the surface. At the air-cooling stage, the surface temperature is essentially
uniform. As soon as cooling water hits the plate surface and the wires, the temperature
field is considerably distorted for both K-plus and K-minus wires. Comparing the curves
in Figures 5.4a and 5.4b may conclude that the K-minus wire disturbs the temperature
field more than the K-plus wire.
Note that there is a thermal conduction along the wire, and the heat flux at the root point
of the wire will be larger than that on the plate surface. More heat is extracted from the
plate surface around the wires than other parts. This effect increases as the conductivities
of the wires increase and as the heat flux on the plate surface rises.
It is worth noting that the deformation of temperature field due to the attachment of the
wire is confined within 0.3 mm for both wires, which implies that the effect of wire
attachment on the temperature field may not be coupled i f the wires are 1 mm or more

97
apart from each other. Therefore, the separate simulations by attaching K-plus and K -
minus wires, respectively, are an appropriate procedure for modelling this effect.

950

900

850

800

750

700
rime=0.1
nme=0.1
650 =10.6 U =10.6

600
=10.7
ft =10.7
11.2 =11.2
550 -11.7 =11.7
=12.2 =12.2
500 =12.7
=12.7

450

400
0 0.1 0.2 0.3 0.4 0 0 0.1 0.2 0.3 0.4 0.5
a D i s t a n c e from Distance from
c e n t r e of wire,mm centre of wire,mm

Figure 5.4 Disturbed field of surface temperature around the wires


a- K-plus wire; b- K-minus wire

Figure 5.5 clearly depicts the effect of TC conduction on the surface temperature history.
In the figure, the reference temperature refers to the one without the attachment of wires.
The disturbance of the surface temperature is picked up in the first few milliseconds after
water impingement, and continues to the rest of the cooling period. Because of the radial
disturbance of the temperature field when the TC wire is attached, the reported K-minus
and K-plus surface temperature in the figure are those for the wire center at the junction.
900

800
o
o
8
3
700

ro
k_

8. 600
E
a>
t-
500

400
8.5 9.5 10 10.5 11 11.5 12 12.5 13

Cooling time, s

-Ref -K-Minus -K-Plus

Figure 5.5 Effects of wire conduction on surface temperatures

98
It is apparent in Figure 5.5 that during the air-cooling, the temperature difference between
the two wire roots is negligible, which means that the measuring junction is in an
isothermal condition. The temperature at the wire roots is slightly less than the reference
temperature. The output of the Seebeck voltage will present the true temperature.
However, such a situation is not the case after water impinges on the surface. During the
water-cooling stage, the temperature difference between the two wire roots is up to about
30 C; this occurs just after water impingement and holds approximately constant
thereafter. This indicates that the isothermal state for the measuring junction is not a valid
assumption. A detailed analysis of this point is presented in Section 5.2.3.
Another issue to be pointed out is that the average value of temperature at the wire root is
always less than that at a point 1 mm or more away from the root. The difference is about
45 C. This means that the measured output is less than the true temperature even if the
isothermal condition is assumed for the measuring junction.

The temperature gradients along the wires are shown in Figure 5.6. A larger temperature
gradient is evident for the K-plus wire. It is also observed that the temperature along the
wire quickly reduces to a very low level because of the very small size and volume.
It is worth mentioning here that the internal temperature is hardly affected by the
attachment of TC on the surface.

o O
o
o k_
~m TinB=0.1
(D~
CO
75
1 i_ - * 10.6
0 Q)
Q_ Q_ -B 10.7

E E 0 10.8
(D

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Distance from surface, mm Distance from surface, mm

Figure 5.6 Calculated temperature gradient along the wires


a- K-plus wire; b- K-minus wire

99
5.2.2 Effect of drilling a hole

In the previous sections, all calculations are carried out using a full-solid model.
However, in practice internal temperature measurements require small holes for the T C
installations. The existence o f such a geometric discontinuity will generally disturb the
internal temperature field as well as the surface temperature distribution.
In this section, such an effect is investigated using a hole-model and a cave-model. For
the two models, hollow cylinders of 7 mm in height and 7 mm in radius are meshed. The
F E models for these two cases are shown in Figure 5.7. The hole-model has the same
radius as the actual hole dimension in the experiment; that is, 0.8 mm in radius while the
depth is 4, 5 and 6 mm to study the disturbance intensity with different depths. The cave
is 0.8 mm in radius and 0.8 mm in height at a distance o f 6 mm from the bottom of the
cylinder. A finer mesh has been assigned around the hole and the cave to capture the
thermal gradient and disturbance of the temperature field. The material properties and
boundary conditions for the two models are the same as those specified in Table 5.1. A l l
surfaces due to the inclusion of the hole or the cave are prescribed as thermally insulated
because excellent insulation measures were taken in the test.

The results from the two models are shown in Figure 5.8 to Figure 5.10. In these figures,
H stands for the hole model, C stands for the cave model and B stands for the full-solid
model; index-0 is used for the surface temperature and index-1 (or 3) are used for the
temperature at 1 mm (or 3 mm) distance beneath the surface corresponding to the hole
depth.

x x
Figure 5.7 Example of F E M model for the effect of hole or cave

100
Figure 5.8 compares the temperature histories of a model with a 6 mm hole and a cave at
6 mm height with those of a full-solid model. It may be seen that the temperature fields
are almost identical for both the hole and the cave cases during the air-cooling or water-
cooling. This implies that the thermal insulation condition for the hole model is
appropriate, and therefore the hole model will be used for our assessments of other
factors.

900
850
800
750
5 700
B-0
E
a; 650 aB-1
I
t- 600 *He -0

550 -1

500
450 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13
8.5 9 9.5 V 10.5 11 11.5 12 12.5 13

Cooling time, s Cooling time, s

Figure 5.8-a Effect of 6 mm hole on Figure 5.8-b Effect of cave at 6 mm on


temperature filed temperature field
B-0: surface temperature without hole B-0: surface temperature without cave at 6 mm
B-1: temperature at 1mm depth without hole B-1: temp, at 1mm depth without cave at 6 mm
H6-0: surface temperature with 6 mm hole C6-0: surface temperature with cave at 6 mm
H6-1: temp, at 1mm depth with 6 mm hole C6-1: temp, at 1 mm depth with cave at 6 mm

Comparing the temperature history in the full model with the corresponding one of the
model with a hole reveals that there is no difference during the air-cooling period.
However, after water impinges on the surface, the difference between the two models
appears and starts to increase with the cooling time (Figure 5.9). The temperatures at the
hole bottom, where an internal TC will be attached, are lower than those at other points
that are at the same height but a few millimeters away. The figure shows a difference of
approximately 40C after 2 seconds from the water impingement. This indicates that the
TC reading will not be accurate.

101
It is important to note that the surface temperature distribution is also affected and that
the surface temperature in the hole case is lower than that without the inclusion of a hole.
The difference is about 25 C after 2 seconds from the water impingement.

900

850 '
tt
800
- Time=8.5
t 750 -10.6
*
= 700 - * 10.7
re s*-x-

|_ 650 _o 11.2

-o11.7
o 600
A -A
I-
- * r - 12.2
550 -e e -O
- e 12.7
500

450
0 0.5 1 1.5 2 2.5 3 3.5

Distance in X direction, mm

Figure 5.9 Disturbed temperature fields at 6 mm

900

850

800

o 750
o
700

so B B-0
a 650
B- 3 4

E
0) -0
h- 600 A H4
H4 -3

550

500

450
8.5 9 9.5 10 10.5 11 11.5 12 12.5 13

C o o lin g tim e , s

Figure 5.10 Effect of 4 mm hole on temperature field


B-0: surface temperature without hole
B-1: temperature at 1mm without hole
H4-0: surface temperature with 4 mm hole
H4-1: temp, at 1mm with 4 mm hole

102
The appropriate depth of the hole is explored here with extensive analyses. One example
is shown in Figure 5.10, and detailed discussions are presented in Section 5.4. It is found
that the surface temperature difference between the full solid model and the hole model
reduces with decreasing hole depth. When the depth is 4 mm for the hole of 0.8 mm in
radius, i.e., with 3 mm thick metal material between the top surface and the bottom
surface of the hole, the effect of the hole on the surface temperature is negligible. Also,
the cooling condition on the top surface is not altered, and the true surface temperature
may be obtained if an appropriate measurement is adopted.
Although the internal temperature field is still disturbed when the depth is 4 mm, the
magnitude decreases to about 15 C after 2 seconds from the water impingement.
A combined model where the surface TC wire is attached (as shown in Figure 5.3) and a
hole is set up (as shown in Figure 5.7a) is used to investigate the comprehensive effects
from both the wire attachment and the existence of a hole. The results are shown in
Figure 5.11. These results verify the expectation that the temperature field measurements
will be quite inaccurate when the TC is attached to the top surface just 1 mm above the
bottom surface of a hole. For example, the surface temperature difference increases to 50
C after 2 seconds from the water impingement, about 10 C higher than that with the
attachment of TC wires.

900

850

800

750

2 650
a>
a.
I
I-
600

550

500

450

400

8.5 9 9.5 10 10.5 11 11.5 12 12.5 13

Cooling time.s
Figure 5.11 Effect of 6 mm hole and K-plus wire

103
5.2.3 Effect of plate material
For an intrinsic T C , the two wires at a separation junction are not directly bonded
together, and they are soldered separately to the cooling surface. Therefore, the steel plate
between these two wires plays the role of a third metal. According to the law of inserted
metals [28], the Seebeck voltage will not change, provided that both wire ends of the
measured junction are at the same temperature. As calculated in Section 5.2.1, the
temperature difference between the two ends (ATi) is about 30-40C.
The output of the TCs with a separation junction is the weighted mean of the two
individual temperatures with the following correlation (see Figure 5.12):

e
AB\ ~ AB e
~ V\ (5-1)

where ABI is the measured output voltage; e B A is the would-be measured output voltage
if both wires have the same temperature which may be taken as the average temperature
of the two wires; V\ is the additional voltage due to a third metal whose ends have
different temperatures.

AT

Wire 2

Figure 5.12 The effect of the inserted third metal

Assuming that the reference temperature is zero, there are:

e =C -AT
AB a = C .(T T )/2
a ]+ 2 (5.2)

v = c A T ; = c . (r,-r )
x 2 (5.3)
where T\ and Ti is the temperatures at the roots of wires 1 and 2, respectively; C is the a

Seebeck coefficient of the TC; C a s is the Seebeck coefficient of the third metal, here, the

steel plate.

104
The absolute error of the measured temperature due to the third metal may be calculated
by the following equation:

C AT C
error = AT = ^AT,
L
(5.4)
C ATa C a
1

From equation (5.4), it is evident that the measured temperature error will become zero
when AT] is zero, that is, when there is no temperature difference between the two wires.
In our experiments, Type-K TC is used and the Seebeck coefficient of the two metal
wires is 30.8 uV/K for K-plus and -8.2 u-V/K for K-minus [29] respectively. Therefore,
C , the Seebeck coefficient of type-K TC is 30.8-(-8.2) = 39 u V / K . The Seebeck
a

coefficient of Fe ( C ) is 14.4 uV/K. Substituting into equation (5.4) with Ari=30^10 C,


as
o

the error of the measured temperature would be about 11-15 C. This implies that the
error due to third metal may not be critical.

By adding the errors calculated above, it can be seen that the total error in the surface
temperature is around 60 C due to the attachment of the surface TC on the surface 1 mm
above a hole for the imbedded TC. The error value is close to that for the lower surface
temperature but is quite different for the high transient period, as shown in Figure 5.1.
This suggests that there may be other factors causing the errors.

5.2.4 Effect of contact w i t h water

Water is a kind of electrical or electrochemical conductor. During the water-cooling, the


plate surface is covered by a layer of cooling water, and the bare TC wire will work as an
electrode, which forms a short circuit between the two TC wires and possibly between
the wires and the plate. "Contact" is, therefore, set up between the bare wires, the
individual wires and the metal surface. This implies that there are other "equivalent-
measuring junctions" other than the one on the metal surface. Therefore, the output of
surface TC may not be the true value even if the effect of wire conduction and the
existence of a hole are excluded. It can be inferred that the output will be lower than the
true one because the "equivalent measuring junctions" are at temperatures ranging
between 20C (the assumed water temperature and ambient one) and the surface
temperature of the plate. A detailed calculation is beyond the scope of this study.

105
5.2.5 Section s u m m a r y

The key points of the above studies are summarized in the following:
1. The accurate and correct output of a TC with a separate measuring junction is assumed
only when the isothermal conditions are guaranteed for both the reference and the
measuring junctions in the TC thermometry.
2. During the air-cooling, the attachment of TC wires on the surface has negligible
influence on the surface temperature distribution. The isothermal condition at the
measuring junction with a separation installation is approximately assured, and the top
surface temperature measurement may be assumed free of installation error.
3. In the case of water impingement cooling of hot steel plates, the conductions of wires
lead to two kinds of effects. First, the surface temperature filed is disturbed and
becomes non-uniform. The average temperature around the measuring junction would
be about 45 C less than the true temperature. Second, the measuring junction with a
separation installation is no longer an isothermal junction. A temperature difference of
about 30 C between the two wire roots may occur as the water impingement starts.
This difference will be held constant thereafter. The plate metal works as the third
wire. This may cause an additional difference of about 10-15 C lower than the
average temperature around the measuring junction. The combination of these two
effects leads to a measurement error of about 60 C lower than the true value. The
attachment of TC on the surface has little effect on the internal temperature 1 mm
below.

4. The disturbance of the temperature field due to the hole for the installation of internal
TC is not significant during the air-cooling. However, a hole with a depth of 6 mm
deforms both the internal and surface temperature fields in the case of water-cooling.
The effect would be to lower the temperatures at the top surface and at 6 mm height.
Without considering the wire conduction effect, the differences of temperatures at the
top surface and at 6 mm height are 25 C and 40 C after 2 seconds of water
impingement, respectively.

5. The deformation of the temperature field is worse when the TC is attached on the top
surface just 1 mm above the bottom surface of the hole.

106
6. The 4 mm hole will slightly affect the internal temperature field only. The hole depth
should be reduced as much as possible in experimental tests.
7. The contacting due to water may contribute to the errors in the surface temperature
measurement during the water-cooling. Some measures such as insulation to these
wires should be taken to eliminate or decrease such an influence.

5.3 Moving Case [61]

This section addresses the effects of the progressing speed of black zone front, heat flux
profile and magnitude, hole depth, and specific heat on the accuracy of temperature
measurements. Most of these analyses are performed using our own F E code, while a
small portion is carried out using the commercial F E package A N S Y S [59]. The plate
material is DQSK and the temperature dependence of thermal conductivity is considered.

5.3.1 Motivation of study

To obtain an overall view of heat transfer characteristics on the top surface, several TCs
should be installed at suitable locations to measure the temperatures. In our tests [27] of
water jet cooling of stationary plates, 8 intrinsic TCs, numbered T C I to TC8, with a fixed
radial distance of 15.9 mm, are used; and the two TC wires are separated approximately
1-1.2 mm. Roughly, the T C I and TC2 are inside the impingement zone, while the others
are outside of it, as shown in Figure 5.13. In such cases, the cooling water would reach
the individual TC wire at a slightly different time instant when the connection line (see
Figure 5.13) between the two wires is inclined or parallel to the water moving direction.
Even if the connection line between the two wires is perpendicular to the water moving
direction, there may still be a very small time difference in reaching the wires because the
water movement can not be perfectly radial. Therefore, the situation in which the
individual TC wire is cooled at a slightly different time instant always exists.

5.3.2 F E Model

1. Geometry

One of the critical issues in the F E modeling of this problem is to set up a typical and
representative geometry of the plate. In our tests the nominal thickness of plates is 7 mm,

107
and the diameter and depth of hole drilling from the bottom are about 1.6 mm and 6 mm,
respectively. These parameters are used in our FE model.
The existence of a hole will generally deform the temperature field. When considering
the effect of water movement, the cooling behaviour is three-dimensional. The initial
studies above show that the effect of a hole may be confined within a region 5 times the
diameter of the hole. Therefore, a block of 20 mm x 20mm x 7 mm with a centred hole of
1.6 mm in diameter and 6 mm in depth is used in the 3D model.

Nozzle TC wire
connection line

Water
Perpendicular Inclined Parallel

Water

Figure 5.13 Cooling water and TCs

In the current work, a variety of 2D models are used to save the calculation costs and
time. In these models, the thickness of the plate is fixed at 7 mm, the width of a two-
dimensional groove is 1.6 mm, and the depth is taken as 6 mm, 5mm and 4 mm,
respectively. These models will facilitate the study of the effect of hole depth.
The elements in 2D models are uniform at the plate's top surface to ensure an unbiased
simulation of the progression of water cooling zones, while they are variable in the
thickness direction, and a relatively denser mesh is adopted around the hole. A typical 2D
mesh is shown in Figure 5.14.

108
2. Specification of boundary conditions

The main objective of the current work is to investiagte the effect of the progression of
water cooling zones on the accuracy of temperature measurement using a TC with a
separation measuring junction. Therefore, it may be assumed that all sides, except the top
surface, are thermally insulated. This may not be practically true, but it should not
significantly influence the results obatined around the TC junction area.

Distanc

ST1 ST2

--- 1T1 IT?


1 mm
tfrff

Figure 5.14 Typical mesh and sampling points

For the 2D model, the heat flux boundary conditions are calculated from a space-time
marching approach, and then applied on the nodes on the plate's top surface. This space-
time marching approach has already been explained in Section 4.4.3. The method to
apply the boundary condition to the top surface in 3D model will be discussed in Section
5.3.3.

E 4

S
x- 3

0 50 100 150 200 250 300

Time Step

Figure 5.15 Types of heat flux profiles

109
3. Heat flux profiles
As discussed above, a formula is used for calculating the new heat flux. Three types of
heat flux profiles, named HT1, HT2 and HT3, are setup to investigate the effect of
possible heat flux changing modes in practice on the magnitude of the temperature
difference, as shown in Figure 5.15.
A l l three types of heat flux can be used to present the cooling process on a stationary
plate. It may be assumed that the heat flux at any location increases before the black zone
front reaches this point, and that the heat flux reaches its maximum value as the point
turns black. Afterwards, the heat flux of type HT1 is kept at the maximum value while it
decreases for types HT2 and HT3. After reaching the maximum, the heat flux of type
HT3 decreases continuously with a similar rate as that of the increasing one until it
reaches a very small value that is again assumed to be equivalent to the air-cooling with
some residual water-cooling. Type HT3 may also be looked at as the case with a moving
plate.
For each type of heat flux profile, three peak values are used, as shown in Table 5.2. The
notation is used in the following sections.

Table 5.2 Heat flux type notation


Notation Type Peak Value (MW/m ) 2

HT1-Q2 2.5
HT1-Q5 5
HT1
HT1-Q10 10
HT2-Q2 2.5
HT2-Q5 5
HT2
HT3-Q10 10
HT3-Q2 2.5
HT3-Q5 5
HT3
HT3-Q10 10

At any time instant, there is a space distribution of heat fluxes. The space distribution
changes step by step, and one example is shown in Figure 5.16. In this example, the water
cooling stage starts at step 50, and 40% of the top surface is already cooled by water at
step 75. At step 120, all areas of the top surface enter the water cooling stage, and the
peak of the heat flux locates at element 20. With the passing of time, the temperatures at
the back part decreases, the heat transfer capacity decreases, and then the peak of heat

110
flux moves forward. At step 175 the heat flux peak moves to element 70. At step 250 the
plate is cooled down for stationary cases, or the water cooling zones move away from this
part of the material for moving cases, and finally all the heat fluxes are at low values.

Step75
B Step 100
S t e p 125
*Step 150
Step175
Step200
Step225
Step250

Element index on top surface

Figure 5.16 Space distribution of heat flux

5.3.3 Results and discussion

In this study, four sampling points are considered, as shown in Figure 5.14. Two points
are on the hole's bottom surface, which is used for installing the TCs. The two points
correspond to the two roots of the separately soldiered TC wires.
Correspondingly, there are two points on the plate's top surface or the cooling surface.
The two surface points may be looked at as the two roots of the separately soldered T C
wires if surface TCs are used.
The paired two points are 1 mm apart from each other. The connection line between the
two points is parallel to the water moving direction.
The temperatures at those four points are defined as IT1 and IT2 for the inner points, and
ST1 and ST2 for the surface points. The temperature difference between IT1 and IT2 is
defined as IDT, and that between ST1 and ST2 is defined as SDT. Since IDT and SDT
may change with time, they are mostly regarded as the maximum values of these
temperature differences during the cooling process in the following analyses.

Ill
Initial simulations have been performed. In those cases, finer TC wires are included in the
mesh model. The results show that the attachment of wires has little impact on the
temperature differences at the un-simultaneous cooling conditions, and that the above
trends hold irrespective of the material properties of the TC wires. Therefore, the TC
wires are not included in the subsequent models.

1. Appropriateness of 2 D model as a replacement of 3 D model

In this section, both 2D and 3D simulations are performed using the commercial F E
package A N S Y S [59]. There are two 3D models used in the analysis: one with a hole of
1.6 mm in diameter, the other with a width-through grove of 1.6 mm in width. The 3D
model with a hole contains 56,140 nodes and 38,029 brick elements, whereas that with a
groove contains 8,389 nodes and 19,065 brick elements. The 2D model has a width of 20
mm and contains only 314 nodes and 520 triangle elements.
A region-jumping approach is used to apply the heat flux boundary condition because
the method described in Section 5.3.2 for 2D models can not be used in 3D models. Here,
the top surface is divided into four regions. At the initial cooling steps, all regions are air-
9 9

cooled (the heat flux is 30,000 W/m ), and then each region is water-cooled (10 M W / m )
sequentially. The two regions covering the two nominal roots of the TC wires are water-
cooled consequently with a time shift of 0.1 second. The equivalent progressing speed of
the water cooling zones is 10 mm/s.

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Cooling time,s Cooling time.s

Figure 5.17 Temperature profile results of 3D and 2D analyses

112
The data show that there is little discrepancy in results from the grove-3D model and 2D
model. So the results from the grove-3D are neglected and not discussed. The
comparisons between the 2D and 3D results are shown in Figure 5.17 and Figure 5.18.

