2009 - Annealing Furnace Model - Effect of Warm Rolled Steel On Recrystallization PDF

http://researchspace.auckland.ac.
nz
ResearchSpace@Auckland
Copyright Statement
The digital copy of this thesis is protected by the Copyright Act 1994 (New
Zealand).
This thesis may be consulted by you, provided you comply with the
provisions of the Act and the following conditions of use:
• Any use you make of these documents or images must be for

research or private study purposes only, and you may not make
them available to any other person.
• Authors control the copyright of their thesis. You will recognise the
author's right to be identified as the author of this thesis, and due
acknowledgement will be made to the author where appropriate.
• You will obtain the author's permission before publishing any
material from their thesis.
To request permissions please use the Feedback form on our webpage.

http://researchspace.auckland.ac.nz/feedback
General copyright and disclaimer
In addition to the above conditions, authors give their consent for the
digital copy of their work to be used subject to the conditions specified on
the Library Thesis Consent Form and Deposit Licence.
Note : Masters Theses
The digital copy of a masters thesis is as submitted for examination and

contains no corrections. The print copy, usually available in the University
Library, may contain corrections made by hand, which have been
requested by the supervisor.
Mathematical Modelling
of an
Annealing Furnace
Nicholas Brian Depree
A thesis submitted in partial fulfilment of the requirements for the degree of Doctor of
Philosophy in Chemical and Materials Engineering, The University of Auckland, 2009.
Abstract
The metal coating line at New Zealand Steel relies on a large electric radiant furnace to
heat steel strip before hot-dip galvanising in a continuous process. The temperature evo-
lution of the strip inside the furnace is vital in ensuring the specified mechanical properties
are achieved for a range of steel products. Ductile products require high temperatures
sufficient to cause recrystallisation of the steel microstructure, while stronger products
must be heated without causing recrystallisation. Strip dimensions and desired proper-
ties are changed often and irregularly during operation, and these changes and associated
furnace control actions cause changes in furnace and strip temperatures and rate of heat
transfer over several different time scales.
Accurate control of temperature is difficult because temperature measurement devices
are strongly affected by reflected radiation in the furnace cavity. The furnace is often
operating during transient temperature conditions, as control actions take effect very
slowly compared to the the rate of change of operational targets. Understanding of the
transient behaviour of this system of interrelated, nonlinear variables can be improved
using modelling to calculate furnace and strip temperatures as a result of control actions
in real time, which cannot otherwise be measured or predicted.
It is shown that a three-dimensional model is capable of accurately calculating furnace
temperatures changing over both time and location, requiring minimal simplification of
the physical system, but is computationally expensive. Radiative heat exchange in the
furnace cavity causes significantly increased temperature along the edges of the steel strip,
which can cause reject product due to localised softening. It was found that furnace ther-
mocouples are strongly affected by reflected radiation, so that furnace wall temperatures
be may significantly hotter than measured.
A simplified, coupled temperature-metallurgical model was shown to accurately calcu-
i
late both furnace and strip temperatures and metallurgical changes, while the 3D model
provides understanding of effects not explicitly modelled in the simplified model. The
simplified model is used for optimisation of furnace operational parameters, to improve
plant throughput and energy efficiency while maintaining desired metallurgical properties,
which is demonstrated by application to common products at NZ Steel.
ii
Acknowledgements
I firstly wish to thank my academic supervisors, Professors John Chen, Mark Taylor and
James Sneyd for accepting me as their student and offering this interesting project for
study. Their advice, insight and knowledge of course proved invaluable in developing my
work and thesis, and their support has allowed me to present this work in conferences
and publications.
I also thank Professor Steve Taylor for his extensive help and support in developing my
mathematical knowledge and solutions. His earlier work on this problem and further sug-
gestions were very useful and influential.
I wish to acknowledge the support of New Zealand Steel, without which this project would
not have been possible. I thank Michael O’Connor for his supervision and guidance on
the direction of the project, knowledge of previous investigations, and technical sugges-
tions and advice. I thank Phil Bagshaw for organising and approving the project and my
candidacy, and Nebojsa Joveljic for his extensive knowledge of the MCL plant, process
metallurgy, and for critically examining my work and assumptions.
Financial support for this project was provided by the Foundation for Research Science
and Technology (FRST), Auckland Uniservices Ltd, and New Zealand Steel, which en-
abled me to pursue this research, and is gratefully acknowledged.
I also wish to thank the staff and students at the Light Metals Research Centre and the
Department of Chemical and Materials Engineering at the University of Auckland for
accommodating me during this study, and providing resources and especially companion-
ship and an enjoyable working environment throughout. Lastly I wish to thank my family
and friends and partner Abby for their support and encouragement.
iii
Contents
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Annealing Furnace Characteristics . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Study Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Literature Review 8
2.1 The Continuous Galvanising Process . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Cold Rolling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Phase Diagram and Crystal Structure . . . . . . . . . . . . . . . . . 12
2.2 Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Recrystallisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Grain Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Annealing Furnace Modelling . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Mathematical Models . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Statistical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.3 Model Based Control Systems . . . . . . . . . . . . . . . . . . . . . 26
2.3.4 Model Type Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Conductive Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2 Convective Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.3 Radiative Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.4 The Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.5 Prior Studies of the NZ Steel MCL Furnace . . . . . . . . . . . . . . . . . 38
2.5.1 NZ Steel Steady State Model . . . . . . . . . . . . . . . . . . . . . 38
2.5.2 MISG Furnace Model . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.5.3 SYDAC Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.5.4 Strip and Thermocouple Temperatures . . . . . . . . . . . . . . . . 47
2.5.5 Requirements For New Model . . . . . . . . . . . . . . . . . . . . . 49
2.6 Metallurgical Calculation Methods . . . . . . . . . . . . . . . . . . . . . . 50
2.6.1 Kinetics of Recrystallisation . . . . . . . . . . . . . . . . . . . . . . 51
2.6.2 State Variable Methods . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.6.3 Statistical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.6.4 Empirical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
iv
2.6.5 Furnace Model Calculations . . . . . . . . . . . . . . . . . . . . . . 55
2.7 Metallurgical Model for NZ Steel . . . . . . . . . . . . . . . . . . . . . . . 56
2.7.1 Model Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.7.2 Metallurgical Model Development . . . . . . . . . . . . . . . . . . . 58
2.7.3 Recrystallisation Trials . . . . . . . . . . . . . . . . . . . . . . . . . 68
3 Practical Temperature Calculation to Improve Control 72

3.1 Difficulty of Measurement and Control . . . . . . . . . . . . . . . . . . . . 72
3.2 Method of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3 Scope of Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.1 Model furnace system in COMSOL software . . . . . . . . . . . . . 74
3.3.2 Examine furnace model behaviour . . . . . . . . . . . . . . . . . . . 74
3.3.3 Simplify equation system . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.4 Build simplified temperature model . . . . . . . . . . . . . . . . . . 75
3.3.5 Create integrated metallurgical model . . . . . . . . . . . . . . . . . 75
3.3.6 Demonstrate optimisation using model . . . . . . . . . . . . . . . . 76
4 Model Methodology and Development 77

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2 Furnace Physical Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2.1 Furnace Heating Method and Equipment . . . . . . . . . . . . . . . 78
4.2.2 Furnace Cooling Method and Equipment . . . . . . . . . . . . . . . 81
4.2.3 Temperature Measurement Equipment and Location . . . . . . . . 84
4.2.4 Hearth Rolls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.3 Furnace Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.2 Furnace Operating Cycles . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.3 Control During Cycle Change . . . . . . . . . . . . . . . . . . . . . 91
4.4 Model theory development . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4.2 Conductive Heat Transfer Calculations . . . . . . . . . . . . . . . . 94
4.4.3 Radiative Heat Transfer Calculations . . . . . . . . . . . . . . . . . 97
4.4.4 Convective Heat Transfer Calculations . . . . . . . . . . . . . . . . 98
4.5 3D Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.5.2 F3DC Model Assumptions and Simplifications . . . . . . . . . . . . 103
4.5.3 F3DC Model Equations . . . . . . . . . . . . . . . . . . . . . . . . 110
4.5.4 F3DC Model Calculation Procedure . . . . . . . . . . . . . . . . . . 111
4.5.5 F3DC Model Outputs . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.6 Simplified Model Development . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.6.2 RDHT Model Assumptions and Operation . . . . . . . . . . . . . . 119
4.6.3 RDHT Model Geometry . . . . . . . . . . . . . . . . . . . . . . . . 121
4.6.4 RDHT Model Calculation Method . . . . . . . . . . . . . . . . . . 122
4.6.5 Radiation Equation Approximation . . . . . . . . . . . . . . . . . . 125
4.6.6 Heat Balance Calculation . . . . . . . . . . . . . . . . . . . . . . . 128
v
4.6.7 Strip Heat Equation Approximation . . . . . . . . . . . . . . . . . . 128
4.6.8 Wall Finite Difference Approximation . . . . . . . . . . . . . . . . . 130
4.6.9 RDHT Model Outputs . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.7 Metallurgical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.7.2 Equation Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.7.3 Recrystallisation Temperature Comparison . . . . . . . . . . . . . . 145
4.7.4 Integration into RDHT Model . . . . . . . . . . . . . . . . . . . . . 153
5 Model Results and Validation 156

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.2 Validation Data Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.3 F3DC Model Results and Validation . . . . . . . . . . . . . . . . . . . . . 159
5.3.1 Validation of Cycle 24 0.4mm x 940mm G550 Strip Model . . . . . 160
5.3.2 Validation of Cycle 15 0.39mm x 905mm G300 Strip Model . . . . . 165
5.3.3 Thermocouple Block Temperature Variation . . . . . . . . . . . . . 167
5.3.4 F3DC Model Strip Edge Heating . . . . . . . . . . . . . . . . . . . 169
5.4 RDHT Model Temperature Results Validation . . . . . . . . . . . . . . . . 172
5.4.1 Model Result Interpretation . . . . . . . . . . . . . . . . . . . . . . 173
5.4.2 RDHT Model Temperature Results . . . . . . . . . . . . . . . . . . 174
5.5 Comparison of RDHT and F3DC Model Results . . . . . . . . . . . . . . . 177
5.6 RDHT-CM Model Recrystallisation Results Validation . . . . . . . . . . . 178
5.6.1 Soft Iron Recrystallised Products . . . . . . . . . . . . . . . . . . . 179
5.6.2 Hard Iron Recovered Products . . . . . . . . . . . . . . . . . . . . . 182
5.6.3 Production Coils With Incorrect Heat Treatment . . . . . . . . . . 183
5.7 RDHT-CM Transient Temperature Results . . . . . . . . . . . . . . . . . . 188
5.7.1 Cycle Change from Hard to Soft Iron . . . . . . . . . . . . . . . . . 190
5.7.2 Other Cycle Change Simulations . . . . . . . . . . . . . . . . . . . 195
5.8 Summary of Model Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
6 MCL Furnace Optimisation 201

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
6.2 Selection of MCL Products . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
6.3 Simulation of Selected Products . . . . . . . . . . . . . . . . . . . . . . . . 203
6.3.1 Product 1. 0.4 940 G550 A 24A . . . . . . . . . . . . . . . . . . . . 203
6.3.2 Product 2. 0.39 905 G300 A 15B . . . . . . . . . . . . . . . . . . . 210
6.3.3 Product 3. 0.55 938 G300 A 14B . . . . . . . . . . . . . . . . . . . 212
6.4 Furnace Optimisation Summary . . . . . . . . . . . . . . . . . . . . . . . . 214
7 Conclusions and Further Work 216

7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
7.1.1 F3DC Model Results . . . . . . . . . . . . . . . . . . . . . . . . . . 216
7.1.2 RDHT Model Results . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7.1.3 RDHT-CM Model Results . . . . . . . . . . . . . . . . . . . . . . . 219
7.1.4 Furnace Optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . 220
7.1.5 Project Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
vi
7.2 Recommended Further Work . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.2.1 RDHT Model Improvements and Applications . . . . . . . . . . . . 222
7.2.2 Improved Transient Solutions . . . . . . . . . . . . . . . . . . . . . 224
7.2.3 Improved Temperature Model Fitting . . . . . . . . . . . . . . . . . 224
7.2.4 Improved Metallurgical Model Fitting . . . . . . . . . . . . . . . . . 225
A Nomenclature 226
A.1 General Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
A.2 Greek Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
A.3 Physical Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
A.4 RDHT-CM Temperature Model Specific . . . . . . . . . . . . . . . . . . . 228
A.5 RDHT-CM Metallurgical Model Specific . . . . . . . . . . . . . . . . . . . 228
B RDHT-CM Model Variable List 229
C RDHT-CM Model Example Code 232
Bibliography 244
vii
List of Figures
1.1 Schematic diagram of the Metal Coating Line process . . . . . . . . . . . . 2

1.2 Furnace schematic diagram showing location of Pyrometers IR1-4 and L1-
L4, Induction Pre-heaters B1-C4 and Contact Thermocouple . . . . . . . . 2
2.1 Comparison of grain structure distortion in cold rolled low-carbon steel . . 11

2.2 The iron-carbon phase diagram . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Property changes during the stages of annealing . . . . . . . . . . . . . . . 14
2.4 Changes in grain structure of typical cold worked steel during annealing . . 15
2.5 Example of transient solution of strip temperature during cycle change . . 21
2.6 Strip temperature and heat transfer coefficient results for direct fired fur-
nace using stepwise model . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 Side and end elevations of similar model for indirectly fired furnace . . . . 23
2.8 Schematic diagrams of multi-layer predictive control schemes implemented
on several continuous annealing lines . . . . . . . . . . . . . . . . . . . . . 27
2.9 Schematic diagram of a predictive control scheme utilising expert system . 28
2.10 Steady state solution for furnace wall and strip temperatures of the MISG
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.11 Predicted temperature variation across the strip width for fixed strip centre
temperature of 500◦ C and wall temperature 900◦ C . . . . . . . . . . . . . . 48
2.12 Predicted temperature variation along thermocouple length for fixed strip
centre temperature of 500◦ C and wall temperature 950◦ C . . . . . . . . . . 48
2.13 Annealing kinetics of an Iron crystal deformed at 70% and annealed at
various temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.14 Typical grain size difference in hot rolled vs warm rolled microstructure . . 65
2.15 Recrystallisation experiment results showing property changes with tem-
perature for 0.4mm Warm Rolled steel . . . . . . . . . . . . . . . . . . . . 68
2.16 Results for 0.4mm Hot Rolled steel where micrographs show recrystallisa-
tion earlier than indicated by softening of steel . . . . . . . . . . . . . . . . 69
2.17 Possible band of selections for recrystallisation start and end temperatures
for 0.75mm warm rolled strip . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.18 Experimentally determined recrystallisation start temperatures for NZ Steel
strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.19 Experimentally determined recrystallisation complete temperatures for NZ
Steel strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1 Typical cross section of MCL annealing furnace heating zone . . . . . . . . 78
viii
4.2 Photograph inside furnace heating zone showing heating elements installed
on roof and floor. The hearth rolls are removed here. . . . . . . . . . . . . 79
4.3 A warped heating element after removal from furnace . . . . . . . . . . . . 80
4.4 Elevation of Zone 20 ’Hot Bridle Zone’ . . . . . . . . . . . . . . . . . . . . 81
4.5 Typical cross section of annealing furnace cooling zone . . . . . . . . . . . 82
4.6 Cross section of gas jet cooler units . . . . . . . . . . . . . . . . . . . . . . 83
4.7 Photograph showing furnace zone thermocouples installed in furnace . . . . 85
4.8 Gold cup and pyrometer measurement comparison in Zone 12 . . . . . . . 86
4.9 Error between gold cup and pyrometer measurements at all pyrometer lo-
cations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.10 Contact thermocouple mechanism installed in transition section between
preheater and furnace entry . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.11 Typical hearth roll assembly in heating section . . . . . . . . . . . . . . . . 89
4.12 List of common furnace cycles for different products depending on grade
and dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.13 Comparison of furnace zone temperatures for a hard iron product and a
soft iron product of the same dimensions . . . . . . . . . . . . . . . . . . . 92
4.14 Example of manual control of linespeed and preheater control during a
cycle change from 0.55mm to 0.95mm gauge strip, causing incorrect peak
strip temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.15 Demonstration of using ‘mirror wall’ along furnace plane of symmetry to
maintain radiative heat transfer inside furnace cavity. . . . . . . . . . . . . 104
4.16 COMSOL model of a single cast nichrome heating element . . . . . . . . . 105
4.17 COMSOL model used to compare radiative intensity from a square heating
block (right) against a serpentine cast heating element (left) . . . . . . . . 106
4.18 End view of F3DC model showing representation of thermocouple using a
protruding rectangular block . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.19 End view of F3DC model for heating zones showing hearth rolls under the
strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.20 End view of F3DC model for cooling zones, showing cooling tubes above
and below the strip, and coarser mesh size on hearth rolls. . . . . . . . . . 108
4.21 End elevation of F3DC model showing irregular mesh spacing through fur-
nace walls, with fine mesh elements adjacent to radiative face. . . . . . . . 109
4.22 Location and types of heat transfer calculations performed in the F3DC
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.23 Summary diagram of F3DC model operation including inputs, calculation
pathway, outputs and assumptions . . . . . . . . . . . . . . . . . . . . . . 113
4.24 Order of operation of the F3DC model components representing three dif-
ferent sections of the furnace . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.25 Sketch of furnace showing location of steady state ‘line’ output data for
furnace wall or strip centre-line temperature . . . . . . . . . . . . . . . . . 115
4.26 Example of temperature variation around the perimeter of the furnace cav-
ity at a fixed x-position . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
ix
4.27 Comparison plotting the transient wall temperature result due to a step
change in heater power by either the full wall temperature profile or by
selection of a single point . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.28 Summary diagram of RDHT model operation including inputs, calculation
pathway, outputs and assumptions . . . . . . . . . . . . . . . . . . . . . . 120
4.29 Schematic diagram showing geometry of the RDHT model, and type and
location of heat transfer equations used . . . . . . . . . . . . . . . . . . . . 122
4.30 Geometry used to calculate view factors between finite rectangles on par-
allel planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.31 Representation of parallel rectangles approximation used to evaluate view
factors between furnace and strip . . . . . . . . . . . . . . . . . . . . . . . 126
4.32 Furnace walls represented as a 2D calculation mesh, including notation for
calculation point spacing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.33 Sketch of furnace showing approximate location of steady state temperature
results for furnace wall or strip temperature profiles . . . . . . . . . . . . . 135
4.34 Sketch of furnace showing transient temperature results for furnace wall or
strip temperature at specified location . . . . . . . . . . . . . . . . . . . . 135
4.35 Example of transient simulation results with power input changes applied
at different times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.36 Fractional recrystallisation (Xv ) against temperature for 25s hold time . . 139
4.37 Calculated recrystallisation compared with experimental yield stress soft-
ening for 0.4mm 100 grade strip . . . . . . . . . . . . . . . . . . . . . . . . 140
4.38 Calculated recrystallisation compared with fractional reduction in mea-
sured upper yield stress (UYS) for 0.4mm 100 grade strip . . . . . . . . . . 140
4.39 Comparison of model results against steel recrystallisation experiments for
two different gauge sizes (a) 0.3mm and (b) 1.75mm of the same chemistry
(grade 100) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.40 Comparison of model results against steel recrystallisation experiments for
two different chemistry grades (a) 210 and (b) 310, of the same gauge size
(1.15mm) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.41 Comparison of model results of recrystallisation start and finish tempera-
tures for grade 110 hot rolled steel against experimental data . . . . . . . . 145
4.42 Scaled yield stress reduction against temperature for recrystallisation trial
of 0.55mm hot rolled grade 100 steel showing selection of recrystallisation
start and stop temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . 146
tures for all gauges of grade 110 hot rolled steel against experimental data 147
4.46 Yield stress softening curves (fractional) for 1.15mm grade 210 steel and
1.55mm grade 310 steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.47 Comparison of model results for grade 210 hot rolled steel against experi-
mental data with alternative 95% temperature selection for 1.15mm gauge 151
x
4.48 Comparison of model results for grade 310 hot rolled steel against ex-
perimental data with alternative 95% temperature selections for 1.15 and
1.55mm gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.49 Cumulative recrystallisation calculation of steel strip against x-coordinate,
showing highest, nominal and lowest likely recrystallisation . . . . . . . . . 155
4.50 Example of transient result from RDHT-CM model . . . . . . . . . . . . . 155
5.1 Variation in applied power to two heating zones after a cycle change from
550MPa 0.4x940mm strip to 300MPa 0.4x905mm strip . . . . . . . . . . . 157
5.2 Variation in power applied to furnace heating zones 1-12 for several steady
state coils of the same specification over a three week period . . . . . . . . 158
5.3 Variation in power applied to furnace heating zones 1-12 for steady state
coils of the same specification processed two years apart . . . . . . . . . . . 159
5.4 F3DC model results for thermocouple temperatures using minimum, max-
imum and average power input values as shown in Figure 5.2 . . . . . . . . 161
5.5 F3DC model result for strip and furnace temperatures for cycle 24 0.4x940mm
product using average steady state input data from Figure 5.2 . . . . . . . 162
5.6 Detail of F3DC model result for wall and thermocouple temperatures for
cycle 24 0.4×940mm product using averaged steady state input data . . . . 163
5.7 F3DC model result for wall and thermocouple temperatures for cycle 24
0.4×940mm product after introducing 10kW heat loss in position of dam
wall between zones 9 and 10 . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.8 F3DC model result for strip and furnace temperatures for cycle 15 0.39mm×905mm
product using averaged steady state input data . . . . . . . . . . . . . . . 165
5.9 F3DC model result for wall and thermocouple temperatures for cycle 15
0.39mm×905mm product compared against recorded furnace zone temper-
ature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.10 F3DC model result showing temperature variation in the thermocouple
blocks for heating zones 1-6 . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.11 F3DC model result showing temperature variation in the thermocouple
blocks for heating zones 7-12 . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.12 F3DC model result showing difference between strip edge and centre tem-
perature for cycle 24 0.4mm×940mm product . . . . . . . . . . . . . . . . 170
5.13 F3DC model result showing strip temperature variation across the strip
width at the end of the heating zones model for products simulated in
Sections 5.3.1 and 5.3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.14 F3DC model result showing strip temperature variation across the strip
width at the end of the heating zones model for products simulated in
Sections 5.3.1 and 5.3.2 with doubled strip emissivity along edge . . . . . . 171
5.15 Wall and strip temperature results from RDHT furnace model compared
against furnace thermocouple and L1 pyrometer readings for two example
soft-steel recrystallised products . . . . . . . . . . . . . . . . . . . . . . . . 172
5.16 Wall and strip temperature results from RDHT furnace model compared
against furnace thermocouple and L1 pyrometer readings for two example
hard-steel recovered products . . . . . . . . . . . . . . . . . . . . . . . . . 173
xi
5.17 Comparison of results from the F3DC model and the RDHT model for
0.4×940mm G550 strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.18 Comparison of results from the F3DC model and the RDHT model for
0.39×905mm G300 strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5.19 Results from the RDHT-CM model, showing the predicted furnace and
strip temperatures and the resulting cumulative recrystallisation with max-
imum and minimum likely tolerance values for a soft iron recrystallised
product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.20 Example of model sensitivity to temperature. The lowest likely recrystalli-
sation increases from 80% to 100% by increasing peak strip temperature
by only 4◦ C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.21 Results from the RDHT-CM model, showing the predicted furnace and
strip temperatures and the resulting cumulative recrystallisation with max-
imum and minimum likely tolerance values for a hard iron recovered product183
5.22 Example of incorrect heat treatment in two coils caused by tripping of the
induction preheaters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.23 RDHT-CM model results for G250 0.55x1200mm coil number G611779
under normal operating conditions showing full recrystallisation . . . . . . 185
5.24 RDHT-CM model results for G250 0.55x1200mm coil number G611782
without preheating showing incomplete recrystallisation . . . . . . . . . . . 185
5.25 Example showing high temperature at end of coil G609702 due to early
change of furnace settings before cycle change . . . . . . . . . . . . . . . . 186
5.26 RDHT-CM model results for G500 1.45 x 1174mm coil under normal op-
erating conditions showing no recrystallisation . . . . . . . . . . . . . . . . 187
5.27 RDHT-CM model results for G500 1.45 x 1174mm coil under incorrect
operating conditions showing partial recrystallisation . . . . . . . . . . . . 187
5.28 Example of power change in a single furnace zone during two consecutive
cycle changes and a linear equation approximation for the RDHT-CM model189
5.29 RDHT-CM model results for G550 0.4 x 902mm coil using steady state
input data before a cycle change occurs . . . . . . . . . . . . . . . . . . . . 190
5.30 Strip temperature measured by L1 pyrometer changing over the cycle change
period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.31 Manual changes applied to furnace induction heater power and linespeed
during cycle change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.32 Changes to power input in heating zones 1-12 during cycle change due to
PID control actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.33 Changes to power input in cooling zones 13-19 during cycle change due to
PID control actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.34 Transient solution strip temperature result at x = 70m compared with
actual L1 pyrometer measurement . . . . . . . . . . . . . . . . . . . . . . . 193
5.35 Transient temperature solutions from Zones 2 and 5 compared with zone
thermocouple measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.36 Transient temperature solutions from Zones 10 and 15 compared with zone
thermocouple measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 194
xii
5.37 Transient strip temperature result and L1 pyrometer measurement for cycle
change from 0.442x940mm G550 to 0.39x905mm G300 . . . . . . . . . . . 196
change from 0.39x905mm G300 to 0.4x940mm G550 . . . . . . . . . . . . . 196
change from 0.39x905mm G300 to 0.55x940mm G550 . . . . . . . . . . . . 197
6.1 Temperature and recrystallisation results for RDHT-CM model simulation

of Product No.1 - 0.4 940 G550 A 24A . . . . . . . . . . . . . . . . . . . . 205
of Product No.1 after increasing linespeed to 140m/min . . . . . . . . . . . 206
of Product No.2 with 140m/min linespeed . . . . . . . . . . . . . . . . . . 210
of Product No.3 with 130m/min linespeed . . . . . . . . . . . . . . . . . . 212
xiii
List of Tables
2.1 Number and type of variables used in neural network models of continuous
galvanising lines by various authors . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Variation of conductivity and heat capacity of steel with temperature [25, 39] 43
2.3 Activation energy of recrystallisation for various steel types [57] . . . . . . 58
2.4 Valid composition range of Bluescope Steel metallurgical model [57] . . . . 62
2.5 Experienced composition range of NZ Steel MCL chemistry grades . . . . . 62
4.1 Material properties of furnace bodies to be used in conductive heat transfer

calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2 Steel thermal properties to suit MCL steel chemistries 100, 110, 210 . . . . 95
4.3 Steel thermal properties to suit MCL steel chemistry 310 . . . . . . . . . . 96
4.4 Physical properties of air at 20◦ C [20, 54] . . . . . . . . . . . . . . . . . . . 99
4.5 Physical properties of nitrogen gas at 927◦ C [25] . . . . . . . . . . . . . . . 100
4.6 Comparison of radiative and convective heat transfer coefficients calculated
for a simplified furnace representation . . . . . . . . . . . . . . . . . . . . . 102
4.7 Zone lengths, number of grid points per zone, and distance between grid
points used in the RDHT model . . . . . . . . . . . . . . . . . . . . . . . . 123
4.8 View factors between furnace strip and furnace floor at increasing grid
position offsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.9 Comparison of fractional reduction in yield strength against fractional re-
crystallisation calculated by metallurgical model . . . . . . . . . . . . . . . 141
4.10 Calculated average activation energy and standard deviation for MCL chem-
istry grades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.11 Measured average composition of each element used in Qrex calculation for
MCL steel grades over Jan - Nov 2007 . . . . . . . . . . . . . . . . . . . . 144
4.12 Comparison of experimental and model calculated recrystallisation start
(10% softening) temperatures . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.13 Comparison of experimental and model calculated recrystallisation stop
(95% softening) temperatures . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.1 Difference between measured and model average temperature for heating
zones, and zone 13 vs. L1 pyro strip temperature for a selection of coils
without preheating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.2 Difference between measured and model average temperature for heating
zones, and zone 13 vs. L1 pyro strip temperature for a selection of coils
with preheating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
xiv
5.3 Nominal (RX) and Minimum (RXmin ) expected recrystallisation results
from the RDHT-CM model for a selection of soft iron products . . . . . . . 180
5.4 Recrystallisation results from the RDHT-CM model showing high sensitiv-
ity to temperature change when in the region of partial recrystallisation . . 181
5.5 Nominal (RX) and Maximum (RXmax ) expected recrystallisation results
from the RDHT-CM model for a selection of hard iron products . . . . . . 182
5.6 Hardness test results (HR30T) from 2 measurement positions on coils be-
fore and during reduction in strip temperature shown in Figure 5.22 . . . . 184
6.1 Factors describing the range of product types processed on the MCL . . . . 202
6.2 The 10 most common products by weight processed on the MCL in the 12
months preceding May-2008 . . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.3 Applied power to each furnace zone, zone target temperature and element
condition applied to the RDHT-CM model for Product No.1 . . . . . . . . 204
6.4 Strip properties and plant operating data applied to the RDHT-CM model
for Product No.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
6.5 Comparison of Product No.1 base case of 130m/min, and with increased
linespeed of 140m/min . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.6 Comparison of Product No.1 existing 140m/min case, and alternative using
preheating instead of increased Zone 1 power . . . . . . . . . . . . . . . . . 208
6.7 Comparison of Product No.1 base case of 130m/min, and with reduced
linespeed of 120m/min . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
6.8 Results for reduced linespeed case to find maximum allowable strip tem-
perature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.9 Comparison of alternative options for Product 2 - 0.39 905 G300 A 15B . . 211
6.10 Comparison of all alternative options for Product 3 - 0.55 938 G300 A 14B 212
xv
List of Abbreviations
GJC Gas Jet Cooler

HR Hot rolled (steel)
L1 ‘Land 1’ Pyrometer (Furnace Zone 13)
MCL Metal Coating Line
MISG Mathematics in Industry Study Group
NZS New Zealand Steel
PID Proportional Integral Derivative (Control)
RX Recrystallisation (of Steel Microstructure)
UTS Upper Tensile Strength (MPa)
UYS Upper Yield Strength (MPa)
WR Warm rolled (steel)
Model Designations
F3DC Full 3D Comsol

RDHT Reduced Dimensionality Heat Transfer
RDHT-CM RDHT - Coupled Metallurgical
xvi
Chapter 1
Introduction
1.1 Background
New Zealand Steel Ltd has been producing galvanised steel strip at its Glenbrook Site
near Auckland, New Zealand, since 1968, and producing steel from locally mined ironsand
since 1970 [2]. It is the country’s sole producer of flat-steel products, producing around
650,000 tonnes per year at the fully integrated steel mill, comprising iron making, steel
making, hot rolling, cold rolling, metal coating and painting plants. The steel mill is
one of New Zealand’s largest industrial sites, with over 1300 employees. New Zealand
Steel is a division of Bluescope Steel, a company publicly listed on the Australian Stock
Exchange.
Much of the final product from NZ Steel is delivered as steel strip with either GALVS-
TEEL or ZINCALUME coating, which is processed in the Metal Coating Line (MCL)
after rolling to shape in the Cold-Rolling Mill. The Metal Coating Line is a continuous
process, where the strip passes through cleaning and pickling sections, followed by heat
treating in the annealing furnace to obtain the desired metallurgical properties.
After heat treating, the strip passes directly into the molten metal pot and is hot-dip
coated, using air knives to regulate coating thickness. The strip is then cooled and skin-
passed to flatten the applied metallic coating. A passivation coating is applied before the
strip is branded, wound and packaged for delivery [1].
Strip forming in the Cold Rolling Mill causes significant distortion and hardening of
the steel microstructure. NZ Steel produces steel grades of varying specification, ranging
1
Figure 1.1: Schematic diagram of the Metal Coating Line process. The annealing furnace
is 160m long [1]
from 250MPa tensile strength ‘Soft Iron’ to 550MPa ‘Hard Iron’ products. The soft irons
require higher temperature heat treatment to fully anneal the steel, whereas the hard iron
products are heat treated at a lower temperature to relieve the internal stresses in the
steel without causing recrystallisation. A full recrystallisation heat treatment significantly
softens the steel and increases ductility, sufficient for forming products without breakage
of the strip, whereas the stress relief process maintains strength but prevents distortion
and stress-corrosion cracking in the strip [29].
This heat treatment is achieved in the annealing furnace as shown in figure 1.1, a 160m
long brick furnace containing electrical heating elements and induction strip preheaters,
through which the strip passes at up to 140m/min. The furnace contains individually con-
trolled heating and cooling zones, where the power input to the heaters is PID controlled
to temperature setpoints in each zone. The cooling zones have lower heating capacity
than the heating zones, and also contain ‘cooling tubes’ to cool the strip to the correct
temperature for coating in the liquid metal pot.
Figure 1.2: Furnace schematic diagram showing location of Pyrometers IR1-4 and L1-L4,
Induction Pre-heaters B1-C4 and Contact Thermocouple
Recrystallisation of the strip is dependent on both strip temperature and duration of

heat treatment, so it is necessary to ensure that soft iron products reach a sufficiently
2
high temperature to allow full recrystallisation of the strip during its residence time in
the furnace. Conversely for hard iron products, it is necessary that the combination of
strip temperature and residence time does not cause recrystallisation.
Ensuring the correct metallurgical outcome is complicated significantly by the diffi-

culty of measuring the temperature of the strip inside the furnace. Radiation pyrometers
in the furnace attempt to measure strip temperature, but are strongly affected by reflected
radiation inside the furnace cavity, and do not give accurate measurements. Improved
knowledge of the temperature evolution of the steel strip inside the radiant furnace would
allow calculation of the expected recrystallisation of the strip, and could greatly improve
control of the furnace by preventing over-heating of hard iron products, and under-heating
of soft iron products.
The very large size and mass of the MCL furnace give it large thermal inertia, taking
a long time to change temperature in response to changes in heating power. When there
are no changes in strip dimensions or desired properties, the furnace can run for long
periods of time in a steady state, but changes to dimensions or required heat treatment
cause the furnace to enter a transient state. The transient state can be of long duration
when the changes have large magnitude, and over the course of normal operations the
furnace is typically in a transient state for half of its total operating time [39].
In cases of large magnitude changes in strip dimensions or required temperature, called

cycle changes, furnace operators manually adjust the linespeed and preheating of the strip,
which are the only controls that can rapidly affect the strip temperature in the furnace.
The operators adjust these to maintain strip temperature in the desired range as read
on the ‘L1’ pyrometer, which can be difficult while the furnace slowly changes temper-
ature. This manual control combined with poor understanding of strip temperature is
the major contributor to product not meeting specification due to incorrect heat treat-
ment. Improved control during both steady and transient operating conditions could
allow increased production, reduced wastage and improved power efficiency, with signifi-
cant financial gains.
While furnace operators control furnace parameters to maintain specified strip tem-
peratures, this does not have an intuitive connection to the quality of product from the
plant. Dynamic calculation of the metallurgical outcome is a more understandable mea-
sure for operating staff to monitor, as it relays the single most important function of the
annealing furnace - ‘is the strip annealed?’. For soft iron products, the strip should always
be 100% annealed, and hard iron products should be 0% annealed, making this measure
3
a binary, ‘yes or no’ outcome, and the times when it is not achieved can be easily seen.
This does not have a gray area such as controlling to a set temperature, where there may
be uncertainty about the allowable deviation from the setpoint, especially in this case
where the accuracy of the measurement is low.
1.2 Annealing Furnace Characteristics
While continuous galvanising is a common industrial process, the MCL annealing furnace
is very rare amongst similar plants, as almost all utilise gas-fired, multi-pass vertical
furnaces, rather than the long horizontal, single pass radiant electric furnace at NZ Steel
[27].
The furnace is divided into 20 individual ‘zones’, comprising 12 heating zones over
the first 68m, and 8 longer cooling zones over the next 89m. Each zone, both heating
and cooling, contains radiant electric heating elements with the supplied power adjusted
by PID control to a temperature setpoint as measured on furnace zone thermocouples,
which are mounted in inconel tubes protruding into the furnace cavity. The cooling zones
have lower heating capacity, and also contain ‘cooling tubes’ through which ambient air
is blown to indirectly cool the strip. Zone 14 only contains four ‘gas jet coolers’, which
blow cooled furnace gases directly on the strip for rapid cooling when required.
The major complicating factor in understanding strip temperature evolution inside the
MCL furnace is the difficulty in temperature measurement. The furnace thermocouples
protrude into the furnace cavity inside tubes, and it is not well understood how the
measured temperature relates to the actual wall or heating element temperature in the
furnace, which controls the heat transfer to the strip.
There are several radiation pyrometers which measure temperature of the strip, but
these have all been shown to be affected significantly by reflected radiation in the furnace
cavity [48]. These pyrometers read strip temperature from 50-250◦ C higher than the
actual strip temperature and are not useful for measurement, other than pyrometer ‘L1’
in zone 13, which is understood to consistently measure only 10-50◦ C too high.
The total heating capacity of the furnace is 4.6MW by radiant electric heating, and
3MW by induction preheaters mounted immediately before the first heating zone, which
can rapidly increase the strip temperature before it enters the radiant furnace. In contrast,
changing strip temperature by adjusting radiant furnace setpoints is very slow, due to the
4
huge size of the furnace which gives it significant thermal inertia. Changing furnace
temperature setpoints typically takes 10 minutes to complete in the absence of other
process changes, such as strip gauge, width changes or speed changes.
The annealing furnace processes a wide range of combinations of strip thickness, width,
tensile strength and coating type. Changing from one product type to another causes the
furnace to enter a transient state until the power inputs and furnace controls settle at a
new steady state. Depending on production requirements and scheduling for the metal
coating plant, the furnace may run at steady state for long periods of time (up to 36hrs), or
short run products may cause it to be in a regularly changing transient state for a similar
duration. Over the operating life of the furnace, it is estimated to be in a transient state
around 50% of the time.
An important feature of the furnace system under study relates to heating elements
failure. Elements may be damaged by a variety of causes such as warpage or being
contacted by the strip in the case of strip breakage. Elements are wired in three electrical
phases, so that damage to any one element reduces zone heating capacity by one third.
The length of time required to shut down the furnace and replace failed elements means
such replacement only occurs approximately every two years, at planned stoppage times
for major maintenance. For this reason, over the operating life of the furnace, the normal
condition is to have one or more heating zones operating at reduced capacity due to failed
heating elements.
1.3 Study Aims
The aim of this project is to investigate how evolution of strip temperature and metallur-
gical properties can be better understood and accurately calculated in a complex system
with many time varying conditions and interacting factors. It will also be investigated
how this can be applied to create a measurable improvement to the operation and control
of this complex system.
Computer modelling techniques will be used to simulate the complex system with
minimal simplification, while preserving the process dynamics. The control of the system
can be improved by using knowledge from this model to create a coupled temperature-
metallurgical model capable of offline or online control and process decision making, to
optimise furnace settings during both steady and transient states.
5
1.4 Methodology
A literature review was conducted to identify previous work done on modelling of similar
furnace systems, as well as furnace control systems and metallurgical and heat transfer
calculations.
A 3D model (F3DC Model) of the furnace system was constructed to allow prediction
of strip temperatures with minimal approximation of the furnace geometry, and containing
most of the physics present in the actual furnace.
The F3DC model was compared against data obtained from the furnace, and calcu-
lations validated against recorded measurable results from typical processing conditions.
The behaviour of the full model was examined to allow simplification of the model while
understanding effects that may not be included in a simplified system.
The F3DC model was reduced to a system of equations and metallurgical calculations,
and further developed to create a fast, coupled temperature - metallurgical model, using a
simplified representation of the furnace geometry (The RDHT-CM Model). This allowed
simulation of typical furnace operating conditions, and alternatives to be investigated to
understand the effect of changes on the metallurgical outcome of the furnace system.
The RDHT-CM Model was used to simulate specific combinations of operating condi-
tions for the most common products of the plant, and specific, realistic alternatives were
identified. These demonstrate the possibility to use the model to increase production,
reduce wastage and increase power efficiency in the annealing furnace.
1.5 Thesis Outline
The furnace system was described in this chapter, and the requirement for improved
knowledge of strip temperature and strip temperature control were explained.
Chapter 2 reviews existing studies on furnace modelling and control of large furnaces,
and the theory and calculations available for the types of heat transfer present in this sys-
tem. The metallurgical process of annealing and recrystallisation is examined, including
calculations and models, and recrystallisation temperature experiments conducted by NZ
Steel.
6
Chapter 3 discusses the main questions to be investigated based on knowledge of the
problem and the literature review. A method of solution is proposed.
Chapter 4 describes all relevant physical details of the annealing furnace system, and
develops the mathematical theory used to model the system. The development of the
F3DC and RDHT models are discussed in detail, including the fitting and integration of
a metallurgical model to create the final RDHT-CM model. Validation of all models is
shown against measurable furnace data.
Chapter 5 demonstrates the results from the modelling efforts, showing how the
RDHT-CM model can be used to optimise settings for high value products at steady
state, and how the model is able to predict furnace behaviour during transient opera-
tions.
Chapter 6 gives conclusions from the work, and recommendations on how the models
can be utilised for further improvement.
7
Chapter 2
Literature Review
This chapter provides a review of the continuous galvanising process as used on the
Metal Coating Line at New Zealand Steel, and the metallurgical changes caused by heat
treatment in Sections 2.1 and 2.2. Prior work on modelling industrial heating furnaces
and model based control systems is examined in Section 2.3. Mathematical equations
governing the modes of heat transfer present in the furnace are described in Section 2.4,
and previous investigations of the NZ Steel Metal Coating Line annealing furnace in
particular are described in Section 2.5. Section 2.6 discusses the kinetics and different
methods of recrystallisation calculation available which may be used in conjunction with
a furnace model. Section 2.7 investigates previous work on empirical metallurgical models
suitable for use in this problem, including recrystallisation temperature trials conducted
by NZ Steel.
2.1 The Continuous Galvanising Process
Galvanised steel is a common industrial product with a global production of 100 million
tonnes per year, produced on over 450 galvanising lines world wide [76]. Although the
steel industry is generally considered to be a mature industry, there has been continuous
development of this process in an effort to increase production efficiency and product
quality to meet increasing demand.
There are a several distinct processes which occur on a continuous galvanising line. In
order of operation, as shown in figure 1.1, these are:
8
• Strip welding and entry loop
• Cleaning and pickling
• Heat treatment furnace
• Hot dip coating with thickness control
• Continuous cooling or quenching
• Tension leveling and skinpassing
• Chemical coating and passivation
• Coiling, cutting and packaging
Continuous processing requires a continuous feed of steel strip to the line, achieved by
welding the ‘head’ end of a new feed coil to the ‘tail’ end of the coil currently on the line.
To accomplish this, an ‘entry loop’ of moving rollers is used, which allows the current
feed coil to sit stationary while the entry loop is consumed, to allow the new feed coil to
be welded on. This provides several minutes for the automatic welding machine to attach
the new coil, after which the new coil starts feeding out onto the line at a raised linespeed
to refill the entry loop.
Before the strip enters the annealing furnace, it is usually cleaned to remove surface
oil from the cold rolling process, and dirt that may have accumulated while waiting in
storage to enter the galvanising line. Alkali cleaning is commonly used to remove these
oil and dirt residues, followed by a hot pickle bath of hydrochloric acid to remove oxides
and lightly etch the surface [52]. This is not done on all continuous galvanising lines, and
some plants have shown acceptable coating performance without this step.
A heat treatment furnace is used to alter the physical properties of the strip by raising
its temperature to a sufficient level to either recrystallise the steel grain structure through
a process of annealing, or else to relieve internal stresses caused by previous processing
without causing bulk microstructural changes. This allows control of the tensile strength
of the steel product to meet varying customer requirements, such as high strength steels
for structural or roofing profiles, and low strength steels for applications requiring high
ductility. The annealing furnace should contain a non-oxidising atmosphere to prevent
oxidation of the steel at raised temperatures. Direct-fired furnaces create reducing condi-
tions by the generation of CO and H2 during combustion of methane gas, whereas radiant
9
furnaces may employ a Hydrogen-Nitrogen ‘HNX’ gas atmosphere to prevent oxidation
[3, 57].
After heating in the annealing furnace, the strip immediately passes into the molten
metal pot to be coated. Typical coatings are either galvanised steel (Zn coated) or Galval-
ume (55%Al-Zn coating), but other coating types are available such as Al, Al-Si or Pb-Sn
[76]. The thickness of coating applied to the substrate steel (coat weight) is controlled by
air knives immediately after the strip leaves the coating pot. The air pressure through
the knives is modulated to achieve the required thickness to meet product specficiation.
After hot dip coating, the strip must be cooled via air cooling or quenching before it
can be further worked. Once cooled, the strip may pass through ‘skinpassing’ or tension
leveling mills. Skinpassing is a light cold-rolling operation, used to improve surface prop-
erties by suppressing coating ‘spangle’ and improve metallurgical properties for further
processing by suppressing the yield point of the steel, which prevents ‘stretcher strain’
defects [30]. A tension leveling mill is used to improve strip flatness, and prevent strip
shape defects such as ‘edge wave’, ‘centre buckle’, ‘longbow’ and ‘crossbow’ [31].
The final processing step in the continuous galvanising line is the application of a
passivation coat to give resistance to surface oxidation during storage. This is commonly
incorporated into or overlaid by a thin organic layer to give resistance to fingerprint
marking, and also reduces the need for lubrication during further processing or forming
[76]. Following this, the strip is wound into a coil, cut to length and either packaged for
dispatch or sent for further processing.
2.1.1 Cold Rolling
The ‘flat rolling’ process has been used for centuries to produce strips and sheets from
larger metal stocks [34]. This is divided into either ‘hot rolling’ when rolling is conducted
above approximately 0.6× the melting point of the metal (TM ), ‘warm rolling’ between
0.3-0.6 TM and ‘cold rolling’ when below 0.3 TM , and most commonly at room temperature
[50].
In most modern steel plants, the as-cast slab is first hot rolled, followed by pickling to
remove surface scale produced during hot rolling, and then either warm- or cold-rolled to
the final strip dimensions. Rolling to final thickness is done in multiple passes, with a total
thickness reduction from the hot rolled strip of up to 80%, which causes very significant
10
distortion of the steel microstructure and elongation of the grain structure as shown in
figure 2.1, which increases the dislocation density in the metal. Cold work increases the
metal strength properties of yield stress and tensile stress, and simultaneously reduces the
ductility. The strain on the metal matrix gives the deformed metal large strain energy,
which can be reduced in the process of annealing by heating to elevated temperatures,
reversing the property changes caused during rolling [50, 46].
(a) 30% Cold Reduction (b) 70% Cold Reduction
Figure 2.1: Comparison of grain structure distortion in cold rolled low-carbon steel [34]
2.1.1.1 Sheet Metal Forming
Sheet metal forming is one of the most widely employed processes in manufacturing, used
for many purposes such as forming automotive bodies, home appliances and building
fixtures [50, 76]. These manufacturing operations often require forming of the metal by
processes such as punching or drawing. Deep drawing in particular makes high demands
on metal properties and can cause breaking or necking during deformation. Residual
stresses in the rolled metal can also cause warpage in the drawn product [50]. Hence the
metal strip properties from the steel plant, including the hot dip galvanised strip produced
on the NZS MCL, must be controlled to suit varying end use requirements i.e. strip for
deep drawing must have sufficient ductility to prevent tearing, and steel for high strength
applications must meet specified yield strength requirements.
NZS produce annealed ductile products with specified yield strength of 250-400MPa,
and ‘full hard’, high strength steel strip in the range 450-750MPa yield strength. These are
locally called ‘soft iron’ and ‘hard iron’ products respectively, and differ in the steel alloy
chemistries used, the magnitude of cold work applied, and the severity of heat treatment.
11
2.1.2 Phase Diagram and Crystal Structure
As shown in Table 2.5, the carbon content found in the steel grades processed on the NZS
MCL plant ranges from 0.026 - 0.194wt%C, with the most common grades used, ‘100’
and ‘110’, having nominal aims for carbon content of only 0.03-0.05wt%C. These are all
hypo-eutectoid steels (< 0.8%C).
Figure 2.2: The iron-carbon phase diagram [9]
The Iron-Carbon phase diagram given in Figure 2.2 shows these steel compositions
will form α + F e3 C upon cooling below 723◦ C i.e. α-ferrite plus iron carbide. The crystal
structure of α-ferrite is body-centred cubic (BCC), while γ-austenite is face-centred cubic
(FCC) [9]. Hence the crystal structure of the NZS steels is BCC during heat treatment
experienced in the MCL annealing furnace, and only the very highest strip temperatures
experienced during full-recrystallisation heat treatments may cause a temporary phase
change if the strip is heated above 723◦ C, which reverts back to BCC on cooling. Note
that no quenching process is present to cause other structures to form such as martensite,
which has body-centred tetragonal (BCT) structure. The pure α-iron that forms around
723◦ C is only present up to 0.02%C, less than any of the NZS steel chemistries.
12
2.2 Annealing
Annealing is a general term used to describe heat treatment of metals by heating to and
holding at a suitable temperature for a period of time to produce desired microstructural
changes. This is usually done to soften the metal and improve workability for further cold
working or machining, which repairs the microstructural damage caused by mechanical
deformation during primary processing [3, 11, 17, 65]. The changes occurring in physical
properties as a function of temperature during annealing are shown in Figure 2.3. When
a metal is heavily cold worked, much of the strain energy is stored in the form of defects
such as lattice dislocations or point defects, imparting it with higher internal energy than
an unworked metal [66].
Heat treatment is carried out in a heating furnace, which can be either a batch or
continuous type, and of many varied designs depending on the heating method, fuel types
used and methods of supporting or moving the load through the furnace. Batch furnaces
are usually required for bulk items such as heavy forgings, or small lots of varying steel
composition or design. Continuous furnaces are more useful for isothermal heat treatments
of long or continuous products such as slabs or strip, as in the NZS MCL annealing
furnace. This thermal treatment of steel strip is included in most continuous hot dip
galvanising lines immediately before galvanising in the liquid metal pot, and design of the
heat treatment cycle also includes the short dipping time in the hot molten metal [3, 76].
Annealing comprises three stages - recovery, recrystallisation and grain growth. Re-
covery includes all changes caused by heat treatment that do not alter the deformed mi-
crostructure. Recrystallisation allows migration of the high-angle deformed grain bound-
aries, and nucleation and growth of a new grain structure [11]. Grain growth occurs after
recrystallisation, and involves the newly formed grains growing in size by consumption of
adjacent grains. The recovery stages causes only a small reduction in hardness, with the
greatest change in metallurgical properties occurring during the recrystallisation stage
[65]. The changes in grain structure occurring during these processes are shown in Figure
2.4. There are many variations of the general annealing process for different purposes,
and include classifications such as full annealing, subcritical annealing, spheroidising, and
normalising [17].
13
Figure 2.3: Property changes during the stages of annealing. [64]
2.2.1 Recovery
The recovery process involves changes in the density and distribution of defects in the
metal lattice caused by cold working, before the onset of recrystallisation. This leads to a
lowering of the strain energy stored in the lattice by processes such as the rearrangement
of dislocations into the grain boundaries and the removal of vacancies from the lattice.
This can result in large changes in electrical resistance, reduction in stored energy and
small reductions in hardness [65].
Recovery occurs when the metal is heated to a temperature below the recrystalli-
sation temperature range. Sufficient heating of the metal provides the energy to allow
the dislocations to be rearranged into lower energy configurations [66]. The amount of
recovery of physical properties that occurs during heat treatment is a function of both
the temperature and time duration of the heat treatment, and the nature of the metallic
material, as some metals show greater ability than others to recover mechanical properties
after deformation. The activation energy for recovery has been shown to change during
14
(a) Cold Worked, Recovered (b) Recrystallised (c) Grain Growth
Figure 2.4: Changes in grain structure of typical cold worked steel at the three stages of
the annealing process [37]
the process, where vacancy migration in the microstructure appears to be the controlling
factor at short recovery times, and self diffusion (dislocation climb) is controlling over
longer recovery durations [64].
The change in yield stress of a steel during recovery can be calculated using an Ar-
rhenius equation such as Equation 2.1 [64].
A = kte−Q/RT (2.1)
Where :
A = Function of change in yield stress (σm − σr )
σm = yield stress after deformation
σr = yield stress after recovery
k = rate constant
t = time of heat treatment
Q = activation energy of recovery
R = gas constant
T = temperature of heat treatment
The high strength, ‘hard iron’ steels produced by New Zealand steel with yield strength
in the 450-750MPa range undergo this type of recovery annealing in the MCL furnace,
and are heated to a maximum of about 560◦ C strip temperature. In many cases, the
peak strip temperature is kept lower than this to ensure no recrystallisation takes place,
in some cases as low as 450◦ C for thin gauge product which may otherwise be susceptible
to localised overheating / softening, and the formation of ‘wavy edge’ defects or similar.
15
2.2.2 Recrystallisation
Heat treatment at a sufficiently high temperature causes an entirely new crystal structure
to grow in the recovered lattice, with new, strain-free grains growing at the expense of
the existing microstructure. If the metal is held in or above the recrystallisation temper-
ature range for sufficient time, the entire cold worked structure will be replaced with the
recrystallised grain structure, driven by the reduction in free energy of the system. The
physical properties of a metal change markedly during the recrystallisation stage. The
tensile strength and hardness of a cold worked metal will be greatly decreased, with a
corresponding increase in ductility as the recrystallised grain structure grows [66, 64].
To initiate recrystallisation, the metal must be heated above the approximate recrys-
tallisation temperature. Because recrystallisation depends on both time and temperature,
the recrystallisation temperature is defined as the temperature that will cause complete
(95% or greater) recrystallisation in a specified time period [64]. Hence, recrystallisation
can start occurring at temperatures below the recrystallisation temperature, but at a slow
rate i.e. requiring a long time to recrystallise fully.
There are a number of factors which affect the required temperature and time to cause
recrystallisation of a metal [66, 64, 62, 42, 44]:
• Amount of deformation (cold work)
• Deformation rate during processing
• Deformation texture
• Initial grain size
• Composition (alloy chemistry) of metal
The rate of recrystallisation is strongly dependent on temperature - at elevated tem-