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
C o o l i n g time.s

Figure 5.18 Comparison of temperature difference in 2D and 3D


Constant heat flux= lOMVvVm ; V=10 mm/s2

Figure 5.17 compares the temperature profile results of 3D and 2D analyses. It is


apparent that the temperatures at the two roots are not the same for all conditions. The
temperature difference is shown in Figure 5.18. It is obvious that the temperature
difference is quite small during air cooling and becomes significant after water cooling.
The temperature difference obtained from the 3D model is, generally, less than that
obtained from the 2D model. However, the discrepancy in temperature difference for the
interior TC is not significant, and the maximum temperature difference is approximately
identical. It is worth noting that the above results are obtained for an input heat flux of 10
MW/m . 2

From the above discoveries and discussions, 2D models are used in all analyses in the
following sections.

113
2. Effect of progressing speed of water cooling zones and depth of hole

The progressing speed of water cooling zones is a critical factor that influences the
temperature difference between the two wire roots. The faster the speed, the shorter the
time difference that the water takes to reach the nominal points of the two wires.
It is, however, not a simple task to obtain this progressing speed. The cooling water is
heated when it spreads outwards, and some of it becomes vapour. The ratio of water to
vapour is difficult to quantify in time and space, and it affects the progressing speed. In
addition, water pressure and plate surface roughness will also affect this speed.
A n alternative way to approximately quantify the progressing speed of water cooling
zones is to rely on the experimental data. As mentioned in Section 4.1.3, the water
movement is not uniform and the average speed changes from 1 mm/s to 6 mm/s. Taking
other factors and test conditions into account, it is assumed that the progressing speeds
are 5 mm/s to 40 mm/s. At those speeds, the root of one wire is cooled 0.2 second to
0.025 second later than the other root.

The effect of the progressing speed is shown in Figure 5.19. In these cases, the heat flux
type is HT3, and the input peak value of 5 M W / m is used.

Figure 5.19 Effect of the progressing speed of water cooling zones


Flux: HT3-Q5
a: Surface; b: Interior

It is clear that the temperature difference between the wire roots both on the surface
(Figure 5.19 a) and inside (Figure 5.19.b) decreases with the increase in the progressing

114
speed for all models with the different hole depths. Such a trend is more evident when the
speed increases from 20 to 40 mm/s. For the surface TCs in the 6 mm-hole model, the
temperature difference is as high as about 71 C when the speed is only 5 mm/s, and is
still about 31 C when the speed increases to 40 mm/s.
The error in the internal TC measurements is much less than that in the surface ones; for
example, with a 5 mm-hole model the interior temperature difference is 40 C while the
surface temperature difference is 61 C.
When the hole depth decreases from 6 mm to 5 mm the surface temperature difference
decreases. However, the situation for the 4 mm-hole model is nearly identical to that for
the 5 mm-hole model, which means that decreasing the depth to less than 4 mm will not
affect the temperature difference between the surface wire roots. This is consistent with
conclusion 6 in Section 5.2.5.
It may be concluded that the depth of the hole has a significant effect on the error; and
that decreasing the depth of the hole will decrease the temperature difference at a
relatively fast rate.
If the water is moving faster enough to cover the two TC wires roughly simultaneously,
there should be no error from the water movement. Based on the above results and
discussion, the following simple correlation is proposed:

A7/ = - ^ - (5.5)
yCi v
'

where AT is the temperature difference, C, V is the progressing speed, mm/s, and C\


and Ci are constants that may be a function of the hole depth, flux magnitude and other
factors.

For the 6 mm-hole model with HT3-Q5, the equation is AT = ^/^ . 06 Based on this

formula, the temperature difference is 6.4 C yet as the moving speed is as fast as 100
mm/s, while the temperature difference is as high as 100C as the moving speed is only 1
mm/s.
A special situation is considered where the water does not move and the front of the
cooling zone is distant from the middle point of two TC wires, say 15, 20, 25 mm, as
shown in Figure 5.14. For these study cases, the sampling time is fixed at 1.5 seconds

115
because the temperature difference steadily increases with time in this condition, and the
heat flux for the cooling zone is kept at 5 MW/m . 2

Figure 5.20 shows the effect of stopping distance. It is interesting to point out that the
temperature difference for the interior TC is always larger than that for the surface one
during the cooling time for this case. And it is clear that the closer the front of water
cooling zones is to the first wire root, the larger the temperature difference will be, and
that the relationship between them is almost linear.

80

75

60
* 55
Q.
I 50
I-

45

40
10 15 20 25 30
Distance, mm
Figure 5.20 Effect of stopping distance
6 mm hole-model

3. Effect of heat flux magnitude

Figure 5.21 and Figure 5.22 show the effect of heat flux magnitude on the temperature
difference in the interior (IDT) for a 6 mm-hole model with the flux type HT3. The
figures indicate that when the progressing speed is fixed at 20 mm/s, the temperature
difference can be neglected during air-cooling. As the cooling water gets closer to the
first wire, the temperature difference starts to show up. The maximum temperature
differences are about 17 C, 32 C and 59 C with respect to different heat flux peak
values. It is quite possible in industry that the peak value of heat flux could be as high as
10 MW/m or more. At such a condition the temperature difference is as high as 59 C
2

even if the two points are only 1 mm apart. This is a huge difference.
Another important point to note is that the maximum temperature difference appears
earlier as the heat flux magnitude increases. For the lower heat flux case, the maximum

116
temperature difference appears at about 0.86 second, which is later than the time instant
when the water reaches the first wire (0.23 second), and is also later than that when the
heat flux reaches its peak at the first wire location (0.73 second). For the medium and
high flux cases, the maximum temperature difference occurs at about 0.72 second, earlier
than the time instant when the heat flux reaches its peak at the first wire location (0.73
second).

70

Cooling time.s

Figure 5.21 Effect of heat flux magnitude


V=20 mm/s; 6 mm-hole model

70

o i ' ' 1 1 1 1 1 1 1
0 5 10 15 20 25 30 35 40 45
Progressing speed,mm/s

Figure 5.22 Effect of heat flux magnitude


and progressing speed
6 mm-hole model

117
It is clear that the temperature difference between wire roots decreases with the increase

o f the progressing speed o f water cooling zones when the heat flux type is fixed. For a

peak value o f 2.5 M W / m , the difference is about 36 C with a speed o f only 5 mm/s,

and is about 26C (10 C less) when the speed is doubled to 10 mm/s. A s the speed

increases to 20 mm/s, the difference becomes about 17 C. When the peak value is

doubled to 5 M W / m , all the values o f the temperature differences are also approximately
2

doubled (the corresponding values are 66, 48 and 32 C). There would be a linear

correlation between the peak value o f the heat flux and the maximum temperature

difference.

4. Effect of the heat flux type and the depth of hole

The effects o f the heat flux type and the hole depth on the temperature difference are

shown in Figure 5.23. In these cases, the peak value o f heat flux is 5 M W / m , and the

progressing speed o f black one front is 20 mm/s.

HHTI
SHT2
HHT3

6MM-S 6MM4 5MM-S 5MM4 4MM-S 4MM4


Model and location

Figure 5.23 Effect o f the depth o f hole and heat flux type
q =5 M W / m ; V=20 mm/s
p
2

It is apparent in Figure 5.23 that there is no significant variation in temperature

differences due to the different patterns o f the heat flux for both the top surface and

interior points. This implies that the maximum temperature difference depends mostly on

118
the peak value of the heat flux, which is the same as the conclusion in the previous
section.
It can also be seen that the temperature difference decreases as the hole depth decreases
(since it takes less time for the cooling water to reach the wire). Take the interior points
for example; the maximum temperature difference is about 32 C for a 6 mm hole. The
corresponding value decreases to 17 C and 10 C when the hole depth decreases to 5
mm and 4 mm, respectively. It may be inferred that the decreasing rate of maximum
temperature difference becomes small when the depth of hole decreases.
In contrast to the situations at the interior points, the maximum temperature difference
between the top surface points slightly decreases from about 42 C for a 6 mm hole to
about 40 C for a 5 mm or a 4 mm hole. This means that the surface temperature
difference is hardly affected by the hole depth, which is again consistent with the findings
in Section 5.1.

Similar trends hold for the time instant when the maximum temperature difference takes
place. It can be easy to find from Figure 24 that the time instants are almost at the same
value for the surface TCs in all cases, and that they steadily decrease with the deepening
of the hole.

6MM-S 6MM-I 5MM-S 5MM-I 4MM-S 4MM-I


Model and location

Figure 5.24 Time instant vs. the hole depth and heat flux type
q =5 M W / m ; V=20 mm/s
p
2

119
5. Effect of the initial temperature and the specific heat

The plate initial temperatures in our current stationary tests have been mostly set to
around 800-900 C. For a moving plate, the plate passes the water jet several times; the
initial temperature at each pass varies and may be as low as 500 C. It is, therefore,
important to study the effect of the different initial plate temperatures. In the following
cases, the heat flux type is HT3-Q5, and the progressing speed of water cooling zones is
assumed to be 10 mm/s.
The numerical cases studied indicate that the initial temperature has little impact for both
the surface and interior temperature measurements, as shown in Figure 5.25. The
maximum temperature difference between the surface TC wire roots only decreases from
around 42 C to 38 C as the initial temperature decreases from 890 C to 610 C. For the
same drop of initial temperature the maximum temperature difference between the
interior TC wire roots differs only by about 2 C.

45
o

600 650 700 750 800 850 900


Intial tern perature,C

IDT *SDT

Figure 5.25 Effect of initial temperature


Flux: HT3-Q5; V=10 mm/s

It is generally known that the temperature field does not change with the initial
temperature, and depends mainly on the boundary conditions and time if the properties of
material are constant. Accordingly, it may be concluded that the above small change of
maximum temperature difference may be attributed to the change of material
conductivity. A linear formula is used to calculate the conductivity, and the value

120
increases with the lowering o f temperature. That is, the average conductivity for a lower

initial temperature case is higher than that for a higher initial temperature case. A higher

conductivity w i l l decrease the temperature gradient, which reflects a small temperature

difference.

The effect o f the specific heat on the maximum temperature difference between the T C

wire roots is depicted in Figure 5.26. In these cases, the heat flux type is also H T 3 - Q 5 ,

and the progressing speed o f water cooling zones is assumed to be 10 mm/s. It can be

seen that the temperature difference for both the surface and interior measurements

decreases approximately linearly with the increase o f the specific heat. Such a trend is in

agreement with the understanding that the higher the specific heat, the larger the heat

capacity and the less the temperature drops.

I
45

V 40
oT
o
| 35

jE

3 30
e
a

| 25

20

450 500 550 600 650 700 750 800 850 900

Specific heat, J/kgC

Figure 5.26 Effect o f specific heat


Flux: H T 3 - Q 5 ; V = 1 0 mm/s

5.3.4 Section summary

It can be concluded that the isothermal condition for the T C measurement does not exist

for the separate roots o f the T C wires, especially when the connection line o f the two

roots is parallel to the water moving direction. The maximum temperature difference

between the T C wire roots w i l l be affected strongly by the progressing speed o f the water

cooling zones and the heat flux magnitude, and mildly or slightly by the heat flux profile,

the hole depth, and the specific heat value.

The temperature difference between the two roots may reach a significant level for both

the surface and interior TCs, especially for the situations in which a deeper hole (closer to

121
the cooling surface) is used to install TCs, and the water moves slowly with a higher
cooling capacity. Therefore, it may be concluded that TCs with a separation measuring
junction should be, generally, avoided in the temperature measurement of water-cooling
in the moving cases, i.e., when the water moves or the plate moves or both.

5.4 Appropriate T C Position

The appropriate distance / and depth h, as shown in Figure 5.27, are investigated using
direct F E analysis in this section. The cooling conditions are designed for the
impingement cooling using one circular water jet. The "null-calorimeter" technique is
examined, and an effective depth concept is suggested through intensive numerical
simulations. Correlations for the minimum TC depth and equivalent depth are provided.
Note that the TC is of the bead type.

5.4.1 General modelling aspects

Axisymmetric conditions are assumed in this study, although this assumption is not
necessary and does not affect the validness of results. Two kinds of FE models are used,
one being the hole model as shown in Figure 5.7, the other being a solid model. As an
example, the solid model is a cylinder of 7 mm in height and 8mm in radius and is used
as reference in most cases for comparisons.
For both models, only the top surface is specified with heat flux as the boundary
condition; and the thermal insulation condition is used on all other surfaces. In the direct
calculations, the applied heat flux may be different constant values during the entire
cooling process; it may be of the heat flux profile shown in Figure 5.27, which is a
simplified version of Figure 4.7 and called a practical input in the following sections.
A transient cooling process is assumed for all cases and the time span is 2 seconds. The
rationale for the span of 2 s is that in all experimental tests the material points are already
cooled down significantly within 2 s after the cooling water hits the top cooling surface.
If not specifically mentioned, the material of plate is the carbon steel DQSK, and the
temperature dependence of conductivity is considered.

It would be too much work to compare all the temperature values during the cooling
process from the two models, and may not be practical and necessary. In this study the

122
temperatures at the cooling end, i.e., at time 2 s, are compared for all direction

calculations. Some point-to-point comparisons show that the above approach can be used,

and no significant difference exists for the entire cooling process.

12

E
I 8

x"
3
= 6
re
f 4
3
Q.
C
~ 2

0
0 0.4 0.8 1.2 1.6 2
Cooling time,s
Figure 5.27 Input o f practical heat flux

5.4.2 Appropriate horizontal distance

In this subsection, the top surface is equally divided into two subregions; and one

sampling point is located in the middle o f each subregion and 1.0 m m beneath the top

surface. A constant heat flux o f 2 M W / m is applied only to the left subregion (relating to
2

T C I ) and thermal insulation is specified on the right subregion (relating to T C 2 ) . The

initial temerature is 900 C and a constant conductivity o f 30 W/m-C is used. Note that

the width o f the surface with the assigned heat flux increases with the increase o f T C

distance. Eventually this width is equal to the T C distance.

The calculation results are shown in Figure 5.28. A s can be seen, the temperature T C I for

the sampling point directly under the specified heat flux decreases dramatically when the

T C distance increases from 1 mm to 10 mm. This is because more material at the two

sides o f the sampling point is involved in the cooling process. A s the T C distance

increases from 10 m m to 16 mm, the temperature T C I shows little change, which means

123
that the cooling process at the points 5 mm or more away from the sampling point has
little influence on the temperature profile at the sampling point.

920 , 1

740 I 1 1 1 '

0 5 10 15 20
TC distance I,mm

Figure 5.28 Temperature vs. TC distance

The tempertaure TC2 at the second sampling point more clearly illustrates the distance
effect. When the cooling source is close to the sampling point, the temperature may
decrease slightly. As the TC distance increases to 10 mm and more there is nearly no
drop in temperature at that sampling point.
This calculation uncovers the following important inferences. As the sampling points are
10 mm or more apart and the top surface is accordingly divided into subregions in inverse
calculation using the F Z M , the heat fluxes may not be interactive; in other words, the
other temperature profiles may not stablize the inverse calculation. Thus, the 2D problem
in reality may be incorrectly handled by the F Z M as several individual ID problems i f
the distance between two TCs is larger than 10 mm. In other words, the distance between
two TCs should be less than 10 mm if a 2D problem should be fairly correctly reflected
by the F Z M .
From these results and those in Section 4.2, it may be concluded that the horizontal
distance of TC should be less than 10 mm to attain a real 2D inverse simulation. That is
the basis for the model geometry setup in Section 3.3.1.

5.4.3 Appropriate vertical depth

124
As mentioned in Section 2.5, it is preferred that a TC be implanted at a depth at which the
surface temperature is nearly unaffected and at the same time the detection of a higher
frequency of boundary condition can be realized.
Figure 5.29 shows the minimum depth of TC if the surface temperature is not to be
influenced. In these cases, different levels of constant heat fluxes are applied on the top
surface. It is clear that for a fixed hole radius the minimum depth increases with the
increase of heat flux, and that the relationship is linear. This is reasonable because the
higher flux represents a stronger cooling capacity, and the more the plate portion is
cooled down the less heat will be transferred to the TC location and to the surface.
As can be seen, the minimum depth increases as the hole radius increases; and the curves
for different hole radii are nearly parallel, which indicates that the correlation between the
minimum depth and the radius is also linear. The following equation presents the bilinear
correlations:

h = 0.081 +1.642 x R + 0.245 x q + 0.087 x R x q (5.6)

where h is the minimum depth in mm, R is the hole radius in mm and q is the applied heat
flux in M W / m . This correlation may be used for the cooling process in which the heat
2

flux does change with time or does not rely on the interaction between the coolant and the
cooled surface.

5
E
E

a>
T3 3
E
3

E 2
E

o
1 2 3 4 5 6 7 8 9

Heat flux,WM/m 2

Figure 5.29 Minimum depth of TC for ideal cooling

125
The heat flux changes with the locations and times for the practical water cooling
process. The minimum depth is rechecked with the heat flux input shown in Figure 5.27,
and the resulting data is presented in Table 5.3 and Figure 5.30. It is clear that the
minimum depth greatly depends on the allowable deviation of the disturbed temperature
from the undisturbed one.

Table 5.3 Minimum depth for practical water cooling


R=0.5 R=0.8 R=l.l R=1.4
0.92% 1.02 2.68 3.69 4.31
0.73% 1.35 3.04 4.02 4.65
0.55% 1.93 3.52 4.35 5.04
0.37% 2.60 4.12 4.95 5.53
0.19% 3.55 5.12 5.83 6.44

If a surface temperature of 1% lower than the undisturbed one is acceptable, the


minimum depth is already up to 2.68 mm for a hole of 0.8 mm in radius. For this depth
the detection sensitivity is around 1.3% for a fluctuation of 10 Hz. From the view point of
the suppression effect, any value of depth greater than 2 mm is not acceptable. For a hole
of 0.5 mm in radius the corresponding values are 1.02 mm of minimum depth of TC and
19% of detection sensitivity for a fluctuation of 10 Hz or 5% of detection sensitivity for a
fluctuation of 34 Hz.

- R=0.5
R=0.8
A R=1.1
B R=1.4

0.0% 0.2% 0.4% 0.6% 0.8% 1.0%


Error percentage
Figure 5.30 Minimum depth of TC for practical water cooling

126
It can be concluded from the above discussion that one blind hole of 1 mm in diameter
(R=0.5 mm) and 1 mm distance from the cooled surface is the best choice. However,
there may be some difficulty in the machining operation for the hole and the installation
manipulation of TC in this hole.

5.4.4 Effective T C depth

The "null-calorimeter" technique is examined through extensive simulations and


comparisons, and the results are shown in Figure 5.31 for the ideal cooling conditions.
The curves reveal that the "null-calorimeter" does exist for the carbon steel; i.e. the
disturbed internal temperature at the TC tip will be very close to the undisturbed surface
temperature if the ratio of the hole radius to the TC depth is properly determined.
The important discrepancy from the conclusions for light metals in the reference [54],
though, is that the ratio is not a constant, and depends on both the TC depth and the
applied heat flux. This ratio increases with the TC depth for any constant applied heat
flux. This means that a relatively thinner remaining disk (referring to the material above
TC) is needed for a higher TC depth. By contrast, the ratio decreases with the applied
heat flux for any fixed TC depth, which implies that a relative thicker remaining disk
makes the null-calorimeter for a stronger cooling capacity. These tendencies are fairly
reasonable from the view point of thermodynamics.

0.9 1.1 1.3 1.5 1.7 1.9 2.1 . 1 2 3 4 5 6 7 8 9

TC depth h,mm Heat flux q, WM/m 2

Figure 5.31 Null-calorimeter techniques for carbon steel

Another important difference may be found in the ratio values. For the cases studied, the
ratio R/h is greater than 1.2, and not fixed at 1.1 as for the light metals reported. The

127
rationale behind this phenomenon may lie in the fact that the conductivity and the
specific heat of light metals are much higher than those of carbon steel, although there
may be a variety of reasons for it.
It may be inferred from these results that the "null-calorimeter" technique is fairly
difficult to apply to the carbon steel, even in the case of the ideal water cooling
conditions. Also, studies using practical cooling conditions are skipped.
It is desired that the hole effect can be considered to attain a higher accurate solution in
inverse analysis. The direct choice is to include the hole in the calculation model.
However, this approach will generally make the analysis being a three-dimensional one.
Here, the "null-calorimeter" idea is extended to define an "effective depth". The
undisturbed temperature at this depth is equal to or closely follows the temperature
measured at the TC tip. It is obvious that the effective depth should be less than the actual
TC depth.

1.015 i 1

0.98 1 1
' ' 1

0 0.5 1 1.5 2
Cooling time.s

Figure 5.32 Coincidence of temperature for practical cooling

The studies in this part focus on one case in which the hole is 0.8 mm in radius and the
actual TC depth is larger than the minimum depth, and the practical heat flux is applied.
The results are shown in Figure 5.32. Figure 5.32 shows the ratio of the undisturbed
temperature at a few depths to the measured one at the TC tip for four different actual
depths for the practical cooling condition. It is readily concluded that the temperature
histories coincide very well in all cases, and that the difference tends to be smaller when
the depth increases. It seems likely that such an effective depth does exist for each case,

128
and the two temperatures can be identical at the engineering level for the whole cooling

period.

The relationship between the actual depth h and effective depth h


acl eff is shown in Figure

5.33. The data signify that a linear correlation may probably exist and can be expressed

with a correlation coefficient of 0.994 in

^=1.08/^,-0.752 (5.7)

2 | , , , .
2.5 2.7 2.9 3.1 3.3 3.5
Actual depth,mm

Figure 5.33 Effective and actual depth of TC

5.4.5 Section summary

The appropriate distance / between TCs and the TC depth h are first investigated using
direct FE analysis. It is concluded that the distance / should be less than 8-10 mm to
attain a real 2D inverse calculation and to improve the stability of inverse calculation, and
that the ratio of TC depth to the hole radius should be larger than a minimum value to
attain a nearly undisturbed surface temperature.
The "null-calorimeter" technique is then examined. It is found that this technique is fairly
difficult to use for the carbon steel, even in the case of the ideal water cooling conditions.
The effective depth concept is suggested through intensive numerical simulations, and a
linear correlation is found between the actual TC depth and the effective TC depth.
Correlations for the minimum TC depth and the effective depth are provided.