peratures, the rate of recrystallisation can be orders of magnitude faster than at the re-
crystallisation temperature. In general, the rate of recrystallisation increases with higher
deformation and faster deformation rates of the metal, and increases inversely to grain
size i.e. smaller grain size before deformation or higher magnitude of deformation will
cause faster recrystallisation. These factors are all related to the density of dislocations in
the metal - increased dislocation density increases the recrystallisation rate and lowers the
temperature required for recrystallisation by virtue of creating more nucleation sites for
16
new grain growth [64, 66]. The chemical composition of the metal affects the activation
energy for recrystallisation - most alloying elements present in usual percentages for low-
alloy steels tend to increase the activation energy and therefore delay recrystallisation,
excepting vanadium and carbon which lower activation energy and promote recrystallisa-
tion [24, 16, 41, 42, 43].
The fully annealed ‘soft iron’ steels produced on the NZS MCL with yield strength in
the 250-400MPa range undergo this type of recrystallisation heat treatment, and are typi-
cally heated to temperatures between approximately 660◦ C and 760◦ C to ensure complete
recrystallisation during the short residence time of the strip in the furnace. Residence
time typically varies between about one and three minutes depending on the strip size
(gauge and width), with most strip sizes able to be run with short residence times i.e. at
high plant line speed.
Recrystallisation of the deformed microstructure is called primary recrystallisation,

which is distinct from the abnormal grain growth or secondary recrystallisation process
that can occur in an already recrystallised microstructure. Primary recrystallisation can
be divided into two separate mechanisms - nucleation where new grains first appear, and
growth where the grains increase in size to replace the deformed microstructure. The
driving force for creation of the new microstructure is the reduction in free energy of
the microstructure, by the reduction of the dislocation network remaining following cold
working and recovery [64].
Nucleation involves an initial crystallite with low internal energy forming in the de-
formed matrix, separated by a high-angle grain boundary. The grain boundary migrates
into the deformed microstructure, by coalescence of the deformed grains, leaving behind
the larger strain-free and undeformed grain with low internal energy. The nucleation
rate of new grains can be either site saturated i.e. constant during recrystallisation, or
time dependent. The growth rate of the new grains is a function of the grain boundary
mobility, and the dislocation density of the strained matrix [24, 64].
2.2.2.1 Laws of Recrystallisation
A set of laws of recrystallisation were proposed by Mehl [45] and Burke and Turnbull [10],
which detail the general behaviours of the recrystallisation process, and how it is affected
by changes in processing or materials parameters. These laws are based on results of
experimental work, and are applicable in most cases, and rational when recrystallisation
17
is considered a thermally activated process driven by stored deformation energy [24, 64].
1. A minimum amount of deformation is required to initiate recrystallisation.
2. Increasing the time of annealing decreases the temperature required for recrystalli-
sation
3. The required temperature decreases as the strain increases
4. The recrystallised grain size depends on the amount of deformation and the tem-
perature of recrystallisation
5. Larger initial grain sizes increase the required recrystallisation temperature for a
given deformation
6. Higher deformation temperatures increase the required recrystallisation temperature

for a given deformation
2.2.3 Grain Growth
Further exposure to temperatures above the recrystallisation temperature after recrys-

tallisation is complete leads to the grain growth stage. This increases the average size
of the grains, by migration of high-angle grain boundaries into the recrystallised matrix.
This is driven by the reduction in surface free energy of the grain boundaries, an overall
reduction in the free energy of the metal lattice. The lowest energy configuration is to
have equal sized hexagonal grains where all grain boundary angles are 120◦ . This means
that recrystallised grains with less than 6 sides will be consumed during grain growth,
while grains with more than 6 sides will grow [64].
This grain growth is normal or continuous grain growth, but after a long duration
it is possible to develop abnormal grain growth, also called secondary recrystallisation,
where some grains start growing preferentially and rapidly and consume other grains.
Secondary recrystallisation is more likely in cases where normal grain growth is inhibited
by the presence of less-stable particles or very large average grain diameters have been
reached [64].
The strip temperature and duration of heat treatment in the MCL annealing furnace
mean that heat treated strips are not likely to experience much grain growth beyond
primary recrystallisation.
18
2.3 Annealing Furnace Modelling
The calculation of temperatures inside furnaces is a common industrial problem. As

well as continuous steel strip production, furnace modelling has also been carried out for
processes such as heat treatment of individual machined or cast metal components for the
automotive and building industries, or small scale furnaces such as ceramic baking kilns
and laboratory scale furnaces.
There are several approaches that have been used to model industrial furnaces of
varying complexity, including furnaces similar to the NZ Steel annealing furnace. The
most common methods used to model these furnaces are either by a system of mathe-
matical equations representing an approximated system, or by purely statistical methods.
3D computer modelling techniques do not appear to have been widely used in furnace
modelling applications.
Mathematical models of industrial furnaces are common, and have been in use for a
long period of time. Early examples of furnace modelling with mathematical techniques
include work by Hottel and Sarofim [22], which are also applicable to continuous steel strip
furnaces similar to the annealing furnace. These approaches require understanding of the
physical processes of heat transfer in the furnace system, and techniques to calculate the
resulting temperatures. These solutions are often achieved using finite element methods
to solve partial differential equations.
Statistical modelling of furnace systems is a completely different approach to furnace

temperature calculation, and does not require understanding of the physical system being
modelled. Techniques such as neural networks or principal component analysis can be
used to predict system outcomes based on any combination of input variables after the
model is fitted to a sufficiently large statistical data set.
Further to the development of models to calculate furnace temperature or product

property outcomes, it is also common to utilise model results to optimise furnace control
or scheduling methods, with the objective of increasing throughput, product quality, or
energy efficiency.
19
2.3.1 Mathematical Models
Mathematical modelling of industrial furnace systems has been a common practice across
many industrial fields, and in the steel industry it is commonly applied to batch reheating
of coils and slabs, and continuous heating furnaces for slabs and uncoiled (flat) strip.
The large majority of modern steel heating furnaces are of the direct-fired type, which
are heated by the combustion of fuel gas or pulverised solid fuel, and transfer heat to the
steel and the furnace walls by both thermal radiation and convection of the hot combustion
gases. The direct-fired type of furnace is more difficult to model than the radiant electric
furnace, as the hot furnace gases play a central role in the heat transfer and must be
considered in the model as well as the radiative heat transfer. Convective heat transfer
in the atmosphere of a radiant electric furnace is of low magnitude and can usually be
omitted.
It is likely that the greater ease of quantitative analysis before powerful computing
technology was available was a large factor in the increasing use of electric furnaces in
the 1960’s, despite the greater overall capital and energy cost compared to direct fired
furnaces [22]. The NZ Steel annealing furnace was commissioned in 1968, however most
other plants owned by Bluescope Steel utilise direct fired furnaces of newer manufacture.
While most modelling effort for steel heating furnaces has been on direct fired furnaces,
these models are still useful to examine and compare with the NZ Steel radiant furnace,
as there are many similarities in the radiant and conductive heat transfer in the furnace.
A common method to model a long continuous furnace has been to approximate the
geometry as two parallel surfaces representing the furnace walls and the steel strip [27, 39]
or as a simple three-surface enclosure with surfaces representing the walls, heaters, and
strip [22]. In all cases, the furnace geometry is represented as a simplified shape, and
radiative heat transfer calculated between the furnace surfaces using radiative view factors
(or shape factors).
Recently, more complex models of annealing furnaces have been developed using finite
element or other methods to represent the geometry more accurately. There is a wide
range of models reported for continuous annealing furnaces, although almost all are for
fuel fired furnaces, either directly or indirectly-fired, rather than electric radiant furnaces.
Several of these models have been examined to investigate the methods of representation
of the complex geometry, the radiative and convective heat transfer, and the methods of
solving the mathematical equations.
20
Marlow [36] modelled a direct-fired furnace on a Bluescope Steel continuous galvanis-
ing line, using a one-dimensional model for both the steel strip and the furnace walls.
Radiation was calculated only in the direction normal to the strip i.e. heat transfer by
radiation was only considered between the strip and the wall immediately opposite it. Be-
cause radiative heat transfer is governed by the furnace wall surface temperature, it was
recognised that a fine calculation mesh spacing is required adjacent to the inner surface
of the wall, but a coarse spacing may be used near the exterior face. The model equations
were solved using a combination of two second-order finite difference methods.
Figure 2.5: Example of transient solution of strip temperature in Marlow model during
‘hard’ to ‘soft’ cycle change [36]
As shown in Figure 2.5, this model was found to be sufficient to predict strip and
furnace temperatures specifically during transient conditions, similar to the transient
‘cycle changes’ that occur in the NZ Steel annealing furnace. The processing conditions in
this furnace are also similar to NZ Steel, processing both ‘hard iron’ (recovered) and ‘soft
iron’ (annealed) products in similar dimensions, although the furnace design, geometry
and control parameters are significantly different.
To obtain the solution to the transient cycle change problem, the initial conditions
to the transient model were created using a steady state solution of the furnace and coil
system before the transient situation occurred. The solution values for temperatures and
heat transfer coefficients at every mesh point in the steady state model are transferred as
input to the transient model, which then calculates transient furnace and strip tempera-
tures based on user-specified control changes such as line speed, strip dimensions or fuel
burner gas flow rates [36].
21
A similar model was developed by Prieto et al [55] for a vertical, multipass, indirectly
fired furnace. In this furnace the combustion gases are contained within radiative heating
tubes, and do not directly transfer heat to the strip by convection. Although the com-
bustion gases are not in contact with the strip, convective heat transfer still occurs in the
furnace atmosphere gases by natural convection in the vertical furnace orientation. The
complex furnace geometry was defined by splitting the furnace into 30 smaller rectangular
shapes, which contained only straight strip paths sectioned from the multi-pass geometry.
(a) Strip Temperature (b) Convective and radiative transfer
Figure 2.6: Strip temperature and heat transfer rate results for direct fired furnace using
stepwise model [55]
Rather than solve using finite differences, this model assumes initial temperature dis-
tributions and then performs a stepwise solution, calculating the heat transfer in each fur-
nace zone sequentially, based on empirical equations for heat transfer coefficients. Simple
radiative view factors are calculated between the four sides of each rectangular section of
the furnace geometry. The solution is iterated until a heat balance is achieved in each zone
based on radiative and convective heat transfer to the strip and the furnace atmosphere,
and heat losses through the furnace walls.
The model results in Figure 2.6(b) show that heat transfer by convection is much
smaller than by radiation. This was dependent on temperature, with convective heat
transfer becoming negligible at higher strip temperatures due to the T 4 dependence of
radiative heat transfer, and the reduction in temperature difference between the strip
temperature and the furnace atmosphere. Convection appears to be significant only at
strip temperatures below around 300◦ C. The heat transfer by natural convection in this
system is likely to be higher than the NZ Steel annealing furnace due to the vertical
orientation of the strip.
The stepwise approach was able to quickly solve the strip and furnace temperatures in
this model, but the iterative solution does not allow calculation of transient effects such
as cycle changes. The model was designed to be easy to implement and run quickly, but
22
the modelling method will not be sufficient to represent transient effects in the NZ Steel
annealing furnace.
A model of an indirectly fired furnace was created by Ramamurthy et al [56], for a

horizontal furnace of similar orientation to the NZ Steel annealing furnace, where the
combustion gases are contained in radiant tubes both above and below the steel strip.
The radiant tube section of the furnace is installed after a direct-fired furnace and before
the cooling section. The indirect radiant tube section of the furnace is used to improve
product surface quality, as the combustion products are not in contact with the steel strip
as in the direct fired section.
It was recognised that a mathematical thermal model of the furnace can be used to
improve throughput by running a series of numerical experiments, where the model allows
detailed simulations of the effect of changing furnace operating conditions and product
dimensions and properties.
(a) Side elevation (top half furnace only) (b) end elevation
Figure 2.7: Side and end elevations of similar model for indirectly fired furnace with
radiant tubes instead of electric heaters [56]
In a manner similar to the model by Prieto et al [55], the heating zones of the furnace
are divided into smaller zones, which are assumed to be isothermal. An energy balance is
applied to each zone, and the resulting equations solved simultaneously to calculate the
temperature and heat flux in each calculation zone. The zones are divided so there is one
radiant tube in each zone, leaving 5 isothermal surfaces to be calculated in each zone -
strip, radiant tube, roof/floor, and two side walls, where each surface is assumed to be a
diffuse gray emitter [56]. The furnace geometry is shown in Figure 2.7, and is similar to
the NZ Steel annealing furnace design.
23
2.3.2 Statistical Models
An alternative approach to creating a physical equation-based mathematical model is to

treat the furnace system as a statistical problem. This requires creating a large data set
containing sufficient variation in all input parameters such as furnace temperatures, strip
size and speed, steel chemistry etc, and all output parameters such as pyrometer temper-
atures and hardness test results. Statistical methods such as neural network modelling
or principal component analysis can be used to create models that predict strip output
properties based on the combination of input properties and processing conditions.
Several recent studies have used neural network models to predict final strip properties
specifically on continuous galvanising lines. Neural networks are some of the most common
artificial intelligence tools now being used to solve industrial problems, mostly using
multilayer feed-forward networks [53]. An advantage to using such tools is that little
understanding of the physical system or transport processes occurring in the system is
necessary to create the model, other than to choose which variables or measurements need
to be included in the data set. These types of models are often used to predict final strip
properties, rather than as physical furnace models, as they do not model any physical
objects or functions directly, and offer no insight into the process [63].
Similar neural network models have been created by Ordieres Mere et al [51], Pernia-
Espinoza et al [53] and Capdevila et al [12], which describe the type of data required to
train the models as shown in Table 2.1. While the models do not predict strip temperature
evolution inside the annealing furnaces, one or more strip temperature values are used to
predict metallurgical outcomes. Two of these models include a peak strip temperature
value in the neural network data set [12, 51], and one model includes several furnace zone
set point temperatures as well as a measured strip peak temperature [53].
Large data sets are required to build and train neural network models, and this is seen
especially in the neural networks that consider the variation in steel chemistry on the
galvanising line. The models considering chemical composition require accurate statistical
data for each alloying element present in the steel for each coil that is heat treated in the
furnace. Other input variables required for each coil will include strip width and gauge,
line speed, cold reduction, furnace temperatures, and may include other factors such as
heating or cooling rates. Several output variables are then predicted by the neural network
based on the combination of input variables.
24
Input Variables
Model 1 Model 2 Model 3
Peak Strip Temp Rolling Temp Strip thickness
Strip Speed Hot Reduction Strip Width
C wt% HR Cooling rate Entry temp
Mn wt% Coiling Temp Zone 1 temp
Si wt% C wt% Zone 3 temp
S wt% Mn wt% Zone 5 temp
P wt% Si wt% Final temp
Al wt% P wt%
Cu wt% S wt%
Ni wt% Al wt%
Cr wt% N wt%
Nb wt% Ti wt%
V wt% V wt%
Ti wt% Nb wt%
B wt% Cold Reduction
N wt% Heating Rate
C equiv. wt% Annealing Temp
Annealing time
Cooling rate
Output Variables
Yield MPa Yield MPa Strip Speed
UTS MPa UTS MPa
Elong. % Elong. %
Authors
Ordieres Mere et al [51] Capdevila et al [12] Pernia et al [53]
Table 2.1: Number and type of variables used in neural network models of continuous
galvanising lines by various authors
25
Considering that several thousand sets of these 10-20 variables are usually required
to train the model, the total amount of data required is very large. For example, the
neural network by Pernia et al [53] utilised over 30,000 data points for each variable
listed, 80% of which is used to train the model, and the remaining 20% used to verify the
model predictions. Each of these neural network models were able to predict the output
variables to acceptable accuracy from the combinations of input variables - the model in
Ordieres Mere et al was able to predict yield strength, tensile strength or elongation with
a mean error of around 3% in each case [51].
It was also shown by Pernia et al [53] that the neural network model could be used
to predict the strip speed required in the furnace to provide sufficient heat treatment to
the strip, given the furnace temperatures, with a mean error of under 6%. In this way,
a neural network model could be used to control the furnace linespeed, if trained with
sufficient data and process variables to encompass all likely processing conditions. Given
that this model utilised 8 variables with 30,000 data points each to model a single steel
chemistry, the amount of data and training required to model variation in several steel
alloying elements and processing conditions such as hot/warm rolling, will be much higher
again.
The main disadvantage of this type of model is that the physical system is not mod-
elled directly, and so there is no capacity to predict any occurrences outside of the data
used for training. The model may produce accurate predictions under normal processing
conditions, but cannot predict new conditions such as equipment failures or new prod-
uct types or dimensions etc without retraining using sufficient data encompassing these
conditions.
2.3.3 Model Based Control Systems
Due to factors such as the long dead time, slow dynamics and thermal interactions in a
continuous annealing furnace, conventional control can be very difficult [79]. To combat
this, several control systems have been proposed which utilise physically-based mathe-
matical models for online control of continuous furnaces, usually of the direct-fired type.
Schematic diagrams of these types of multi-layered control scheme are shown in Figure
2.8. These model based control systems involve a combination of a physical model to
predict strip and furnace temperatures in steady state or dynamic conditions, and one or
more tunable control systems that adjust parameters to the physical model and optimise
furnace settings based on model outputs [74, 75, 79, 77]. As well as optimising furnace
26
operations, physically based furnace models have also been used to optimise scheduling
of products heat treated in the furnace to maximise throughput while maintaining the
required heat treatments [60].
The physical models included in the control systems are mostly of similar nature to the
physical furnace models described earlier, using simple radiative heat transfer calculations
to predict heat transfer from the furnace to the strip under steady state conditions only
[74, 77].
Rather than modelling the temperature of strip and furnace bodies using heat transfer
equations during dynamic conditions, these models usually predict dynamic temperature
changes by curve fitting to measured temperature responses. The strip and furnace tem-
perature responses are similar to a first order response to a change in one parameter,
with nonlinear gains or additional dead times before or after the first order behaviour
[74, 79]. A steady state model is used to predict the temperature before a disturbance
occurs, such as a change in strip speed or fuel flow rate, as illustrated in Figure 2.8(b).
The approximated response equations are then used to predict strip temperatures as they
change to the next steady state during this disturbance.
(a) Yoshitani et al [79] (b) Ueda et al [74]
Figure 2.8: Schematic diagrams of multi-layer predictive control schemes implemented on

several continuous annealing lines
Once steady state and dynamic strip temperature prediction is established through
physical equations or statistical models, a variety of methods are used to control either
furnace temperatures or strip speeds etc to maintain correct heat treatment or maximise
throughput. Yoshitani et al [79] utilise a sophisticated combination of self-tuning control
schemes to compensate for rapid changes in operating conditions, and parameter estima-
tion schemes to compensate for slow changes in plant characteristics as shown in Figure
27
2.8(a). The main objective of the control model is to look ahead in the product schedule
to foresee upcoming product changes, predict the best method of adjusting furnace heat-
ing and linespeed to minimise temperature disturbances and maximise throughput, and
apply control changes at the calculated best time. This package has been implemented
in several Japanese steel plants successfully, and has been shown to enable a production
increase of 3-5% over conventional PID control [79].
Another approach to model-based control is the ‘expert system’ used by Yahiro et al

as shown in Figure 2.9, where the model makes control change decisions based on a set
of rules. These rules are knowledge based, derived from the experience of plant operators
and supervisors, and translated into empirical rules or bounding limits on the control
system. Rather than the control system reacting directly to changes in control variables,
the expert system can examine other system variables and make changes based on known
upcoming events such as strip gauge changes, or uncommon events such as equipment
failure in order to minimise temperature deviations.
Figure 2.9: Schematic diagram of a predictive control scheme utilising expert system [77]
It was found that this expert system was able to reduce the frequency of manual
intervention by operators to one fourth of that with a conventional control system. Other
advantages of the expert system are the integration of the knowledge of experienced
operators, and the standardisation of judgement criteria for control actions [77]. This also
allows the control system to make intelligent decisions when there are multiple options to
correct control variable deviations i.e. in the case of a strip temperature drop, the control
system can choose between either increasing furnace heating power, or decreasing strip
28
speed based on the expert knowledge. This expert system refers to the rules to work as a
switch to choose between routines to either change the linespeed or to change the furnace
temperature as in Figure 2.9.
2.3.4 Model Type Selection
Due to several reasons, an analytical mathematical model is the best choice for this
project, and the full development of this model is shown in Chapter 4.
As covered in Section 3.3, a 3D mathematical computer model is first developed, which

is capable of accurate simulation of the furnace geometry and physical system, and can
show detailed results regarding the spatial variation of furnace and strip temperatures in
all directions. There are several spatial effects of particular interest, such as the temper-
ature variation of furnace thermocouple tubes, and differential strip edge heating. These
effects have operational consequences, and are important to investigate to improve under-
standing of furnace behaviour and control. This sort of spatial temperature calculation is
beyond the capability of a statistical or process control model, as it requires direct calcu-
lation of temperature based on geometry and heat transfer equations. It is impossible to
model statistically, as these temperatures cannot be measured in the actual furnace, and
hence cannot be included in the model data set.
In order to provide rapid solutions suitable for real-time simulation, the 3D model is
simplified further. While it would be feasible to develop a statistical model such as a
neural network to approximate the 3D model solutions, this approach is not ideal, as it is
not adjustable or maintainable by plant engineers in the future, one of the requirements
to satisfy NZ Steel. It would also only be applicable to simulations run under current
furnace conditions, and any design changes in the future may cause further training of
the model to be required.
In this case, a simplified mathematical model is developed, as shown in Section 4.6.

This model requires many approximations and simulates a highly simplified furnace ge-
ometry to solve furnace and strip temperatures rapidly, but is capable of calculations
within an acceptably small margin of error. This is validated against both actual furnace
operational data, and the 3D computer model developed in Section 4.5.
29
2.4 Heat Transfer
Modes of heat transfer are generally divided into three basic modes - Conduction, Con-
vection and Radiation [59]. All three of these basic modes are present in the physical
system described by the MCL annealing furnace, and their relative contributions must be
investigated to determine the system of equations to be solved when calculating overall
heat transfer within the furnace.
2.4.1 Conductive Heat Transfer
The process by which heat is transferred through a body by diffusion is heat conduction.
Unlike radiation which requires no transfer medium, conductive heat transfer requires
physical contact between conducting bodies. On the microscopic level, conduction is a
consequence of high energy molecules or atoms interacting with lower energy molecules. In
electrically non-conductive bodies, heat conduction is due to atomic lattice motion, and in
electrically conductive bodies also includes motion of free electrons [7]. At a macroscopic
level, conductive heat transfer can be described by Fourier’s Law in one dimension:
dT
q = −kA (2.2)
dx
where q is the rate of heat transfer (heat flux) in the x-direction, k is the thermal con-
ductivity of the body material, A is the cross-sectional area of heat transfer and dT
dx
is the
temperature gradient in the positive x-direction [7].
Conductive heat transfer is a dominant method of heat transfer within the furnace
system, as it is present at all times within all bodies in the system. Examples of heat
conduction in the MCL furnace are heat transfer through the furnace walls, as they have a
hot inner surface approaching 1000◦ C, and cool outer walls near the ambient temperature
around 50◦ C, causing heat to conduct through the wall from hot to cold. Other bodies
strongly affected by heat conduction include:
• Hearth Rolls
• Thermocouple Blocks
• Steel Strip
30
Conduction is a major contributor to the overall heat transfer calculation of the furnace
system, and controls the temperature of the furnace and strip bodies based on the amount
of heat being delivered to each body via radiation from the heating elements.
2.4.2 Convective Heat Transfer
Convection is often considered as a distinct mode of heat transfer, although it is reliant

on conductive heat transfer as its driving force. Convection relates to the flow of heat
from a boundary to a moving fluid, where the movement of the fluid causes the heat to
be physically transferred, either by buoyancy forces moving the fluid or from an external
source such as a fan or pump. If the fluid flow is induced by a pump or other method, the
heat transfer is forced convection, and if the fluid flow is caused by changes in the fluid
density due to the temperature change, it is natural or free convection. While the fluid
flow causes heat transfer in the fluid, the basic mechanism is conductive heat transfer into
and within the fluid [59]. Convective heat transfer to or from a boundary is described by
Newton’s Law of Cooling:
q = hA(Tw − Tf ) (2.3)
where q is the rate of heat transfer (heat flux) at the boundary, h is the heat transfer
coefficient, A is the area of the boundary, Tw is the wall (boundary) temperature, and Tf
is the temperature of the fluid. The definition of Tf changes according to the situation -
in unbounded convection such as a hot plate immersed in a large fluid body, Tf is usually
taken as the bulk temperature of the fluid away from the boundary, or T∞ . If the fluid is
flowing in channels or in heat exchanger tube bundles etc, Tf is taken as the mean fluid
temperature Tm [59].
Convective heat transfer is present in the MCL furnace system in the gas atmosphere
inside the furnace cavity, and also on the outer faces of the furnace in contact with the
ambient atmosphere. The H2 /N2 atmosphere in the furnace is mechanically stirred by
the movement of the strip and rollers, so heat transfer can be considered to be by forced
convection, and the heat transfer from the outside walls is by natural convection only.
The heat transfer coefficient over a surface varies with position, due to changes in
adjacent fluid temperature or speed, so the coefficient used to calculate the bulk heat
flow from the surface is an integrated value over the entire surface [7]. The heat transfer
31
coefficient over a surface is given in terms of the nondimensional Nusselt Number, Nu:
hL
Nu = (2.4)
k
where L is the characteristic length of the surface, and k is the thermal conductivity of the
convecting fluid. Numerous correlations exist to calculate the Nusselt number depending
on convection type and boundary geometry, such as for natural convection from a vertical
plate as in the furnace exterior side walls [14]:
2
0.387Ra1/6

N u = 0.825 + (2.5)
[1 + (0.492/P r)9/16 ]8/27
This uses the nondimensional Rayleigh, Prandtl and Grashof numbers:
µcp υ
Pr = = (2.6)
k α
Ra = Gr · P r (2.7)
3
gβL (Tw − T∞ )
Gr = (2.8)
υ2
υ = fluid kinematic viscosity