129
5.5 Conclusions

The reliability of TCs with a separation measuring junction in the temperature


measurement of water jet cooling has been extensively investigated using F E analyses.
Two kinds of situations have been considered: one for a simultaneous cooling of two
wires of TCs in the impingement zone; the other for a sequential cooling of two wires of
TCs in the parallel zone. It is found that the conduction of surface TC wires in a water jet
cooling process has a significant effect on the measured temperature, and that the un-
simultaneous cooling of the two wires of the internal TCs produces pronounced effects.
The disturbance of the temperature field due to the introduction of a small hole for the
installation of an internal TC has also been studied, which shows similar but less
pronounced effects on the surface measurement. A n increased distortion of the
temperature field is evident when TC is attached on the top surface directly above the
bottom surface of the hole. It is concluded that TCs should not, generally, be installed on
the surface and TCs with a separation measuring junction should not be used in the
temperature measurement of water-cooling in moving cases, i.e., when the water moves
or the plate is not stationery or both.
When the measured internal temperatures are used for inverse calculations by the F Z M ,
the horizontal distance of TCs should be less than 10 mm for a real 2D problem. Also,
there is a minimum depth for an embedded TC at which the surface temperature may not
be disturbed by the inclusion of a hole for the embedded TC, and thus the cooling
behaviour may not be affected. The proposed "effective depth approach" may be a
possible method to include the hole effect in the inverse calculations of heat flux without
the hole in the model.

130
Chapter 6

H E A T TRANSFER BEHAVIOR UNDER A CIRCULAR W A T E R J E T

In this chapter, the flux zoning method (FZM), the flux marching method (FMM) as well
as the effective depth approach (EDA) are used to determine the heat fluxes and surface
temperatures of stationary and moving plates cooled by a circular water jet in an industry
scale test facility. A detailed analysis of the results reveals the complex nature of the
cooling process and provides appropriate procedure and certain numerical bases for the
simulation of the problem [62].

6.1 Experimental Tests

The heat transfer in water jet impingement is a complex phenomenon that depends on
many factors, such as water temperature, plate surface temperature, jet velocity, as well
as the physical properties of both the water and the plate. It is generally believed that on
the top surface, various cooling modes, such as film boiling, transition boiling, nucleate
boiling and single convective heat transfer, may occur simultaneously.
In our experimental tests, the effects of steel grade, water temperature, water flow rate,
and moving speed on the heat transfer modes have been investigated using the facility
described in Section 2.1.3. In all tests carried out, only one nozzle is used.

6.1.1 Tests using stationary plates

For the tests using stationary plates, the test plates are heated in the furnace to the
predetermined temperatures, discharged from the heating furnace, and transferred to the
test position. The nozzle is closed before the test plates are positioned at the
predetermined location where the water hits the centre of the test plates.
The cooling water is then turned on and impinged onto the plate's surface. The cooling
times to room temperature varies greatly for different water temperatures. During the
whole cooling process, the water flow rate and water temperature are maintained
constant.

131
The configuration of TCs shown in Figure 2.6 is used. The plate's temperatures at all
measurement points are recorded with a rate of 100 Hz during the whole cooling period.
Note that the initial plate temperatures are the ones measured at the very instant when the
plates are pulled out from the heating furnace. The time for translating and setting up the
plate from the heating furnace to the predetermined position is normally not constant, so
the plate's starting cooling temperature may be slightly different even if the initial
temperature is the same.
Twenty-seven tests are selected from the more than thirty tests conducted using stationary
plates. They are numbered and listed with their corresponding parameters in Table 6.1.
The hydrodynamic parameters at the stagnation point, such as the jet velocity and the
water saturation temperature for these tests, are listed in Table 6.2. They are calculated
with the equations presented in Section 2.1.1.

Table 6.1 Experimental tests using stationary plates

Water Water flow rate,


Steel temperature, Q , 1/min
w

T r 15 30 45
30 #1 #2, #3 #4
40 #5 #6 #7
50 #8 #9 #10
DQSK
60 #11 #12
70 #13, #14 #15 #16
80 #17 #18 #19
30 #20, #21 #22, #23 #24
SS316
50 #25 #26 #27

Table 6.2 Hydrodynamic parameters at the stagnation point

No. Q , 1/min
w D ,
n mm V ,
n m/s Dj, mm Vj, m/s Ps,Pa T at,C
S

1 15 19 0.88 7.6 5.49 116,389 103.6


2 30 19 1.76 10.6 5.70 117,556 103.9
3 45 19 2.64 12.6 6.03 119,502 104.4

In addition to Figure 4.2, Figure 6.1 and Figure 6.2 show another two examples of typical
measured internal temperature profiles using stationary plates (the measured temperatures

132
refer only to the internal ones in the following sections). The cooling conditions for these
two tests are almost identical, but the steel grade is different.

1000

#3 Cooling time.s

Figure 6.1 Example A of measured temperatures


From left to right: Ti to T 8

DQSK; Q = 30 1/min; T = 30 C; D = 19 mm
w w n

1000

Cooling time.s

Figure 6.2 Example B of measured temperatures


From left to right: T, to T (T, and T are overlapped)
8 2

SS316; Q = 30 1/min; T = 30 C; D = 19 mm
w w n

133
Note that the curves from left to right present the temperatures at locations TCI to TC8.
Also, the data shown in the figures and used in the inverse calculations is smoothed and
filtered by using the running filtration approach stated in Section 5.1. Careful checks and
comparisons show that all the features of the actual cooling curves are retained in the
smoothed profiles and that the peak temperature gradients are not affected. This
smoothing technique is also used to filter out the noises in the measured temperatures
using moving plates.
It is proven that the random errors in the internal temperatures are reduced to 1 C after
the filtration operations. That implies that the inversely calculated heat fluxes and surface
temperatures have an accuracy of 90% according to the numerical test results presented
in Chapter 3.
It is clear that the cooling curves of the temperatures at locations TC 1 and TC2 are almost
identical for the whole cooling time for all tests using stationary plates. From a
thermodynamical viewpoint, these two locations fall within the boundaries of the
impingement zone.
It is apparent from the two figures that, at the air-cooling stage, the plate temperature is
fairly uniform, and it decreases gradually and almost linearly due to radiation and air
convection heat transfer. As soon as the water impinges onto the plate, the temperatures
at locations T C I and TC2 have an immediate and remarkable drop, and then the gradient
decreases from a temperature of around 300-400 C. With the spreading of cooling water
on the plate surface, the temperatures at locations TC3-TC8 begin to decrease. Although
the starting temperatures of the quick drop at those locations outside of the impingement
zone becomes smaller with the increased distance from the stagnation point, the
temperature curves at those locations also change their slopes to much smaller values
when the temperature reaches about 300-400C.

6.1.2 Tests using moving plates

The tests using stationary plates are supplemented with initial tests using moving plates.
After the plate is heated to the desired initial temperature, it is discharged from the
furnace and accelerated to the nominal test speed at which the plate passes through the
cooling section. A specially designed mechanism realizes the plate's movement.

134
A single-pass test is a more conservative test in which the plate is simply brought to a
stop after leaving the cooling section. However, a multiple-pass test has been proven
possible for moderate speeds, and permits a significant improvement of test efficiency. In
the multiple-pass case, the plate is quickly decelerated after leaving the cooling section,
and then quickly accelerated in the reverse direction to the nominal test speed. The plate
enters the cooling section again with a lower initial temperature. This procedure can be
repeated until the plate is cooled to the water temperature. Note that the nozzle is always
open when the plate passes the cooling section.

The combined deceleration and acceleration period outside the water-cooling section can
be designed such that a sufficiently homogeneous temperature distribution throughout the
plate can be re-established; i.e., the surface chilling is eliminated by the heat conduction
from the plate's hot center. The plate's moving speed is around 0.3 m/s and 1.0 m/s in
these initial moving plate tests.
The test plates are also instrumented with up to 16 internal intrinsic TCs, which are
positioned about 1 mm below the plate surface at different locations across the plate. A n
example of an employed configuration with 8 TCs is shown in Figure 6.3.

TC2

Centre line
of the plate TCI
TC3

Moving
direction
-31.8 mm -31.8 mm

Figure 6.3 Example of TC arrangements for moving plates

Figure 6.4 shows an example of temperature profiles measured at the central location
TCI in the moving plate tests. Figure 6.5 is an enlarged portion showing the temperature
recovery. It is clear that the initial temperature (cooling start temperature) at the current

135
cooling pass is equal to or slightly higher than the recovered temperature at the prior pass,

and that the temperature between two passes is fairly constant for the period, showing a

full recovery to uniform temperature distribution through thickness.

Note that the temperatures at locations TC2-TC4 are almost identical, which is important

for the modelling to estimate the heat flux and surface temperature.

50 100 150 200


Cooling time.s

Figure 6.4 Example of measured temperatures in moving plate tests


DQSK; Q = 30 l/min; T = 30 C; D = 19 mm; V = 0.3 m/s
w w n p

650
\
600

r
o
o 550
of
3 500
sv
V
*- V"-
Q.
E 450
a>
400

350
50 55 60 65 70 75
Cooling time.s

Figure 6.5 A n enlarged portion of Figure 6.4

136
6.2 Validation and Application of the EDA
In this section, the E D A is checked for its suitability with 18 tests. The analyses are
confined to the stagnation point. In the calculations, sincere efforts have been made to
match the maximum heat fluxes using the ID model with the E D A with those using the
2D model (Section 6.3), by adjusting the effective depth.
The results are shown in Table 6.3 and Figures 6.6, 6.7 and 6.8. In Table 6.3, the heat
flux is the maximum heat flux for the entire cooling period, and the temperature is the
one corresponding to the maximum heat flux. In addition, the notations " l D w E D A " and
" l D w t E D A " mean the results from the models with the E D A and without the E D A ,
respectively. These two notations are also used in the relevant figures. It may be seen that
it is possible to find an effective depth at which the maximum heat fluxes as well as the
temperatures are very close to each other. It is also clear that the maximum heat flux and
the temperature from the ID model, without considering the hole effect, will be different
from those from 2D: higher maximum heat fluxes and lower temperatures.

Table 6.3 Comparison of results using different approaches

21D l D w EDA lDwtEDA


Test Flux, Temp, Flux, Temp, Flux, Temp,
MW/m 2
C MW/m 2
C MW/m 2
C
#1 17.0 170.50 17.6 168.51 23.4 -4.98
#2 20.1 -58.21 20.3 -60.63 27.3 -245.53
#3 20.0 170.99 20.6 175.42 28.3 -41.67
#4 14.2 280.19 14.4 308.51 19.5 126.61
#5 11.9 351.16 12.1 385.85 16.9 225.08
#6 10.0 412.34 10.1 436.00 13.6 297.11
#7 11.5 267.93 12.0 286.16 16.0 280.14
#8 13.4 256.87 13.8 304.57 19.6 201.13
#9 10.4 403.89 10.7 429.79 14.5 286.00
#10 12.8 177.45 12.5 214.75 16.6 66.11
#20 12.0 346.79 12.2 428.11 18.6 229.62
#21 14.4 365.43 14.0 371.43 23.1 110.39
#22 13.0 361.72 13.4 424.46 19.9 230.86
#23 11.3 252.86 13.4 403.99 22.4 152.66
#24 11.4 450.13 11.5 441.04 22.3 127.91
#25 9.7 455.91 10.0 364.58 15.7 347.55
#26 12.6 399.06 12.4 400.42 20.1 162.60
#27 13.1 350.40 13.4 350.77 19.8 223.77

137
Figures 6.6 and 6.7 show the examples in which the boiling curves from 2D and ID
models with the E D A are nearly identical during the entire cooling period. It can be
suggested from this result that the E D A is an appropriate option for considering the hole
effect without including the hole in the model.

16.0
14.0
CM 12.0
E
g 10.0
s
* 8.0
1 6.0
1
4.0
2.0
0.0
0 200 400 600 800 1000
Surface temperature,C
Figure 6.6 Boiling curves using different modellings for test #6

The two figures also show that the boiling curves from ID without the E D A are above
those from 2D modelling, indicating that the heat fluxes from ID without the E D A are
generally larger than those from 2D. However, the boiling curves from ID without the
E D A still reflect all the features of heat transfer modes like those from 2D modelling, and
do not produce erroneous heat transfer phenomena. The latter is informative.

16.0

14.0

12.0

I 10.0
I 8.0
S 6.0
i
4.0
2.0
0.0
0 200 400 600 800 1000
Surface temperature,C

Figure 6.7 Boiling curves using different modellings for test #9

138
These two examples suggest that the E D A may be considered as another option to
account for the hole effect. However, several difficulties arise when applying the E D A in
the extensive investigations of the experimental tests. First, it is in some cases difficult to
find an effective depth, and the inverse calculation results are very sensitive to the value
of the effective depth. Figure 6.8 shows one example for this situation. For this test, the
boiling curve from ID analysis without the E D A is still above that from 2D modelling,
and the peak value of the heat flux from ID without the E D A nearly doubles that from
2D modelling. This result is consistent with the above two cases. However, whatever the
effective depth value, the boiling curves from ID with the E D A are not close to that from
2D modelling. Note that the difference between two adjacent values is only 0.07 mm for
the four trial effective depths. With this small variation in the depth value, the difference
in the maximum heat flux is about 7-10%, showing the high sensitivity of the calculated
heat flux to the effective depth. This unsuitability and sensitivity of the E D A occurs for
both the carbon steel and the stainless steel.

25.0 | 1

2D
20.0
IDwtEDA
IDwEDA-1
- e - 1DwEDA-2
I 15.0 A 1DwEDA-3

- O - IDwEDA-4

0 200 400 600 800 1000

Surface t e m p e r a t u r e , C

Figure 6.8 Sensitivity of results to the effective depth, #24

The second difficulty is that the correlation shown in equation (5.7) is not valid for all
real cooling tests; i.e., there is no general valid correlation between the actual depth and
the effective depth at which the maximum heat flux is close to that from 2D, irrespective
of the difference in the boiling curve shape. Figure 6.9 shows that the effective depth
hardly relates to the actual depth with a fixed relationship. However, it is apparent that
the effective depth is on average about 40% less than the actual depth.

139
Actual depth, mm

Figure 6.9 Relationship between the actual depth and effective depth

Even for the stagnation point, to finding an appropriate effective depth takes much effort
and time because the value changes from case to case. It is possible that the effective
depth depends on both the actual depth and the cooling parameters. It may be impractical
to find such a correlation.
The last difficulty is a conflicting one. The intention of developing the E D A is to take the
hole effect into account without including the hole in the model, especially for the holes
outside the impingement zone. In this regard, the calculation conditions should be close
to those outside the impingement zone. However, it is not possible to realize such
calculations from the numerical simulation side. In fact, the actual cooling conditions
outside the impingement zone are unknown and are to be determined. Therefore, no
appropriate boundary conditions can be used in the direct simulations. Second, the
correlation between the actual depth and the effective depth obtained from the
simulations at the stagnation point, i f the correlation could be obtained, may not be
suitable to the points in the parallel zone.

From the above discussion, it may be concluded that the proposed E D A in Section 5.4.4
may not be an effective and general approach used in the water jet cooling process.
However, an alternative approach can be obtained from the above studies on the E D A . In
this alternative approach, the used depth for the TC location is 30% less than the actual
depth in the inverse calculations. The value of 30% is conservative and is based on the
data in Figure 2.9. Although this reduction of the TC depth may not have a strong basis

140
and may not create an accurate account of the hole effect, it will definitely reduce the
deviation of the calculated boiling curves from the true one.
This approach is nevertheless still called an effective depth approach (EDA).
The application of the modified E D A is illustrated by the following example. In this
example, the results from the F Z M with and without the E D A are compared. A n overall
view of the heat fluxes on the top surface of a stationary stainless steel plate is obtained
first using the F Z M without the E D A , as shown in Figure 6.10. In this calculation, a full
2D axisymmetrical F E model has been used. The model has 115 mm radius and 7 mm
thickness. The model has 2430 quadratic elements. The elements are uniform in the
radial direction and vary from fine on the top surface to coarse on the bottom surface in
the thickness direction. In the calculation, the right hand surface, i.e., the outside of the
cylindrical disc, is assumed thermally insulated, and a constant heat flux equivalent to air
cooling is applied to the bottom surface. A constant average conductivity value of 20
W/m-C is used, and phase transformation heat is not considered.

The F Z M is used in which the top surface is divided into eight sub-regions; each sub-
region is symmetric with respect to the location of the corresponding TC. The heat fluxes
for all nodes in each individual sub-region are assumed to have the same value; i.e., there
are only eight heat fluxes corresponding to the eight measurement locations. The actual
values of heat fluxes at the eight points corresponding to measurements of temperatures
are determined using inverse analysis. Only the data at the current time step is used, the
regularization parameter is fixed at le-10, and only several iterations are required to get
convergent results for each time step.
The results in Figure 6.10 reveal all the features of its corresponding temperature curves.
Note that the heat fluxes at all locations have relatively close values of about 14 KW/m
for the air cooling stage, which is typical due to the combination of radiation and air
convection cooling.
As the water hits the plate, the heat fluxes at locations 1 and 2 instantaneously jump to
very high values. The flux profiles at these two locations are almost identical in shape
with a slight difference in the peak values for all performed tests and times; the peak
value of the heat flux at the second location is generally higher than that at the stagnation
point. This is a point to be further studied. The difference in peak heat fluxes may come

141
from the slight difference in the measured temperatures, which ascertains the sensitivity
of the inverse problem. It is also noted that the peak value of the heat flux is generally
50% higher for the stainless steel than that for the carbon steel.

20
18
16
~ 14

0 20 40 60 80
Cooling time.s

Figure 6.10 Overall view of heat flux on the top surface, Test #25

The heat fluxes at locations 3-8 jump at different time instants proportional to the
distance from the stagnation point. The shapes of heat flux profiles are, again, almost
identical. The magnitudes of the peak heat fluxes at these locations are much lower than
those at locations 1 and 2; the former are 3 times less than the latter.

142
The modified E D A is also applied to this test whereas the TC depths in the calculation
are 30% less than the actual ones. The results are shown in Figure 6.11. In this figure, the
peak values of the heat flux at 8 locations are compared. It is clear that the values with
the E D A are lower than those without the E D A . It is interesting to note that the
percentages at the first two locations in the impingement zone are around 72% while
those in the parallel zone are nearly fixed around 84% irrespective of the individual
locations. This discovery may be very informative and may provide means to simplifying
the calculations.
For the stagnation point (location 1), the peak value from the modified E D A is 9.6
M W / m and is very close to those from the 2D hole-model and the actual E D A (Table
2

6.3). This point shows that the modified E D A may compensate for the hole effect. At the
same time, the peak value from the full model is 13.2 M W / m and is close to those from
the ID model without the E D A . This discovery again reveals that the F Z M may not
correctly reflect the real heat transfer in the water jet cooling.

6.3 Heat Transfer in the Impingement Zone of Stationary Plates [62]

In this section, the discussion is confined to the heat transfer behaviour in the
impingement zone or at the stagnation point because the internal temperature at the
stagnation is the most reliable one and is not affected by the water movement. Under this
condition, the F Z M can be easily applied.

6.3.1 Initial investigations

Initial investigations are conducted on Test #15. As discussed above, the temperature at
locations TCI and TC2 are nearly identical during the entire cooling period for all tests.
So, it can be assumed that both of them are inside the impingement zone. And it is further
assumed that the impingement zone extends to the centre between TC2 and TC3; i.e., the
radius of the impingement zone is 24 mm. Thus, the ratio of the diameter of the
impingement zone to the nozzle diameter is about 2.53.
Two-dimensional axisymmetrical conditions are assumed in the inverse calculations, and
the holes are not included in the mesh. The measured temperatures at locations TC 1 and
TC2 are used. The right-hand surface, i.e., the outside of the cylindrical part, is assumed

143
to be thermally insulated, and a constant heat flux equivalent to air cooling is applied to
the bottom surface. A constant weighted averaged conductivity value of 35 W/m-C
(considering that the critical heat transfer occurs above 400 C) is used for the whole time
span, and phase transformation heat is not considered.
Two cases are studied. In the first case, the top surface is divided into two subregions
corresponding to the two measured temperature histories, and two different heat flux
components are specified on the subregions. In the second case, the top surface is only
one subregion and only one heat flux component is prescribed on the plate's top surface.
The second situation means that the unknowns are less than the temperature
measurements, and so an overdetermined inverse problem exists, which would be,
generally, more stable. In these two cases, different numbers of future time steps are
adopted to investigate the effect on stability. The results obtained from this model are
shown in Figure 6.12 and 6.13. In the two figures OF and 3F indicate the heat flux with 0
and 3 future steps.

Cooling time.s Cooling time.s

Figure 6.12 Heat flux in the impingement zone


Two components of heat flux assumed
a: Component 1; b: Component 2

144
0 KU-
10.5 11 11.5 12 12.5
Cooling time.s

Figure 6.13 Heat flux in the impingement zone


One heat flux assumed

Figure 6.12 is for the first case and shows the effects of the number of future steps on the
inversely calculated heat fluxes in the impingement zone for the instant of impingement
and shortly thereafter. There is a sharp jump in the heat flux values at the instant of
impingement, followed by a much smaller but sharp decrease to a value that steadily
decreases at a slow rate. This indicates that similar cooling conditions exist for a smaller
area in the centre of the jet that may be considered as a single cooling zone, the
impingement zone. This conclusion may be confirmed by the results shown in Figure
6.13 for the second case.
Analyses also show that there is no considerable difference between the heat fluxes using
various numbers of future steps, especially when the number is larger than 3. The
insensitivity of these results to the number of future steps may indicate that the random
errors in the internal temperatures are filtered out due to the average smoothing technique
discussed above.
It can be concluded that the simplified model may be efficiently used to study the heat
transfer behaviour in the impingement zone.

6.3.2 Heat transfer behaviour

According to the initial investigations, the hole-model shown in Figure 5.7a has been
used in this analysis. The radius of the small geometric portion of the plate is fixed at 8
mm, and its height and the hole dimensions correspond to the values at each test. The

145
temperature dependence of the plate's conductivity is considered. In all calculations, only
the data at the current time step is used. The values of the regularization parameters are
10% of the diagonal values of the sensitivity matrices, and the convergence criteria are
fixed at 5% and 1 C.

1. Effect of water temperature


Figure 6.14 depicts the calculated surface temperature profiles at different water
temperatures at a water flow rate of 15 1/min for the carbon steel. It should be noted that
the starting time has been adjusted by shifting to allow the curves to be clearly separated
for better comparisons, and that the time interval for two adjacent points on the curves is
the same, that is, 0.1 s. It is clear that the surface temperature drops sharply in several
time steps to around 300-200 C as the cooling water of 40 C or 50 C impinges onto
the plate surface, and that the cooling rate is as high as 2000-2400 C Is. After this fast
dropping stage the cooling process starts to level off. For a water temperature of 60 C,
much more time is needed to cool the surface from around 880 C to 200 C, which can
be verified by the point density at the higher temperature part of the curves.