α = fluid thermal diffusivity
β = fluid coefficient of thermal expansion
L = characteristic length
The amount of heat transfer occurring by convection is proportional to the heat trans-
fer coefficient and the temperature differences between surface and fluid, as well as the
area of the surface. The relative impact of convection for both exterior walls and internal
furnace cavity walls will be considered in this thesis and is given in Section 4.4.4.
2.4.3 Radiative Heat Transfer
Radiative heat transfer or thermal radiation involves the movement of heat without physi-
cal contact or a carrier medium as required for conductive or convective heat transfer. All
materials continuously emit and absorb electromagnetic radiation with their surroundings,
32
by raising or lowering their molecular energy levels [47]. Radiation phenomena are usu-
ally classified according to their characteristic wavelengths - thermal radiation occurring
between 0.1 - 100 µm, the strength and wavelength of the emitted radiation dependent
upon the fourth power of the temperature of the material [33].
The amount of radiation emitted or absorbed by an object is dependent upon the

surface properties of the object. A blackbody is defined as a perfect emitter and receiver
of radiation, across all temperatures and wavelengths - at any given wavelength, no surface
can emit or receive more radiation than a blackbody [13]. The amount of radiation emitted
by a blackbody is given by the Stefan-Boltzmann Law :
qr
Eb (T ) = = σT 4 (2.9)
A
qr = Heat flux by radiation, W

A = Area of blackbody emitting the radiation, m2
T = Absolute temperature of the area A, deg K
σ = Stefan-Boltzmann constant
= 5.67 × 10−8 W/m2 .K 4
Most real surfaces do not behave as blackbodies, and may emit or absorb radiation
preferentially as function of wavelength, temperature and direction. Dimensionless factors
are used to relate the properties of a real surface to a blackbody, defined as:
energy emitted by a surface

Emissivity, =
energy emitted by a black surface at the same temperature
absorbed part of incoming radiation

Absorbtivity, α=
total incoming radiation
reflected part of incoming radiation

Reflectivity, ρ=
total incoming radiation
For a real surface, these properties may vary both with the wavelength of incoming
radiation, and the direction of incoming radiation. For engineering applications, it is
usually accurate enough to choose suitable average values of these properties, called gray
body properties, and to assume no directional dependence i.e. to assume that the surface
33
is a diffuse emitter. A non-diffuse surface is called specular. For diffuse, gray surfaces,
these properties are related by:
=α=1−ρ (2.10)
2.4.3.1 View Factors
The amount of radiation heat transfer between any two surfaces is dependent on the
geometry and orientation of the two surfaces relative to each other. The relative amount
of diffuse radiation leaving surface Ai and reaching surface Aj is called the view factor
FAi −Aj also called a shape factor or configuration factor [33].
The view factor between two areas Ai and Aj is defined as:
diffuse energy leaving Ai intercepted by Aj

FAi −Aj =
total diffuse energy leaving Ai
There is also a law of reciprocity:
Ai FAi −Aj = Aj FAj −Ai (2.11)
The view factor between two finite areas can be difficult to calculate, and requires the
solution of a double area integral. An extensive collection of solved integrals for common
simple geometries has been catalogued by Howell [23]. There also exists numerous software
packages used to numerically calculate view factors for more complicated geometries. For
simple geometries, it is possible to calculate view factors using other methods, such as the
crossed-strings method for two dimensional geometries [47]. This method can be of use
in this case, as the long MCL annealing furnace has a constant cross section, and view
factors calculated on the 2D section can be used as a simple approximation for the whole
geometry.
2.4.3.2 Heat Transfer in Gray Enclosures
While heat transfer inside enclosures consisting of black surfaces is easy to calculate, most
real situations involve non-black surfaces which complicate calculations due to reflection
between surfaces - these calculations can be exceedingly difficult unless the surfaces are
considered to be gray, diffuse emitters [13]. Because the surfaces in the enclosure reflect as
34
well as emit radiation, the calculation of heat transfer between surfaces involves the total
radiation energy leaving each surface, the radiosity, J. The amount of radiation reaching
a surface is called the irradiation, G.
The radiosity per surface area is equal to:
Ji = (radiation emitted by surface i) + (radiation reflected by surface i)

= i Ebi + ρi Gi (2.12)
where Ebi is the blackbody emissive power of surface i = σTi4 . The net radiative exchange
between any two surfaces is equal to:
Qi−j = (radiation leaving i reaching j) - (radiation leaving j reaching i)

= Ai Ji Fi−j − Aj Jj Fj−i
= Ai Fi−j (Ji − Jj ) by applying the law of reciprocity
= Ai Fi−j σ(Ti4 − Tj4 ) (2.13)
2.4.3.3 Three Surface Enclosure
Hottel and Sarofim [22] describe a simple radiative transfer equation for a ‘three-zone
enclosure’ as being a very good first approximation to an enclosure involving only surface
radiation, such as electric heating furnaces. The three surfaces are an isothermal grey heat
source, A1 representing the heating elements, an isothermal grey sink A2 representing the
strip or other heated item, and a radiatively adiabatic surface Ar which approximates
the furnace refractory walls. Note that this does not allow for heating or cooling of the
refractory walls, but this may be a good approximation for short time periods where the
wall temperature does not appreciably change.
The radiative heat exchange formula developed for this situation is:
1
Q̇1−2 = ρ1 ρ2 1 (J1 − J2 ) (2.14)
A1 1
+ A2 2
+ F1−2 + 1
1 + 1
F1−r F2−r
where Q̇1−2 is the rate of heat exchange between the grey surfaces A1 and A2 given the
view factors F1−2 , F1−r and F2−r between each surface, and the radiosity of the two grey
surfaces J1 and J2 .
35
2.4.4 The Heat Equation
Many physical processes, such as acoustics, fluid dynamics, and specifically heat transfer
can be described using partial differential equations. The heat equation is a parabolic
PDE that describes heat conduction in a solid, in one, two or three-dimensions. These
PDEs can be solved by approximation using finite element or finite difference schemes to
replace the partial derivatives with a system of algebraic equations [15] and solved either
directly (the explicit solution) or iteratively (the implicit solution).
A simple representation of the heat equation in one dimension is [19, 71]:
∂ 2T

∂T
cp ρ =k +Q (2.15)
∂t ∂x2
cp = material heat capacity

ρ = material density
k = material thermal conductivity
x = length dimension
T = temperature
Q = heat generation rate
Solving the heat equation also requires boundary conditions to be specified at each
boundary of the calculation domain, which depend on the physical mechanism or heat
transfer type present at each boundary [19, 71]. The three main types of boundary con-
dition useful for heat transfer calculations are prescribed temperature, insulated boundary
and Newton’s Law of Cooling, which is the equivalent of convective heating or cooling at
the boundary.
The prescribed temperature boundary condition is used where a temperature at a

boundary is fixed at a single value, or as a function of time, where T (0, t) represents
temperature or heat T at position x = 0 at any time t.
T (0, t) = TB (t) (2.16)
It is possible to specify the heat flux, φ(t), at the boundary if it is known, or in the
case of insulated boundaries, where no heat may flow through the boundary, the flux is
36
specified to equal zero:

∂T
−k (0, t) = φ(t) (2.17)
∂x

∂T
−k (0, t) = 0 (insulated boundary) (2.18)
∂x
In the case of convective heat transfer at a boundary, Newton’s Law of Cooling is used
to specify the heat flux using a heat transfer coefficient h, which uses a linear combination
of T and ∂T∂x
[19].
∂T
−k (0, t) = −h [T (0, t) − TB (t)] (2.19)
∂x
37
2.5 Prior Studies of the NZ Steel MCL Furnace
Bluescope steel has previously conducted comprehensive modelling work on the Port Kem-
bla MCL6 annealing Line in the mid 1990s, and since about 2003, modelling work has
been carried out on the NZS MCL annealing furnace which is the subject of this project.
There are four separate areas of investigation on this, viz.
1. NZ Steel steady-state model
2. Mathematics in Industry Study Group (MISG) investigation
3. Sydac control model
4. Further University of Auckland investigations
The original NZ Steel steady state model was developed by Jinks and reported inter-
nally to Bluescope Steel in 2004 [27]. Further to this, the problem of furnace modelling
was presented to the Mathematics in Industry Study Group 2004 (MISG) and the results
published by McGuiness and Taylor [39]. The results from these investigations were used
in the construction of another steady state model by a contracting Australian company,
Sydac Pty Ltd, with the intention of eventually enabling model-based process control with
a transient model which was not pursued. Further to these investigations, mathematicians
at the University of Auckland investigated several specific problems of interest identified
in the MISG work.
2.5.1 NZ Steel Steady State Model
The initial modelling effort on the NZ Steel MCL annealing furnace was conducted by
D.Jinks of Bluescope Steel Research, beginning in 2003. The objective of this model
was to calculate strip temperature, and the power input required to meet furnace zone
temperatures in each of the 20 furnace zones under static operating conditions only, for
the full range of strip dimensions and furnace and line setups in use. This model was
completed and reported internally to Bluescope Steel in 2004 [27], and implemented by
NZ Steel production engineers and technologists in an offline capacity.
The proposed use of the steady state model was to calculate strip temperatures, and
allow the calculation of linespeed and furnace temperature setpoints. This would enable
38
optimisation of the furnace operations by finding the most efficient heating schemes to
use in the furnace for different products, and the estimation of heating required for new
products or products operating under new conditions, such as when increasing plant
linespeed. While reasonable results were obtained for some product types, this model was
unable to satisfactorily predict temperatures and power requirements for many products
[27]. This may be due to some of the simplifications and assumptions used in the model,
which are detailed here and may be improved upon in this project.
Much of the initial research and development work on this NZ Steel model is very
useful to this project, as many different factors were considered in depth by the designer.
In fact, the desired output of this project is a model similar to this steady state model,
but capable of transient calculations, with improved calculations accurate across the full
range of products and conditions possible in the furnace.
A significant simplification in this model is in the approximation of the furnace geom-

etry - the walls and strip of the furnace are approximated as infinite, parallel planes, and
the furnace is divided into one calculation element per furnace zone i.e. 20 calculation
points are used to cover the 160m furnace length. The radiation law used is poor in that
it does not take into account variation of the strip width with different products, which
has been seen in practice to affect heat transfer to the strip significantly. Instead, a cali-
bration factor (f2 ) in the radiation equation is used to match strip temperatures against
different product types.
The radiation equation used is:
2σ(w + δ). (Ts4 (i − 1) − Tz4 (i))

Qws(i) = f2 1 1 .∆x (2.20)
s
+ z
− 1
Where:
Qws = Heat transfer wall to strip
w = Strip width
δ = Strip thickness
Ts , Tz = Temperature of strip, zone
s , z = Emissivity of strip, zone
∆x = length of furnace zone
39
Note that the strip width is used to calculate the total heat transferred to the strip
only, but does not change the view factor between furnace and strip. The view factor is
assumed to be unity i.e. the strip receives all radiation emitted from the furnace zone.
Another major simplification is the heat loss modelling through the furnace walls,
which also uses a simple equation with a tuning factor (f1 ) to match operational data.
The equation used was:
p
Qwe = f1 .k.(Tz − Te ).∆x (2.21)
d
Where:
Qwe = Conductive heat loss through walls
k = Conductivity of wall refractory brick
p = Nominal wall perimeter
d = Nominal wall thickness
Te = Ambient environment temperature
These two simplifications proved problematic in the calibration of the model. The
wall heat loss calibration factor f1 in particular could not be well fit across the range of
products using a heat balance method. For some products, no value for f1 could be found
to give a correct heat balance at all, and after further examination it was decided to leave
the factor at unity i.e. the heat loss equation could not be calibrated satisfactorily. For
this reason, it will be desirable to model the furnace walls without relying on a calibration
factor to fit an approximate equation.
Fitting the calibration factor f2 for the radiative heat exchange was similarly difficult
in this model. The factor was fit separately in four sections along the furnace length,
corresponding to points where strip temperature could be more accurately measured with
a gold-cup pyrometer probe [48]. A wide spread of f2 values was found in each of these
sections, and also very large differences between the sections of up to an order of magnitude
(0.2 vs 2.0), without any obvious explanation. The calculated nominal uncertainty in strip
temperature calculation (to 95% confidence) due to this variation in calibration factor
was ±90K in heating zones 1-9, and up to ±32K in the other furnace zones. Many
reasons were suggested as possible contributions to the cause of this variation, and it was
concluded that a significantly more sophisticated representation of the physical system
40
would be required to properly calculate the radiative heat transfer [27].
One of the main improvements suggested included refining the calculation mesh to
up to 1000 elements along the furnace, rather than calculating each whole zone as one
element. Other major improvements would be to model the zone heating elements sepa-
rately to the furnace walls to allow for different element and wall heating, and to include
the thermal mass present in the hearth rolls in the calculations.
The NZ Steel steady state model also included metallurgical calculations based on
the metallurgical model developed for Port Kembla MCL6 by Renshaw [57]. This allows
calculation of strip recrystallisation at each temperature calculation point in the furnace
model, but this recrystallisation calculation is obviously dependent on the accuracy of
the strip temperature calculations as well. A similar method could be used to integrate
a metallurgical model into the furnace model to be developed in this project, but it is
also noted that the MCL6 metallurgical model is not accurate at NZ Steel due to the
differences in steel chemistry as shown in Section 2.7.
2.5.2 MISG Furnace Model
New Zealand Steel engineers presented the problem of modelling the MCL furnace at the
Mathematics in Industry Study Group (MISG) in January 2004. The work from this
study group was expanded and published by McGuinness and Taylor [39]. The objectives
of the workshop exercise were chosen by NZ Steel to be [49]:
• Develop mathematical model of non-steady state furnace conditions (applicable ap-

proximately 50% of the time).
• Investigate accuracy of the existing steady state model.
• Predict transient strip temperatures for actual production schedules with changes
in product dimension, steel grade and furnace temperature settings
• Couple the temperature model to a metallurgical model (if time allows)
The published work from MISG considers many areas of interest to this project, and
lays out ‘important mathematical and physical principles from which a more detailed
model may be constructed’ [39]. Specifically, this work develops equations for the radiative
heat transfer, strip heat transfer, furnace zone power balance, and heat flow through
41
furnace walls. It also considers time and length scales of the furnace system, the radiative
effects on thermocouples, and the steady state solutions to the equations.
The radiative heat transfer equations were developed from first principles, and ap-
plied to a furnace geometry approximated as a long box of constant cross section, with
isothermal wall surfaces. This approximate geometry is likely to be much more accurate
than the assumption of infinite parallel planes used in the NZ Steel steady state model,
as shown in Equation 2.20, as it considers the width of the strip when calculating view
factors from adjacent walls to the strip. The radiative heat transfer equation is:
2ws σ(Tw4 − Ts4 )

qws = s (1−w ) w
(2.22)
1+ w p
Where:
qws = Radiative heat transfer from walls to strip, per unit length, W/m
w = width of strip, m
p = half perimeter of walls (height + width), m
w , s = emissivity of walls and strip
Tw , Ts = Temperature of walls and strip, K
This requires that heat transfer by radiation is only between faces in the same cal-
culation segment, i.e. radiation is insignificant in the direction along the furnace length
due to the gradual temperature changes in this direction and the small view factors. This
uses the following view factors between the furnace walls and strip [39]:
FSW = 1 (strip to walls) (2.23)

FW S = w/p (walls to strip) (2.24)
FSS = 0 (strip to strip) (2.25)
FW W = 1 − w/p (walls to walls) (2.26)
This set of view factors implies that all radiation leaving the strip is intercepted by
the furnace walls (FSW ), which is a sensible assumption as the strip is entirely enclosed
by the walls, and there is no line of sight from any part of the strip to any other part
(FSS = 0). The amount of radiation leaving the walls and intercepted by the strip is
equal to the fraction of wall perimeter obscured by the strip width (FW S = wp ), and
42
the radiation leaving part of the wall and received by a different part of the wall is the
remaining fraction (FW W = 1 − wp ).
The strip is modelled using the heat equation for conduction, with an additional term
for advection of heat down the length of the furnace, corresponding to the strip speed
through the furnace. It is simplified to include heat transfer by advection only, as the
rates of conduction are negligible compared to the speed of strip movement [39]:

∂T ∂T q
ρ s Cs +v = (2.27)
∂t ∂x wh
Where:
ρs = Strip density, kg/m3
Cs = Strip heat capacity, J/kg.K
T = Strip temperature (average over thickness), K
t = Time, s
v = Strip speed, m/s
x = Distance in direction down length of furnace, m
q = Radiative heat transfer received (Eqn. 2.22)
w = Strip width, m
h = Strip thickness (gauge), m
Modelling the strip also requires knowledge of the physical properties of the steel as
they change with temperature, which have been investigated. These properties given in
Table 2.2 are curve-fitted using polynomial approximations in the model equations. Note
that these are generic values for medium carbon steel. Values for conductivity and heat
capacity better suited to the particular steel chemistries in use at NZ Steel are given in
Section 4.4.2.
T (K) 300 400 600 800 1000

k (W/mK) 60.5 56.7 48.0 39.2 30.0
Cp (J/kg.K) 434 487 559 685 1169
Table 2.2: Variation of conductivity and heat capacity of steel with temperature [25, 39]
43
Heat transfer into the furnace walls is considered as a heat balance between the heat
supplied to the electric heaters, and the heat transferred to the strip by radiation, as the
heating elements and the inner furnace walls are considered as lumped isothermal objects.
The result of the heat balance is represented by Eqn. 2.28:
P − qws
Φ= (2.28)
2p
Where:
Φ = Heat flux into wall per unit length, W/m
P = Power supplied to heat elements per unit length, W/m
qws = Heat transfer from walls to strip from Equation 2.22
p = Furnace half perimeter as in Equation 2.22
The wall temperatures are then modelled using a simple one-dimensional heat conduc-
tion equation, using boundary conditions of a specified heat flux (Φ) at the inner face, and
heat loss by natural convection at the external wall face. The one-dimensional problem
considers heat transfer in a radial direction only i.e. perpendicularly to the direction of
strip travel. A disadvantage of this approximation is that conduction in the longitudinal
direction may become important due to the high thermal mass of the furnace walls, which
is not captured in the one-dimensional model. A two-dimensional wall model would better
show the effect of the hot furnace walls resisting rapid temperature change.
Dimensional analysis was used to predict time and length scales for the furnace wall
heating, predicting that for bulk changes in furnace temperature, the furnace as a whole
would respond with a time constant of around 320 hours. However, inner furnace face
temperatures change much more rapidly in response to changes in electric heater power,
and the calculated length scale for response to changes in furnace temperature was a 20mm
depth of the furnace brick wall, with a time scale of only 40 minutes, which is comparable
to times seen in practice for the recorded thermocouple temperature to steady after a
setpoint change [39].
A steady state solution to the model equations was found by solving the differential
equations using MATLAB software, as shown in Figure 2.10. The results were found
to be similar to those predicted by the NZ Steel steady state model, although predicted
strip temperatures were slightly lower due to the increase in strip heat capacity with
44
Figure 2.10: Steady state solution for furnace wall and strip temperatures of the MISG
model [39]
temperature, which was not modelled by NZ Steel.
Another area investigated during MISG was the temperature recorded by the furnace
thermocouples, which was also further investigated by Wang [39, 78]. It is recognised that
the thermocouple temperature is affected by radiative heat transfer from all of the furnace
and strip surfaces. By approximating the furnace geometry above the strip as a rectan-
gular box with a protruding thermocouple tube, view factors were calculated between
the strip, walls, and thermocouple. It was calculated that the cooler strip temperature
had a strong effect at the bottom of the thermocouple tube, so that the thermocouple
temperature would roughly be an average of the strip and wall temperatures. Furnace
wall temperature influence increased proportionally to the distance squared, so that half
way up the thermocouple tube, the ceiling temperature had four times the weight of the
strip temperature, meaning recorded temperature was strongly affected by positioning of
the thermocouple inside the tube [39].
2.5.3 SYDAC Model
Subsequent to MISG 2004 and the NZ Steel steady state modelling work, the Australian
engineering company Sydac Pty. Ltd. were engaged to develop steady state and transient
models, with the possibility of future development to allow model-based control of the
MCL furnace. A comprehensive steady state model was developed, including a demon-
stration of the possibility of transient solutions, but the work did not progress beyond this
point. The model was written using proprietary MatrixX and Systembuild software, but
the development of the model equations and assumptions made were detailed in commer-
cial reports to NZ Steel. The particular equations developed may not be disclosed here,
45
but some of the assumptions and general findings will be discussed [68, 69, 70].
As with the previous modelling efforts, this model was designed to accomplish several
main tasks. These were: to allow determination of the most appropriate machine setup
parameters for all strip products, allow any ‘what-if’ studies to be simulated, allow de-
termination of alternative settings in the case of failed equipment (heating elements in
particular), and to predict strip temperatures along the full length of the MCL furnace.
A metallurgical model was not included in the SYDAC modelling effort.
One major difference with the SYDAC model is that it was designed to be transient in
nature from the beginning, which allows both transient and steady state simulations using
the same model [70]. This means a steady state solution is found by solving a transient
equation system until it converges. The time required to solve this depends on the quality
of initial conditions, but the time restriction imposed by NZ Steel of less than 10 seconds
required many simplifications of the model equation system.
The radiation equations in the model were simplified to meet this time demand, and it
was assumed that radiative heat transfer only occurs between the strip and furnace walls
in the same segment - the furnace was divided into calculation segments (mesh points)
1.83m long, being the distance between hearth rolls. It was found that radiative heat
transfer from a strip segment (the ith segment) to wall segments adjacent to the opposite
segment (the i − 1 and i + 1 segments) was only 10% of the heat transfer between strip
and wall in the same (ith ) segment. This is an important assumption, as reducing the
complexity of the view factors between furnace and strip segments dramatically reduces
the number of calculations required at each time step for the radiative heat transfer
calculations.
The SYDAC model followed other investigations, and does not include heat transfer
due to convection of the furnace atmosphere. This can be included using simple heat
transfer coefficient approximations on the furnace and strip heat transfer faces, but is
ignored due to solution time requirements. Other assumptions agreeing with the MISG
assumptions were that conduction in the strip body in all directions can be ignored, as
the temperature gradients are small in the thickness and width directions, and the main
source for propagation of temperature in the long direction is the movement of the strip
down the furnace [70, 39].
The modelling of the furnace walls showed that an irregular mesh spacing was conve-
nient, to allow the temperature of the inner face to change rapidly in response to radiative
46
heat transfer using a small mesh spacing, but the low thermal conductivity of the walls
and slow rate of temperature change allows a much coarser mesh spacing away from the
inner face. A spacing of 1mm was used at the inner face, and 20cm adjacent to the outer
face.
The SYDAC model also includes simulation of the cooling tubes in the cooling sections
of the furnace. It was found that the cooling tubes have very little direct effect on the
temperature of the strip, and remove energy mostly from the furnace zone (walls and gas
atmosphere). This leads to a slow temperature response of the strip using cooling tubes.
The model also recognised the difficulty in matching furnace thermocouple tempera-
tures with any calculated value, as they are affected by furnace zone, atmosphere, and
strip temperatures. The thermocouple temperatures were predicted using a weighted av-
erage of the wall and strip temperatures, and it was found that only weights of 0.88× wall
temperature + 0.12× strip temperature produced feasible model results [70].
The model was developed to a point capable of reasonable prediction of steady state
strip and furnace temperatures, with improved complexity over the original NZ Steel
model as it includes the effect of cooling tubes and gas jet coolers. The transient model
was not developed further, and its accuracy has not been substantiated.
2.5.4 Strip and Thermocouple Temperatures
The distribution of furnace temperatures due to radiation were investigated by Wang

and Taylor [73] at the University of Auckland. These are mathematical investigations of
simplified furnace geometries, involving calculation of view factors between all surfaces,
and are able to make predictions of temperature variation across the strip width, heat
flux distribution on the furnace walls, and temperature variation along the length of the
thermocouple probe.
Two results from these investigations are of specific interest - the strip and thermo-
couple temperature variations. By simplifying the system as having isothermal furnace
walls at 900◦ C, and a base strip temperature of 500◦ C, it was found that the strip edge
temperature was 7◦ C higher than at the centre. The temperature distribution result of
this calculation is shown in Figure 2.11. It should be noted that actual strip temperatures
may be even hotter at the edges, as the rolling process causes the strip edges to be thinner
than at the centre.
47
Figure 2.11: Predicted temperature variation across the strip width for fixed strip centre
temperature of 500◦ C and wall temperature 900◦ C [73]
Experience of MCL technologists suggests that the strip temperature difference pre-
dicted is certainly present, and may be underestimated here. Several strip types experience
problems of ‘wavy edges’ due to overheating of the strip edges, despite strip temperature
measured by the pyrometer being low enough not to cause recrystallisation in the bulk of
the strip. To prevent this wavy edge defect, strip target temperature had to be reduced
by as much as 40 − 50◦ C for some products.
Figure 2.12: Predicted temperature variation along thermocouple length for fixed strip
centre temperature of 500◦ C and wall temperature 950◦ C [78]
The temperature variation along the length of the thermocouple tubes was investi-
gated by a student of Wang [78], also using approximated geometry of a square furnace
cavity with a rectangular thermocouple protruding into the cavity to a distance of 20cm
48
from the strip. The temperatures of furnace walls and strip were fixed at 950◦ C and 500◦ C
respectively. Under these conditions, it was found that the bottom end of the thermo-
couple could be cooled by the strip underneath by up to 20◦ C, as the temperature of the
lower tip of the thermocouple was 930◦ C. The temperature change is plotted in Figure
2.12, showing that the whole thermocouple is cooled slightly below the wall temperature,
with the cooling effect stronger closer to the strip. The temperature increases to the wall
temperature of 950◦ C in a very small distance at the top of the thermocouple, as the
internal heat transfer is much slower than the rate of heat transfer by radiation.
2.5.5 Requirements For New Model
While several investigations into MCL modelling have been carried out previously as
described in Sections 2.5.1 to 2.5.4, they do not fulfil the requirements for NZ Steel. To
improve on prior investigations, a new solution must be transient, predictive and also
improve understanding of the spatial effects of the radiation in the furnace cavity.
A new model must be able to calculate the transient temperature behaviour of the
furnace and strip, as transient temperature changes are a major cause of incorrect heat
treatment and hence reject production. The MCL furnace is a complex system with many
interacting, non-linear, time-varying conditions, and effectively modelling these aspects is
vital to ensure a transient temperature solution is both accurate and feasible in terms of
calculation power and time.
The model must also predict furnace and strip temperatures with sufficient accuracy
and speed of solution to allow it to be used for optimisation or furnace control purposes.
This will enable selection of furnace control measures in response to any changes in op-
erating or input conditions, such as strip dimension change, partial heating failure etc,
which maintain the correct heat treatment.
To improve upon the calculation methods used in prior investigations, the new model
should seek to calculate and understand the spatial effects of radiative heat transfer in the
furnace cavity in three dimensions. This includes the effect on the measured thermocouple
temperatures compared to actual furnace temperatures, and the temperature profile of
the strip including increased heating at the strip edges.
49
2.6 Metallurgical Calculation Methods
While there are calculation methods available for the stress relief and grain growth stages
of the annealing process, calculation of the recrystallisation stage is most important when
considering the MCL annealing furnace. It is a guideline for heat treatment that hard
iron products should not show any recrystallisation of the microstructure, or at most 5%
recrystallisation. Soft iron products should be fully recrystallised, or at least above 95%
recrystallisation as it has been measured that as little as 10% unrecrystallised material
can result in a significant loss of ductility [57].
Actual measurement of the recrystallised fraction of the steel is difficult and time con-
suming, and is not practical to conduct for each coil of steel in the production environment.
Measurement of recrystallisation requires examination of the steel microstructure to de-
termine the fraction that has been recrystallised (Figure 2.4(b))and the fraction that still
appears as cold worked elongated grains (Figure 2.4(a)). This is commonly done by man-
ual observation under optical microscope, such as using the ‘point counting’ technique
using an eye-piece grid to count the number of grains on the grid intersections that ap-
pear recrystallised. This is a subjective measurement, and is likely to be accurate to only
around ±20%.
Rather than recrystallisation measurements, NZS test the steel hardness and yield
strength to ensure product quality, which is more useful than recrystallisation measure-
ments, as the product is sold in specified yield strength ranges, with the other properties
of hardness, coating weight, elongation percentage etc tightly controlled. Recrystallisa-
tion however is the physical process by which these properties are affected during heat
treatment, and so it is this process which can be modelled and calculated to indicate the
changes likely in the physical properties.
There are a number of different approaches that can be taken when modelling recrys-
tallisation processes [63]:
• State Variable Methods
• Statistical Methods
• Empirical Methods
These methods provide very different ways of calculating the annealing behaviour of
a metal during heat treatment, and utilise different measurements and variables of the
50
heat treating process. Some are coupled to process histories, and others simulate the
underlying physical processes occurring.
2.6.1 Kinetics of Recrystallisation
Theory of the kinetics of recrystallisation describing the nucleation and growth processes
is described by the JMAK model, developed independently by Johnson and Mehl [28],
Avrami [4] and Kolmogorov [32]. The general form of this equation is written as [11, 24,
57, 64]:
Xv = 1 − exp(−Btn ) (2.29)
Where:
Xv = Volume fraction of material recrystallised
B = Kinetic parameter related to growth and nucleation rates
t = time
n = Avrami exponent
The Avrami exponent is related to the growth mechanism in the material, where:
1 < n < 2 = One-dimensional (wire) recrystallisation

2 < n < 3 = Two-dimensional (plate) recrystallisation
3 < n < 4 = Three-dimensional (lump) recrystallisation
The Avrami exponent is a function of the metal alloy and physical system, and can be
determined through experiment by plotting ln[ln(1/(1 − Xv ))] vs ln t, where n is the
slope of the line as shown in figure 2.13 [24]. For most common materials, the Avrami
exponent is in the range 1 < n < 2 [11].
The kinetic parameter, B, is dependent on a number of factors, including the temper-

ature at which recrystallisation occurs, as well as the chemical composition and strain of
the metal, as described in the ‘laws of recrystallisation’ in Section 2.2.2.1. The kinetic
parameter determines the start point i.e. the onset time of recrystallisation in Figure
2.13.
The JMAK model assumes that nucleation sites in the metal are randomly distributed,
and so can be unsuitable in cases of inhomogeneous recrystallisation. This can occur in
51
Figure 2.13: Annealing kinetics of an Iron crystal deformed at 70% and annealed at
various temperatures [24]
metals with very large initial grain sizes, as the nucleation sites will be preferentially
situated along the grain boundaries, leaving large spaces without nucleation sites. Metals
with smaller grain size and higher strain will have high density of grain boundaries, and
hence many likely sites for nucleation throughout the metal.
The JMAK model is commonly applied using a normalised time, t, as a fraction of

the calculated time required for recrystallisation, typically t0.5 or t0.95 , being the time
required for either 50% or 95% recrystallisation respectively. In this case, the JMAK
equation appears as in equation 2.30 [57, 62]:
n
t
Xv = 1 − exp −B (2.30)
tx
The time for 50% or 95% recrystallisation can be calculated by a number of meth-
ods detailed later, but is commonly done using a form of the Arrhenius Equation for
heat-activated processes. Equation constants are fitted empirically, based on a series of
experiments covering the range of varying strip properties and processing conditions, and
have been extensively studied for C-Mn steels and HSLA steels. An equation such as
given in equation 2.31 is commonly used [24, 62]:

Qrex
tx = AD0s −n Z −k exp (2.31)
RT
52
Where:
Qrex = Activation energy for recrystallisation
R = Gas Constant
T = Temperature
D0 = Initial grain size
= Strain (Cold Reduction)
Z = Zener-Hollomon parameter related to strain rate
A, s, k, n = Empirical constants
2.6.2 State Variable Methods
State variable methods for predicting microstructure evolution during metal processing
or heat treatment involve selecting a specific factor that can be correlated to either mi-
crostructure or physical properties of the metal, and predicting changes during processing
by differential equation laws.
A ‘surrogate’ state variable method uses state variables that are indicative of the
underlying microstructural change, rather than a physically measurable quality [63]. An
example is predicting the flow stress of a metal with differential equation laws using
the instantaneous strain values [61]. This can be used to predict the softening of a
metal during rolling, using mathematical equations fit to test data, but does not have
any physical basis. This offers no insight into the physical processes occurring, and is
dependent on fitting the equations to specific metal or alloy samples [63].
Alternatively, ‘physically based’ state variable methods can be used, where the vari-
able has direct physical meaning during the process. Dislocation densities in the metal
matrix can be measured using electron microscopy, and differential equations used to fol-
low the changes over complex process histories. An advantage of this is that the model
and measurements can be carried explicitly from one processing stage to the next. Auto-
mated electron backscatter diffraction (EBSD) techniques allow measurement with high
precision, but are time consuming and costly.
53
2.6.3 Statistical Methods
Statistical methods can be used to relate processing conditions to metallurgical outcomes

without consideration of the physical processes occurring during annealing. There are
many types of advanced statistical methods useful for this type of analysis such as artificial
neural networks or methods to perform nonlinear regression on large and complex data
sets of process variables [63].
A large data set containing all relevant product and processing information is created,
which for a continuous galvanising line includes furnace parameters such as strip entry
temperature, strip speed and the main steel chemistry components such as weight percent
of Carbon, Manganese, Silicon, Aluminium and others. This is combined with the output
parameters such as yield strength, tensile strength and elongation. Neural Network models
have been used to predict results of continuous heat treatment, with accuracy improving
with the size and quality of the data set used to train the model [51].
One disadvantage of statistical methods is they can be distrusted by engineers and

metallurgists in particular, as they have no basis in metallurgical theory, and provide no
insights into the process behaviours [63]. These methods have been shown to be useful
however, and can be used to find trends and relationships in complex data sets.
2.6.4 Empirical Methods
Empirical methods for predicting recrystallisation of steels are widespread, and established
empirical approaches typically use variations of the same equations with a calculated
activation energy of recrystallisation and measured values such as grain size and applied
strain. These methods do not consider the evolving microstructure, and rely on a series of
laboratory or process experiments to empirically fit the equation parameters [57, 62, 63].
Much of the work in this area focuses on modelling dynamic recrystallisation during hot
rolling of steel strip, but the equations have also been used to model static recrystallisation
between roll passes, or during post-rolling heat treatments, such as in continuous annealing
furnaces [16, 57, 62, 42, 44].
The established approach for these empirical methods uses two main equations - one to
calculate tx , the time required for x% recrystallisation of the metal under given processing
conditions, and one to calculate Xv , the percentage recrystallised. Other equations may
54
also be used to calculate an approximate grain size during recrystallisation [8, 24, 40, 57,
62, 72].
The accuracy of empirical methods can be very good, but requires a large range of
tests or experiments to be performed that cover the expected ranges of strain, strain rate,
grain size, temperature and metal chemistry to fit the empirical constants. This approach
is widely used for flat rolled steels, but has been shown to be inaccurate for other deformed
metals with more complex strain paths, such as aluminium alloys [63].
2.6.5 Furnace Model Calculations
Extensive work has been done in the area of steel recrystallisation calculations during
hot rolling, which includes both dynamic recrystallisation during rolling, and the static
recrystallisation between roll passes. This work also covers a range of steel types, including
C-Mn steels similar to those used at NZ Steel. The majority of these recrystallisation
calculations are of the empirically fitted type, using the JMAK model (2.30) and Arrhenius
equation (2.31), and many encompass similar values of strain and grain size as NZ Steel
also.
A metallurgical model was constructed by Bluescope Steel for ‘Metal Coating Line 6’
at the Port Kembla works, which uses the same basic equations derived from hot rolling
recrystallisation, and is empirically fitted to laboratory test data [57]. This model has
proven to be serviceable, and is a useful start to construct a model for NZ Steel, although
there are several differences in processing conditions and steel chemistry which need to
be considered. A range of recrystallisation tests have been conducted using NZ Steel
steel chemistries, which will be used to fit a new metallurgical model using the empirical
methods described in section 2.6.4.
55
2.7 Metallurgical Model for NZ Steel
This section investigates the specifics of empirical metallurgical models that have been
developed previously, and their suitability to apply to the steel type and process in use
at NZ Steel. The characteristics of other empirical models and the type of equations
used are shown in Section 2.7.1. Development of a model suitable for the MCL furnace
is discussed in Section 2.7.2, and a description of recrystallisation trials conducted by NZ
Steel is given in Section 2.7.3.
2.7.1 Model Basis
The basis for these types of empirical model is work by Sellars and others [6, 8, 62],
which specifically cover plate rolling in a hot strip mill using C-Mn steels, and is also
applicable to the static recrystallisation occurring after deformation is complete. As
described in Section 2.2.2, static recrystallisation is a function of time and temperature
of heat treatment of the deformed structure, and is affected by the processing conditions
such as prestrain before annealing.
The original Sellars and Whiteman model uses two main equations - Equation 2.32 is
comparable to the Arrhenius equation (Equation 2.31) and Equation 2.33 is comparable
to the JMAK equation (Equation 2.30) [62].
Qrex
tx = A D0s −n Z −k exp (2.32)
RT
Where :
tx = time required for x% recrystallisation at given temperature
Qrex = activation energy for recrystallisation, J/mol
= strain (cold reduction)
D0 = initial grain size, µm
Z = Zener-Holloman parameter
A, s, n, k = Empirical constants
3
In this work, the empirical constants took on the values s = 2, n = 4, k = 8
, A =
3.54 × 10−21 s0.625 µm−2 [62].
56
n
t
Xv = 1 − exp −B (2.33)
tx
B = −ln(1 − x) (2.34)
Where :
x = Fraction of recrystallisation e.g. 0.5, 0.95
tx = time required for ‘x’ recrystallisation at given temperature
t = actual time at given temperature
n = Avrami Exponent
For x = 0.5, B = 0.693, and for x = 0.95, B = 2.996.
The Zener-Holloman parameter is a function of strain rate and temperature during hot
rolling. These equations agree well with the theory of recrystallisation in that the time for
recrystallisation, tx , decreases with increasing prestrain, strain rate, and temperature, and
is affected by the steel chemistry in the form of the activation energy, Qrex i.e. increasing
the strain, strain rate or temperature will increase the rate of recrystallisation due to
increased dislocation density in the metal.
The fraction of recrystallised material, Xv , is calculated using the Avrami (JMAK)

equation, with a constant Avrami Exponent of n = 2. This value is taken from observa-
tions of recrystallisation in both austenitic stainless steel and α-iron, and is equivalent to
2-D or plate growth in the metal. This constant value was found to fit the general scatter
of empirical testing, but can be less accurate for very low strain amounts and very low
recrystallised fractions [6].
There are many variations of these forms of equations derived by other researchers, but
all follow the same method, using a calculation for the x-fraction recrystallisation time tx ,
and a formula for the volume fraction recrystallised, Xv , relating the recrystallisation time
to the actual time experienced at temperature by the metal. The empirical constants and
exponents derived by other researchers are often similar, but dependent on the particular
conditions used for the experiments.
The empirical constant of most interest is the activation energy for recrystallisation,
Qrex , as this is strongly dependent on the chemical composition of the steel [21, 40, 44],
and it is necessary to find values also applicable to the range of steel chemistries in use
57
at NZ Steel. This is similar to the activation energy for grain boundary self diffusion in
iron, and increased by the chemical additions to the steel. Carbon is seen as the only
alloying element to lower the activation energy and accelerate the recrystallisation kinetics,
whereas all other typical alloying elements increase Qrex and slow the recrystallisation [44].
The value of Qrex can vary markedly with different steel types and compositions as shown
in Table 2.3.
Steel Type Activation Energy kJ/mol

Rimmed 122 - 143
Al Killed 157 - 305
Ti Stabilised 312 - 475
Nb Stabilised 270 - 370
HSLA 300 - 1300
Table 2.3: Activation energy of recrystallisation for various steel types [57]
The steel chemistries used at NZ Steel fit in the ‘Al Killed’ category, but the activation
energy for these steels must be well defined, in order to show the effect of variations in
steel chemistry across the different grades of steel used, and this is one of the main tasks
in the construction of the metallurgical model.
These model equations are similar to those used by Renshaw [57] in the Port Kembla
MCL6 model, with empirically fitted constants and exponents using a series of recrystalli-
sation temperature trials. One specific area in which the MCL6 model does not suit NZ
Steel is in the differences in steel chemistry, as the NZ steel contains different amounts
of Vanadium and Nitrogen in particular, which are expected to increase the activation
energy and delay recrystallisation. This is observed in practice, as the MCL6 model
under-predicts recrystallisation temperatures when used at NZ Steel.
2.7.2 Metallurgical Model Development
This section discusses the equations and individual factors required to construct a met-
allurgical model suitable for NZ Steel to predict recrystallisation in the MCL furnace.
Development of a suitable metallurgical model involves the formulas described in Section
2.7.1, and investigation of empirical constants used for similar systems. A limited series of
laboratory-scale heat treatment experiments previously conducted by NZ Steel are avail-
able to fit the model to, by variation of the empirical constants. The main factors of the
metallurgical model that are investigated are:
58
• Form of Model Equations
• Activation energy (steel chemistry)
• Grain size
• Cold Reduction
• Avrami Exponent
• Strain Rate
The specific form of the model equations to be used is investigated in Section 2.7.2.1.
The activation energy of the steel is examined in Section 2.7.2.2, and the grain size of the
steel in Section 2.7.2.3. Cold reduction of the MCL steel is given in Section 2.7.2.4, the
strain rate in Section 2.7.2.5, and the Avrami exponent in the equations found in Section
2.7.2.6.
2.7.2.1 Form of Model Equations
Considerable developmental work has been undertaken by various authors since the equa-
tions proposed by Sellars et al [6, 62] to calculate the recrystallisation time and amount
of recrystallisation of a metal during heat treatment. The Zener-Holloman parameter is
commonly excluded, and replaced with the strain rate during rolling for calculations of dy-
namic recrystallisation. It is removed completely for static recrystallisation calculations,
or may become included in the empirical constants in the formula [71].
Medina et al [40, 42, 44] have developed these equations specifically to refer to static
recrystallisation of hot deformed steels, using similar C-Mn chemistries as well as mi-
croalloyed steels containing Ti, Nb and V. The ‘improved model’ [44] uses the following
development of the Sellars equations for 50% recrystallisation i.e. x = 0.5.
Qrex
t0.5 = A p ˙q Ds exp (2.35)
RT
n
t
Xv = 1 − exp −0.693 (2.36)
t0.5
where A, Q, p, q, s and n are fitted empirically based on a series of laboratory tests.

Some of these empirical factors are constant, whereas others are variables dependent on
59
temperature, grain size or activation energy. The values for strain and strain rate still
apply during static recrystallisation, and refer to the prior strain before recrystallisation.
Increasing the strain or strain rate increases the dislocation density in the steel and
increases the number of sites for the nucleation of new grains, although this has less effect
than the grain size [44].
These equations are very useful to apply to the NZS MCL annealing furnace, as they
contain factors representing the variation in steel types and processing conditions that
occur in the coating line. For example, the cold reduction (strain) of strips on the MCL
changes with steel grade and gauge, with the heavy gauge and high strength strips having
less cold reduction. Lower cold reduction reduces dislocation density, meaning a higher
temperature or longer time is required to achieve recrystallisation. These equations can
be used to predict the effect of cold reduction on the time/temperature profile required
to recrystallise the strip.
The Bluescope Steel model on MCL6 at Port Kembla uses a different calculation, but
considering many of the same factors as the model by Sellars et al [57]:

1 3.77 Qrex
ln = 49.54 + 0.23 − (2.37)
t0.95 RT
Xv = 1 − exp −3(PQ )0.733

(2.38)
where PQ is a factor for heat treatment under non-isothermal conditions, and equal to 1
when the actual equivalent time reaches the equivalent time for 95% recrystallisation.
These equations are different to the Medina et al ‘improved model’, as different meth-
ods were used for statistical analysis of the empirical constants. The Medina equations
are more useful for this application however, as they also encompass the effect of changing
steel grain size. NZ Steel use two different feed types to the cold rolling process imme-
diately before the metal coating line, being either hot rolled or warm rolled. The warm
rolling process results in significantly larger grain sizes than hot rolling, and hence the
warm rolled feed is more difficult to recrystallise in the annealing furnace, due to fewer
nucleation sites for recrystallisation [44, 64]. This effect should be captured in a new NZS
model, which is not possible using the MCL6 model equations [57].
The MCL6 model is fitted only to those steel types and situations experienced at the
Port Kembla plant, and is not likely to fit well when extrapolated to situations outside of
these operating conditions and material properties. The basic metallurgical equations to
60
be used in the MCL annealing furnace model will therefore follow the Medina ‘improved
model’, initially starting with the already developed empirical constants, and then adjust-
ing these constants to better fit the steel types used at NZ Steel, based on recrystallisation
experiments conducted by plant technologists in Section 2.7.3.
2.7.2.2 Activation Energy
It has been widely shown that activation energy of recrystallisation, Qrex , is strongly
dependent on steel chemistry, and so will vary depending on the type of steel used. NZ
Steel uses similar chemistry grades as in the MCL6 model with several variations, and is
similar to other low-carbon, Al-killed, C-Mn steels.
Many researchers have previously used constant values of Qrex for metallurgical calcu-
lations of a single steel chemistry, such as Qrex = 230, 000 J/mol for a C-Mn steel [6, 40].
Recent work has since progressed to include a variable Qrex calculated using weight per-
centage of varying alloying elements. Much work on this has been conducted by Medina
et al [41, 42, 43, 44], creating a comprehensive calculation including contributions from
many common alloying elements as follows:
Qrex = 148636 − 71981[C] + 56537[Si] + 21180[M n]

+ 121243[M o] + 64469[V ] + 109731[N b]0.15 J/mol. (2.39)
where [C] is the weight percentage of carbon etc. This correlation is useful, as it includes
most of the chemical elements that vary between NZ Steel grades used on the Metal Coat-
ing Line. It does however exclude important elements aluminium and nitrogen present in
NZS steel.
Titanium can also have a large effect on the activation energy of the steel, but is no
longer considered in the most recent work, as the actual Ti concentration is very difficult to
quantify [44]. An earlier model included Ti in the form of an addition to the Qrex formula
of +76830[T i]0.213 , which may give an indication of the rough magnitude of the effect
from Ti. This effect can be very large when Titanium is present in quantities typically
used in Ti-Microalloyed steels, but the NZ Steel chemistries have very low percentages of
Titanium, usually under 0.001 wt%.
There are many forms of correlation for Qrex proposed which include different alloying
elements or combinations of elements [42, 43, 67], but none appear to include Nitrogen,
61
and only the MCL6 model includes Aluminium. As these types of empirical correlation
are developed using statistical regression approaches, including the full range of alloying
elements present in the steel would require a very large sample size encompassing statisti-
cally significant variation in each of these elements. The sample steels used in the Medina
models do not include either a large enough amount of Al and N, or enough variation of
these to be statistically significant in the Qrex correlation.
The correlation for activation energy proposed by Renshaw in the MCL6 model is
given by Equation 2.40 [57].
Qrex = 2311[M n]−1.306 − 71277[Al]0.536 − 407396 J/mol. (2.40)
In this model, it was acknowledged that many other elements including C, N, Si, P
and Ti can have significant effect on the recrystallisation kinetics, but the variation in the
samples used in this testing was not sufficient to quantify statistically. This shows that
the particular correlation developed for any metallurgical model is strongly dependent on
the quality and scope of empirical testing. It was acknowledged that calculations from
this model would only be accurate over the range of chemistries used to construct the
model as shown in Table 2.4.
C N Mn Al P Si Ni Cr Cu
min .03 .001 .19 .008 .01 0 0 0 0
max .15 .006 .75 .075 .025 .02 .025 .035 .03
Table 2.4: Valid composition range of Bluescope Steel metallurgical model [57]
This is compared with the range of steel compositions used at NZ Steel on the metal
coating line. The MCL uses four distinct chemistry grades (100, 110, 210, 310). Table
2.5 lists the actual measured minimum and maximum elemental composition across any
of these grades over the 11 months January to November 2007.
C N Mn Al P Si Ni Cr Cu
min .026 .002 .14 .02 .004 .001 .015 .001 .01
max .194 .0094 .7 .08 .029 .02 .022 .045 .06
Table 2.5: Experienced composition range of NZ Steel MCL chemistry grades
It can be seen that there is much overlap in the ranges, but the grades used at NZ
Steel can be outside the valid range of the Renshaw model for the amount of carbon,
nitrogen, manganese and aluminium in particular, for either the maximum and minimum
62
concentrations. These elements are all important to the development of the model, as
they have relatively high concentrations compared to other elements, high variability
across the range of steel grades used, and have all been shown to have strong effect on
the recrystallisation kinetics of steel [41, 44].
The two elements with the highest concentrations are carbon and manganese, yet the
Renshaw model was not able to statistically separate these elements due to the interaction
between C and Mn, forming C-Mn dipoles [24, 57]. This could be an important factor,
as carbon is shown to reduce the recrystallisation temperature, as opposed to all other
elements, which retard recrystallisation and increase the recrystallisation temperature.
carbon in particular can be well outside the valid range of the Renshaw model. In the
case of grade 310 steel, the measured concentration range over 11 months was 0.154 - 0.19
wt%, which is entirely outside the valid range.
The steels used at NZS also contain other elements in low concentrations that can
have a strong effect on the recrystallisation temperature, being vanadium, titanium and
niobium. Vanadium in particular is important, as it has the largest concentration of these
elements of up to 0.011 wt%, whereas Nb and Ti are both under 0.001 wt%. Vanadium
is present in large concentration in the ironsand feed material used at NZ Steel, but is
removed during the steel making process, and as such is only present in small amounts in
the final steel chemistry.
The model proposed by Medina et al. considers vanadium specifically in the C-Mn
steel calculations. Nb is also considered, although the factor used is empirically fitted for
microalloyed steels containing much higher concentrations of Nb, in the range 0.007 - 0.093
wt% [44]. A previous model also considered Ti in higher concentrations in microalloyed
steels in the range 0.02 - 0.08 wt%, well above the 0.001% usually found in MCL grade
steels, but was removed from later models.
The Medina model is used as a basis for the calculation of activation energy for re-
crystallisation (Qrex ) in this investigation at it better covers the range of elements present
in the NZS MCL steel, and specifically covers Carbon concentration, which has a strong
effect on recrystallisation, and opposite to the other alloying elements.
The effect of nitrogen on the recrystallisation temperature of the steel is a known issue
at NZ Steel, as there can be significant variation in N content, which has been seen to
cause problems in many recent cases. There are currently no known studies quantita-
tively calculating the effect of N content on the activation energy for recrystallisation.
63
Other studies have examined the effects of specific elements on recrystallisation kinetics
qualitatively [38], but authors frequently generalise models for the same groups of steels
without specific calculation based on chemical composition [41].
Steels in present studies have not had enough content or variation of nitrogen to
quantify this exactly, so useful future work would be to conduct a series of recrystallisation
temperature tests specifically to study the effect of nitrogen, but this would require careful
control of the elemental composition of the steel to ensure the effect was not masked by
variation of other elements, and is likely to be difficult and time consuming.
2.7.2.3 Grain Size
Grain size is known to have a strong effect on the recrystallisation of steel, with smaller
grain sizes accelerating recrystallisation, as the grain boundaries serve as preferential
nucleation sites for new grains [24, 62, 64]. Most empirical calculation methods of recrys-
tallisation therefore include a grain boundary term to calculate tx , the time required for
x% recrystallisation.
The Bluescope Steel MCL6 model does not include a term for grain size, but specifically
mentions that the grain size variation of the samples used was not known, and this effect
contributed to the model error [57]. Recrystallisation calculations for use at NZ Steel
should certainly include the effect of grain size variation however, due to the different
feed types used on the Metal Coating Line.
Feed coils supplied to the MCL may be of either the hot rolled or warm rolled type.
Most products have traditionally been hot rolled, but a significant amount of warm rolled
feed is now utilised to take advantage of the different microstructural properties caused by
the different rolling process [5]. The warm rolling process allows less grain refinement of
the microstructure during rolling, resulting in a larger grain structure as shown in Figure
2.14.
The hot rolled grain structure has an approximate grain size of ASTM size 10, whereas
the warm rolled is ASTM grain size 8. These grain sizes correspond to 11.2µm and 22.5µm
average grain diameter respectively. These grain sizes are only approximate, and do not
represent the range of diameters present in any particular sample, or the distribution of
sizes, but do give an indication that the warm rolled steel grains are approximately twice
the diameter of the hot rolled steel grains. A more accurate indication of the range and
64
(a) Hot Rolled (b) Warm Rolled
Figure 2.14: Typical grain size difference in hot rolled vs warm rolled microstructure,
200x magnification [5]
distribution of grain sizes is difficult, as neither NZ Steel nor any of the Australian steel
mills owned by Bluescope Steel undertake regular grain size measurements. It is possible
to measure the size more accurately using mean intercept length measurements, but these
are only done by these steel mills during product complaint investigations, meaning there
is no history of typical grain sizes or size variation for normal product steels.
Empirical models of static recrystallisation usually include a grain size term (in µm)
with an exponent empirically fitted. The Medina ‘improved model’ used as the basis
for the metallurgical model in this project includes the term Ds in the equation for tx
(equation 2.35). Laboratory experiments gave the value of the exponent s = 0.996 ≈ 1
in the original model, and later refined to s = 1.09 ≈ 1 in the improved model [42, 44],
which were empirically fit to a group of C-Mn steels similar to those used at NZ Steel.
Other authors have found different relationships for the grain size, often of the form D2
i.e. s = 2 [6, 62, 71].
2.7.2.4 Cold Reduction
Cold reduction is the thickness reduction applied to the hot rolled strip during cold rolling.
Hot rolled strips at NZ Steel are between 2.0 and 4.5mm for strip products processed on
the metal coating line, and are cold rolled to a final thickness between 0.3 and 2.25mm
over several passes, giving the strips on the MCL cold reduction (engineering strain, )
between 0.39 < < 0.86. The thickness of hot rolled strip to be used is specified for each
final gauge thickness and required strength, allowing high strength products to receive less
65
cold reduction than low strength strips, imparting good properties for processing of the
high strength product, which may otherwise be prone to cracking if high cold reduction
is applied.
Increasing cold reduction has an effect on the recrystallisation temperature of a steel