0 10 20 30 40 50

Nominal cooling time.s

Figure 6.14 Water temperature vs. surface temperature at Q - 15 1/min


w

The cooling curve at a water temperature of 80 C is quite different from the first two.
When the cooling water hits the plate surface there is a fairly long and gentle cooling
process, which implies that the cooling capacity is low. As the surface temperature

146
reaches around 450 C a faster cooling stage appears, and the surface temperature drops

to around 200 C in several time steps.


The cooling curves at different water levels at a water flow rate of 30 1/min reveals a
similar trend, as shown in Figure 6.15.

1000
900
y 800
e 700
g 600
tu
| 500
400
300
= 200
07
100
0
0 10 20 30 40
Norn inal c o o l i n g t i m e , s

Figure 6.15 Water temperature vs. surface temperature at Q = 30 1/min w

Figure 6.16 shows the effect of water temperature on the cooling behaviour at a water
flow rate of 15 1/min for the carbon steel. Generally, the heat flux increases by decreasing
the water temperature. The highest value is around 14 M W / m for a water temperature of
2

147
40 C and 6 M W / m 2
for a water temperature of 80 C. It may be concluded that
increasing the water temperature significantly reduces the heat extraction from the plate.
Water temperature also affects the cooling pattern or thermodynamics. At a lower water
temperature (<50C), the heat flux monotonically increases with the decrease of surface
temperature until a value of about 300 C and then again monotonically decreases with
the decrease of surface temperature. This may indicate that there is no film boiling
regime at the beginning, and that the heat transfer is due to transition boiling and then it
shifts to nucleate boiling. The reason for this phenomenon is the sharp drop of the surface
temperature at the beginning and the incapacity to heat the water to the vaporization
temperature. It may be concluded that the surface around the stagnation point is
immediately wetted by water at lower water temperatures.
In contrast to the situation at lower water temperatures, there is a film boiling regime
when the water temperature is higher than 60 C. When the cooling water of higher
temperatures impinges onto the surface, it is easily heated to its boiling temperature and
the heat flux is kept at a lower value. The duration of film boiling depends mainly on the
water temperatures in the test conditions. From a comparison of the boiling curve for 60
C water temperature with that for 80 C, it is clear that the higher the water temperature,
the longer the film boiling period and the lower the heat flux for this period.
From the above discussion it may be postulated that the maximum water temperature for
immediate wetting is around 60 C, which is in good agreement with that in reference
[16] where such a temperature is 68 C.

It is worthwhile to point out the difference of the film boiling mode shown in Figure 6.16
from that corresponding to the water pool cooling. For the film boiling in water pool
cooling, the heat flux decreases with the decrease of the surface temperature to a
minimum heat flux value. This feature was also reported in reference [15] for water jet
cooling. In the case shown in Figure 6.16, the heat flux during the film boiling period
does not change much, and is kept at nearly a constant value. This trend is more obvious
when the water temperature is 80 C, i.e., the subcooling is around 20 C.
The temperature at which the heat flux reaches its minimum value and the heat transfer
shifts from film boiling to transition one is commonly referred to as the Leidenfrost point.

148
12.0

200 400 600 800 1000

S u r f a c e tern p e r a t u r e , C

Figure 6.17 Effect of water temperature at Q = 30 1/min w

According to Figure 6.16, the Leidenfrost point is around 510 C and 430C for a water
temperature of 60 C and 80 C, respectively. Also, the figure indicates that there is a
single forced convection regime when the surface temperature is lower than about 150 C
where the slope of the boiling curve changes from a mild one to a sharp one.
Experiments with different flow rates of 30 1/min and 45 1/min have been conducted and
the results are shown in Figures 6.17 and 6.18. These figures show trends similar to those
discussed above.

16.0

is
o>
X

200 400 600 800 1000

S u r f a c e tern p e r at ure,C

Figure 6.18 Effect of water temperature at Q = 45 1/min w

1 4 9
2 . Effect of water flow rate
The impingement velocity is proportional to the water flow rate when the distance of the
nozzle exit to the plate and the nozzle diameter are fixed, as in the conditions in this
study. Thus, the effect of the impingement velocity can be represented by the effect of the
water flow rate, as shown in Figure 6.19 to Figure 6.23.
The figures indicate a somewhat ambiguous phenomenon. The water flow rate has little
effect on the heat transfer for all cases in which the water temperatures are 30 C, 40 C,
70 C and 80 C (Figure 6.19 through Figure 6.22); but it has a noticeable effect when the
water temperature is 50 C (Figure 6.23). This is a characteristic to be carefully checked.

21.0
^ \ CS-Tw=30
18.0

I i Qw=15
15.0
AQw=30
E p
Qw=45
| 12.0
T i' 1 \ V ^ 1
^
X
^ 9.0 X i
1
I \ \ ^^t^
M
; / \ \
6.0
T

3.0

0.0
200 400 600 800 1000

Surface temper at ure,C

Figure 6.19 Effect of water flow rate at T = 30 C w

It may be seen from Figure 6.19 and Figure 6.20 that the heat transfer regimes are,
sequentially, transition boiling, nucleate boiling and convective boiling when the surface
temperature decreases from 880 C to the room temperature. There are some differences
among the maximum heat flux values and the temperatures at which the maximum heat
flux occurs when the water temperature is 30 C. This difference does not appear when
the water temperature is 40 C.

It may be generally stated that the heat flux should increase with the increase in the jet
velocity because a higher velocity increases the pressure as well as the saturation
temperature of water at the stagnation point and, therefore, the subcooling increases. In
the test conditions presented, when the water flow rate varies from 15 to 45 1/min, i.e., an

150
increase of 3 folds, the jet velocity increases only from 5.49 to 6.03 m/s, and both
pressure and saturation temperature show modest changes. Such small changes should
not cause significant changes in the heat flux.

200 400 600 800 1000


Surface temperature,C

Figure 6.20 Effect of water flow rate at T = 40 C


w

200 400 600 800 1000

Surface temperature,C

Figure 6.21 Effect of water flow rate at T = 70 C w

The above discussion may also be applied to the heat flux values at higher water
temperatures, as shown in Figure 6.21 and Figure 6.22. The difference among the boiling
curves at lower and higher water temperatures lies in the existence of film boiling

151
regimes. For the film boiling regime, the heat flux decreases slightly with the decline in
surface temperature. There is no clear and defined relationship between the Leidenfrost
point and the water flow rate at different water temperature levels.

6.0
-t CS-Tw=80
5.0
- Qw=15

E 4.0 6 Qw =30

Qw=45
x 3.0

ra
2.0

1.0

0.0
200 400 600 800 1000

Surface temperature,C

Figure 6.22 Effect of water flow rate at T = 80 C w

Although the boiling curve for the water flow rate of 30 1/min shows the same
combination of the heat transfer modes as those for the water flow rate of 15 1/min or 45
1/min, its magnitude is much less than the values at the corresponding points. Whether
this may be attributed to measurement errors or to a complex relationship at the higher
temperature is an issue to be explored in future work.

CS-Tw=50

-Qw=15
E -A Qw=30
S Qw=45
x"
3

ra
a>
X

200 400 600 800 1000

Surface temperature,C

Figure 6.23 Effect of water flow rate at T = 50 C w

152
Figure 6.23 shows the boiling curves at the stagnation point for the water temperature of
50 C at different flow rates for the carbon steel. It is clear that there is no film boiling
regime, and that little effect of the water flow rate can be observed at the beginning of the
transition boiling region and the later period of the nucleate boiling. The higher effect of
the water flow rate occurs when the surface temperature varies from 600 to 200 C. It is
to be noted that this kind of effect is, generally, against the commonly accepted trend;
i.e., the increase of the water flow rate does not increase the heat transfer capacity. This is
quite abnormal, and further investigation would be needed to verify the phenomenon.
Figure 6.24 clarifies the effect of the water flow rate on the heat transfer behaviour at the
water temperature of 50C for the case of stainless steel SS316. It is clear that the
maximum heat, flux at the water flow rate of 30 1/min or 45 1/min is higher than that at 15
1/min and that little effect can be seen for the overall cooling process for both flow rates
of 30 1/min and 45 1/min.

14.0
/>s. SS-Tw=50
12.0
Qw=15
~ 10.0
E
6Qw-30
| 8.0 Qw=45
x"
c 6.0
CO

4.0

2.0

0.0

0 200 400 600 800 1000

Surface temperature,C
Figure 6.24 Effect of water flow rate at T = 50 C for SS316
w

From the above discussions it may be concluded that the water flow rate generally has
little effect on heat flux magnitude at all water temperature levels or subcooling levels.
This may be attributed to the consideration that the large variation of water flow rate has
a slight effect on the jet impinging velocity as well as the pressure and the saturation
temperature. Also, it should be reported that some discrepancies are evident, and further
investigation is needed.

153
3. Effect of steel grade
The effect of steel grade on the heat transfer behaviour is considered by the change in
material properties, especially the heat conductivity in this study. Figure 6.25 shows this
kind of effect at the water temperature of 30 C. It can be seen that for two levels of water
flow rates, the heat fluxes with SS316 are always smaller than those with the carbon steel
DQSK. The DQSK has generally higher conductivity than SS316, and the difference in
conductivities increases with the decline in temperature, as shown in Figure 6.26 (refer to
equations (2.8) and (2.9)). When the temperature is around 500 C, the conductivity of
DQSK is nearly twice that of SS316. It incites the cooling effect of water on the plate
surface and instigates a faster decrease of the temperature at the target location. On the
other hand, the heat from the bottom part is also easily transferred to the target location,
and this tends to increase the target temperature. To reduce the temperature at the target
location to the same value, a higher heat flux may be needed for cooling the material with
higher conductivity. This may be the reason why the heat fluxes for the SS316 are always
smaller than those for the carbon steel DQSK for the water temperature of 30 C.

25.0

Q CS-Tw 30-Qw 30
20.0

E
| 15.0

tro 10.0

5.0

0.0
0 200 400 600 800 1000
Surface temperature,C

Figure 6.25 Effect of steel grade at T = 30 C


w

154
70

0 I , , , , 1
0 200 400 600 800 1000

Tem p e r a t u r e , C

Figure 6.26 Temperature-dependent conductivity

The effect of steel grade on the heat flux at the water temperature of 50 C is shown in
Figure 6.27. In contrast to the correlation at 30 C, for the same levels of water flow rates
as those in Figure 6.23, the heat fluxes for the SS316 are mostly higher than those for the
carbon steel DQSK. As discussed above, the cooling capacity depends mainly on the
water temperature and not the water flow rates. With the increase of the water
temperature, the cooling efficiency decreases. This means that the interior temperature
decreases at a much lower rate with higher water temperatures than with lower water
temperatures, which can be verified by the point densities at the starting stages of cooling
in Figure 6.23 and Figure 6.25. Also, this indicates that the cooling is more confined at
the top surface, and it is easier to cool down the surface with a higher conductivity than
with a lower conductivity; i.e., less heat flux is needed to reduce the surface temperature
to a given value for the material with higher conductivity than for that with lower
conductivity.

6.3.3 Section summary

The heat fluxes and surface temperature at the stagnation points or in the impingement
zone of stationary plates in the circular water jet cooling is successfully determined by
using the F Z M and a small model. A n analysis of the data shows that the heat transfer

155
behaviour at the stagnation point is mainly and greatly affected by the water temperature,
slightly affected by the steel grade and hardly affected by the water flow rate. When the
water temperature is lower than 60 C the cooling process is governed by the transition
boiling and nucleate boiling. When the water temperature is higher than 60 C, there is
evidence of the existence of a film boiling regime. The features of boiling curves and the
typical values obtained are in close agreement with other resources.

14.0

0 200 400 600 800 1000


Surface tem perature,C

Figure 6.27 Effect of steel grade at T = 50 C


w

6.4 Heat Transfer in the Parallel Zone of Stationary Plates

Generally, the heat transfer in the parallel zone is different from that in the impingement
zone. The results shown in Figure 6.10, which are obtained using the F Z M without the
E D A , show that the maximum heat fluxes at different locations in the parallel zone are
fairly close to each other and are all much smaller than those in the impingement zone.
In this section, the focus is on the heat transfer at location TC3 for stationary plates. The
F M M is used together with the E D A . The two reasons why only the heat transfer at
location TC3 is studied are that the heat flux profiles at different locations in the parallel
zone are nearly identical, and that a great effort is needed to apply the F M M .
For all calculations in this section, a 2D axisymmetrical condition is adopted and a single
mesh model is used, whereas 896 quadratic elements are used to discretize a geometry of
64 (radius) x 7 (thickness) mm. The data at the first four locations are used in the inverse
calculations. There are uniformly 64 elements in the radius direction. For the carbon

156
steel, the temperature dependence of conductivity is considered, and an average value of
20 W/m-C is used for the conductivity of the SS316.

6.4.1 Case study

Test #25 is studied in detail. The results are shown in Figures 6.28, 6.29 and 6.30.
Figure 6.28 compares the cooling curves from different approaches. As expected, the
cooling curves with using the E D A are always below the corresponding ones without the
E D A for both methods of specifying heat flux. With the F Z M , the peak value of heat flux
with the E D A is 20% lower than that without the E D A , while the former is around 30%
lower than the latter with the F M M . This illustrates that the E D A has more influence with
the F M M than with the F Z M . This is reasonable because the F M M is point-oriented
while the F Z M has an averaging feature.
The cooling curves either from the F Z M or the F M M show similar heat transfer modes.
That means that the results without the E D A can be used to obtain a qualitative
correlation, which may turn to a quantitative one with an appropriate coefficient.
However, the cooling curves from the F M M present a heat transfer mode slightly
different from the F Z M cooling curves. These curves show that the heat flux increases
from the start of cooling to a temperature, and then decreases. This temperature is about

157
400 C for the F Z M and 200-300 C for the F Z M . This point should be carefully checked
with the other test results.

Location index

Figure 6.29 Comparison of peak heat fluxes at the first four locations, #25

The comparison of peak heat fluxes at the four locations from the F M M is presented in
Figure 6.29. Similar to those shown in Figure 6.20, the percentages at the two points in
the impingement zone are nearly the same and lower than those at the other points in the
parallel zone. In fact, F M M results are similar to F Z M ones for the impingement zone.

158
Figure 6.30 illustrates the difference in the heat transfer modes at the locations TCI and
TC3. The first conclusion is that the difference in heat flux magnitudes at the two
locations is not remarkable as shown by the F Z M , in Figure 6.10, as well as in the
reference.

16

0 200 400 600 800 1000


Surface temperature,C

Figure 6.31 Comparison of cooling curves, #3

The second point is that the difference may be due to the difference in surface. It can be
seen that the surface temperature at TCI at the very beginning of water cooling is around
900 C while this temperature is around 720 C at TC3. For this test, the surface
temperature at TC3 decreases due to a combination of effects: a longer air cooling period,
the transversal temperature gradient before the water reaches this location, and the
filming boiling cooling before the location is rewetted. For some tests, the surface
temperature at TC3 decreases to a lower value mainly due to the film boiling cooling
process before this location is rewetted, as shown in Figure 6.31.

6.4.2 Parameter effects

In this section, the effects of the cooling parameters on the cooling behaviour at TC3 are
briefly discussed using a limited number of examples with typical experiment conditions:
the flow rate is fixed at 30 1/min and/or the water temperature is fixed at 30 C.
Therefore, the conclusions may be confined to the tests discussed and are not general.

159
1. Effect of water temperature

The effect of water temperature on the heat transfer at TC3 with a water flow rate of 45
1/min is shown in Figure 6.32. It is may be seen that the water temperature has little
influence on both the mode and the magnitude of heat transfer behaviour, an observation
which is totally different from that at the stagnation, whereas the heat transfer is affected
greatly by the water temperature.

1000

Surface temperature,C

Figure 6.32 Boiling curves at TC3 by F M M for #24 ( T = 30C) and #27 ( T = 50C)
w w

The rationale behind this phenomenon at TC3 may lie in the local water temperature.
When water impinges onto the hot surface and spreads streamwise, it is heated to a higher
temperature. It can be inferred that the water temperature at TC3 and afterwards have
similar values irrespective of the initial one, and that the water temperature difference at
TC3 will decrease. For the two tests, the initial water temperature difference is only 20
C; thus, the local water temperature difference at TC3 will be definitely less than 20 C
and may not cause a big discrepancy in heat transfer at this location.

2. Effect of water flow rate

Figure 6.33 shows the effect of flow rate on the heat transfer at TC3 for SS316 at a water
temperature of 30 C.
In contrast to the conclusion discussed in Section 6.2, in which the water flow rate has
little effect on the heat transfer at the stagnation point, the water flow rate affects both the

160
mode and the magnitude of heat transfer in the parallel zone. There is a cooling stage of
film boiling for the flow rate 30 1/min, while the film boiling seems not to exist for the
flow rate 45 1/min. The increase in the water flow rate means that more fresh water floods
location TC3, and this process also progresses faster. Therefore, location TC3 may be
rewetted much faster with a higher water flow rate than with a lower water flow rate.
Correspondingly, the heat flux with a flow rate of 45 1/min is generally higher than that
with a flow rate of 30 1/min. These two trends are consistent and compatible.

3. Effect of steel grade (Conductivity)

The effect of steel grade on the heat transfer at TC3 is illustrated in Figure 6.34. For these
two tests, the flow rate is fixed at 30 1/min and the water temperature is fixed at 30 C.
The two cooling curves show the same cooling behaviour, and the cooling modes are as
complete and typical as those shown in Figure 2.1. This means that the steel grade has no
or little effect on the cooling modes.
However, the Leidenfrost point is somewhat lower for the SS316 than for the DQSK; and
the maximum heat flux for the DQSK is about 2 times that for the SS316. This is because
the conductivity of the D Q S K is higher than that of the SS316. Higher conductivity will
greatly increase the heat transfer from the bottom hotter part to the top lower part, which
increases the heat flux.

161
These trends at TC3 are similar to that at the stagnation point for the same cooling
conditions.

14

Surface ternperature,C

Figure 6.34 Boiling curves at TC3 by F M M for #3 (DQSK) and #22 (SS316)

It can be concluded that the water flow rate is the key factor in the parallel zone and it
affects both the mode and the magnitude of heat transfer, and that the conductivity
mainly influence the heat flux values.

6.5 Heat Transfer for the Moving Plates

First, the modelling for moving plates should be discussed. As discussed above, the
impingement zone is about 2.5 times the nozzle diameter for the stationary plates.
Although the cooling zone changes to a pearl-shape for the moving plates, the transverse
dimension of the cooling zone is not significantly affected. Locations T C I and TC2 in
Figure 6.3 are still in the instant impingement zone; and there is no transverse heat flow
around T C I . Moreover, as analyzed in Section 4.2, the material points are nearly cooled
by a moving pattern for the moving plates; and the speed of the moving pattern can be
assumed to be the plate's moving speed. As the plate's moving speed is greater than 0.3
m/s, two points that are 16 mm apart in the longitudinal direction are cooled down with a
time difference of 0.053 s or less. This means that the cooling conditions for the two
points are nearly identical, and there are no evident of longitudinal temperature gradients.

162
The temperature profiles at location T C I , TC2 and TC3 verify this analysis. Therefore, it
would be appropriate to simplify the heat transfer at TCI as 2D (thickness and
longitudinal direction) or ID (thickness direction). In the following example, 2D
modelling is used.
Figure 6.35 gives an example of the heat fluxes obtained from moving plate experiments,
which provide new insight into heat transfer on ROT and constitute a crucial
advancement for developing more realistic heat transfer correlations for industrial ROT
cooling. In this example, the water temperature is 30 C and the water flow rate is 30
1/min, which are all typical values for the cooling on ROT.
It is clear that increasing the plate speed leads to a decrease in heat fluxes for a given
water temperature and water flow rate. For example, as the initial temperature is around
260 C, the peak heat flux is 1.85 M W / m for the plate's moving speed of 0.3 m/s while
2

it decreases to 1.2 M W / m for the plate's moving speed of 1.0 m/s.


2

3.5
V=0.3 m/s
3.0
V=1.0 m/s D


N 2.5

E
| 2.0

t= 1.5 I
u
n
0)
X 1.0

0.5

0.0
100 200 300 400 500 600 700
Initial t e m p e r a t u r e , C

Figure 6.35 Effects of the plate's speed and initial temperature on the peak heat flux
Q = 30 1/min, T = 30 C, DQSK
w w

The effect of the plate's initial temperature on the peak heat flux is not monotonic. As the
initial temperature is lower, the peak heat flux increases with the increase of the initial
temperature and reaches a maximum. Afterwards, the peak heat flux decreases with the

163
increase of the initial temperature. This trend is retained for the two levels of the moving
speed. However, the critical temperature at which the peak heat flux reaches its
maximum seems at a lower value for a higher speed. The critical temperature is around
400 C for the speed of 1.0 m/s while it is about 500 C for the speed of 0.3 m/s.
With an extrapolation, the peak heat flux may be only 2.0 M W / m for the speed of 0.3
2

m/s as the initial temperature is 850 C. Recall that the peak heat flux is around 20.6
M W / m for the stationary plate with an initial temperature 850 C at the same cooling
2

conditions (Q =30 1/min and T =30 C). It may be concluded from the two values that
w w

the motion of the plate has a great influence on the heat transfer.
The above trends are verified by other tests.

6.6 S u m m a r y

Experimental test data are successfully analyzed by the developed approaches. The F Z M
is used for the investigation of the heat transfer behaviour at the stagnation point of
stationary plates while the F M M and E D A are used for the parallel zone. The F Z M and
E D A are used for the moving plate tests.
Results show, that the heat transfer behaviour at the stagnation point is mainly and greatly
affected by the water temperature, slightly affected by the steel grade and hardly affected
by the water flow rate. When the water temperature is lower than 60 C the cooling
process is governed by the transition boiling, nucleate boiling and convective boiling.
When the water temperature is higher than 60 C, there is evidence of the existence of a
film boiling regime.
It is apparent that the water flow rate is the key factor in the parallel zone and it affects
both the mode and the magnitude of heat transfer, and that the conductivity mainly
influence the heat flux values. However, the initial water temperature has little effect on
the heat transfer in the parallel zone, which suggests that local water temperatures may be
required.
It is found by analyzing the moving plate tests that increasing the plate speed leads to a
decrease in heat fluxes for a given water temperature and water flow rate.

164
Chapter 7

SIMULATION OF RUN-OUT T A B L E COOLING PROCESS

The cooling process of steel strips on an industrial ROT is simulated using ID model.
The accuracy of the ID model simulations is discussed with a comparison to the field
measurements [64]. Transverse 2D modelling is used to investigate the effect of non-
uniform heat flux in the width direction on the final coiling temperature. The necessity of
longitudinal 2D modelling is also investigated.

7.1 Simplication of Governing Equation

The general governing equation for conduction heat transfer problems of a moving
volume in Lagrangian reference may be written in the form [57]:

DT
V(kVT) + q= cn (7.1)
Dt

where k is the conductivity, T is the temperature, q h


is the heat generation per unit

volume, p is the density, c is the specific heat, and t is the time. It should be noted that
p

the material properties k, c , and p, are usually temperature dependent.


p

Moving Direction

Figure 7.1 Coordinate system used for modelling

The differential operators for a Cartesian coordinate system (Figure 7.1) are defined as:

Y7 r d -a -a (7.2)
V =; + j + k
dx dy dz

165
wherei,j and k are the unit direction vectors along the x,y and z direction, (x,y, z) is the

space point position with respect to the Lagrangian coordinate reference, and (w , %, w )
x z

are the components of velocity vector u at the point (x, y, z).


For the controlled cooling process of the hot strip on ROT, the ranges of coordinate
geometry can be assumed as the following:

O<X<L

-b/2<y<b/2 (7.4)
-hl2<z<hl2

where L is the total length of the cooling zone from the last finishing stand to the
downcoiler, and b and h are the width and thickness of the strip, respectively.
To apply the governing equation to the controlled cooling process of the hot strip, a
number of asumptions and measures should be made [65-77].
1. It isassumed that the strip does not move vertically up and down and laterally (in
practice, such movements eventually exist and may have effects on the cooling results),
that is:

u =u =0
y z (7.5)

2. It isassumed that the strip is continous, and that no distinction is made between the
leading end, trailing end or central portion of the strip, or that the strip is infinitely long;
and that the initial temperatures as the strip enters the cooling zone are the same; and that
the process operates at a steady state.
From the above assumption, the temperature profile at any fixed location (temperature
gradient at a fixed spatial position along the ROT) on ROT does not change with time,

^ =0 (7.6)
dt

The above two assumptions were adopted in all published literature and are used in this
study. Under these assumptions, the differential equation for the ROT cooling process
becomes:

166
d dT. d ,, dT. d dT. _ dT
(k) + (k) + (k) + Q = c pu p (7.7)
dx dx dy dy dz dz dx

Equation (7.7) presents a three-dimensional steady state heat transfer problem, and
theroretically there would be no problem to getting the numerical solution. Some
researchers [78] did simulations based on those simplications. The results showed that
this kind of 3D simulation has modest effect on the accuracy of the coiling temperature
prediction. However, the solution procedure is very time-consuming and not economical
because there would be more than 100 thousand elements in the length direction, even
with large ratio of element dimensions in the length direction to those in the width and
thickness. Therefore, more simplications are required to render a more economical
simulation without sacrificing too much accuracy.
3. Most researchers introduced another assumption that the heat transfer or flux in the
strip width direction is neglected:

dT_
0 (7.8)
dy

This assumption is only valid for the uniform boundary condition and temperature along
the strip width. Because the water on the top surface will flow out transversely for both
laminar and curtain cooling, the heat flux or heat transfer coefficient along the strip width
may change.
On the other hand, the flow rate along the strip width may be intentionally designed to be
non-uniform, slightly higher in the middle, decreasing with the distance from the centre
line of the strip, in order to get a final uniform coiling temperature.
These effects are, however, ignored, and by adopting this assumption there is:

d ., dT. d .. dT. _ dT
(k) + (k) + Q = c pu p (7.9)
dx dx dz dz dx

4. It can be easily proved that the heat transfer due to the bulk motion in the moving
direction (x-direction) is much larger than the heat conduction in the same direction. A
comparison is carried out using Peclet number, Pe:

n convection ux .
Pe = = ty (7.10)
diffusion oy

167
where a is the diffusion coefficient, and l is the typical strip length in the cooling zone
c

under one jet.


The Peclet number is greater than 20,000, i.e. much greater than unity, for typical hot
strip mill conditions; therefore, the heat conduction in the moving direction may be
neglected,

l>- (7
- n)

dx ox

The above assumption renders equation (7.9) to take the simple form of:

d dT dT
(k) + Q = cpu (7.12)
dz dz ox
Equation (7.9) or (7.12) still describes a 2D heat transfer problem in the length and
thickness directions. A novel approach is developed in this work to study the intensity of
longitudinal heat transfer [66] and will be discussed in Section 7.3.
5. A commonly used approach to further simplify the calculation is the parameter
replacement under the assumption that the strip velocity is constant. This approach is in
essence a coordinate transformation In the form:
x = ut x (7.13)
It is noteworthy that the time is estimated from the beginning of the cooling process, and
that the coordinate system becomes a moving one, i.e., the Euler coordinate. The
temperature is material-point oriented.
Substitution of equation (7.13) into equation (7.12) or (7.9) leads to:

d dT. ^ dT t
(k ) + Q = c p p (7.14)
dz dz dt
d dT. d dT. dT , . n
(k) + (k) + Q = c p-- (7.15)
dy dy dz dz dt
p

Combining the above assumptions, the heat transfer problem becomes a transient ID
(equation (7.14)) or 2D (equation (7.15)) with changing boundary conditions. Note that

in equations (7.14) and (7.15) is the material derivative of temperature with respect

to time.

168
Thus, while the temperature distribution through the thickness (ID) or on the transverse
cross section (2D) at each time instant must be solved simultaneously, the solution in the
moving direction can be solved sequentially, which greatly simplifies the calculation and
reduces the C P U time.

7.2 Modelling and Implementation

There are several issues to be resolved to achieve an appropriate and accurate simulation
of the cooling process of steel strip on ROT. In the following, the handling of the main
ones is addressed.

7.2.1 Numerical method

Both the finite difference and the finite element methods were used to numerically solve
equation (7.14) or (7.15) to simulate the temperature field and to predict the coiling
temperature in literature [70-75].
In this work, F E M is used. For ID model, a uniform mesh is used, and the discretization
in the thickness direction can be automatically realized with the input of the strip
thickness and the number of elements (nodes) required [65]. For the transverse 2D
modelling, it is assumed that the cooling conditions are of the same profiles for the top
and bottom surfaces and are symmetric to the width centreline. Thus, only part of the
cross section is discretized.

7.2.2 Initial condition

A literature survey reveals that a uniform initial temperature assumption is most widely
adopted; that is, there is no temperature difference through the thickness (ID) or through
the thickness and width (2D).

T =T C t = 0, -h!2<z<hl2
or -hl2<z<hl2-bl2<y<bl2

If the boundary conditions on the top elements and the bottom ones are the same, this
assumption will again reduce the element numbers required for discretization.

169
Althrough it is very practical, this assumption neglects the fact that the initial temperature
is far from uniform, due to the different cooling conditions on the top surface and the
bottom surface during the rolling process prior to the cooling process.
For our ID model, the initial temperatures through the thickness may be different. When
implementing this issue, two temperatures are always required, one for the top element,
one for the bottom element. Linear interpolation is used to assign the initial temperatures
for all other elements (nodes).

7.2.3 Boundary condition

1. Literature review

The reliablity of the simulation to ROT cooling will depend, to a great extent, on the rigor
and the accuracy of the mathematical descriptions of the boundary conditions. On the
other hand, it is a difficult issue to accurately determine the boundary conditions. Many
approaches have been explored and results have been published. In general, they may be
classified into three methods.
Some researchers used a very simple approach in which a variety of constant HTC values
was assumed from the start to the end of the cooling process. In this approach, the target
is normally the coiling temperature and, not the time history profile of temperature [67-
71]. Various values of H T C were assumed by different researchers; e.g., 464-1394
W/m -C (1670 to 5020 kJ/m -h-C) in references [67-68] and 2500^1500 W/m -C in
2 2 2

[69-71]. The HTC value for the best prediction of the coiling temperature was assumed
the one for the processing condition and used for later analysis, e.g., the simulation of
microstructure evolutions. Although they are very practical, these approaches may
underestimate the actual cooling rates and their effects on the microstructure evolution
and finally on the properties.
For the second method, different HTC values were adopted for the cooling zones directly
around the cooling headers and the cooling zones between cooling headers [72-75]. For
example in reference [72], the HTCs were 5500 W/m -C and 100 W/m -C for the
2 2

impingement zone and the parallel zone, respectively. Therefore, it is necessary to


determine the boundaries between the different zones and the H T C values or their
distributions for these zones.

170
The last approach is to use H T C correlations based on thermodynamic principles that
consider the interaction between water and hot surfaces [77-78], or based on empirical
equations [79-84]. In these equations, the local water temperature, strip surface
temperature, impingement angle and moving speed of the strip may be required to
compute the local H T C values. For example, Guo et al. [79-81] used the following
equation to determine the absolute maximum HTC of each pair of headers:

K = K^f-r ( f r &v {^-r (7.17)

where V , t, T is the speed, thickness, surface temperature of the strip, respectively, Q


p s w

is the header flow rate; the corresponding values with the subscript "0" are the reference
ones. K and ci to C4 are curve-fitting constants determined by the operation data and
experimental ones. Obviously, the reliability and precision of this approach depend
greatly on those of the correlations for HTC.

2. Division of the cooling surface

A hybrid approach is conducted in this study. First, the cooling surface is divided into a
few subregions. Second, different correlations are used for each subregion.
As shown in Figure 7.2, the cooling surface for an active or working jetline (jetline is on
and water impinges onto the strip) is divided into a maximum of 5 cooling zones: air
cooling zone-B, parallel water cooling zone-B, impingement water cooling zone, parallel
water cooling zone-F and air cooling zone-F. However, there is only an air cooling zone
for a dead jetline (jetline is off and no water flows out).

Air Cooling Parallel Impingement Parallel Air Cooling


Zone-B Zone-B Zone Zone-F Zone-F

Figure 7.2 Cooling zones under one active jetline

The actual divisions of the cooling zones and their corresponding lengths depend on the
on-off status of two adjacent jetlines.

171
The critical issue is to determine the length of the impingement zone. Two assumptions
are made for the sake of calculation simplicity in this study. The first one is that for any
active jetline, the impingement zone is symmetrical to the jetline and its total length is
proportional to the nozzle diameter. The ratio of the impingement zone length to the
nozzle diameter is defined as the width coefficient in this study.
The second assumption is that there is no air cooling zone for a working jetline. Thus, the
parallel zone extends to the middle of two adjacent jetlines, and the length of one parallel
zone is equal to half of the distance between two adjacent jetlines minus the impingement
zone width. Because the jetline distances may be different, the lengths of the parallel
zone-B and zone-F may not be equal even if the impingement width is the same.
This setup is schematically shown in Figure 7.3. If the jetline /' is active, the characteristic
length Z, hai
C is divided into three subregions: the parallel zone-B: 0.5(Xj-Ijm j),
P the
impingement zone Z,j j
mg and the parallel zone-F: 0.5(Zi+i-Zj i).
m g

If the jetline / is not active, all Z ,

J uu
c n a i belongs to the air cooling zone.

* Lj * * Lj+i *

1 ' 1 !
1
1 |""Limpi"-|

^0.5Li*-h*~' i+i*~ 5 L

~*"Lchai 0.5(Li+Lj+i)^
=

Figure 7.3 Design of the cooling zone lengths

3. Correlations used in this study

i) Air cooling zone

While the strip surface is exposed to the air, the heat loss is due to forced convection and
radiation. The heat transfer may depend on the moving speed and the surface
temperature. A correlation is set up and used based on the equation in the reference [79]

172
for the H T C of forced air convection with various strip speeds and the equation in the
reference [80] for the HTC of radiation:

Kir = -5 V
4
p
0926
+ [-3038 + 0.9547; + 2.14 x 10" (7; + 273) ]/(7; - 2 0 ) 8 4
(7.18)

where is the H T C in W/m -C, V is the strip speed in m/s, and T is the surface
2
p s

temperature in C.
The relationship of HTC to the surface temperature at two levels of the plate's speed is
shown in Figure 7.4. It is evident that the H T C is mainly determined by the surface
temperature while the plate moving speed has only a slight effect on the HTC.

600 I : 1

100 300 500 700 900


Surface temperature,C

Figure 7.4 HTC in air cooling zone

ii) Impingement zone and parallel zone

The empirical model for H T C in the reference [83] is used to calculate the H T C in the
impingement zone. The formula is:

h imD = 2186.7 (T /1000) (V 120)~


s
25 4

P P
(7.19)
= 2.292xl0- -r -F; 4 2 5 0 4

where h, mp is the HTC in the impingement zone in W / m C , T is the surface temperature


2
s

of the strip in C, and the V is the strip speed in m/s.


p

173
Figure 7.5 illustrates the HTC profiles. It can be seen that the surface temperature is the
critical factor affecting the HTC, and that the strip speed has a larger effect when the
surface temperature is at a higher level.
For the common operation conditions on ROT, i.e., when the strip speed is 6-10 m/s and
the surface temperature is 500-900 C, the H T C varies from 500-2700 W/m -C. This
2

range seems fairly reasonable in a comparison with the data in the literature.

4500

4000

3500

3000
O
"g 2500

O 2000
l-

1500

1000

500

100 300 500 700 900

Surface temperature,C
Figure 7.5 HTC in impingement cooling zone

It is found that equation (7.19) with a modification to the coefficient is also the best
choice of the HTC correlation for the parallel zone in this study. The coefficients are
presented in Section 7.4.1.
When the whole thickness is modelled, both the top and bottom surfaces should have the
appropriate boundary conditions; when half the thickness is modeled, the centre is
assumed thermally insulated and appropriate boundary condition are applied on the top
surface. In this study, the second option is used because there are no appropriate
correlations for the bottom impingement cooling zone.

7.2.4 Temperature dependences of thermophysical properties

The thermophysical properties of material may change with temperature. In general, the
specific heat decreases with the decrease of temperature while the conductivity decreases
before phase transformation and increases again thereafter.

174
There are two approaches to deal with this kind of dependence. In most cases, the
changes were neglected, and average values were used. Another approach is to use
correlations of properties as function of the temperature. In such cases, the simulation
will be nonlinear and iterative solutions are required [70-71, 74-75].
In our implementation, the thermal conductivity and specific heat are considered as
temperature dependent are considered and specific available data may be used to find the
best fit for the properties.

7.2.5 Latent heat

During the cooling process on ROT, the deformed austenite will change to other phases at
some temperatures. The latent heat is released due to this phase transformation and may
affect the coiling temperature and finally the properties of the strip.
Some researchers have neglected or ignored the effect of the latent heat [85] whereas
others have included the effect. In a more appropriate simulation, the transformation of
austenite to ferrite and pearlite is usually taken account into, and the effect of the
corresponding latent heat on the cooling temperature profile is analyzed [70-71].
In our implementation there is an option to consider latent heat in the analysis. In this
case, the heat generation per volume is calculated by:

Q = HF (7.20)

where H is the mole of the material considered, W/kg; F is the transformation rate of
austenite to ferrite and pearlite per volume and will be calculated in the module for
microstructural evolution.

7.2.6 Microstructure evolution and property prediction

It is becoming customary to integrate models for microstructural evolution and property


prediction in heat transfer analyses [83-88]. Most of these models have paid attention to
the transformation of austenite to ferrite and pearlite, and to the grain sizes and proportion
of ferrite and pearlite or bainite. Properties such as yield strength and toughness may be
calculated from the compositions of phases. Some models [86] utilized more complicated
equations to take account of the dislocation change and its effect. Few reports
investigated the distributions of microstructures and properties along the strip width
and/or length [83].

175
The model developed in this study is intended to predict the composition of
transformation products and the mean ferrite grain size as well as the mechanical
properties of the steel considered. Equations describing the transformation of austenite to
ferrite, pearlite, and bainite would be implemented and applied under the cooling
conditions of strip on ROT. The correlations for calculating the product properties will
also be included in the final full version of this program. These correlations are not
presented here because they are lengthy, and this study focuses on the prediction of the
coiling temperature.

7 . 2 . 7 Program structure

A l l of the above issues have been implemented in an F E M code to simulate the cooling
process on ROT. This code for ID modelling is called RealROT. The RealROT is
written as modular interactive program. The code is composed of a read-in module, a
ROT data module, an initial condition module, a material module, a H T C correlation
module, a microstructure module and an F E M module.

1. Read-in module

This module is designed to be interactive and user friendly. Main parameter inputs to the
program include: the material type identification number, the strip thickness, the top and
bottom surface temperatures, the element (node) number (even number) required to
discretize the strip, and whether the heat generation should be considered or not. In
addition, the user is required to specify the heat flux or the H T C condition as the
boundary condition. The residual strain is also input here, i f applicable.

2 . Initialization module

This module performs meshing, specification of initial temperature and time stepping.
The half strip thickness will automatically be discretized into elements (nodes) that are of
same size. It was proven that 40 elements are enough to get reasonably accurate result.
The module also calculates the initial temperatures of the strip after the strip is meshed.
The calculation is based on the two input temperatures and the linear interpolation.. In
addition, the residual strain is assigned to each node.

176
The common method for time stepping is first to determine the total number of time steps
or the analysis duration and then to calculate the time step size. Existing approaches
generally use equal time steps sizes. This has the deficiency of need to track the location
of the target material slice and to determine in which zone the target slice exists. In
addition, the time step size should be small enough to avoid jumping over the water
cooling zones, especially the impingement ones. Normally on such cases, the total
number of time step is huge and the calculation time is then considerable.
A novel approach is used here for time stepping. First, five time step numbers for
different cooling zones are defined as follows:

NFAIR: number of time steps for the entry air-cooling zone


NLAIR: number of time steps for the exit air-cooling zone
NBAIR: number of time steps for the air-cooling zone if a jetline is not active
N P A R A : number of time steps for the parallel cooling zone i f a jetline is active
NIMPG: number of time steps for the impingement cooling zone i f a jetline is active
Thus, the total number of time steps depends greatly on the number of active jetlines and
on the time steps for the parallel and impingement water cooling zone. In addition, each
time step is identified with a flag to indicate its belonging to the different zones. With this
approach, tracking is not needed and no water cooling zone will be surpassed.
Second, each number of time steps can be determined according to the accuracy
requirement and the strip speed. In fact, the number of steps for the air cooling zones can
be a small value, which greatly reduces the total number of time steps. Therefore, for this
approach, the time step size is not constant.

3. R O T data module

This module deals with the ROT setup, including the ROT length, the jetline number, the
jetline on-off status, the jetline distance, the nozzle number and diameter for each jetline,
and the water flow rate for each jetline.
The common approach is first to set up the water banks, including the number and
distance, and then to set up the jetlines under each water bank, including the number and
distance of the jetlines. In this program, instead, the water bank information is not
handled; i.e., ROT directly consists of the jetlines. Numerically, this approach saves one
tag to identify the belongings of the jetlines. Practically, this creates an advantage to

177
study the spacing o f jetlines on the cooling effect independently and provide more

information for the design o f new R O T .

In this program, the jetline distance means the spacing between the current jetline and the

previous one. The jetline distance for the first jetline is the distance between the first

jetline and the finishing m i l l stand.

There is a special arrangement for the length o f the cooling zone for the first and the last

jetline i f they are active. For the first jetline, the Parallel-Zone-B is equal to the Parallel-

Zone-F, which is calculated, while the Parallel-Zone-F is set up to the Parallel-Zone-B,

which is also calculated.

4. Material module

This module calculates the thermodynamic properties o f the materials, including heat

conductivity, heat capacity and density. The constant values or correlations for the

material properties are implemented in this module. A l l the correlations are assumed

being linear. This module is called each time step when the temperature dependences o f

the properties are considered.

5. HTC correlation module

In this module, H T C is performed and used as boundary condition to avoid strip

temperature lower than the ambient (water) temperature.

6. Microstructure module

The models for the microstructure evolution and final mechanical properties are input i n

this module. The composition o f the final microstructure, i.e. the percentages o f ferrite,

pearlite and bainite, is estimated. This module is not discussed in this chapter.

7. F E M module

This module assembles all the necessary data, converts them to F E M format, solves the

F E M equation, and outputs the required data.

178
7.3 Studies on ROT Cooling

The developed program is used to study the industrial R O T cooling process and some

typical results are presented i n the following section.

7.3.1 Case study

In this section, simulations are performed using the RealROT to the practical cooling

processes on a full-scale R O T . The predicted coiling temperatures are compared with the

measured ones.

1. ROT setup

Figure 7.6 schematically shows a typical industrial R O T . This R O T consists o f five top

banks and five bottom banks. Each bank has six jetlines. In the following, only the data

about the top banks are presented.

The entry length ( L I ) and the exit one (L3) are 8.46 m and 70.6 m , respectively. The

distance between two adjacent banks (L2) is 2.74 m ; and the distance between jetlines

within one bank ( b l ) is 1.37 m. The strip speed is up to 16 m/s.

For each top jetline, there are fifty-one circular nozzles o f 30 m m i n diameter. A l l o f the

nozzles impinge the cooling water vertically onto the strip surface. The spike (distance)

of the top nozzle exit to the strip surface is 1.79 m. The cooling water has a temperature

of 20-40 C and a flow rate o f 39.2 1/s per jetline.

The starting and ending temperatures o f the cooling process are measured by pyrometers

at the exit o f the finishing m i l l ( F M ) and at the entry o f the down coiler (DC). The ending

temperature o f the cooling process is usually called the coiling temperature.

CT
Bunk No.
Bank No.S

HilrilUlfH 1 iJUPHflU
IS !! I!!l II! Illl III I iii till illl Illll!!l Illlilil l!!l DC
III !! I!!l II! IIH III Ilil mi Illl
l!!| l!!l Illl Illl nn mi l!!l
II II mi in mi in /mm
L_J L_J
b2 1
hi l-'M: finishing mill
DC: clown coilcr
l-'MT: finishing mill exit temperature
CT: coiling temperature

Figure 7.6 Schematic illustration o f a typical R O T

179
For the products with the same thickness, the cooling temperature profile along ROT and
the coiling temperature can be changed and controlled by adjusting the strip speed and
the on-off status of the jetlines (banks) for the different temperatures at the exit of the
finishing mill stand. Generally, the bank state is coded to present the combinations of the
on/off of jetlines. The code used in the particular case examined is shown in Table 7.1.