as given in Section 2.2, as the cold reduction changes the dislocation density. Higher
cold reduction increases dislocation density in the cold rolled strip, creating more sites
for nucleation of new grains and accelerating the recrystallisation process [64]. Most
metallurgical calculations of recrystallisation include the effect of cold reduction (strain)
with a term in the tx equation (equation 2.35), similar to the factor for grain size.
The Medina ‘improved model’ equation for recrystallisation, Equation 2.35, includes
the term p , where is the cold reduction, and p is an empirically fit exponent. The value of
the constant was calculated over a range of steel types, and was between −2.3 < p < −1.5.
The variation of this exponent was found to be dependent on the grain size, and the final
model used the correlation p = −4.3D−0.169 [44].
Other authors find various values of the strain exponent, p, but all are of similar
magnitude. Fernandez et al find variation in the exponent from −4 to −2 for microalloyed
steels [16] and Tang et al [71] use a fixed value of p = −2.5 for C-Mn steels. Application of
the correlation used in the ‘improved model’ generates values of the exponent of p = −2.9
for 10µm grain diameter hot rolled steels, and p = −2.6 for 20µm grain size warm rolled
steels, which are both inside the range of likely values proposed by other authors.
2.7.2.5 Strain Rate
The strain rate during rolling can be important during dynamic recrystallisation calcula-
tions, but often omitted from the recrystallisation equations during static recrystallisation,
such as in the NZ Steel MCL furnace [71], and was not included in the model developed
for Bluescope Steel MCL6 [57].
Cold rolling mills typically impart strain rates of up to 1000s−1 [58]. This factor may
be included in the overall model empirical factor, A, if required upon examination of the
NZ Steel recrystallisation experiments.
66
2.7.2.6 Avrami Exponent
As discussed in Section 2.6.1, the ‘Avrami Exponent’, n, is the exponent in the JMAK
model (Equations 2.29, 2.30) related to the metal alloy system, which is experimentally
determined in empirical models, and usually between 1 > n > 2 for most common ma-
terials [11]. The early work on recrystallisation of steel during hot rolling by Sellars [62]
used a constant value of n = 2, using observations on both austenitic stainless steel and
α-iron, which represents plate-like or 2-Dimensional growth [4].
This exponent can be fitted to recrystallisation temperature experiments conducted

by NZ Steel, but it is worthwhile to examine the values for n found in other experiments.
Several researchers have found n to be a function of temperature - in one study of NiAl
intermetallics, n was found to decrease from 1.8 at 610◦ C to 1.0 at 725◦ C [21], which
has also been used to effectively fit steel recrystallisation behaviour. This indicates that
the dimensionality of recrystallisation may decrease at higher temperatures i.e. as recrys-
tallisation rate increases at elevated temperature, the grains tend to grow in ‘acicular’ or
needle shapes [4], and more spherical or lineal shapes at low temperatures.
The Medina ‘Improved Model’ also found a variation in n due to temperature [44],
however these experiments found the reverse relationship to that above, where n decreased
at lower temperatures. The value found for n varied between 0.62 < n < 1.24. One reason
these values may be lower is the much larger grain size (100 − 200µm) of the steel in these
experiments. Fernandez et al [16] examined this exponent in relation to grain size in
microalloyed steels, and found a value approaching n = 2 for small grain sizes of 20µm
diameter, similar to the grain size of strip in the NZ Steel MCL. Other authors also found
values of either n = 1 for C-Mn steels, or n = 2 for low-alloy steels [16].
Due to the variation in this exponent found amongst literature for different materials
and conditions, it appears that examination of NZ Steel recrystallisation experiments is
the best method to obtain a value for n. This exponent may depend not only on the steel
chemistry used, but also the microstructural properties in the steel strip, and the size,
shape and processing temperature of the strip in the furnace. Most other experiments have
found values between 1 < n < 2, as expected, and so a value fit to NZ Steel experiments
it is unlikely to be outside this range.
67
2.7.3 Recrystallisation Trials
In 2003, technologists at NZ Steel performed a series of recrystallisation experiments

using production grade strip, which may be used to fit a new metallurgical model to
the steel types used in the MCL furnace. In these trials, steel samples cut from cold
rolled coils were heated in a laboratory scale furnace to a range of target temperatures
to find the temperatures at which softening and recrystallisation started and stopped. In
each trial, the steel was rapidly heated from room temperature to the target temperature
in 20-25 seconds, held at temperature for 20-25 seconds and rapidly cooled again in 20
seconds. A typical set of results in Figure 2.15 shows the changes in steel properties
against temperature for 0.4mm gauge, warm rolled steel.
Figure 2.15: Recrystallisation experiment results showing property changes with temper-
ature for 0.4mm Warm Rolled steel
These results can be compared with Figure 2.3 to see the stages of recrystallisation
occurring as temperature increases. Before recrystallisation starts, there is a gradual
reduction in hardness (HRB), ultimate tensile strength (UTS) and upper yield strength
(UYS), and an increase in total elongation (TE%), which correspond to recovery of the
strained microstructure. In this example, recrystallisation appears to start around 590 −
600◦ C, when these properties start to rapidly change. The end of recrystallisation is
harder to estimate as there may be significant further softening due to grain growth, but
appears to finish around 650 − 670◦ C.
68
A major area of uncertainty in these trials is the manual examination of the recrys-
tallisation curves to select the temperatures at which recrystallisation starts and stops.
In some cases, this can be seen in microscopic examination, where micrographs indicate
recrystallisation is occurring much earlier than indicated by the change in hardness or
strength such as shown in Figure 2.16.
Figure 2.16: Results for 0.4mm Hot Rolled steel where micrographs show recrystallisation
earlier than indicated by softening of steel
There are other sources of uncertainty in the results, such as discontinuities in some
softening curves, especially seen on the lightest gauge tested (0.33mm) where softening
occurs very rapidly. Other results also show extended periods of recovery softening before
recrystallisation, and it can be difficult to select representative recrystallisation onset and
finish temperatures. The accuracy of selected recrystallisation start or finish temperatures
appears to be around ±10 − 15◦ C, as there can be a wide band of possible temperature
selections to suit recrystallisation start or finish temperatures as shown in Figure 2.17.
The full series of recrystallisation experiments encompass much of the range of steel
types used at NZ Steel on the metal coating line, from 0.3 to 2.2mm gauge thickness,
warm and hot rolled strip, and all four chemistry grades 100, 110, 210 and 310. The
majority of experiments deal with grade 110 hot rolled feed across the full range of gauge
thicknesses, but select gauge ranges have been tested with grade 100 warm rolled, and
grade 210/310 hot rolled strip as well. The resulting recrystallisation start and finish
temperatures for the full range of tested strips are included here in Figures 2.18 and 2.19
respectively.
69
Figure 2.17: Possible band of selections for recrystallisation start and end temperatures
for 0.75mm warm rolled strip
The main observation from these charts is there is a steady increase in required re-
crystallisation temperature as the gauge thickness is increased for all strip types. The
warm rolled strip appears to require around 10 − 15◦ C higher temperature for recrystalli-
sation than the 110 grade hot rolled strip, and the 210/310 grade hot rolled strips require
around 15 − 25◦ C higher temperature also. These effects are due to the changes in time
required for recrystallisation (t0.5 ), as described in Section 2.7.2. The cold reduction of
NZS steels decreases as the gauge size increases, and the hot rolled steels have a larger
grain size than the cold rolled steels. As the time and temperature of the recrystallisation
experiments is fixed, this results in less recrystallisation with larger gauge size or for hot
rolled steels compared to cold rolled steels, and increases the temperatures required to
start or complete recrystallisation as shown in Figures 2.18 and 2.19.
70
Figure 2.18: Experimentally determined recrystallisation start temperatures for NZ Steel
strip
Figure 2.19: Experimentally determined recrystallisation complete temperatures for NZ

Steel strip
71
Chapter 3
Practical Temperature Calculation

to Improve Control
This chapter describes the overall investigation conducted in this project. The underlying
problem and desired outcome is explain in Section 3.1, and a method of solution proposed
in Section 3.2. The scope of the research, and the specific steps undertaken to find this
solution are then detailed in Section 3.3.
3.1 Difficulty of Measurement and Control
The fundamental problem that creates difficulty in the control of the NZ Steel annealing
furnace is an observability problem. The evolution of strip temperature inside the furnace
is not well understood and cannot be accurately measured, as the measurement devices
available are strongly affected by reflected radiation inside the furnace cavity. The strip
temperature is a direct driver of the resulting mechanical properties of the steel, due to
the recovery and recrystallisation processes occurring. The frequency of product changes
and large thermal inertia of the furnace cause it to be in a transient temperature state
for much of its operating time, with changes occurring over several time scales. The
long time scale of control actions means strip temperature control is very difficult during
transient periods, which results in strips processed at temperatures too high or low, giving
undesirable mechanical properties.
72
The major aim of this project is to identify and demonstrate how the evolution of
strip temperature and mechanical properties can be better understood and accurately
calculated in this non-linear system with complex geometry, time-varying conditions and
many interacting factors. It will be shown that this can be done in a practical way, which
requires real time solution using furnace operational conditions. This will be followed by
investigating how this improved knowledge of furnace behaviour can be applied to result
in a measurable improvement in the control and outcome of the annealing process.
3.2 Method of Solution
It is hypothesised that understanding of the strip temperature evolution and the effect of
parameter changes can be improved using modern computer modelling techniques that
require minimal simplification and approximation of the process, while still preserving
the process dynamics. This knowledge can be applied to calculate strip temperatures
and expected changes in the metallurgical condition under any combination of processing
conditions.
It will be demonstrated that improved knowledge and calculation of temperature evo-

lution can be used to improve control of an industrial furnace. This can be done by
creating a coupled thermal-metallurgical model of the furnace system, which is capable
of simulating operating conditions and predicting the process outcomes to optimise oper-
ational settings, and should also be suitable for application to online control and process
decision making.
3.3 Scope of Research
In order to improve the understanding of temperature evolution within this complex

system, and improve the overall control of the furnace during both static and dynamic
temperature conditions, the following plan of investigation will be used:
73
3.3.1 Model furnace system in COMSOL software
By using the COMSOL 3D multiphysics modelling package, a model can be developed

with minimal simplification of the actual furnace system. It is possible using this software
to represent the full geometry of the furnace, including complicating factors such as the
thermocouple tubes and their relationship to the temperatures of the strip and the furnace
walls, and the spatial variation in furnace and strip temperature in all dimensions due to
proximity of the walls and the radiative view factors. It is possible to capture most of the
important physical behaviours of the furnace system, which are otherwise impossible to
observe using available measuring equipment, and very difficult to represent mathemat-
ically without the use of 3D modelling software. The model development is illustrated
and explained in Section 4.5.
3.3.2 Examine furnace model behaviour
Examination of the behaviour of the COMSOL model can provide insight into physical
phenomena occurring in the annealing furnace, some of which may be anticipated by
NZ Steel technologists or plant operators, but cannot be measured due to constraints
of the furnace system. For example, differential heating across the width of the strip
can be examined in the model, but it is impossible or prohibitively expensive to mount
equipment in the furnace capable of accurately measuring this temperature difference.
This has been investigated using mathematical techniques [39, 73] for a highly simplified
system, but can be investigated using a significantly more complex and accurate geometry
using COMSOL software.
Other physical behaviours that are investigated using the F3DC model include temper-
ature variation around the furnace walls, the behaviour of the furnace heating elements,
hearth rolls and thermocouples. It was proposed in the MISG study that an improved
model of the furnace system would require representing the furnace as a large number of
isothermal surfaces which exchange thermal radiation with each other [39]. A very large
number of calculations would be required to find view factors between all of these partic-
ipating surfaces, which can be best done using 3D software to analyse this geometry, and
may allow greater model complexity, which would be practically impossible to calculate
manually. Furnace model results and behaviour are shown in Section 5.3.
74
3.3.3 Simplify equation system
Once the behaviour of the furnace model has been observed and understood, it will be
investigated how to best simplify the model equation system. This is required to create
a model capable of rapid solution of furnace temperatures, necessary to enable practical
investigation of multiple options for furnace settings and control strategies. While a
complex 3D model should provide more accurate solutions, the solution time required is
prohibitive for this type of investigation, where many alternative settings will be trialled to
find optimal control parameters. A guideline from NZ Steel provided for the development
of the ‘Sydac’ model was that a solution should take less than 10 seconds, using without
specialised computer hardware (Desktop PC with 3GHz P4, 512MB RAM, Windows XP)
[69]. The complete equation system developed for the simplified RDHT model is shown
in Section 4.6.
3.3.4 Build simplified temperature model
Once a simplified system of equations has been developed for the furnace model, they
can be programmed into MATLAB programming language using an appropriate solution
scheme. The accuracy of the simplified model will be checked against the F3DC model,
and validated using furnace temperature measurements taken from plant operations across
a wide range of operating conditions and strip product types. The model should solve
the equations rapidly, be able to calculate solutions for both steady state and transient
temperature situations, and be capable of being utilised for online control and temperature
prediction in real time if required. The solutions from the RDHT model and comparison
against operational data is shown in Section 5.4.
3.3.5 Create integrated metallurgical model
Investigation of metallurgical models and heat transfer calculations will be conducted to

create a model capable of predicting the metallurgical outcome from the strip temperature
result. This is vital to convert predicted temperatures into a measure of the success of
the heat treatment of the strip in the furnace. These metallurgical predictions will also
be compared and validated against strip recrystallisation temperature experiments and
other specific cases investigated by NZ Steel. Full development and validation of the
metallurgical model is demonstrated in Section 4.7.
75
3.3.6 Demonstrate optimisation using model
The ability of the temperature and metallurgical calculations of the model to improve
process control and optimise process settings will be demonstrated by the application
to specific furnace operating conditions. By simulating a number of alternatives, it will
be possible to identify areas where a measurable improvement in production, product
quality or energy efficiency can be realised by NZ Steel. It will also be shown how the
model is able to predict temperatures under transient conditions, and demonstrate the
possibility for online furnace control and prediction. Use of the model results is shown
in Chapter 6, where they are applied to a number of specific products at NZ Steel to
find alternative operating conditions that allow higher throughput, power efficiency, or
improve the reliability of heat treatment.
76
Chapter 4
Model Methodology and

Development
4.1 Introduction
In this chapter, the full development of the furnace and metallurgical models is demon-
strated. The physical design of the furnace, specific important details, and methods of
operation of the annealing furnace and wider MCL plant are shown in sections 4.2 and 4.3.
Mathematical formulas relating to heat transfer in the furnace and steel are constructed
in section 4.4, and these are applied to a full 3D computer model as shown in section 4.5.
The reduction of this model and creation of the simplified model is given in section 4.6,
and the development of a metallurgical model fitted to NZS metallurgical testing data is
described in section 4.7.
4.2 Furnace Physical Design
The furnace used in the MCL plant to heat treat steel strip immediately before galvanising
as shown in Figure 1.2 is approximately 160m long, and is divided into 20 separately-
controlled ‘zones’, being 12 heating zones, 7 cooling zones and zone 20, the ‘hot bridle’
zone. The heating zones comprise the first 67m, each zone on average 5.6m long, whereas
the cooling zones are 81m long, with an average length of 11.6m each. Zone 20 is a special
case, enclosing the area where the strip descends into the molten metal coating pot at the
77
exit from the furnace. The overall flow of product through the plant before and after the
furnace is shown by the plant schematic in Figure 1.1.
The furnace heating method and mechanical design is detailed in Section 4.2.1, and
the cooling methods and devices are covered in Section 4.2.2. Temperature measurement
methods and locations are given in Section 4.2.3 and the furnace hearth rolls are shown
in section 4.2.4.
A cross-section of the heating zones of the annealing furnace is shown in Figure 4.1.
The furnace was installed in 1968, and is constructed from insulating refractory bricks
enclosed in a steel shell. Electric radiant heaters are installed both on the furnace floor,
and suspended from the roof.
Figure 4.1: Typical cross section of MCL annealing furnace heating zone, showing (a)
refractory brick lining, (b) suspended radiant heating elements, (c) hearth rolls, (d) floor
mounted radiant heating elements and (e) steel shell
4.2.1 Furnace Heating Method and Equipment
The strip in the annealing furnace is heated by two separate methods - radiative heat
transfer inside the furnace cavity using the radiant heaters shown in Figure 4.1, and by
preheating using induction heaters immediately before the furnace entry.
78
Radiative heating is accomplished with cast nichrome electric elements wired on 3
electrical phases. In the heating zones, there are 5 rows of elements mounted on the
floor of the furnace, and 4 rows of elements suspended from the roof on nichrome hooks.
The cooling zones also contain heating elements, as described in Section 4.2.2, but have
significantly lower heating capacity, with three rows of elements on the furnace floor only.
The heating capacity of each heating zone is approximately 346kW, cooling zones 13-15
have approximately 115kW capacity, and zones 16-19 have 200kW capacity each.
Figure 4.2: Photograph inside furnace heating zone showing heating elements installed
on roof and floor. The hearth rolls are removed here.
The heating elements visible in Figure 4.2 have a serpentine shape to give extended
surface area. The heating elements are prone to damage or failure during normal operation
due to several causes. Repeated heating and cooling over an extended period of time
can lead to strong warping of the elements, causing electrical shorting between adjacent
elements as shown in Figure 4.3. Elements may also be physically damaged or electrically
shorted in the case of breakage of the moving strip, which may become melted to the
elements if it runs slack. Damage to one element causes an immediate loss of one heating
phase to the heating zone, reducing heating capacity by one half in that zone. If this
occurs, it is possible to rewire the neutral phase to the star point, recovering heating
capacity to two-thirds of normal. If two phases of elements are damaged or unusable, no
heating power will be available in that zone.
79
Figure 4.3: A warped heating element after removal from furnace
Heating elements are time consuming and expensive to replace, and this is only done
approximately every two years during major maintenance down time. This means that
damage to heating elements can cause furnace heating to be running at decreased capacity
in one or more heating zones for a long period of time. A generalisation is that the MCL
furnace is usually operating with reduced electrical heating capacity in one or more zones.
A major intended use of the furnace modelling is to calculate how to compensate for failed
heating elements - damaged elements in one zone may be counteracted by increasing
heating power to adjacent zones to maintain the desired peak strip temperature.
The heating elements used in Zone 20 are of a different design and installation. These
‘ELRAD’ heaters are mounted horizontally inside cylindrical ceramic brackets, some of
which also contain gas transport tubes for maintaining the H2 − N2 furnace atmosphere.
These ’HNX’ ELRADS and the heating-only ELRADs are mounted as shown in Figure
4.4.
The other available method of heating the strip is a series of 4 induction pre-heaters
that were retrofitted to the furnace in 2004. There are four pre-heaters mounted imme-
diately prior to the strip entry, as shown in Figure 1.2. These can rapidly increase the
strip temperature by up to around 300 − 350◦ C, depending on the steel width, gauge,
linespeed and chemistry. It is possible to measure the strip temperature after the in-
duction pre-heaters and immediately before the radiant furnace entry using the ‘contact
thermocouple’ as described in Section 4.2.3.3.
80
Figure 4.4: Elevation of Zone 20 ’Hot Bridle Zone’ showing layout of ELRAD heaters and
thermocouples
4.2.2 Furnace Cooling Method and Equipment
The purpose of the cooling zones of the furnace shown in Figure 1.2 are to reduce the strip
temperature to the optimal range for coating in the hot metal pot, which is typically 450◦ C
for galvanised products, or 550◦ C for Zincalume coated products. Cooling is required to
reach these temperatures for ‘soft iron’ recrystallised products, which may have peak strip
temperatures exiting the heating zones of up to 760◦ C measured on the L1 pyrometer.
The cooling devices used to reduce the strip temperature are the ‘cooling tubes’ described
in Section 4.2.2.1, and the ‘gas jet coolers’ described in Section 4.2.2.2.
Products with a lower peak strip temperature such as the ‘hard iron’ recovered prod-
ucts may not require any cooling, and in some cases may require some heating to reach
the correct strip temperature for coating. Recovered products may be run with peak
strip temperatures as low as 450◦ C, and will require further heating in the cooling zones
in the case of Zincalume coating. For this purpose, the cooling sections of the furnace
also contain heating elements, but at much lower capacity than the heating sections.
81
Figure 4.5: Typical cross section of annealing furnace cooling zone, showing (a) upper
cooling tube, (b) upper cooling tube air exit, (c) upper cooling tube air entry, (d) hearth
roll supporting strip, (3) lower cooling tube assembly, (e) lower cooling tube air entry and
(f) floor mounted heating elements
The cooling sections of the furnace also have a different cross-section, as shown in
Figure 4.5, with thinner refractory walls. The typical thickness of walls in the heating
section is 0.34m, whereas the cooling section walls are 0.23m thick. The outer dimensions
are the same, giving the cooling sections a more spacious furnace cavity.
4.2.2.1 Cooling Tubes
The furnace cooling tubes, as shown in Figure 4.5 are mounted through the furnace walls
both above and below the strip. Ambient air is circulated through an annular pipe in the
cooling tube assembly, and returned through the exit pipe. This is an indirect cooling
method, as the cool air circulating through the tubes directly contacts only the inner
surface of the cooling tube. The cooling of the tube increases radiant heat transfer to
the tubes by increasing the temperature difference. The extracted heat then exits the
system in the airflow return, removing heat from the furnace zone walls, strip, and other
equipment exchanging radiative heat with the cooling tubes.
82
The cooling tubes are densely spaced in the furnace cavity, mounted at approximately
530mm centre line spacing, and tubes have an outer diameter of 178mm each. There were
originally over 260 of these cooling tubes, but as per Section 4.2.2.2, the cooling tubes
have been completely removed from Zone 14, and replaced with gas jet coolers.
4.2.2.2 Gas Jet Coolers
Cooling Zone 14 uses four ‘gas jet coolers’ in place of the cooling tubes in other cooling
zones. These use a series of pipes through the furnace body, through which cold air
is blown. Jet nozzles on the pipes cause the cold air to be blown directly onto the
moving strip, cooling it very rapidly. The air in the furnace body in these cooler sections
is extracted and cooled via air/water heat exchangers of 480kW cooling capacity each.
These coolers allow the strip to be cooled much quicker than by using the indirect radiant
cooling tubes in the remainder of the cooling section.
Figure 4.6: Cross section of gas jet cooler units [35]
A cross-section of the gas jet cooler units is shown in Figure 4.6. Note that there is no
refractory brick lining in this section of the furnace, which consists of the four sequential
steel-shelled cooler units, lined with insulating board and ceramic fibre blanket. Using
the gas jet coolers, it is possible to cool the strip by over 130◦ C in the 8m of strip travel
through this zone, depending on the strip mass throughput and size [35].
The gas-jet coolers are commonly used for the ‘soft iron’ fully recrystallised products
only, where the strip exit temperature from the heating zones is well above the temperature
range required for coating. The coolers are operated in a cascade fashion, where a single
83
cooler is run up to nearly 100% capacity before additional coolers come online. When the
coolers are not in use, the fans must still be run at 3% speed to keep the bearings cool.
The cooler units also contain small electric radiant heating units, to compensate for heat
loss through the walls which are less insulating than the rest of the furnace - the heating
units may be turned on to maintain strip temperature when the coolers are not in use.
4.2.3 Temperature Measurement Equipment and Location
Temperature measurement in the MCL furnace is complicated by the radiative heat ex-
change inside the furnace cavity, where reflected radiation strongly affects temperature
readings. There are three different temperature measurement methods in use in the fur-
nace. The temperature of the furnace zones is measured by thermocouples inside the
furnace cavity, described in Section 4.2.3.1. The temperature of the strip is measured by
radiation pyrometers as shown in Section 4.2.3.2, and the temperature of the strip after
preheaters and before entry to the radiant furnace is measured by a retractable contact
thermocouple, which is shown in Section 4.2.3.3.
4.2.3.1 Furnace Zone Thermocouples
Furnace ‘zone’ temperatures are measured by a series of thermocouples mounted inside

inconel tubes protruding into the furnace cavity as shown in Figure 4.7. These thermocou-
ple tubes protrude into the furnace approximately half the distance from the furnace roof
to the level of the strip, and as such receive radiative heat transfer from the furnace walls,
roof, strip and heating elements in an unknown proportion. The temperature measured
by these thermocouples is taken as the overall furnace ‘zone temperature’.
Three thermocouples are mounted next to each other in each of the heating and cooling
zones. Only one of these thermocouples is used as the temperature control point for each
zone PID controller. One thermocouple is connected to a temperature alarm system,
and the third thermocouple is connected to a separate control system that can be used
manually in case of complete computer failure.
Thermocouples may become damaged due to failure of the inconel sheaths allowing
ingress of the furnace atmosphere gas, which may be up to 1000◦ C or higher in the furnace
heating zones. Failure may occur by the sheaths being impacted and bent in the case
of a strip breakage in the furnace. In the case of an obvious failure of the thermocouple
84
Figure 4.7: Photograph showing furnace zone thermocouples installed in furnace
it may be removed and replaced during production, as the sheaths protrude through the
furnace roof and can be accessed externally. Thermocouples may also be inaccurate if the
probe itself is in an incorrect position inside the sheath. Thermocouple probes should be
inserted fully into sheaths, making contact with the bottom inner surface of the pipe. Air
gaps or insulating material in the sheath may cause inaccuracy and slow response time of
the thermocouples.
4.2.3.2 Pyrometers
As shown in Figure 1.2, there are many pyrometers mounted in the MCL furnace which
measure the strip temperature. These pyrometers are designated IR1-IR4 for those of
‘Ircon’ manufacture, and L1-L4 for those more recently installed of ‘Land’ manufacture.
The Ircon pyrometers are installed in heating zones Z9 and Z12, cooling zone Z15, and
the hot bridle zone Z20. The Land pyrometers are installed in pairs, with two in cooling
zone Z13, and two at the end of the gas jet coolers in Z14.
The pyrometers in the furnace have a very low degree of accuracy, due to radiative
interference and inefficiency of the installed shielding. All of the pyrometers are installed
in sighting tubes that project through the walls of the furnace, which have a flange at the
end nearest the strip to attempt to exclude stray radiation from the furnace cavity affecting
the pyrometers. Cool H2 − N2 gas is injected at the bottom of the sight tube to also assist
with reducing radiative interference, although all other Bluescope Steel MCL furnaces
use full water jackets on pyrometer sight tubes [48]. The installation of pyrometers in
heating sections is also unusual practice, as these areas are likely to have very high levels
85
of radiative exchange and potential for interference. Pyrometers are usually installed in
‘transition’ sections of furnaces, where there is no strong heating or cooling of the strip
to create radiative interference.
Figure 4.8: Gold cup measurements in Zone 12 showing the difference between online
pyrometer measurements and gold cup measurements taken at the same time [48]
The accuracy of the L1-L4 pyrometers was measured in 2003 using a ‘gold cup py-
rometer’ device, which utilises a small gold cup on the end of a long probe to exclude all
reflected radiation, and measure the strip temperature while the gold cup is in contact
with the moving strip. This measurement method is difficult and time consuming, and
can only be carried out for trial purposes. It is not feasible to use as an online, continuous
measurement as the gold cup device rapidly corrodes in the hot furnace environment, and
can be damaged by strip welds passing through the furnace.
The Ircon IR1-IR4 pyrometers tested in the gold cup trials were found to read at
least 80◦ C too high for light gauge recrystallised product, and up to 400◦ C too high
for heavy gauge recovered product [48]. An example of common differences in gold cup
measurements and the online pyrometer measurements for one zone is shown in Figure
4.8 [48].
The more recently installed Land pyrometers have proved more accurate, particularly
the ‘L1’ pyrometer installed in the first cooling zone, Z13. Subsequent gold cup mea-
surements shown in Figure 4.9 found this pyrometer to read typically around 10 − 50◦ C
higher than gold cup measurements across the range of products tested [26]. This is the
only online temperature measurement of the strip inside the furnace that can be used for
model validation purposes. Its position immediately after the heating zones allows it to
be used as an estimation of the peak strip temperature.
86
Figure 4.9: Error between gold cup and pyrometer measurements at all pyrometer loca-
tions [26]
While this pyrometer is the most accurate available for model validation purposes,
the error in its measurement is still relatively high. For a strip temperature increase of
around 400◦ C for a recovered steel in the furnace, a measurement variation of 10 − 50◦ C
could represent up to 12% measurement error at this location. This is far from ideal,
but there is no more accurate online measurement available for all coils. Improvement of
radiative shielding of the other pyrometers is not feasible at present, and is likely to have
only limited benefit in the heating zones where the strip temperature is of most interest.
4.2.3.3 Contact Thermocouple
The strip temperature after the induction preheaters and before entry to the radiant
furnace is measured using a retractable ‘contact thermocouple’ as shown on the furnace
schematic in Figure 1.2. This measurement of initial strip temperature is vital to the
accuracy of strip temperature calculations in the radiant furnace, especially in the case
of strong preheating of the strip. When the preheaters are not operating, the entry
temperature is likely to be around 40 − 50◦ C, depending on strip gauge and width, after
passage through the hot caustic cleaning baths. With preheaters operating there may be
great variation in the strip input temperature, which can be as high as 400◦ C.
87
The contact thermocouple is mounted in the transition section between the last pre-
heating furnace and the entry seal rolls of the radiant furnace. It is stored retracted
inside a nitrogen cooled bay above the furnace section, and is lowered onto the strip pass-
ing below using a hydraulic cylinder, as detailed in Figure 4.10. The thermocouple itself
is a flexible metal strip protruding below a stainless steel trolley on ceramic rollers that
sits on the moving strip when lowered. The trolley sits on the strip for a fixed period of
time, currently set at 5 seconds, before being automatically retracted back into the cooled
bay. The contact thermocouple unit is only activated manually by MCL operators when
required under plant operational guidelines.
Figure 4.10: Contact thermocouple mechanism installed in transition section between