Table 7.1 Bank state code


Code Down
2 6 12 14 22 28 30 54 60 62 118 124 126
number stream
1 On On On
2 On On On On On On
Jet 3 On On On On On On On On On
Line 4 On On On On On On On On
5 On On On On On On On On On On On On
6 On On On On On On On On On

Examples of the strip production data are shown in Table 7.2. Selected from 1160
production cases, these examples show the diversity of production parameters. For these
1160 examples, the starting cooling temperature varies from 861 C to 938 C, and the
strip thickness varies from 2.03 mm to 7.61 mm. The strip speed changes from 3.52 m/s
to 10.45 m/s. The strip material is DQSK.

Table 7.2 Examples of cooling parameters


Finish Water Water Strip Thick- Bank State Code
No. Temp. Temp. Flow Speed ness BK1 BK2 BK3 BK4 BK5
C C Rate m/min mm T B T B T B T B T B
2 884.61 25.78 F 627.1 2.03 124 124 62 62 0 0 0 0 30 30
5 879.12 24.34 F 619.0 2.03 124 124 126 126 2 2 0 0 14 14
17 893.77 26.37 F 625.7 2.03 124 124 126 126 14 14 0 0 14 14
34 881.26 28.29 F 592.5 2.17 124 124 126 126 6 6 0 0 14 14
40 873.63 25.78 F 467.1 2.29 124 124 . 2 2 0 0 0 0 14 14
55 893.95 28.77 F 598.4 2.29 124 124 126 126 30 30 0 0 30 30
68 877.42 26.10 F 557.4 2.42 124 124 126 126 126 126 14 14 14 0
79 876.86 26.31 H 506.3 2.54 124 124 30 30 0 0 0 0 14 0
86 895.96 25.19 H 585.8 2.79 124 124 126 126 62 62 0 0 14 0
99 890.02 23.16 H 451.5 2.92 124 124 62 62 0 0 0 0 6 0
130 907.12 24.18 H 516.9 3.26 124 124 126 126 126 126 6 6 14 0
149 901.75 25.89 H 467.5 3.56 124 124 126 126 62 62 0 0 14 0
155 891.03 22.41 H 424.8 3.68 124 124 126 126 6 6- 0 0 14 0
160 901.46 29.25 H 393.3 4.19 124 124 126 126 2 2 0 0 126 126
165 891.14 21.13 H 303.9 5.08 " 124 124 126 126 14 14 0 0 6 6
178 904.03 24.71 H 343.5 5.46 124 124 126 126 126 126 62 62 14 0
192 902.04 26.15 H 225.6 7.61 124 124 126 126 126 126 0 0 14 14
BK: Bank; T: Top jets; B: Bottom jets; F: Normal flow rate; H: High flow rate

180
2. Results and discussion

A l l 1160 cooling processes are simulated by the developed program. In the simulations,

half o f the thickness is discretized with 40 elements. For reducing the calculation time,

only 2 time steps and 4 time steps are assigned for each impingement zone and for each

parallel zone, respectively. A l s o , the number o f time steps is 2 for the air cooling zone

between jetlines, and it is 10 for both o f the entry and exit air cooling zones. The number

of the total time steps is only 178 to 280 depending on the on-off status, as shown in

Table 7.3. The smallest value of time step is 0.0022 s. Numerical experiments showed

that using more time steps has little or no effect on the predicted coiling temperatures

(Table 7.4).

The width coefficient o f the impingement cooling zone to the nozzle diameter is fixed at

1.5. Moreover, the H T C factor for the parallel cooling zone is found to be 0.15 and 0.25

for the lower and higher flow rate, respectively. The temperature dependence of

conductivity is considered while the latent heat is not considered.

Table 7.3 Examples o f calculation results

Measured Calculated
Strip Time
Strip Coiling Coiling
Thickness Step T IT
No. Temp, Temp,
mm No.
T C T ,C
C

2 2.03 192 664.31 651.0133 1.02042


5 2.03 200 639.21 640.6026 0.99783
17 2.03 216 641.96 641.9504 1.00002
34 2.17 208 642.76 643.0853 0.99949
40 2.29 152 639.95 646.4410 0.98996
55 2.29 232 651.39 649.4726 1.00295
68 2.42 264 649.56 627.0100 1.03596
79 2.54 176 638.64 630.2401 1.01333
86 2.79 232 644.25 639.1925 1.00791
99 2.92 176 638.88 640.4161 0.99760
130 3.26 256 640.20 632.1535 1.01273
149 3.56 232 637.65 643.3785 0.99110
155 3.68 208 641.49 642.5314 0.99838
160 4.19 224 642.91 642.7898 1.00019
165 5.08 208 640.36 637.2108 1.00494
178 5.46 280 639.65 630.0060 1.01531
192 7.61 240 639.38 629.7765 1.01525

181
Table 7.3 shows the calculation results of the coiling temperatures for the production
examples shown in Table 7.2. It is apparent that the predictions are in good agreement
with the measured coiling temperatures.
Figure 7.7 illustrates the ratios of the measured coiling temperatures to the calculated
ones for the first 200 production cases among 1160 cases. It is clear that the prediction is
within 8%, which is a reasonable and acceptable engineering range. Note that the
prediction accuracy for all 1160 cases is also within 10%.
The general prediction accuracy postulates that the setup of the modelling and the
coefficients are practical.

1.10

1.08

1.06

1.04

1.02
o
t 1.00
H
0.98

0.96

0.94

0.92

0.90
50 100 150 200
Case index

Figure 7.7 Prediction accuracy

Figure 7.8 shows the profiles of the top temperature and centre one along the thickness
for Case 192. For this cooling case, four banks are active, which can be easily observed
from the cooling curves. It is clear that for each of the water cooling zones (each active
bank and jetline) the centre temperature is higher than that at the top surface. It is also
evident that the top surface recovers between two banks, and that the recovery is much
more obvious after the first bank and the third one. The fourth bank is not working and
the strip experiences air cooling; therefore, the top surface quickly recovers to the centre
one. The top temperature and centre one are nearly identical after the strip enters the exit
air cooling stage.
The predicted coiling temperature is approximately 629.78 C and is lower than the
measured average one of 639.38 C. However, this level of accuracy is acceptable.

182
Figures 7.9 and 7.10 illustrate the HTC values and the cooling behaviour around the first
active bank (Bankl), respectively. At the air cooling stage, the H T C is small; the
temperature difference between the top and the centre is small. When the strip enters the
first parallel cooling zone, the H T C increases sharply and thus the surface temperature
remarkably decreases accordingly. Note that the centre temperature decreases slightly.

3500

3000 <

*
<
<
< >
2500 f '

p
"g 2000

O 1500
I-
1000

500

0
\
1.5 2 2.5 3 3.5 4 4.5
Cooling time.s

Figure 7.9 HTC profiles around the first bank for Case 192

183
As expected, the HTC increases to a higher value as the strip enters the first impingement
cooling zone, and causes the surface temperature to drop suddenly. Because the
impinging cooling time is very short, the cooling effect is confined within a limited depth
beneath the surface. The heat transfers from the slightly cooled centre part to the surface,
and then the surface temperature recovers. This sharp drop and recovery of the surface
temperature repeats for each active jetline.
After the water cooling stage, the strip enters the spacing between two banks where air
cooling is dominant. The surface temperature promptly comes back nearly to the centre
temperature.
A l l of the cooling phenomena are reasonable and consistent with those in the literature.

920

900

O 880
P
3 860
2
g_ 840
E

820

800

780
1.5 2.0 2.5 3.0 3.5 4.0 4.5
Cooling time.s

Figure 7.10 Temperature profiles around the first bank for Case 192

7.3.2 Parametric study

In this section, Case 192 is chosen to study the effects of various parameters on the
coiling temperature, such as time stepping, impingement zone width (nozzle diameter),
production parameters, material properties, and ROT configuration.
1. Time step and impingement zone width
It is generally expected that time stepping may affect the simulation results and accuracy.
As shown in Table 7.4, the predicted temperatures in this case have only a slight

184
difference as the total number of time steps increases as much as 8 times. This means that
the size of time step (below the ones chosen for the simulation) has little effect on the
predicted temperature. Therefore, the results in Section 7.4.1 can be assumed being
independent of the time stepping. In addition, Set 1 in Table 7.4 can be used for other
parameter studies.

Table 7.4 Effect of time step on the coiling temperature for Case 192

Set NFAIR NLAIR NBAIR NPARA NIMPG TOTAL Tc, C


1 10 10 2 4 2 240 629.78
2 10 10 10 10 10 720 629.90
3 10 10 20 20 20 1420 630.00
4 100 100 20 20 20 1600 629.96
5 250 250 20 20 20 1900 629.96

The impingement zone width has dual-meanings. First, it is a numerically assigned value
for a given nozzle and it may be adjusted to better predict the coiling temperature.
Second, the impingement zone width may also be linked to the nozzle diameter and it
should generally increase with the increase in nozzle diameter.
The effect of the impingement width on the coiling temperature is shown in Figure 7.11.
It is obvious that the coiling temperature decreases nearly linearly with the increase of the
impingement width. This is reasonable because the H T C in the impingement zone is
much higher than that in the parallel zone.

635.0

620.0 I 1 ' 1 1 1 1
0.5 1.0 1.5 2.0 2.5 3.0 3.5

Width coeff

Figure 7.11 Effect of impingement width


on the coiling temperature for Case 192

185
Note that the increase of the impingement zone width implies a decrease of the parallel
zone length under the current setup of the cooling zone divisions. This may be the reason
why the predicted coiling temperature decreases only 10 C as the impingement zone
width triples.
Based on the above finding and the general ratio of the impingement zone width to the
nozzle diameter, the coefficient of 1.5 is reasonable and is used for the other parameter
studies.

2. Production parameters

The production parameters inluded here are the starting cooling temperature, strip speed
and thickness, and flow rate. The production parameters are artificially changed to
investigate their effects on the coiling temperature and to check the reliability of the
developed program.
Figure 7.12 illustrates the effect of the starting cooling temperature on the coiling
temperature. Simulations show that the coiling temperature decreases from 661 C to 572
C with the decrease of the starting temperature from 1000 C to 750 C, and that the
relationship can be correlated as linear for the common temperature range. Every drop of
50 C in the starting temperature will lead to a decrease of 18 C in the coiling
temperature. This means that a certain decrease of the coiling temperature cannot be
realized by reducing the starting temperature with a same magnitude; i.e., the influence of
the decrease in the starting temperature is diminished.

680

560 I 1 ' ' ' ' ' 1

700 750 800 850 900 950 1000 1050

Starting tem perature,C

Figure 7.12 Effect of the starting temperature


on the coiling temperature for Case 192

186
The strip speed is a critical parameter in the strip production. In general, the cooling time
decreases as the strip moves faster. The effect of the strip speed on the coiling
temperature is depicted in Figure 7.13. For the fixed on/off status of the jetlines, the
coiling temperature increases significantly from 585 C to 700 C as the strip speed
changes from 3.0 m/s to 5.5 m/s. It is clear that the strip speed has a significant effect on
the strip coiling temperature.

720

560
2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0

Strip s p e e d , m/s

Figure 7.13 Effect of the strip speed


on the coiling temperature for Case 192

700

450
1.61 3.11 4.61 6.11 7.61 9.11 10.61
Strip t h i c k n e s s , m m

Figure 7.14 Effect of the strip thickness


on the coiling temperature for Case 192

Similarly, for the fixed on/off status of the jetlines, the coiling temperature increases
considerably from 473 C to 660 C as the strip thickness changes from 3.11 mm to 9.11

187
mm, as shown in Figure 7.14. The total heat amount is proportional to the thickness and
is larger for a thicker strip than for a thinner one! As the heat that can be extracted by the
cooling apparatus is mainly determined by the on-off status, the coiling temperature is
higher for the thicker strip.
The water flow rate can be another important parameter affecting the coiling temperature.
In the developed program, this parameter is not explicitly present in any correlation. It is
found by the simulations that to obtain consistent and satisfactory predictions a larger
coefficient should be used for the H T C correlation for the parallel zone when a higher
water flow rate is used. This is consistent with the discovery in Section 6.4. For Case 192,
the coiling temperature is 629.78 C as the water flow rate is of the higher level. If the
water flow rate is changed to the lower level, the coiling temperature becomes 677.38 C
with an increase of about 50 C.

3. Material properties

Material properties refer to the strip conductivity and specific heat in this discussion.
Their effects on the coiling temperature are shown in Figures 7.15 and 7.16. In these
investigations, constant values are input.

635.0

632.5
O
O

S 630.0
3
e

I 627.5

jf 625.0
o
o

622.5

620.0

20 30 40 50 60

Conductivity, W/mC

Figure 7.15 Effect of the strip conductivity


on the coiling temperature for Case 192
Figure 7.15 shows that the conductivity has a slight effect on the coiling temperature. The
coiling temperature decreases only by 10 C as the conductivity triples. It is interesting to
point out that the coiling temperature is 627.17 C with a constant conductivity of 30

188
W/m-C, which is just the average value for the temperature range of 100-1000 C.
Recall that the coiling temperature is 629.78 C as the temperature dependence of
conductivity is considered. Thus, an average value may be a good option to replace the
temperature dependence. This approach may greatly reduce the calculation time, and is
used in the 2D modelling.

670 , ,

610 I 1 1 1 '

440 480 520 560 600

Specific heat, W/kg

Figure 7.16 Effect of the strip specific heat


on the coiling temperature for Case 192

In all calculations except those in Figure 7.16, the specific heat is fixed at 480 W/kg. The
specific heat of the DQSK has a weak temperature dependence and does not change
much. The values of the specific heat in Figure 7.16 are artificial and may not be
practical.
While the increase of the conductivity slightly reduces the coiling temperature, the
increase of the specific heat increases to some extent the coiling temperature. This is
because a higher specific heat represents a higher heat amount in a unit volume. This
result is similar to the case of increasing the strip thickness.

4. R O T configuration

In this section, the jetline distance b l and the bank distance L2 are changed to investigate
the effect of ROT configuration on the coiling temperature.
Two situations are designed for changing the jetline distance b l . In the first one all jetline
distances are changed while in the second one only the jetlines within the first bank is
adjusted to new values. The results are shown in Figure 7.17.

189
It is clear that the coiling temperature decreases when the distances b l increase. This is
because in this program the increase of jetline distance is equivalent to the increase of the
parallel zone lengths. The water cooling time increases as the parallel zone lengths
increase. Therefore, the coiling temperature decreases. The nearly linear correlations also
reveal that the decrease of the coiling temperature is mainly caused by the increase in the
cooling time.
Comparing the two curves leads to another conclusion that the contribution from the first
bank to the reduction of the coiling temperature is about 26.9%, only slightly higher than
25%. Recall that four banks are active and 20 jetlines are working for Case 192 (Tables
7.1 and 7.2). Five jetlines in the first bank are active and are exactly one quarter of the
total working jetlines. The reason the contribution of the first bank is slightly higher may
be because the surface temperature at the first bank is at the highest level for the four
banks and the HTC is higher.

700

Distance b1, m

Figure 7.17 Effect of the distance b l


on the coiling temperature for Case 192

The increase of the bank distance may create two phenomena: first, the air cooling time
increases; second, the top surface temperature may recover close to the centre
temperature, which increases the cooling efficiency during the subsequent water cooling
process. Both of the phenomena result in a decrease in the coiling temperature. The curve
shown in Figure 7.18 verifies this analysis. The reason why the change of the coiling
temperature is very small is that the increase of the cooling time is only about 1 s.

190
A l l of the findings discussed above are consistent and reasonable. This means that the
developed program is reliable and can be used to predict the coiling temperature and the
profiles of cooling temperature with an acceptable accuracy.

640 I 1

620 I ^- ' ' 1


2.1 2.4 2.7 3.0 3.3
Distance L 2 , m

Figure 7.18 Effect of the distance L2


on the coiling temperature for Case 192

7.4 Investigation of Necessity of 2D Simulation of R O T Cooling

It is proposed that the simulation of ROT cooling be performed using 2D or 3D


modelling to uncover the effect of heat transfer in the longitudinal (moving) direction
or/and in the transverse (width) direction on the temperature profile and to get a more
accurate prediction of the coiling temperature. In this section, both 2D transverse and 2D
longitudinal modelling are conducted to assess this matter.

7.4.1 2D transverse modelling

1. Modelling aspects

One of the basic requirements for the ROT cooling system is that the cooling must be
uniform in every location of the top and bottom surfaces, edge and center, and leading
and trailing ends. Setting appropriate transverse gradient for water flow from centre to
edge along with edge masking devices may greatly reduce or prevent transverse
temperature gradient and buckling leading to edge weaves. However, the cooling along

191
the width cannot be uniform in practical productions. A difference of 10% between the
HTCs at the width centre and at the edge was reported and used in simulations in
literature [83].
For the cooling with multiple jets in one jetline, there is an interaction zone between two
adjacent jets except for the parallel and impingement zones. The local HTCs are
definitely different between two adjacent jets.
Though the width of strips depends on the rolling facilities and the product specifications,
it is, generally, in the order of one meter and it is not practical to model the whole width.
Note that the distance between two adjacent jets in one jetline is usually 200-400 mm and
that the temperature gradient may be changed by artificially adjusting the magnitude of
the H T C difference. Therefore, a width of 150 mm is assumed in the modelling by
assuming that the average jet distance is 300 mm and the cooling is symmetrical to the
middle of the two jets.
Case 192 is simulated with 2D modelling. Thus, a cross-section of 150x3.805 mm is
discretized with 150 uniform elements in the width direction and 10 fine-to-coarse
elements in the thickness direction.

Jet centre
Symmetry
Middle of two jets
Symmetry

HTC distribution

S
3

a
2
o

O
O 10 20 30 120 130 140 150

Width, mm

Figure 7.19 2 D transverse modelling

192
The H T C on, the top surface is still calculated based on equation (7.18) with a linear
correlation in the function of the width. The nominal H T C difference is assumed being
2.5% to 15%> across the 150 mm. The other three sides are assumed thermally insulated
because all of them are have symmetrical conditions. The above arrangements are
schematically shown in Figure 7.19. Note that the dimension sizes are not to scale.
The reason why the difference in HTC is nominal is that the actual HTC also depends on
the surface temperature. The surface temperature is higher at a point with a smaller HTC
coefficient than at a point with a larger H T C coefficient. Therefore, the actual H T C
difference is slightly less than the nominal one. The time stepping scheme is chosen to
Set 1 in Table 7.4.

2 . Results and discussion

The simulation results are shown in Figure 7.20. In this graph, there are two curves for
the coiling temperatures, one with higher HTCs (H-HTC) and the other with lower HTCs
(L-HTC).
It should first be mentioned that the coiling temperatures at the two corner points are
629.64 C if the HTC is uniform. Recall that the coiling temperature is 629.78 C by ID
model with 100 elements in the thickness direction. These two temperatures are very
close, which implies that the number of elements (higher than 40) has little effect on the
predicted coiling temperature, and that 40 elements are enough to get an accurate result
with respect to the discretization.

625
0,0 2.5 5.0 7.5 10.0 12.5 15.0
HTC d iffere nee,%
Figure 7.20 Effect of transverse HTC difference
on the coiling temperature for Case 192

193
As expected, the coiling temperatures at the top corner point with higher HTCs are
always lower than those at the top corner point with lower HTCs. This is reasonable
because more heat is extracted away when H T C is higher. It is clear that as the H T C
difference increases from zero to 15%, the coiling temperatures at the top corner point
with higher HTCs do not evidently change while the coiling temperatures at the top
corner point with lower HTCs increases from 629.64 C to 651.92 C. It may be inferred
that the transverse heat transfer is not active.

0 2.5 5 7.5 10 12.5 15


HTC difference,%

Figure 7.21 Effect of transverse HTC difference


on the coiling temperature for Case 192

The difference between the two coiling temperatures is shown in Figure 7.21. The coiling
temperature difference is about 20 C when the H T C difference is 15%. These two
values uncover very important messages.
First, it is possible that there is a 15% difference in the HTCs between the jet centre and
the middle of two jets. In this case, the local coiling temperature difference reaches 20
C. The importance lies in that the overall coiling temperature difference across the
width is also around 20 C because the H T C difference changes periodically for every
two adjacent jets.
Second, it is assumed that in practice the 15% difference in the HTCs is for a half width
of the strip of 1.0 m. In this case, the overall coiling temperature difference across the
width may be less than, or at maximum equal to, 20 C.

194
A temperature difference of 20 C is acceptable for a 15% H T C difference. In steel strip
production a variety of measures is adopted to keep the cooling temperature across the
width as uniform as possible. The H T C difference would be less than 15% and cause less
change in the coiling temperature across the width. Therefore, it can be concluded that
the 2D transverse simulations has little impact on the final coiling temperature.
Figure 7.22 shows the temperature distributions on the cross-section at the time step 100
and 240 for a 15% H T C difference. It clearly shows that the temperature becomes more
and more uniform over time.

Figure 7.22 Temperature distributions on cross-section


Upper: time step 100; bottom: time step 240

7.5.2 2D longitudinal modelling

1. Modelling

The geometry and mesh stated in Section 7.5.1 are also used here with the difference that

the length direction of the model is along the longitudinal direction. The choice o f the

195
length of 150 mm is because the impingement zone is 45 mm in length in the above
reference ID calculations.
Four cases are studied, as shown in Table 7.5. For all one and two jetline cases, the entire
model is in the parallel cooling zone-B and its right side is 250 mm apart from the jetline
centre for the first time step. Then it moves through the impingement zone. It stops as its
right side passes the jetline centre with a distance of 250 mm. The movement of the
model is illustrated in Figure 7.23. . There is no heat transfer assumed due to bulk
motion.

Table 7.5 Longitudinal modelling parameters and results

Case 7 5 2 - 1 7 5 2 - 2 7 5 2 - 3 7 5 2 - 4

Initial Temp, C 902 902 902 650


Moving speed, m/s 1.0 10.0 1.0 1.0
Moving distance, m 0.5 0.5 1.0 1.0
Time step 500 500 1000 1000
Jetlines passed 1 1 2 2

-ve T iff, C*
d -2.6249 -0.5998 -2.6249 -1.6663
1 jetline
st

+ve T ff, C*
di 2.2107 0.5084 2.2107 1.4276

-ve T iff, C*
d
- - -2.4808 -1.5900
2 n d
jetline
+ve T iff, C*
d
- - 2.1031 1.3667

Final T diff , C* 0.0172 0.0063 0.0000 0.0108


* More significant digits are intentionally kept for comparison

The bottom of the model is assumed to be thermally insulated. The critical issue is how to
assign the boundary conditions on the left side and right side. Thermal insulations were
assumed for both sides in reference [83]. Our numerical studies also showed that the
influences of the boundary conditions on these two sides are extremelly localized and
will not affect the temperature at the middle of the piece. Even though this appraoch may
not reflect the reality on ROT it is still used in this study.

196
Impinge-
Parallel zone-B Parallel zone-F
ment

,p^ g di t
Start posit i
+ u r e r a e n

i" positioTT
1

End position

Figure 7.23 Movement of model piece

3500

3000

2500

o
8 2000
E

O 1500
x

1000

500

0.0 0.2 0.4 0.6 0.8 1.0


Cooling time,s

Figure 7.24 HTC values for the two-jetlines cases

The H T C condition is applied to the top surface. The time-space shift scheme discussed
in Section 4.4.3 is adopted and is not repeated here. Based on the H T C shown in Figure
7.9, two constant values of 800 W/m -C and 3200 W/m -C are assumed here for the
2 2

parallel and impingement zone, respectively. Figure 7.24 shows the HTCs on the first,
middle and last nodes (F, M , and L in the figure) for the two-jetlines cases.
Note that the temperatures at the middle node on the top surface in the 2D models are
compared with those from the ID model with the same H T C profiles. This is more

197
reasonable because the effect of the alternative change of heat flux in the longitudinal
direction can be included in the temperature profile at the middle node, and because the
influence of the boundary conditions at the left and right sides is diminished.

2. Results and discussion

The surface temperature differences from ID and 2D modelling are also shown in Table
7.5. It is apparent that there is very little discrepancy between 2D predictions and ID
ones. Some detailed discussion follows.
Figure 7.25 shows the 2D-temperature and lD-2D-temperature difference for Case 752-1.
Note that all the simulations have shown that the ID temperature and 2D temperature are
nearly identical for the middle nodes for the entire cooling time.

910

900

890

o 880
a>
3 870
2
2. 860
E
3 850
a>

| 840

^ 830
CM

820

810

800
0 100 200 .300 400 500
T i m e step index

Figure 7.25 2D-temperature and lD-2D-temperature difference for Case 752-1

For all cases, the middle point enters and leaves the first impingement zone at time step
305 and at time step 355, respectively. It is clear that the temperature decreases smoothly
before the material enters the impingement zone, and there is no difference between the
temperatures from ID and 2D predictions.
The temperature drops sharply as the material enters the impingement zone.
Correspondingly, the temperature difference T2D-T1D suddenly increases from zero to a
negative value of 2.63 C, which is the maximum temperature difference. This can be
explained as follows. For the 2D model, the left side of the middle point is still in the

198
parallel zone as the middle point enters the impingement zone; and a temperature gradient
appears and longitudinal heat transfer occurs from left to right (Figure 7.23), which leads
to a 2D temperature lower than the ID one. As more material points enter the
impingement zone, the temperature gradient around the middle point promptly decreases
and then the temperature difference decreases.
In contrast, the temperature starts to recover as the middle point leaves the impingement
zone and enters the parallel zone-F. At this time instant, more heat is transferred from the
right side of the materials, which have already recovered, and makes the 2D temperature
at the middle point recover faster. Thus, the 2D temperature is higher than the ID one and
the temperature difference T2D-T1D is positive. Note that the positive temperature
difference is 2.21 C. Because the temperature in the parallel zone-F is lower than that in
the parallel zone-B, the value of the temperature gradient at the left side is higher than
that at the right side. Therefore, the positive temperature differences are always less than
the negative ones. This feature is verified by all cases.
Again, as more and more points enter the parallel zone-F, the temperature gradient
decreases and the difference between 2D and ID diminish. By the end of the simulation,
the temperature difference T2D-T1D is only 0.0172 C, which has no effect on the
prediction of the coiling temperature.
Simulations show that the maximum temperature difference T2D-T1D decreases with the
increase of the strip speed. The reason is that the alternative change of temperature
gradient changes in a shorter time and the longitudinal heat transfer has modest influence
when the strip speed increases.
Similarly, the maximum temperature difference T2D-T1D decreases with the decrease of
the initial temperature. That is because, the temperature gradients are smaller when the
temperature is lower.
It is worth examining the two-jetlines cooling process. There are two reasons for setting
up the jetline distance at 500 mm. First, it can reduce the calculation time in 2D
modelling. In fact, a longer parallel zone contributes little to the temperature difference
T2D-TID- Second, it is based on industry ROT parameters. For some ROTs, there are two
jetlines in a header. The distance for these two jetlines is around 500 mm.

199
It can be inferred from the data in Table 7.5 that a continuous two-jetlines cooling will
not enhance the longitudinal heat transfer.

The 2D two-jetlines cooling simulations have an implicit importance. For the one-jetline
cooling, the initial temperatures are uniform across the section. However, as shown in
Figure 7.26, after the strip is cooled by the first jet during the two-jetline cooling process,
the temperatures at the top surface and at the centre are not the same, and there is a larger
temperature gradient along the thickness direction. This means that the initial temperature
condition for the second jetline cooling is not uniform. This non-uniformity does not
affect the temperature difference T2D-T1D.
The data of Case 752-4 suggest important information for the ROT cooling. The final
cooling stage on ROT takes place around 650 C, and is a fine-tuning for the right coiling
temperature. Usually, one or two jetlines are active. Before this final cooling stage, the
strips experience a longer air cooling period; their temperatures are generally uniform
through the thickness. The results show that there is no difference between ID and 2D
modellings.

200
7.5 Summary

A special F E program is developed for the simulation of the ROT cooling process. 1160
cooling processes of steel strips on an industrial ROT conditions are sucessfully
simulated using ID modelling. The predictions of the ID model simulations are of an
accuracy of 10% with a comparison to the field measurements. A l l of the findings in
the parametric studies are consistent and reasonable. It is concluded that the developed
program is reliable and can be used to predict the coiling temperature and the profiles of
the cooling temperature with an acceptable accuracy.
The effect of non-uniform heat flux in the width direction on the coiling temperature is
investigated using transverse 2D modelling. The coiling temperature difference may be
up to 20 C when there is a 15% H T C difference across the width. In steel strip
production a variety of measures is adopted to keep the cooling across the width as
uniform as possible. The H T C difference would be less than 15% and cause less change
in the coiling temperature across the width. Therefore, it can be concluded that the 2D
transverse simulations may not be necessary.
It is found by the longitudinal modellings that the longitudinal heat transfer is fairly
localized and concentrated around the entrance and exit of the impingement zones. It is
also found that the temperature differences between ID and 2D modellings are less than 3
C and tend to be zero. A l l of these suggest that the longitudinal 2D modelling is not
necessary for simulating the ROT cooling process.

201
Chapter 8

CONCLUSIONS AND FUTURE W O R K

8.1 Conclusions

The controlled cooling process of steel strips on ROT is the critical stage for obtaining a
desired combination of mechanical properties. Although it has been used in the steel
industry for decades, this cooling process is mainly still a trial-and-error procedure as
new products are to be developed, because the thermal events during the cooling process
are very complex and a fundamental understanding of this process is not yet complete.
A group of researchers at U B C has been performing both experimental investigations
using a test facility of industrial scale and numerical simulations on this cooling process
in order to develop novel properties of steel strips.
This study engages in acquiring reliable heat transfer data and developing an effective
and robust FE simulation of the cooling processes on ROT. Several aspects of this work
have been finished to realize the objective. These are summarized in the following points.

1. 2D F E program for direct heat transfer analysis

A general 2D F E heat transfer analysis program has been developed. The 2D F E program
is able to analyze both state steady and transient heat conduction problems. Various
boundary conditions such as specified surface temperature, surface and nodal flux,
convection and radiation as well as heat generation can be considered, all of which can be
functions of location and time. In this program, the temperature dependence of thermal
properties can be taken into account.
In addition, a choice of implicit and explicit solution algorithm, consistent and lumped
capacitance matrix is available. Frontal solver is adopted for the non-symmetric stiffness
matrix. Besides, the program is capable of using a mixture of element types. Test cases
showed that the developed F E program has compatible accuracy with the commercial
program A N S Y S [58].

202
2. 2D F E program for inverse heat conduction analysis

A n algorithm for inverse heat conduction analysis is developed and implemented into the
above 2D F E program. In this method, the least-squares technique, future time step
technique, zeroth order regularization and sequential function specification as well as
iterative technique are used. The sensitivity matrix is only calculated once at the
beginning of the iterative procedure and used for all steps. The iterative technique can
compensate for the use of the regularization parameter and the simplification in
sensitivity matrix calculation to reduce the C P U time while maintaining accuracy. Both
absolute and relative norms are used to obtain a reasonable convergent solution.

3. Characterization of impingement cooling

The division and progression of water cooling zones on the top surface of hot plate
during water jet cooling is analyzed in detail. The cooled surface under one free water jet
can be generally divided into the impingement zone and the parallel zone. The cooling
zones are symmetrical about the stagnation point for a stationary single circular jet. Only
the material points in the impingement zone are simultaneously cooled; any material
point outside of the impingement zone is sequentially cooled. The further the point from
the stagnation point, the later it is cooled. The progressing speed of the water cooling
zones is much lower than the jet velocity and is of ~10 mm/s for the tests described in
this study. For the moving plate case, the cooling zones are only symmetric about the
plate moving direction and become a pearl shape or a comet shape whose long axis
follows the moving direction. The velocity of water can be neglected, and the plate is
cooled by a moving pattern of cooling water. The moving speed of the cooling pattern is
fairly close to the plate's moving speed.

4. Flux Zoning Method (FZM)

The commonly used Flux Zoning Method (FZM) is clarified and first used to specify the
heat fluxes on the cooled surface. The approach is to divide the target surface into several
subregions and each subregion corresponds to one temperature measurement and one
heat flux value that is calculated using the proposed inverse algorithm.
Parametric study results of the inverse calculations using the F Z M approach indicate that
the fluctuation of inversely calculated heat flux due to random error in the measurement

203
data may be damped by increasing the number of future steps and regularization
parameter. It is shown that the presented algorithm and procedures are very simple, and
effective in recovering the heat flux history around the impingement zone. It has been
found that the FZM may not be an appropraite method for the cooled surface outside of
the impingement zone because it does not consider the water movement.

5. Time-space shift scheme

A time-space shift scheme has been designed to assign the heat flux boundary condition
for the moving water situations. The heat flux history on the precedent nodes is shifted to
the subsequent nodes according to the assigned speed and element length. This scheme
means a space-time-marching approach and implies that the cooling condition for all
points in the parallel zone is almost identical except for a time shift and small
temperature drop due to the longer air cooling period.

6. Flux Marching Method ( F M M )

A new approach named Flux Marching Method (FMM) is developed for considering the
water movement. A main feature of the F M M is the essence of wave propagation. In this
method, the cooling surface is still divided into several subregions and each subregion
corresponds to one temperature measurement. The heat flux at the first node for each
subregion is inversely calculated from the internal measured temperature, and functions
as the impetus of the wave and the flood of the heat flux would sweep the points in each
individual subregion. Numerical tests show that the F M M provides a more accurate
recognition of input heat flux for the parallel zone in the inverse calculations if the flux
marching speed matches correctly with the progressing speed of water cooling zones.

7. Investigations of temperature measurement errors

Two kinds of situations have been considered for the assessments of measurement errors
associated with both surface and imbedded TCs with separation measuring junction: one
for the simultaneous cooling of the two wires of TCs referring to those in the
impingement zone, and the other for the sequential cooling of the two wires of TCs in the
parallel zone. In these numerical tests, the input heat flux is of the order of practical
cooling. For the sequential cooling simulation, the space-time shift scheme is used.

204
It has been found that the conduction of surface TC wires in a water jet cooling process
leads to two kinds of effects on the measured temperature in the case of water
impingement cooling. First, the surface temperature field is disturbed and becomes non-
uniform and the average temperature around the measuring junction would be lower than
the true temperature. Second, the measuring junction with a separation installation is no
longer an isothermal junction. The plate metal works as the third wire and may cause an
additional deviation in measurements. The combination of these two effects leads to a
measurement error of about 60 C lower than the true value. It has also been found that
the attachment of TC on the surface has little effect on the internal temperature 1 mm
below, and that the attachment of TC in the hole has little effect on the surface
temperature.

The disturbance of the temperature field due to the introduction of a small hole for the
installation of interior TC is similar but has less pronounced effects on the surface
measurement. A n increased distortion of the temperature field is evident when TC is
attached to the top surface directly above the bottom surface of the hole.
Numerical tests show that the non-simultaneous cooling of the two wires of internal TCs
has pronounced effects. It has been found that the isothermal condition does not exist for
the separate roots of TC wires, as the connection line of the two roots is parallel to the
water moving direction. The maximum temperature difference between the TC wire roots
will be affected strongly by the progressing speeds of water cooling zones and heat flux
magnitudes, and mildly or slightly by heat flux profiles and the depths of the hole and
specific heat values. The temperature difference between the two roots may reach a
significant level for both surface and interior TCs, especially for the situations in which a
deeper hole (close to cooling surface) is used to install TCs and the water moves slowly
with a higher cooling capacity.
It may be concluded that TC with separate measuring junction should be, generally,
avoided in temperature measurements of water cooling in moving cases; i.e., when the
water moves or the plate is not stationery or both.

205
8. Investigation of appropriate locations of T C s

Appropriate distance / between TCs and TC depth h are investigated using direct F E
analysis in this study. It has been found that the distance / should be less than 8-10 mm to
attain a real 2D inverse calculation and to improve the stability of inverse calculation
when the measured temperatures are used for inverse calculations by the F Z M . It is also
postulated that there is a minimum depth for the embedded T C , at which the surface
temperature may not be disturbed by the inclusion of the hole for the embedded TC and
the cooling behaviour may not be affected.

9. Effective depth approach

An effective depth approach (EDA) is suggested through intensive numerical


simulations. A linear correlation is found between the actual TC depth and the effective
TC depth for the actual depth of 1 mm to 3 mm. This approach is a possible method for
including the hole effect in the inverse calculations of heat flux, without including the
hole itself.

10. Investigation of heat transfer behaviour in the impingement zone

The developed inverse analysis program and techniques are used to determine the surface
heat flux and heat transfer behaviour during the water jet cooling process using both
stationary and moving plates. The F Z M and F M M for specifying the heat flux
distribution on the cooling surfaces as well as the E D A are used in those analyses.

The heat transfer at the stagnation point or in the impingement zone of stationary plates
cooled by an industry scale circular water jet cooling is successfully determined using 2D
models with the F Z M . Analysis results show that the heat transfer behaviour at the
stagnation is mainly and greatly affected by water temperature, slightly affected by the
steel grade and hardly affected by water flow rate. When the water temperature is lower
than 60 C the cooling process is governed by transition boiling, nucleate boiling. When
the water temperature is higher than 60 C, there is evidence of a film boiling regime. The
features of boiling curves and the typical values obtained are in good agreement with
other resources.

206
11. Investigation of heat transfer behaviour in the parallel zone

The F M M and E D A are used for the parallel zone. It has been found that the water flow
rate is the key factor in the parallel zone, that it affects both the mode and the magnitude
of heat transfer, and that the conductivity mainly influences the heat flux values.
However, the initial water temperature has little effect on the heat transfer in the parallel
zone.

12. Investigation of plate motion effect

The F Z M and E D A are used for the moving plate tests. It has been found that increasing
the plate speed leads to a decrease in heat fluxes for a given water temperature and water
flow rate.

13. ID simulation of R O T cooling process

Special FE programs are developed to simulate the ROT cooling process. The ID F E
modular program is an interactive one and can be referred to special designs of ROT by
the input files. It is user friendly and convenient to implement microstructure evolution
equations for different steel grades, formula or empirical models for heat transfer, and
correlations between cooling parameters, microstructure and mechanical properties.

1160 cooling processes of steel strips on an industry ROT are sucessfully simulated using
the ID modelling. The predictions of ID model simulations are of an accuracy of 10%
when compared to field measurements. The developed program is reliable and can be
used to predict the coiling temperature and the profiles of cooling temperature with an
acceptable accuracy.

14. Investigation of necessarity 2D simulation of R O T cooling process

The effect of non-uniform heat flux in the width direction on the coiling temperature is
investigated using a transverse 2D modelling. The coiling temperature difference may be
up to 20 C when there is a 15% HTC difference across the width. It has been found that
the heat transfer in the longitudinal direction is fairly localized and concentrated around
the entrance and exit of the impingement zones, and that it has little effect on the
prediction accuracy. These observations suggest that both 2D transverse and 2D

207
longitudinal modellings would have slight but not significant effect on the ROT cooling
process.

8.2 Suggestions and Future Work

1. T C installation

To realize more reliable temperature measurement in experimental tests and then to


obtain more accurate inverse calculation results in the future, several modifications to TC
installations should be conducted.
First, surface TCs should not be used on the cooled surface since it may cause significant
measurement error and disturbs the heat transfer behaviour. This modification has been
performed in the moving plate tests.
Second, TCs with separation measuring junction should be avoided in water jet cooling
even if it is installed inside of the plate and opposite to the cooled surface. It is not
guaranteed that the connection line of the two wires is perpendicular to the water moving
direction. TCs should be, generally, of the bead type.
Third, the imbedded bead-type TCs should be installed in a carefully designed hole. The
hole diameter should be as small as possible, and its distance to the cooled surface should
have a minimum value to avoid disturbing the surface temperature profile and the heat
transfer behaviour.
Finally, the distance between two TCs should be less than 10 mm for stationary plates if
the F Z M will be used for inverse calculations. The results from the F Z M may be a good
approximation to real ones with a smaller distance. By contrast, the distance between two
TCs in the moving direction for the moving plate may be larger than 10 mm even i f the
F Z M is used.

2. Single nozzle tests using stationary plates

Further tests using a single nozzle should be designed to investigate the thermodynamics
in the impingement zone and to determine the impingement zone size and the progressing
speeds of water cooling zones because they are of importance to the moving plate cooling
and to the application of the F M M in inverse calculations. Thus, more temperature
measurements should be conducted in one test. With an increased number of
measurements the stagnation point may be also accurately located.

208
Further experimental studies should extend to heat transfer using lower initial
temperature of plates, by changing the nozzle geometry and orientation, jet profile and
velocity, different surface roughness and plunging jet.

3. Multiple nozzle tests

The cooling condition using one nozzle is mostly for academic interest and is far from the
real situation. To make the experimental condition closer to the real ROT conditions, at
least three nozzles in one jetline should be used. For stationary plates, TC can be
arranged in the scheme shown in Figure 8.1. Besides, extra TCs should be arranged along
lines that are perpendicular to the jetline. It is better to install two groups of TCs as
shown in Figure 8.1a. This setup can also be used for moving plates. In this case, the
plate moving direction is perpendicular to the jetline and TCs should be in the upstream
side.
With this setup, the interaction on the cooled surface that does not occur in the case of
single jets can be included and investigated. The interaction is mostly associated with
adjacent impinging jets and refers to the collision of the surface flow, i.e., the wall jets.
This interaction is of increasing importance when the jets are closely spaced, and when
the distance from the nozzle exit to the plate is small, and the jet velocity is large.
A better scenario is shown in Figure 8.1b. With such a design the cooling conditions will
be close to those on real ROT. And the obtained data may be of great accuracy and useful
for the simulation of the ROT cooling process.

TCs
O TCs s

\..\
)
(
-m >-
* * * ^J
(
J
Nozzles^
O
)

Moving direction Jetline Moving direction Jetlines


a b

Figure 8.1 Extra TCs and nozzles for future tests

2 0 9
The nozzles should be opened to stabilize the flow rate before the plate arrives
underneath the nozzles. The flow rate, water pressure (water height level in the tank)
should be kept constant as much as possible during the cooling process. This means that
there should be a source of water supply to the tank during the experiment.

4. New IHCP models

New IHCP models are to be developed to determine the surface heat fluxes and surface
temperatures from the internal temperatures measured in tests using higher moving plates
and multiple jets.
The existing IHCP model has been tested for the single circular jet with stationary or
slow moving plates. It is necessary to extend the model for cases using higher moving
plates whose speeds are comparable to those used in industrial ROTs. The possibility of
modelling the moving case with one single jet as a quasi-2D problem should be
investigated and the effect of moving speed should be studied.
Significantly different heat fluxes may be generated when using multiple jets, and may
create large temperature gradients across the jetline between two jets. A new 3D model
for multiple jets should be developed, tested and optimized. The model should consider
the interaction between two or more jets. Special numerical tests, which are similar or
equivalent to the physical tests, should be used to validate 3D F E IHCP modelling.

5. Analyses of thermal stresses on ROT cooling

Another area for future work is to develop a 2D/3D F E model to investigate the thermal
stresses due to the large temperature gradients that occur in the jet impingement cooling
process, especially when multiple jets and/or strips/plates of larger thickness are used.
The 2D modelling should be able to predict the thermal stresses under the cooling
conditions using one single jet, either stationary or moving plate. The 3D model should
be able to predict the thermal stresses under the cooling conditions using multiple water
circular jets or one single water curtain jet with non-uniform water capacity setup.
This study will help provide information for balancing water on the top and bottom
surfaces at an optimum ratio and for setting the appropriate transverse water amount
gradient from centre to edge to prevent operation problems and/ or no-uniform
mechanical properties.

210
6. Water jet hydrodynamic simulation

Appropriate heat transfer modelling requires information about velocities, temperatures


and phase changes on the plate surface. The physical experiments entail many problems
such as cost, time, efficiency and accuracy. It is not possible to experimentally investigate
the vast number of parameters in this complex process. Therefore, an important area of
future work in the numerical aspect is to develop FE models for the hydrodynamics of the
flow. These models would enable an efficient and inexpensive parametric study on the
various parameters affecting the flow and pressure conditions of water jet and then
predict the heat transfer modes along the interactive surface under conditions similar to
industry practice. This study can significantly improve the understanding of the heat
transfer of jet impingement and compliment the experimental data.