preheater and furnace entry
Similar to the gold cup pyrometer, the contact thermocouple cannot be operated
continuously due to the possibility of damage to either the unit or the strip. The heat
in the transition section can cause damage to the trolley roller bearings over time, and
failure of the bearings and jammed wheels can cause marking or damage to the moving
strip. The unit must also avoid the passage of welds in the strip, as these rough areas
can also damage the wheels and thermocouple. When activating the unit, operators must
88
ensure that no strip welds are about to enter the furnace i.e. shortly after feed coil
changeovers. Extended operation in the hot environment can also damage the pneumatic
system, causing failure to operate, or failure to actually contact the strip when the unit
is initiated.
Using the contact thermocouple, it is possible to measure the input temperature of

the strip accurately. The contact thermocouple unit does require regular maintenance of
the hydraulic system and thermocouple trolley to ensure good contact is made with the
strip, and no damage of the strip occurs.
4.2.4 Hearth Rolls
The hearth rolls are electric motor-driven rolls extending through the furnace cavity, and
are used to both support the strip and move it through the furnace, in conjunction with
the other powered drive rolls in the MCL plant. A typical hearth roll assembly is shown
in Figure 4.11 as per the original design of hearth rolls in the furnace heating zones.
Figure 4.11: Typical hearth roll assembly in heating section
The hearth rolls are approximately 120mm diameter, constructed from 12mm steel
with a hollow interior. The design has been changed since originally installed to include
tapered ends, and new coating types for increased resistance to corrosion and mechanical
damage. There are 83 hearth rolls spaced a constant 1.83m apart in the heating and
cooling sections of the furnace, not counting the hot bridle zone.
The hearth rolls are primarily of interest in the mathematical modelling of the MCL
furnace due to their significant thermal mass, which may have an effect on the transient
thermal behaviour of the system. The rolls are exchanging heat radiatively with the
89
furnace walls and heating elements and the strip, and also transfer heat to the strip via
conduction. Because the strip is not rigid, the contact area of the strip on the rolls is
defined by a wrap angle.
4.3 Furnace Control
4.3.1 Introduction
Description of the control methodology of the MCL furnace is concentrated on two sep-
arate methods - the ‘cycle’ method of adjusting furnace settings for different products
running at steady state as described in Section 4.3.2, and how the furnace is controlled
during changes between different product cycles as shown in Section 4.3.3.
4.3.2 Furnace Operating Cycles
There are a wide variety of steel product types and sizes processed regularly in the NZS
MCL plant - a total of 750 combinations of strip width, gauge, tensile strength and
coating type were processed in the 12 months preceding May 2008, as detailed in Section
6.2. Rather than assign control setpoints to each individual product, the furnace control
settings are stored in a number of ‘cycles’. Groups of product types are run with a
common cycle, designated with a cycle number and letter variation e.g. Cycle 5A, Cycle
24B etc. There are approximately 20 cycles in regular use, but cycle numbers and settings
are under regular revision by NZS technologists to solve particular operational issues. An
example of the cycles in common use is shown in Figure 4.12.
The product grade in Figure 4.12 refers to the minimum desired tensile strength, in
MPa, of the steel strip after heat treatment. For example, G500 strip must have at least
500MPa yield strength or higher, in accordance with international standards. Grades GC
and G1-G350 are ‘soft iron’ recrystallised products heat treated at high temperatures,
and grades G450-G550 are ‘hard iron’ products undergoing recovery heat treatment only.
Each furnace cycle contains information giving the setpoints for all heating and cooling
equipment in the furnace. For example, each furnace cycle contains the temperature
setpoints for all 20 furnace zones, power setpoints for the four induction preheaters and
four gas jet coolers, and the strip linespeed. Each of these settings can be manually
90
Figure 4.12: List of common furnace cycles for different products depending on grade and
dimensions
changed by operators during production if required to compensate for problems such as

failed heating elements etc.
The furnace zone temperatures are maintained by controlling the heating element
power input using individual PID controllers in each zone, tied to the zone temperature
setpoints. As described in Sections 2.5.4 and 4.2.3.1, the temperature measured by the
zone thermocouples is not well understood, and does not relate to any particular temper-
ature of the furnace walls, strip or elements. Hence the zone temperature setpoints are an
arbitrary temperature that has been found to meet the required heating of the strip, and
the actual inner wall temperatures may be significantly different to the control points.
These furnace cycles have proven to be effective for controlling the furnace for any
product operating continuously over a long period of time i.e. once the furnace system has
reached a steady state. Furnace cycles are often adjusted iteratively by plant technologists,
using learning from experience by plant operators and mechanical test results of product
steel, such as cases where problems like strip cracking or wavy edges have occurred.
4.3.3 Control During Cycle Change
The major difficulties in furnace control occur when changing between cycles, particularly
in the cases of large magnitude temperature change from a hot cycle for recrystallised
91
product (‘soft iron’), to a cooler cycle for recovered product (‘hard iron’) or vice versa.
The soft iron products appear in the list of furnace cycles in Figure 4.12 as grades G1-
G350, and the hard iron products are grades G450-G550.
Figure 4.13: Comparison of furnace zone temperatures for a hard iron product (G550)
and a soft iron product (G300) of the same dimensions (0.4 x 1220mm)
When changing furnace cycles from hard to soft iron or vice versa, there can be a
significant change in the zone setpoint temperatures. As shown in Figure 4.13, the typical
furnace temperatures for a hard iron and a soft iron of the same product dimensions may
differ by 100 − 200◦ C in each heating zone, although cooling zone temperatures are likely
to have a smaller difference. The large thermal mass and slow reaction time of the furnace
system means that the changes to these furnace setpoints initiated at the time of cycle
change take a long time to come to equilibrium i.e. the furnace is in a transient thermal
condition, which may require an hour or more to reach a thermal steady state.
During the time the furnace heating zone temperatures are changing and are in a
transient state, the peak strip temperature as measured on the L1 pyrometer is being
monitored by the furnace operators. To maintain the strip temperature within acceptable
margins for the product type being processed, the operator manually adjusts both the
plant linespeed and the furnace induction preheaters. These two parameters are the only
methods by which the strip temperature can be rapidly adjusted to compensate for the
changing zone temperatures. Maintaining strip temperature accurately over these periods
can be difficult for operators to achieve, as there can be a large number of adjustments
required, and there are no guidelines or automatic systems to assist this process - the
92
control of strip temperature during transient periods relies upon the experience and ability
of operators as there is no standardised procedure followed.
An example of the difficulty of manual control during cycle change in shown in Figure
4.14, where strip gauge changes from 0.55mm to 0.95mm, but the required strip tem-
perature stays constant (unchanged G300 strength). At the time of cycle change, it is
necessary to either slow down the linespeed or increase preheating to maintain strip tem-
perature with the thicker gauge product. In this example, operators have strongly reduced
both linespeed and preheater power, which have opposing effects and the net result is an
unexpected spike in strip temperature. The duration of the temperature excursion is
around 8 minutes long, and at an average 110m/min this will affect around 900m of strip
passing through the furnace. In this case, increased temperature on a soft-iron strip is
unlikely to cause major problems, but would be very damaging to mechanical properties
of a hard-iron strip, causing undesirable recrystallisation.
Figure 4.14: Example of manual control of linespeed and preheater control during a cycle
change from 0.55mm to 0.95mm gauge strip, causing incorrect peak strip temperature
A further complication is that cycle changes can occur in rapid succession, due to short-
length runs of the less-common products. A very common product such as 0.4×940mm
strip may run continuously for 24hrs or longer, but uncommon or less-popular products
may run as short as 5-10 minutes for a single coil before the next cycle change. The
furnace schedulers arrange the order of coils so there is as little change between successive
coils as possible, but there are periods of time when many short-run products are required,
and furnace setpoints change frequently for a full 8 hour shift or longer, making manual
control by operators exceedingly difficult, and hence inaccurate.
Coil types are arranged into groups so there will not be rapid changes from hard- to
93
soft-iron and back again, but changing coil sizes may be constantly moving temperature,
preheating or linespeed setpoints by a small magnitude. Large-scale temperature change
from hard-soft or soft-hard iron types is likely to occur only one to two times per day, but
as seen in Figure 4.14, even cycle changes without variation in strip temperature setpoints
can be problematic with manual control. Subsequently, based on historical plant data it is
estimated that under normal operations the MCL furnace may be in a transient thermal
state for up to 50% of operating time [39].
4.4 Model theory development
4.4.1 Introduction
All three modes of heat transfer are present within the MCL furnace system, and cal-
culation of these can be used to predict furnace and strip temperatures for any given
operating condition. These modes of heat transfer are described in Section 2.4, and it is
here shown how each of these modes applies to a furnace model, and how they are rep-
resented mathematically to enable the calculation of furnace and strip temperature. The
modes of heat transfer considered are conduction in Section 4.4.2, convection in Section
4.4.4 and thermal radiation in Section 4.4.3.
4.4.2 Conductive Heat Transfer Calculations
Conductive heat transfer occurs in all the solid bodies present in the furnace, and is most
important to calculate the temperature of the furnace walls, heating elements, steel strip,
hearth rolls and thermocouples. In each of these bodies, heat is either received or lost from
the surface by convection or radiation, and conduction causes the heat flux to propagate
through the solid body.
As described in Section 2.4.4 for a one-dimensional body, conduction in a three-

dimensional body can be represented by the heat conduction equation:
∂ 2T ∂ 2T ∂ 2T

∂T
cp ρ =k + + +Q (4.1)
∂t ∂x2 ∂y 2 ∂z 2
The physical properties of density, ρ, conductivity, k, and heat capacity, cp are required
94
to calculate heat transfer by conduction within the furnace bodies. The values used for
these properties in the furnace modelling are listed in Table 4.1. As many of these
properties vary with temperature, a function is used rather than a single value. To allow
for the temperature dependence of conductivity and heat capacity these variables may
be placed inside the divergence as shown by Equation 4.15 used in the F3DC model.
Note that thermocouple tubes are modelled using physical properties of steel rather than
inconel for simplicity.
Body Density, ρ Conductivity, k Heat Capacity, cp

kg/m3 W/m.K J/kg.K
Nichrome Elements 8200 16.6 450
Refractory Bricks 497 kref ractory 850
Hearth Rolls 7850 ksteel cp,steel
Thermocouples 7850 ksteel cp,steel
Steel Strip 7850 ksteel cp,steel
Table 4.1: Material properties of furnace bodies to be used in conductive heat transfer
calculations
The conductivity of the ‘K-23’ type insulating refractory brick is given in manufac-
turer’s data, and approximated with the function:
kref ractory = 0.0615 + 0.000137 × T (K) W/m.K (4.2)
The heat capacity and conductivity of steel depend both on temperature and the
composition of the steel. The steels used at NZ Steel for MCL products fall into two
ranges - the values given in Table 4.2 are closest to chemistry types 100, 110 and 210 with
nominal composition 0.06%C and 0.4%Mn. The values in Table 4.3 are most suitable for
steel chemistry 310, with a nominal composition of 0.23%C and 0.6%Mn.
Temperature Heat Capacity Conductivity

◦
C J/kg.K W/m.K
20 - 65.3
100 482 60.3
200 520 54.9
400 595 45.2
600 754 36.4
800 875 28.5
1000 - 27.6
Table 4.2: Steel thermal properties to suit MCL steel chemistries 100, 110, 210 [17].
95
Temperature Heat Capacity Conductivity
◦
C J/kg.K W/m.K
20 - 51.9
100 486 51.1
200 520 49
400 599 42.7
600 749 35.6
800 950 26
1000 - 27.2
Table 4.3: Steel thermal properties to suit MCL steel chemistry 310 [17].
Quadratic equations are fitted to the conductivity and heat capacity data against
temperature for the steel types in Table 4.2. A separate equation has not been used for
steel chemistry 310, as the differences are relatively minor for this type, and the usage
of this steel chemistry on the MCL is low. The heat capacity correlation is given by
Equation 4.3 and the conductivity by Equation 4.4.
cp,steel = 0.0003275 × T 2 + 0.101 × T + 396.4 J/kg.K (4.3)

ksteel = 0.00002739 × T 2 − 0.0827 × T + 87.5 W/m.K (4.4)
Where T is degrees Kelvin
4.4.2.1 Simulation of Strip Movement
Rather than simulate strip movement through the furnace by moving the calculation
elements in the strip, it is convenient to simulate it by moving the internal heat of the
strip body down the furnace at a rate equal to the speed of movement of the strip. This
greatly simplifies modelling efforts for both the full 3D and simplified models. This is
done by adding an advection term to the conduction heat equation in the strip body. The
conduction heat equation for the strip in three dimensions is now as shown in Equation
4.5, where x is the direction of strip movement down the furnace and v is the speed of
movement (the plant linespeed).
∂ 2T ∂ 2T ∂ 2T

∂T ∂T
cp ρ +v =k + + +Q (4.5)
∂t ∂x ∂x2 ∂y 2 ∂z 2
96
4.4.3 Radiative Heat Transfer Calculations
The radiative heat transfer is given by the Stefan-Boltzmann Law, Equation 2.9. Cal-
culating heat exchanged between any two or more surfaces via radiation requires further
information in the form of the surfaces’ emissivities, and the view factors between sur-
faces [47]. In the case of the F3DC model, view factors are directly calculated by the solid
angles between each individual calculation element, whereas the simplified RDHT model
requires the selection of a view factor correlation suitable to represent the furnace system,
such as used in Equations 2.14 and 2.22, which are functions of the approximation made
to the furnace geometry.
Selection of emissivities to use in the modelling work is difficult, as surface emissivities

are very difficult to measure in practice, and vary greatly due to surface conditions that
may also change with age. It is often difficult to measure emissivities more accurately
than to within 0.2 [47]. Jinks [27] details several values and functions for emissivity of
plain steel strip used in other modelling works, which vary between 0.1 and 0.25 at strip
operating temperatures, however the value used in both the NZ Steel model (Section
2.5.1) and the MISG investigation (Section 2.5.2) for the MCL furnace is s = 0.2, which
will be followed here. Note that this value of emissivity is representative of a moderately
reflective surface, which does not easily absorb radiative energy.
The other main values of emissivity required are for the furnace walls, heating elements,
hearth rolls and thermocouples. These surfaces are likely to have much higher emissivities
than the strip, as these items are permanently installed inside the furnace, and hence
become dirty or worn with age, increasing the emissivity. The NZS and MISG models both
use a value of w = 0.9 for the furnace inner walls, which appears reasonable considering
the condition and colour of the refractory bricks as shown in Figure 4.2. A lower emissivity
of hr = 0.8 is used for the hearth rolls and tc = 0.8 for the thermocouples, as they are
metallic objects and more reflective than the bricks. The emissivity of the heating elements
was found by iteration in the full 3D model to be elem = 0.7.
97
The radiative heat transfer is represented as a boundary condition to the heat equation
(Equation 4.1) as follows.

∂T
−k (0, t) = Qrad (4.6)
∂x
i.e.

∂T
−k (0, t) = 1 σA1 F1−2 (T14 − T24 ) (4.7)
∂x
4.4.4 Convective Heat Transfer Calculations
4.4.4.1 General
Convective heat transfer in the MCL furnace system exists on most surfaces to some
extent, but is of most significance on the external walls of the furnace exposed to the
ambient atmosphere, and on the strip surface where movement through the furnace at-
mosphere creates forced convective heat transfer. As shown in Section 2.4.2, the rate
of heat transfer by convection at a boundary is given by Equation 2.3. Convective heat
transfer coefficients are calculated for the furnace outer walls in Section 4.4.4.2 and the
strip in Section 4.4.4.3.
Convective heat transfer is represented mathematically as a boundary condition to the

heat conduction equation (Equation 4.1) as follows.

∂T
−k (0, t) = Qconv (4.8)
∂x
i.e.

∂T
−k (0, t) = hc A1 (T1 − T∞ ) (4.9)
∂x
4.4.4.2 Convection on Furnace Walls
As there is no forced flow of air around the external furnace walls, natural convection can
be assumed to take place. Simple correlations can be used to calculate hc , the natural
convection heat transfer coefficient, for the furnace geometry. This results in coefficients
for the vertical outer walls, and the horizontal roof of the furnace. The lower horizontal
boundary of the furnace floor is shielded from airflow by the furnace support system, so
98
will experience very little heat transfer by convection.
Property Value
Density, ρ 1.16 kg.m−3
Coeff. Thermal Exp. β 0.00341 K−1
Thermal Conductivity, k 0.0251 W.m−2 K−1
Viscosity, µ 1.85e−5 Pa.S
Kinematic Viscosity, υ 1.51e−5 m2 s−1
Prandtl Number, P r 0.705
Table 4.4: Physical properties of air at 20◦ C [20, 54]
The convective heat transfer coefficient for natural convection on the outer furnace
vertical walls is calculated using physical properties of air at 20◦ C as given in Table 4.4.
Typical furnace wall temperatures as measured by NZS are around 50 − 60◦ C, giving a
typical ∆T = 30◦ C for ambient air temperatures around 20 − 30◦ C. Assuming the vertical
height of furnace walls L = 2m, Equation 2.8 gives Gr = 3.5 × 1010 .
N u = a(Gr.P r)m (4.10)
Equation 4.10 gives a correlation for the Nusselt Number for natural convection [54]. In
the case of natural convection on a vertical surface with Gr.P r > 109 , the constants
a = 0.13 and m = 1/3. This gives N u = 379.
Equation 2.4 is used to calculate hc , the natural convection heat transfer coefficient.
Under these conditions, the heat transfer coefficient hc = 4.8 W/m2 on the vertical furnace
walls. This is increased to hc = 10 W/m2 in the models for all furnace exterior walls,
including horizontal surfaces to allow for increased heat loss due to additional factors,
such as less than ideal condition of furnace insulation materials, and conductive heat loss
through the furnace superstructure which are otherwise exceedingly difficult to measure
or calculate. This assumed value gives much closer agreement to measured temperatures
than the lower calculated value.
4.4.4.3 Convection on Strip Faces
Further calculations are required to estimate the convective heat transfer coefficient on the
strip surface inside the furnace caused by its movement through the furnace atmosphere.
Previous works as described in Section 2.5 have found the heat transfer by convection to
be negligible when compared with heat transfer by radiation inside the furnace cavity.
99
The relative heat transfer coefficients by convection and radiation are calculated and
compared here for a simplified case to confirm this.
Density, ρ 0.2851 kg/m3

Thermal Conductivity, k 0.0718 W/m2 .K
Viscosity, µ 4.45×10−5 Pa.S
Prandtl Number, P r 0.748
Table 4.5: Physical properties of nitrogen gas at 927◦ C [25]
Heat transfer coefficients on the furnace and strip inner walls are calculated by con-
sidering the top half of the furnace as a duct-flow situation, where the strip constitutes
the bottom surface of the duct. The air movement is considered as flow through the duct
at 140m/min rather than the duct surface (strip) moving through stationary air. The
Reynolds number for this airflow is calculated using equation 4.11, using gas properties
of nitrogen at 927◦ C as in Table 4.5.
ρU Dh
Re = (4.11)
µ
Where:
4A
Dh = Hydraulic Diameter =
P
Using calculated area and perimeter of the furnace cross section above the strip (cross
section given in Figure 4.1):
4 × 0.987m3
Dh = = 0.879m
4.49m
0.2851kg/m3 × 2.33m/s × 0.879m

Re = = 13140
4.45 × 10−5
The Sieder-Tate Equation, Equation 4.12 [54] for turbulent duct flow (Re > 10,000)
is used to calculate the average Nusselt Number:
N u = 0.023 Re4/5 P r1/3 (4.12)

N u = 41.16
100
The convective heat transfer coefficient is calculated from this Nusselt number in
Equation 4.13.
hconv × Dh
Nu = (4.13)
k
41.16 × 0.0718
hconv =
0.879
hconv = 3.36 W/m2 .K
4.4.4.4 Comparison with Radiative Heat Transfer
To estimate the effect of convective heat transfer as compared with radiative heat transfer,
the radiative heat transfer to the strip surface is calculated for a highly simplified geometry
of a flat plate (strip) in a black environment (furnace), using strip width = 1m, view factor
from strip to walls Fsw = 1, and strip emissivity s = 0.2.
The net radiative heat transfer between the strip and walls for this simplified situation
is given by Equation 4.14 [54], using assumed wall temperature T1 = 950◦ C in the heating
zones, and strip temperature T2 = 100◦ C near strip entry, from the strip body A1 to the
furnace walls:
Q1−2 = A1 σ1 T14 − F1−2 σT24

(4.14)
Qs−w = 1m × 5.67 × 10−8 0.2 × 3734 − 1 × 12234

Qs−w = −126630 W per m length
This heat flux can be used to find the heat transfer coefficient due to radiation, which
depends strongly on the temperature difference. Assuming the same wall and strip tem-
peratures of 950◦ C and 100◦ C respectively:
Qw−s = hrad × A × ∆T
126630
hrad =
(1223 − 373)
hrad = 149 W/m2 K
This value of hrad = 149 W/m2 K is assumed for the start of the heating zones, where
the temperature difference between furnace walls and steel strip is high. The radiative
heat transfer reduces strongly as the temperature difference reduces, and hence a different
101
hrad is calculated for the cooling zones. Using assumed values of Tstrip = 700◦ C and
Twalls = 500◦ C, the radiative heat transfer coefficient hrad = 50 W/m2 K.
Temp Radiative Convective Ratio