211
Bibliography

[1] M . Korchynsky, Development of "controlled cooling" practice, Accelerated Cooling


of Steel, P. D. Southwick (ed.), 1986, pp. 3-14.
[2] G. Tacke, H . Litzke and E . Raquet, Investigations into the efficiency of cooling
systems for wide-strip hot rolling mills and computer-aided control of strip cooling,
Accelerated Cooling of Steel, P. D. Southwick (ed.), 1986, pp. 35-54.
[3] S. J. Chen, A . A . Tseng and F. Han, Spray and jet cooling in steel rolling, Heat
Transfer in Metals and Containerless Processing and Manufacturing American
Society of Mechanical Engineers, Heat-Transfer, V o l . 162, 1991, pp. 1-11, or
International Journal of heat and fluidflow, Vol. 13 No. 4, 1992, pp. 358-369.
[4] D. H . Wolf, F. P. Incropera and R. Viskanta, Jet impingement boiling, Advances in
Heat Transfer, Vol.23, 1992, pp. 5-132.
[5] D. A . Zumbrunnen, F. P. Incropera and R. Viskanta, Method and Apparatus for
measuring heat transfer distributions on moving and stationary plates cooled by a
planar liquid jet, Experimental Thermal and Fluid Science, Vol. 3, 1990, pp. 202-
213.
[6] B . W. Webb and C. F. Ma, Single-phase liquid jet impingement heat transfer,
Advances in Heat Transfer, Vol.26, 1995, pp. 105-217.
[7] V . Olden and M . Raudensky, Water spray cooling of stainless and C-Mn steel, Steel
Research, Vol. 69 No.6, 1998, pp 240-246.
[8] T. Ochi, S. Nakanishi, M . Kaji and S. Ishigai, Cooling of a hot plate with an
impinging circular jet, Multi-Phase flow and heat transfer III, proceedings of the
third multi-phase heat and flow transfer symposium workshop, Miami Beach,
Florida, U S A , Apr 18-20,1983, pp. 671-681.
[9] S. Ishigai, S. Nakanishi and T. Ochi, Boiling heat transfer for plane water jet
impinging on a hot surface, Sixth International Heat Transfer Conference, Toronto,
Ontario, Canada, August 1978.
[10] D. T. Vader, F. P., Incropera and R. Viskanta, A method for measuring steady local
heat transfer to an impinging liquid jet, Experimental Thermal and Fluid Science,
Vol. 4, 1991, pp. 1-11.

212
[11] D. T. Vader, F. P. Incropera and R. Viskanta, Convective nucleate boiling on a
heated surface cooled by an impinging, planar jet of water, Journal of Heat
Transfer, Vol. 114, 1992, pp. 152-160.
[12] D. H . Wolf, F. P. Incropera, R. Viskanta, Local jet impingement boiling heat
transfer, International Journal of Heat and Mass Transfer, Vol. 39, 1996, pp. 1395-
1406
[13] S. J. Chen, J. Kothari and A . A . Tseng, Cooling of a moving plate with an
impinging circular water jet, Experimental Thermal and Fluid Science, V o l . 4,
1991, pp. 343-353.
[14] F. Han, S. J. Chen and C. C. Chang, Effect of surface motion on liquid jet
impingement heat transfer, Fundamentals of Forced and Mixed Convection and
Transport Phenomena, American Society of Mechanical Engineers, Heat Transfer,
Vol. 180, 1991,pp. 73-81.
[15] S. J. Chen and S. K . Biswas et al, A., Modeling and analysis of controlled cooling
for hot moving metal plates, Monitoring and Control for Manufacturing Processes,
American Society of Mechanical Engineers, Production Engineering Division
(Publication) PED, Vol. 44, 1990, pp. 465^174.
[16] J. Stevens and B. W. Webb, Local heat transfer coefficients under an axisymmetric,
single-phase liquid jet, Journal of Heat Transfer, Vol. 113, 1991, pp. 71-78.
[17] D. A. Zumbrunnen, R. Viskanta and F. P. Incropera, The effect of surface motion on
forced convection film boiling heat transfer, Journal of Heat Transfer, V o l .
111,1989, pp.760-766.
[18] J. Chen, T. Wang, and D. A . Zumbrunnen, Numerical analysis of convective heat
transfer from a moving plate cooled by an array of submerged planar jets, Heat
Transfer in Metals and Containerless Processing and Manufacturing American
Society of Mechanical Engineers, Heat-Transfer, Vol. 162, pp. 25-34.
[19] J. Filipovic, J. Viskanta and F. P. Incropera, Thermal behavior of a moving steel
strip cooled by an array of planar water jets, Steel Research, Vol. 63 No. 10, 1992,
pp. 438-446.
[20] J. Filipovic, J. Viskanta and F. P. Incropera, Cooling of a moving steel strip by an
array of round jets, Steel Research, Vol. 65 No. 12, 1994, pp. 541-547.

213
[21] N . Hatta and Y . Tanaka, A numerical study on cooling process of hot steel plates by
a water curtain, ISIJ International, Vol. 29 No. 8, 1989, pp. 673-679.
[22] N . Hatta and H . Osakabe, Numerical modeling for cooling process of a moving hot
plate by a laminar water curtain, ISIJ International, Vol. 29 No. 11, 1989, pp. 919
925.
[23] Z. Liu, Experiments and mathematical modeling of controlled run-out table cooling
in a hot rolling mill, Ph. D. Thesis, the University of British Columbia, 2001.
[24] Z. L i u , D. Fraser and I.V. Samarasekera, Experimental study and calculation of
boiling heat transfer on steel plates during run-out table operation, Canadian
Metallurgical Quarterly, Vol.41, 2002, pp. 63-74.
[25] Z. Liu, D. Fraser, I.V. Samarasekera and G. T. Lockhart, Experimental observation
and modeling of thermal history within a steel plate during water jet impingement,
Canadian Metallurgical Quarterly, Vol.41, 2002, pp. 75-86.
[26] A . T. Hauksson, Experimental study of boiling heat transfer during water jet
impingement on a hot steel plate, Master Thesis, the University of British Columbia,
2001.
[27] Q. Meng, Experimental study of transient cooling of a hot steel plate by an
impinging circular jet, Master Thesis, the University of British Columbia, 2002.
[28] L . Michalski, K . Eckersdorf and J. McGhee, Temperature Measurement, John
Wiley & Sons Ltd., 1991, pp.334-360.
[29] R. E. Bentley, Theory and Practice of Thermoelectric Thermometry, Handbook of
Temperature Measurement, Vol. 3, Springer-Verlag, 1998.
[30] R. J. Moffat, The gradient approach to thermocouple circuitry, Temperature-Its
Measurement and Control in Science and Industry, Reinhold, New York, 1962, pp.
33-38.
[31] D. K . Hennecke and E. M . Sparrow, Local heat sink on a convectively cooled
surface-Application to temperature measurement error, International Journal of
Heat and Mass Transfer, Vol. 13, 1970, pp.287-304.
[32] J. E. Park, K . W. Childs, G. M . Ludtka and W. Chu, Correction of errors in intrinsic
thermocouple signals recorded during quenching, National Heat Treat Conference,
Minneapolis, M N , July 26-31, 1991, pp. 309-318.

214
[33] F. White, Accuracy of TCs in radiant-heat testing, Experimental Mechanics, Vol.2,
1962,pp.204-210.
[34] M . H . Attia and L. Kops, Distortion in thermal field around inserted thermocouples
in experimental interfacial studies, Journal of Engineering for Industry, Vol. 108,
1986, pp. 241-246.
[35] M . H . Attia and L . Kops, Distortion in thermal field around inserted thermocouples
in experimental interfacial studies - Part 2: Effect of the heat flow through the
thermocouple, Journal of Engineering for Industry, Vol. 110, 1988, pp. 7-14.
[36] M . H . Attia and L. Kops, Distortion in thermal field around inserted thermocouples
in experimental interfacial studies - Part 3: Experimental and numerical verification,
Journal of Engineering for Industry, Vol. 115, 1993, pp. 444^149.
[37] M . H . Attia, A . Cameron and L . Kops, Distortion in thermal field around inserted
thermocouples in experimental interfacial studies - Part 4: End effect, Journal of
Manufacturing Science and Engineering, Vol. 124, 2002, pp. 135-145.
[38] T. C. Tszeng and V . Saraf, A study of fin effects in the measurement of temperature
using surface-mounted thermocouples, Journal of Heat Transfer, Vol. 125, 2003,
pp. 926-935.
[39] D. L i , Boling water heat transfer during quenching of steel plates and tubes, Ph.D.
thesis, the University of British Columbia, 2003.
[40] J. V . Beck, B . Blackwell and C. R. S. Clair Jr., Inverse Heat Conduction: Ill-posed
Problem, Wiley-Interscience Publication, New York, 1985.
[41] O. M . Alifanov, Inverse Heat Transfer Problem, Springer, Berlin, 1994.
[42] N . Al-Khalidy, A general space marching algorithm for the solution of two-
dimensional boundary inverse heat conduction problems, Numerical Heat Transfer,
Part B, Vol. 34, 1998, pp.339-360.
[43] R. A . Khache and Y . Jarny, Determination of heat sources and heat transfer
coefficient for two-dimensional heat flow-numerical and experimental study,
International Journal of Heat and Mass Transfer, Vol. 44, 2001, pp. 1309-1322.
[44] S. K . Kim, W. I. Lee, Solution of inverse heat conduction problems using maximum
entropy method, International Journal of Heat and Mass Transfer, Vol. 45, 2002,
pp. 381-391.

215
[45] E. Videcoq, D. Petit, Model reduction for the resolution of multidimensional inverse
heat conduction problems, International Journal of Heat and Mass Transfer, V o l .
44,2001, pp. 1899-1911.
[46] C. F. Weber, Analysis and solution of the ill-posed inverse heat conduction
problem, International Journal of Heat and Mass Transfer, V o l . 24, 1981, pp.
1783-1792.
[47] B. R. Bass, Application of the finite element method to the nonlinear inverse heat
conduction problem using Beck's second method, Journal of Engineering and
Industry, Transaction A S M E , Vol. 102, 1980, pp. 168-176.
[48] A.S. Osman, Investigation of transient heat transfer coefficients in quenching
experiments, Journal of Heat Transfer, Transaction A S M E , V o l . 112, 1990, pp.
843-848.
[49] S. Kumagai, S. Suzuki, Y. R. Sano and M . Kawazoe, Transient cooling of a hot slab
by an impinging jet with boiling heat transfer, ASME/JSME Thermal Engineering
Conference, Vol. 2 A S M E 1995, pp. 347-352.
[50] H . Louahia-Gualous, P. K . Panday and E. A . Artioukhine, Inverse determination of
the local heat transfer coefficients for nucleate boiling on a horizontal pipe cylinder,
Journal of Heat Transfer, Vol. 125, 2003, pp. 1087-1095.
[51] H . K . Kim, S. I. Oh, Evaluation of heat transfer coefficient during heat treatment by
inverse analysis, Journal of Material Processing Technology, Vol. 112, 2001, pp.
157-165.
[52] M . Pietrzyk, J. G. Lenard, A study of heat transfer during flat rolling, International
Journal for Numerical Methods in Engineering, Vol. 30, 1990, pp. 1459-1469.
[53] J. V . Beck and H. Hurwicz, Effect of thermocouple cavity on heat sink temperature,
Journal of Heat Transfer, Vol.2, 1960, pp. 27-36.
[54] C. T. Kidd, High heat flux measurements and experimental calibrations, NASA
Conf Publish, No.31, 1992, pp.31-50.
[55] F. X u and M . S. Gadala, A simple FE-inverse heat conduction algorithm and its
application to the determination of surface heat flux during water jet cooling, 5 th

International Symposium on Multiphase Flow, Heat Mass Transfer and


Energy Conversion, Xi'an, China, 3-6 July 2005.

216
[56] M . S. Gadala and F. X u , A FE-based sequential algorithm for the determination of
heat flux during impingement water cooling, International Journal of Numerical
Methods for Heat & Flow Fluid, in press.
[57] K . J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall,
1982.
[58] R. W. Lewis, The Finite Element Method in Heat Transfer Analysis, Wiley-
Interscience Publication, New York, 1996.
[59] A N S Y S , User's Manual, Swanson Analysis Systems Inc., 2002.
[60] F. X u and M . S. Gadala, Investigation of error sources in temperature measurement
using thermocouples in water impingement cooling, Experimental Heat Transfer,
Vol. 18, No.3/July-September, 2005, pp. 153-177.
[61] F. X u and M.S. Gadala, Study on the application of thermocouple with separate
measuring junction in temperature measurement in water-cooling, submitted to
Numerical Heat Transfer-B.
[62] F. X u and M.S. Gadala, Heat transfer behavior in the impingement zone under
circular water jet, International Journal of Heat and Mass Transfer, accepted in
Feb., 2006.
[63] G. Wei, X . Dong and J. Wang, Study on convective heat transfer coefficient of high
temperature plate during water curtain cooling, Kang-T'ieh-Iron and Steel-(In
Chinese), Vol. 29, 1994, p 22-26.
[64] F. X u and M.S. Gadala, A novel FE approach to simulate the run-out table cooling
of steel strips, submitted to Steel Research.
[65] R. Colas, Modelling run out table cooling, Manufacturing Science and Engineering
American Society of Mechanical Engineers, Production Engineering Division
(Publication) PED, Vol. 68-2, 1994, A S M E , New York, N Y , USA, pp. 611-617.
[66] H . Dyja and P. Korczak, The thermal-mechanical and microstructural model for the
F E M simulation of hot plate rolling, Journal of Materials Processing Technology,
92-93 (1999), pp. 463-467.
[67] M . Suehiro and K . Sato, Mathematical model for predicting microstructural changes
and strength of low carbon steels in hot strip rolling, THERMC-88: International
conference on physical metallurgy of thermomechanical processing of steels and

217
other metals: Proceedings, June 6-10, 1988, Keidanren Kaikan, Tokyo, Japan, pp.
791-798.
[68] M . Suehiro, T. Senuma and H . Yada et al, Application of mathematical model for
predicting microstructural evolution to high carbon steels, ISIJ International, Vol.
32 No. 3,1992, pp. 433^39.
[69] A . Prasad, S. Jha and N . S. Mishra, Modelling of microstructural evolution during
accelerated cooling of hot strip on the run-out table, Steel Research, Vol. 66, 1995,
pp. 416-423.
[70] A . Kumar and C. McCulloch, Modelling thermal and microstructural evolution on
run-out table of hot strip mill, Material Science and Technology, Vol. 7, 1991, pp.
360-368.
[71] C. A . McCulloch, Characterization of the cooling and transformation of steels on a
run-out table of hot-strip mill, Master Thesis, the University of British Columbia,
1989.
[72] M . Packo, H. Kusiak and M . Pietrzyk, Modelling water cooling of steel strip during
hot rolling, Steel Research, Vol. 64, 1993, pp. 128-131.
[73] M . Pietrzyk, Finite element based model of structure development in the hot rolling
process, Steel Research, Vol. 61, 1990, pp. 603-607.
[74] R. K . Kumar, S. K . Sinha and A. K . Lahiri, Modelling of the cooling process on the
run-out table of a hot strip mill - a parallel approach, Conference Record IAS Annual
Meeting (IEEE Industry Applications Society), Vol. 4, 1996, pp. 2563-2571.
[75] R. K . Kumar, S. K . Sinha and A. K. Lahiri, Real time simulator for the run-out table
of hot strip mills, Conference Record IAS Annual Meeting (IEEE Industry
Applications Society), Vol. 4, 1996, pp. 2547-2554.
[76] V . H . Hernandez-Avila, Heat transfer model of the hot rolling run-out table cooling
and coil cooling of steel, Master Thesis, the University of British Columbia, 1994.
[77] V . H . Hernandez-Avila, Modelling of the thermal evolution of steel strips cooled in
the hot rolling run-out table, PhD Thesis, the University of British Columbia, 2000.
[78] J. Filipovic, J. Viskanta and F. P. Incropera, A parameter study of the accelerated
cooling of steel strip, Steel Research, Vol. 63, 1992, pp. A96-A99.

218
[79] R. M . Guo, Heat transfer of laminar flow cooling during strip acceleration on hot
strip mill run-out tables, Iron and Steelmaker, Vol. 20 No. 8, Aug 1993, pp. 49-59.
[80] R. M . Guo, Modeling and simulation of run-out table cooling control using
feedforward-feedback and element tracking system, IEEE Transactions on Industry
Applications, Vol. 33, 1997, pp. 304-311.
[81] R. M . Guo and S. T. Hwang, Investigation of strip cooling behavior in the run-out
section of hot strip mills, Journal of Materials Processing and Manufacturing
Science, Vol. 4, Apr 1996, pp. 339-351.
[82] M . M . Prieto, L . S. Ruiz and J. A . Menendez, Thermal performance of numerical
model of hot strip mill run-out table, Ironmaking and Steelmaking, Vol. 28, 2001,
pp. 474-^180.
[83] C. G. Sun, H . N . Han and J. K . Lee, A finite element model for the prediction of
thermal and metallurgical behavior of strip on run-out table in hot rolling, ISIJ
International, Vol. 42, 2002, pp.392-400.
[84] C. A . Muojekwu, Modeling of thermomechanical and metallurgical phenomena in
steel strip during hot direct rolling and run-out table cooling of thin-cast slabs, PhD
Thesis, the University of British Columbia, 1998.
[85] M . Miyake, J. M . Too and I. V . Samarasekera, Mathematical modelling of hot
rolling of steel strip with microstructure analysis, Simulation of Materials
Processing: Theory, Methods and Application, Mori (ed.), 2001 Sweits &
Zeitlinger, Lisse, pp. 275-280.
[86] C. M . Sellars, Modelling microstructural development during hot rolling, Materials
Science and Technology, Vol. 6, 1990, pp. 1072-1081.
[87] H . Yada, Prediction of microstructural changes and mechanical properties in hot
strip rolling, Accelerated Cooling of Steel, P. D. Southwick (ed.), 1986, pp. 105-
119.
[88] G. Uetz, G. Woelk and T. Bishops, Influencing the formation of the steel structure
by suitable temperature control in the run-out sections of hot-strip mills, Steel
Research, Vol. 62, 1991, pp.216-232.

219
Appendix A

Finite Element Matrix Equations for Conduction Heat Transfer

The governing equation for conduction heat transfer in three dimensional solid is given
by:

dT_ d (' dT dT_


pc kx + k ^ +- (A.1)
dt dx dx dz dz

where k , k , k are the thermal conductivities in x, y and z directions, respectively, T is


x y 2

the temperature, q is the internal rate of heat generated per unit volume, / is time, p is
b

the density and c is the specific heat.


The above equation may be subject to the general form of the boundary conditions of:

- K ^ = q + h(T -T ) + ecr(T* - T?)


s
f M
(A.2)
dn
where n is the normal to the boundary.
The input heat flux is considered only on pertinent surfaces, i.e., surfaces that have

specified boundary conditions and are given by:

Concentrated or distributed flux (q");

Convection B C : h(T - T ), h is the convection HTC, 7^is the fluid temperature and7^. is
f s

the surface temperature;

Radiation B C : sa(T* -T*), is the emissivity of the surface, cr is the Stefan-

Boltzmann constant, T sr is the absolute temperature of surface and T is the known


r

temperature of the external radiative source.


The sources of nonlinearity in equation ( A . l ) may be due to temperature dependent
material properties or dependency of the boundary conditions on temperature.
Now, we assume an approximation function for the temperature given by:

r(x) = jv,(x)7; (A.3)

220
where N (x) is the approximation function with / varying from one to the number of
t

nodes per element, the vector x has the components x, y and z, and 7} are the nodal

temperature. Note that we consider only a typical element and assume that the normal
finite element assembly procedures apply.
Applying Galerkin approach and utilizing Gauss theorem, the equivalent finite element
equations representing equations ( A . l ) and (A.2) may be written in the following final
form:

[c\ [r}+ IK ] [K ] [K,1 {r}={Qf + fc}' + for + for


C + h +
(A.4)

where the definitions of terms in equation (A.4) are summarized in Table A . l .


Considering the general nonlinear and transient case of equation (A.4), the solution may
be realized by re-writing the equation at time (t+At) and iteration (i) in the following
form:

T +' '(K + K* + K ) ' 'T


(/) +A c R +A (/)
= ('+*Q6 *Q* / A/QA f A/Q,yo
+ + + + + + ( A - 5)

It may be shown that using Newton-Raphson approximation and the a-method, the
solution to the above equation yields:

-i(i-i)

(K +K +K ) +
C H

\aAt)
R
C AT ^ 0
(Q*+Q> (Q + Q ) - (Q +q J
c c
(A.6)

where the definition of the new terms introduced here are given in Table A.2.

221
Table A . l Definition of the terms in the general F E heat conduction equation

Term Definition

[C]= [pcM[N\iV Thermal capacity matrix

[KA=[[B] [KWV T
Thermal conductivity matrix

[K ]= [h[N][N\lS
h Thermal conductivity matrix due to convection B C

[K ]=r ^ [N'][N'\iS
K Thermal conductivity matrix due to radiation B C

{Q} -[ Mdv b
q
B
Heat flux vector due to internal heat generation

{Q} = [q [N ] dS
s S S
Heat flux vector due to input surface flux

{Q} = [hT [N] dS


h
F Heat flux vector due to convection B C

{ } = [KT\N*] dS
Q
R
Heat flux vector due to radiation B C
Vector of global nodal temperatures and temperature
{T}, {t}
gradients, respectively

K= Ecr(T?+Tl)(T +T ) r sr Equivalent heat transfer coefficient due to radiation


\dT\
< =^-(m=[B]{T} e
Partial derivative of temperature
\dx\ ox
' K o o "
[K} = k y 0 Element conductivity matrix
sym kz

Approximation or shape function , n is the number of


[/V]=[/V, N - N]
nodes per element
2 n

222
Table A.2 Definition of the terms in heat transfer equation

Term Meaning Expression


Nodal heat flux
contribution due to **
convection B C , nonlinear Note: [""h^W'tf-dS^K*''"
**
and transient effects
Nodal heat flux
/+AfQA '- ( | )
_ f '+Al / .(/-l) SN
R
N S^ + A f T _ /+A ( ( i - l ) j
T ^

contribution due to
l+A/Qr '"' 1 1

radiation B.C., nonlinear Note: [ 'K^W N -dS^K^


,+A T s

and transient effects


Nodal heat flux
contribution due to
I + A / Q C (
' - ' >

conductivity, nonlinear Note: J B ' ^ K ^ B dV=' "K "


R
+ cl n

and transient effects


Nodal heat flux
contribution due to
' *q
+ c ( M )
= j/ '(pc)
+A (M)
N R
N - ^ T ^ - ' T J / A / ] - ^
r+A< " c '
thermal capacity,
( _ l )

nonlinear and transient


effects

223

You might also like