Difference H.T Coefficient H.T Coefficient hr : hc
Heating Zones 850◦ C 149 W/m2 K 3.36 W/m2 K 44 : 1
Cooling Zones 200◦ C 50 W/m2 K 3.36 W/m2 K 15 : 1
Table 4.6: Comparison of radiative and convective heat transfer coefficients calculated for
a simplified furnace representation
Comparing the heat transfer coefficients by radiation and convection in the heating
and cooling zones, it is seen in Table 4.6 that heat transfer by radiation is much stronger
than by convection, by a ratio of 44:1 at the start of the heating zones, and 15:1 in a
typical cooling zone. It is clear that radiation is the dominant heat transfer mechanism in
this approximation, being 94% of the total in the cooling zones, and 98% in the heating
zones where the strip temperature calculation is most important. This is consistent with
previous investigations of similar furnaces [27], and validates the simplification to ignore
convective heat transfer to the strip surface in further furnace modelling.
102
4.5 3D Model Development
4.5.1 Introduction
The full 3D model is created using COMSOL Version 3.3 finite element modelling package,
and is hereafter referred to as the ‘Full 3D Comsol’ Model, or F3DC Model. It consists
of three separate models, being one model for the heating zones as shown in furnace
schematic Figure 1.2 (zones 1-12), and two models for the cooling zones - one for zones
13-14 only, and another for the remaining zones 15-19. Zone 20 (The hot bridle zone) has
not been modelled here, as the geometry and heating equipment are significantly different
to the rest of the furnace, requiring different modelling techniques, as well as being of
little significance in terms of the peak temperature and metallurgical changes in the steel
product.
The F3DC model contains most of the complex internal geometry of the furnace, in-
cluding hearth rolls, cooling tubes, thermocouples and heating elements, although several
simplifications are required to enable practical solution time of the model as described
in Section 4.5.2. The type of heat transfer equations being solved, and their location
in the 3D model are shown in Section 4.5.3, and the method of solving these equations
for steady state or transient solutions is described in Section 4.5.4. The type of output
available from the model is described in Section 4.5.5.
4.5.2 F3DC Model Assumptions and Simplifications
The two main simplifications required to enable solution of the F3DC model due to
computer power and software capabilities are the use of a vertical plane of symmetry
down the centre line of the furnace as described in Section 4.5.2.1, and the exclusion of
conductive heat transfer between bodies inside the furnace.
Difficulties in creating the calculation mesh required adjacent objects to be physically

separated so as to prevent contact between the object’s meshes. For example, the blocks
representing the thermocouple tubes do not touch the furnace roof, and the hearth rolls
do not touch either the furnace walls or the strip. In each case, furnace objects were
separated by a 1mm gap between the bodies to prevent conduction.
103
This simplification is undesirable, but will have little effect on furnace temperatures
at steady state, as most heat transfer is by radiation, which bypasses any such gaps. The
area this is likely to have the most effect is in conduction from the hearth rolls to the
strip, which is not included. The hearth rolls still transfer heat to the strip by radiation
strongly due to their proximity and temperature difference, and still behave as a thermal
mass inside the furnace during transient temperature conditions. It is likely that some
small errors are caused by this simplification, although the total length of strip in contact
with the hearth rolls is small. The small amount of missing area for conduction will have
little effect on the overall temperature response of the furnace, which is governed by the
high mass of the furnace walls.
4.5.2.1 Model Symmetry Plane
The main simplification made in creating the furnace geometry is to use a vertical plane
of symmetry down the length of the furnace. This is commonly done for conductive heat
transfer modelling to halve the number of calculation elements. This is more difficult
when considering radiative heat transfer, as any calculation element on one side of the
line of symmetry will exchange heat with areas on both sides of the line of symmetry.
(a) Full furnace geometry (b) Plane of symmetry and

mirror wall
Figure 4.15: Demonstration of using ‘mirror wall’ along furnace plane of symmetry to
maintain radiative heat transfer inside furnace cavity.
To maintain the correct radiative transfer, a ‘mirror wall’ is created to cover the
open furnace cavity, which uses an emissivity of m = 0 to completely reflect all incident
104
radiation. This scheme is demonstrated in Figure 4.15, showing that incident radiation
on the strip inside the full cavity is the same as a half-cavity with a mirrored face on the
open side.
4.5.2.2 Heater Element Representation
The electric radiant heating elements in the MCL furnace are of a complex shape, as
shown by the model of a single cast heater element in Figure 4.16. The large number
of these heating elements in the furnace results in a prohibitively large number of mesh
elements to model this shape - a coarse approximation required 364 mesh elements in
COMSOL, which over the whole furnace is 660,000 mesh elements.
Figure 4.16: COMSOL model of a single cast nichrome heating element
To keep the number of mesh elements manageable, it is possible to approximate this

serpentine heating element as a plain rectangular block. This was done in the COMSOL
model by creating a separate simulation where the dimensions of a rectangular block were
adjusted until it gave the same incident radiation on an adjacent strip as a fully-modelled
serpentine heating element with the same heat (power) input. This model is shown in
Figure 4.17.
It was found that a square block element 180mm wide x 23mm high gave the same
incident radiation to the strip per metre length of heating element as the actual cast
nichrome heating elements. These square heating blocks are the same thickness and
length as the real elements, but somewhat thinner in width, as the square blocks have
more surface area directly facing the strip than the cast elements i.e. a better element
to strip view factor for radiative heat transfer. These dimensions give the square heating
blocks approximately 3× the volume of the cast elements, and hence the density of the
elements in the model was multiplied by a factor of 0.336 to give the block elements the
same thermal mass as the real cast elements.
105
Figure 4.17: COMSOL model used to compare radiative intensity from a square heating
block (right) against a serpentine cast heating element (left)
4.5.2.3 Thermocouple Representation
In order to validate the F3DC model using real furnace measurements, it is vital to
include the thermocouples in the 3D model geometry. As discussed in Section 2.5.4,
the temperatures measured by the thermocouples can be significantly different to the
temperatures of the furnace walls due to the radiative heat transfer inside the furnace
cavity. Modelling thermocouple units of the same size and location as in the real furnace
allows comparison of the temperatures measured on the model with those recorded by
furnace thermocouples, and allows validation of the predicted variation of temperature
along the thermocouple unit itself [39, 78].
Figure 4.18: End view of F3DC model showing representation of thermocouple using a
protruding rectangular block
106
As shown in Figure 4.18, the thermocouple is modelled as a single rectangular block,
representing the middle of the three thermocouples at the ends of each zone. The ther-
mocouple is not connected to the furnace roof and so exchanges heat solely by radiation.
The temperature on the bottom face of the rectangular block is examined and compared
with the temperature measured by the furnace thermocouple probe.
4.5.2.4 Hearth Roll and Cooling Tube Representation
The F3DC model includes objects representing each individual hearth roll, with the same
external dimensions and locations as in the real furnace. The real furnace hearth rolls are
hollow, made from 12mm rolled steel plate, as per Figure 4.11. Modelling hollow hearth
rolls is a significant increase in complexity over solid cylinders, and a large increase in
number of calculation mesh elements would be required considering the number of rolls in
the furnace. The rolls have instead been modelled using solid cylinders, with the density
reduced to maintain the correct thermal mass in the furnace. The location of the hearth
rolls and representation in the F3DC model are shown in Figure 4.19.
Figure 4.19: End view of F3DC model for heating zones showing hearth rolls under the
strip
The cooling tubes in the cooling zones are represented similarly, as shown in Figure
4.20, with the hollow tubes and annular spaces for airflow not modelled and the density
of the body reduced to maintain the correct thermal mass. Note that there are a large
number of cooling tubes in the cooling zones compared to the number of hearth rolls. This
required a very coarse mesh size to be used for the cooling tubes, and also the mesh size on
107
the hearth rolls to be increased over that used in the heating zones (compare with Figure
4.19). A very coarse mesh size is needed, as increasing the number of mesh elements in
the radiative furnace cavity exponentially increases the number of calculations required,
as every mesh element in the cavity exchanges radiation with every other element. Only
the very coarse mesh shown in Figure 4.20 is able to provide practical solution times for
the model.
Figure 4.20: End view of F3DC model for cooling zones, showing cooling tubes above and
below the strip, and coarser mesh size on hearth rolls.
Heat extraction from the cooling tubes is modelled in COMSOL as a fixed rate of heat
loss from the cooling tube bodies. These are cooled from within, and extract heat from
the furnace system by radiative heat exchange with the hotter furnace walls and moving
strip. As with the hearth rolls, the modelled cooling tubes do not touch the furnace walls
to conduct heat, and only transfer heat by radiation within the furnace cavity.
4.5.2.5 Mesh Size and Spacing
Selection of the optimum mesh size for the furnace model is vital to ensure solution
times and computer power requirements are practical, whilst maintaining accuracy of the
solution. Radiative heat transfer in particular makes this difficult, as every mesh element
inside the furnace cavity exchanges radiant heat with every other element that it can ‘see’
i.e. the view factors between each and every element must be calculated in the model
solution. This is emphasised in the mesh size required for the furnace cooling tubes, which
must be very coarse to ensure a solution is possible at all, as shown in Figure 4.20.
108
Figure 4.21: End elevation of F3DC model showing irregular mesh spacing through furnace
walls, with fine mesh elements adjacent to radiative face.
Because radiative heat transfer is a surface effect only, it is useful to use an irregularly-
spaced mesh with fine mesh spacing adjacent to the radiating surfaces. Optimising the
mesh spacing allows a significant saving in calculation time, where coarse mesh elements
can be used in the body of the walls. The mesh spacing used in this study is shown in
Figure 4.21, with 4 mesh elements through the furnace walls, where the element at the
inner radiative face is one-tenth the size of the outer-most element.
It was found that this was the coarsest mesh that did not affect the steady state tem-
perature results. A very fine reference mesh with 10 elements at the same 10:1 thickness
ratio was used for comparison, and the strip temperature solution as shown in Figure
4.21 was different by less than 0.1◦ C. Coarsening the mesh spacing to 3 elements was the
first model to cause a temperature difference of over 1◦ C, and hence the 4 mesh element
spacing was used as the coarsest acceptable spacing.
109
4.5.3 F3DC Model Equations
The F3DC model solves equations for transient heat conduction in each of the furnace
body subdomains, as well as convection and radiation on selected subdomain boundaries.
Figure 4.22 shows the approximate location of each of the types of heat transfer calculated
in the model.
Figure 4.22: Location and types of heat transfer calculations performed in the F3DC
model
Conductive heat transfer in three dimensions is represented in COMSOL with the heat
equation as described in Section 4.4.2,
using nomenclature
shown in Equation 4.15, where
∂T ∂T ∂T
∇T is the divergence of T i.e. ∇T = ∂x , ∂y , ∂z [19]. This equation is active in all of
the subdomains in the furnace model.
∇ · (−k∇T ) = Q (4.15)
An additional term is added to Equation 4.15 in the subdomain representing the mov-
ing strip. Rather than moving the subdomain mesh elements in the model, an advection
term is added to move the internal heat inside the strip subdomain at a prescribed rate.
This equation active in the strip subdomain is shown in Equation 4.16, where v is the
speed of strip movement, the MCL linespeed.
∇ · (−k∇T ) = Q − ρcp v · ∇T (4.16)
110
There are three distinct types of equation applied to the subdomain boundaries in
the F3DC model, which are either insulation, convection or radiation. An insulation
boundary does not permit heat transfer through in either direction, and hence is used
on all furnace walls which lie on the plane of symmetry i.e. where the same heat flux
approaches from each side of the plane. These boundaries are represented by Equation
4.17, which is the same as that described in Equation 2.18 for an insulating boundary.
−n · (−k∇T ) = 0 (4.17)
Boundaries on which a heat flux is present such as convective heat transfer use Equa-
tion 4.18, which is similar to Equations 2.17 for a specified heat flux, and 2.19 for a
specified heat transfer coefficient. This allows specification of either a constant heat flux,
or a constant heat transfer coefficient. All convective heat transfer boundaries in the
F3DC model use q0 = 0 i.e. no set heat flux. Heat transfer coefficients are as described
in Section 4.4.4.
−n · (−k∇T ) = q0 + h(Tinf − T ) (4.18)
Boundaries on which radiative heat transfer occurs use two equations, representing
both the irradiation, G, or radiative heat energy reaching the boundary, and the radiosity,
J, the radiative heat energy leaving the boundary. These are shown as Equations 4.19
and 4.20 respectively. The view factors between each mesh element are calculated in
the COMSOL software using a ‘hemicube’ method, where rays are projected from each
element, which identifies obstructions in the furnace cavity.
−n · (−k∇T ) = q0 + h(Tinf − T ) + (G − σT 4 ) (4.19)

(1 − )G = J0 − σT 4 (4.20)
4.5.4 F3DC Model Calculation Procedure
The F3DC model can be used to solve either the steady state solution to a set of given
input conditions, or can deliver the transient solution over time when a change is made
to existing input conditions. The model itself always runs in a fully transient state,
starting from defined initial conditions and the calculation proceeding for a specified time
duration. If a sufficient number of time-steps (simulated time duration) is allowed, the
111
calculations will approach the steady state solution i.e. where there is no further change
in subsequent calculations. It has been found by testing that a simulation time of 1 × 105 s
is sufficient to find the approximate steady state solution - from any given conditions
changes in temperatures in the furnace will have abated after approximately 27 hours of
simulated time.
To find the steady state solution for the heating zone model, there are several initial
conditions that must be specified as shown in Figure 4.23. These are:
1. Strip dimensions (width × gauge)
2. Strip entry temperature (post induction preheaters)
3. Strip speed (MCL linespeed)
4. Zone power input (%) into each zone
The overall method of operating the F3DC model is shown in Figure 4.23. This shows
that a steady state solution for the whole furnace system can be calculated using the
above variables and the combination of all three 3D models. The order of operation of
these models is shown in Figure 4.24. These must be solved in order, as the exiting strip
temperature from the first model (heating zones 1-12) is required as input to the second
model (cooling zones 13-14), and similarly for the final model (zones 15-19).
Once the steady state solution is found, a ‘disturbance parameter’ can be introduced
i.e. a change made to one of the input conditions used to calculate the steady state
solution. Using the full 3D steady state solution as the initial condition, the disturbance
parameter can be changed in the COMSOL interface, and the model re-solved over a new
time period to follow the effects on the furnace as the temperatures change over time. For
example, the power to a heating zone can be changed, and the model solved to show the
zone and strip temperatures changing due to the zone power change.
The transient solution can be calculated for any desired time duration. A long time
duration of several hours or longer can be used to approach the next steady state of
the furnace, or a short time duration of several minutes can be used to see the immedi-
ate impacts on the strip and furnace temperatures of control actions such as increasing
linespeed.
The F3DC model is calculation-intensive, as each of the three separate models took
2-3hrs to solve for the large number of time steps required to reach steady state. The need
112
Figure 4.23: Summary diagram of F3DC model operation including inputs, calculation
pathway, outputs and assumptions
to run these models sequentially as in Figure 4.24 means that the steady state solution
of the entire furnace takes a long time to calculate. It is instead possible to use only
the heating zones model for steady state and transient calculations, as the peak strip
temperature is likely to be achieved at the end of the heating zones, which is the value
of most interest to modelling plant conditions. The heating zones model (zones 1-12) can
be run independently from the cooling zone models, but the cooling zone models require
input from the previous zone models.
4.5.5 F3DC Model Outputs
The advantage of creating a full three dimensional model is that the temperature results
can be seen and extracted for any point, line, face or body at any of the stored time steps.
Time step length and total simulation duration can be specified as required before the
113
Figure 4.24: Order of operation of the F3DC model components representing three dif-
ferent sections of the furnace
model is solved, such as storing data every 10s over 30min duration, or every 10min for
6hrs duration etc.
Extracting temperature results from the model requires selection of a particular point,
line or area of data. Line data is useful for examining the temperature profile down the
length of the furnace at a position on the wall or strip at steady state, as given in Section
4.5.5.1. Steady state results may also include area selections to measure the average
temperature or heat flux over a surface such as the furnace side walls. Point data is most
useful for examining the transient temperature changing over time at a particular location
as described in Section 4.5.5.2.
4.5.5.1 F3DC Model Steady State Results
The steady state solution of the F3DC model can be examined as a 3D rotatable image,
however this is not easily compared against plant measurements or other model results.
114
Extracting the temperature data as a line down the length or across the width of the
model is the easiest for comparison, as alternative cases can be easily plotted on the same
axis. Extracting this data relies on the position chosen on the 3D geometry to place the
‘sampling line’, as shown in Figure 4.25.
Figure 4.25: Sketch of furnace showing location of steady state ‘line’ output data for
furnace wall or strip centre-line temperature
The sampling location is easily chosen for strip temperatures, as the strip is long and
thin i.e. approximately a flat plane. The most useful temperature results to extract are
either the strip centre-line temperature, which is the lowest temperature of the strip,
or the strip edge temperature, which is the hottest strip temperature at any length (x)
position. A line across the strip width at a chosen x-position is used to show the effect of
differential edge heating of the strip, where the edges are hottest due to the proximity of
the furnace walls, as described in Section 2.5.4.
A meaningful sampling location for the wall temperature is more difficult to select, as
there is variation in the temperature around the perimeter of the furnace cavity at any
given x-location, as shown in Figure 4.26. In any single heating zone, the temperature
of the inner furnace wall can vary by as much as 50 − 100◦ C. As these wall tempera-
tures cannot be directly measured in the real furnace, it is more useful to examine the
temperature measured at the bottom face of each of the thermocouple blocks shown in
115
Figure 4.26: Example of temperature variation around the perimeter of the furnace cavity
at a fixed x-position
Figure 4.18. The temperatures extracted from these locations can then be meaningfully
compared with thermocouple measurements from the real furnace, as the thermocouples
should be in contact with the inner bottom surface of the protective sheath.
It is also possible to extract and plot surface data, such as the temperature along the
full length of the strip or furnace walls. The radiative flux is another useful result that
can be plotted over an area, and can give a visual indication of the effect of the furnace
geometry on the radiative heat transfer. Similar to plotting the full 3D result, a 2D area
plot is difficult to compare numerically against other alternatives. The results can be
averaged over the area however, which can be easily compared against other models.
When examining the steady state results from the 3D furnace model, the most useful
results are taken from the lines down the centre and edge of the strip, which will be referred
to as the strip centre temperature profile and the strip edge temperature profile. The other
major result to examine is the temperature of all furnace thermocouples, extracted as
a single value at the bottom of each thermocouple, for comparison with real furnace
measurements. These points are referred to as the model thermocouple temperatures.
4.5.5.2 F3DC Model Transient Results
Once the furnace model has been solved to a steady state solution, it can be used as a
base point to solve a furnace situation which changes with time. The steady state solution
represents the heat balance for any particular combination of furnace setup and strip sizes
and properties. From here, it is possible to change any number of the model parameters,
such as power input to any zone, strip speed, strip width or gauge etc. The model will
116
then solve the problem iteratively for any specified time duration. The temperature results
can then be plotted to show the response of wall or strip temperatures to the change in
parameters.
Figure 4.27: Comparison plotting the transient wall temperature result due to a step
change in heater power by either the full wall temperature profile or by selection of a
single point
Examining the results for the transient solution to a change in operating parameters
is best done by choosing a single point to follow over time. A line plot can be used to
plot the whole strip or wall temperature profile over several chosen time steps, but the
rate of change of temperature is very difficult to gauge. Figure 4.27 shows the comparison
between plotting either the full wall temperature profile at several time steps during the
transient solution, and plotting a single point changing continuously over the same time
duration.
This example shows the transient response due to a sudden step change in heater
input power to one heating zone. The wall temperature profile change shows the spike
in temperature due to the power change and the length of the furnace affected by the
change, but it is difficult to gauge the rate of change. The single point plot gives a good
indication of the rate and magnitude of change, but does not show the effect on the rest
of the furnace i.e. how much of the furnace is affected by this change.
117
The best method to examine model results for transient furnace simulations is therefore
a combination of both of these plots, where the furnace wall and strip temperature profiles
are examined at steady state, and at conclusion of the transient simulation to gauge
the extent of the temperature effect. Plotting single points of interest, such as on a
thermocouple or the peak strip temperature should be used to gauge the rate of change,
and the time required to approach the next steady state.
Results from the F3DC model and validation with actual furnace measurements during
operation are shown in Section 5.3.
118
4.6 Simplified Model Development
4.6.1 Introduction
While the F3DC model accurately represents the furnace geometry and radiative heat
transfer in three dimensions, the complexity of calculations requires a long solution time
of up to 9hrs for the full furnace model. The F3DC model cannot be used to improve
control of the strip quality, production rate or energy efficiency, as it would be impractical
to simulate large numbers of alternative options in a production situation to find the
optimum settings. The purpose of the simplified model is to solve furnace and strip
temperatures using a simplified equation system, which can be done rapidly to enable
furnace control optimisation.
The model is simplified by reducing the complex 3D furnace geometry to a 2D ap-

proximation for the furnace body, and a 1D approximation of the moving strip, linked by
radiative heat transfer calculations and energy balances. This simplified model is hereafter
referred to as the ‘Reduced Dimensionality Heat Transfer’ model, or the RDHT model.
The assumptions required to create the RDHT model, and the method of operation
are detailed in Section 4.6.2. The simplification of the furnace geometry is shown in
Section 4.6.3, and the method of solution of model equations shown in Section 4.6.4.
The approximated radiative heat transfer equation is given in Section 4.6.5, and the heat
balance between the heat transfer to the strip and walls given in Section 4.6.6.
The equations required to model the strip are derived in Section 4.6.7, and the finite
difference approximation made to represent the furnace walls is derived in Section 4.6.8.
The available output types from the RDHT model are described in Section 4.6.9.
4.6.2 RDHT Model Assumptions and Operation
The overall operation of the RDHT model is very similar to the operation of the F3DC
model, as it requires similar types of inputs, and gives similar outputs with reduced
complexity. Figure 4.28 gives a summary of the calculation steps in the RDHT model,
including the inputs required, the outputs available, and the main assumptions made.
Many assumptions are made to reduce the equation system for the RDHT model,
relating to either the heat transfer equations, the geometry of the system modelled, or
119
Figure 4.28: Summary diagram of RDHT model operation including inputs, calculation
pathway, outputs and assumptions
the physical properties of the bodies modelled. The main assumption in the heat transfer
equation is the simplification of the radiation equation, which is detailed in Section 4.6.5.
The heat transfer is also simplified by not modelling other solid bodies inside the furnace
system, such as the hearth rolls, cooling tubes or thermocouples. The furnace is modelled
as a single object, interacting with the strip by radiation only, and the model does not
consider conductive heat transfer through the hearth rolls to the strip.
The heat removal from the furnace system by the gas jet coolers and cooling tubes
is not modelled explicitly including radiative heat transfer to the cooling tubes, but is
considered as a heat sink in the furnace walls in each zone. The radiant heating elements
are also not included individually, and the power applied to each zone is instead modelled
as heat source in the furnace zones.
Because the thermocouple temperatures are not individually calculated in the RDHT
model, an assumption must be made to compare the calculated wall temperatures against
120
furnace thermocouple temperatures to validate against measured operational data. This
is investigated further in Section 5.3.3.
A major difference between the operation of the RDHT model and the real furnace is
the control of the radiant heating elements. In the real furnace, these are PID controlled
to a furnace zone temperature setpoint, but in the RDHT and F3DC models, they are
necessarily modelled as a given power input. This is unavoidable, as it is necessary to
know the amount of power input to each zone to calculate the heat balance to the furnace
walls. Including PID control of this power input would require further modelling and
tuning to match the NZS furnace control computers. This PID control is liable to change
either tuning or design in the future, but the amount of power supplied is a measurable
quantity that will always be available.
4.6.3 RDHT Model Geometry
Simplification of the 3D geometry is the main method by which the equation system is
reduced for the RDHT model. This greatly reduces the complexity of the heat trans-
fer calculations for 3D conduction in the furnace bodies, and 3D radiation between all
surfaces inside the furnace cavity. To achieve this, the furnace walls are reduced to a
2D approximation representing the thickness of the walls, but not variation around the
perimeter of the furnace cavity. The strip is reduced from a 3D body to a 1D line, as the
strip is very thin compared to its width and length i.e. approximately two-dimensional,
and is easily simplified further. This does not allow estimation of geometry effects such
as the edge heating of the strip, but these can be examined using the F3DC model, as in
Section 5.3.4.
Figure 4.29 shows the geometry scheme used in the RDHT model, with a 2D model
of the furnace walls and 1D model of the strip, as well as the type and location of heat
transfer calculations used. This can be contrasted with the geometry and calculations
shown in Figure 4.22 for the F3DC model.
Rather than constant grid spacing as used in the F3DC model, the RDHT model uses
a unequal grid spacing for both the strip and furnace walls. The length of each zone is
divided by the nearest integer to give approximately 1m long calculation mesh spacing in
each zone, which makes the mesh lengths add up to exact zone lengths. This allows the
zone power inputs to be easily discretised over the number of calculation points for each
zone. An equal grid spacing would mean zone power inputs would be ‘smeared’ across
121
Figure 4.29: Schematic diagram showing geometry of the RDHT model, and type and
location of heat transfer equations used
the divisions between zones. The Zone lengths, number of grid points and grid lengths
are shown in Table 4.7. Including a point at x = 0, this adds to 158 mesh points along
both the furnace walls and strip in the x-direction.
4.6.4 RDHT Model Calculation Method
The overall model calculation method is shown in Figure 4.28. The MATLAB software
package is used to programme the model equations, solution methods and numerical and
graphical outputs. Once the required inputs and calculation parameters are specified, the
model equations are solved iteratively until reaching the defined number of time steps.
Matrices of the strip enthalpy, strip temperature, wall temperature, power inputs and
heat fluxes are stored during the calculation. The calculation domain is divided into 158
mesh points in the x dimension (furnace length), giving grid length δ ≈ 1m length for
each calculation cell as per Table 4.7.
The time division into N number of steps of k duration is chosen carefully to ensure
stability of the iterative solution. To maintain stability of the explicit solution to the
partial differential equations for the strip and wall temperatures, the Courant-Friedrichs-
Levi (CFL) condition must be met [18]. In the case here with variable grid spacings, the
122
Zones Zone Number Grid
Length, m Grids Spacing (δ), m
1 5.395 5 1.08
2-8 5.461 5 1.09
9 5.414 5 1.08
10 6.07 6 1.01
11 6.147 6 1.02
12 6.299 6 1.05
13 9.195 9 1.02
14 9.271 9 1.03
15 9.424 9 1.05
16 13.24 13 1.02
17-19 13.354 13 1.03
20 15.6 15 1.04
Table 4.7: Zone lengths, number of grid points per zone, and distance between grid points
used in the RDHT model
smallest spacing (δ) is used.
kv
0< ≤1 (4.21)
δ
This states that the distance travelled by the strip during the course of one time division
must not be greater than the length of one x division, hence the number of time divisions
is affected by the speed of advection. The time divisions must be shortest for the fastest
speeds of strip movement, but can be increased for slower-moving strip calculations.
The model matrices used for the MATLAB steady state solution are named as follows:
H(x), H2(x) = 1D strip enthalpy, J

ST(x), ST2(x) = 1D strip temperature, K
TW(x,y), TW2(x,y) = 2D wall temperature, K
Q(x) = radiative heat transfer to strip, W/m
PHI(x) = heat flux to furnace walls, W/m
POWER(x) = furnace applied power at each grid point, W/m
The strip enthalpy and strip and wall temperature values are not stored at every time
step due to the large number of time steps required and memory restrictions. Only the
values at current time and the next time step are stored, and they are replaced in each
123
calculation step. The current time step values are stored in H, ST, T W and the next
time step values stored in H2, ST 2, T W 2. At the end of calculations for each time step,
H, ST, T W are replaced by H2, ST 2, T W 2 for the next time step calculation.
The y-coordinate in the T W wall temperature matrix represents the thickness through
the wall, where y = 0 is the inner face of the furnace cavity. The Q and P HI matrices are
re-calculated at each iteration and use 1-dimensional vectors to reduce memory usage.
The P OW ER vector is constant for a steady state solution, or may be time-dependent

for a transient solution where the furnace input power changes over time. The vector holds
the zone power inputs to the furnace, discretised into 158 grid points along the furnace
length. The power inputs are the actual applied power per zone (Watts) divided by the
number of calculation points per zone as in Table 4.7. The applied power is calculated
by the maximum available power in each zone, depending on the element condition as
in Section 4.2.1, multiplied by the percentage power applied to each zone as recorded in
operational data.
The matrices used in the transient solution are changed slightly:
ST(x,t) = 1D strip temperature over time

TWY0(x,t) = Wall temperature at y = 0 over time
In the transient simulation, strip temperature is stored at every time step in order to
allow examination of strip temperature changes over time. The wall temperature is also
saved at every time step, but only the value at y = 0, the inner wall temperature in
the radiative cavity. This can be tracked over time, and compared against thermocouple
temperatures.
The strip heat equation is calculated using the H(x) enthalpy matrix, and the enthalpy
converted to a temperature at each iteration, which is needed to calculate the radiative
heat transfer, and also to output results as a meaningful value. The general iterative
calculation procedure at time t is:
1. Calculate Q flux to strip using existing strip and wall temperature at time t
2. Solve strip heat equation at time t + 1 using Q heat input
3. Convert strip enthalpy to temperature at t + 1
4. Calculate heat flux to walls value PHI
124
5. Solve wall heat equations to give new wall temperatures at t + 1
6. Move to next time point t + 1
Iterating for a large number of time steps give the steady state result for strip and
wall temperature. A transient calculation is performed by running the model again using
the steady state result as the initial guess for the new solution.
The radiative heat transfer from the furnace to the strip is calculated using equations
shown in Section 4.6.5, and the heat balance for the each calculation grid calculated as
in Section 4.6.6. The movement of heat in the strip is calculated using the heat equation
given in Section 4.6.7, and the two-dimensional conduction equations for the heat in the
furnace walls is given in Section 4.6.8.
4.6.5 Radiation Equation Approximation
The calculation of radiative heat transfer in the furnace is a major simplification in the
RDHT model, compared to the F3DC model. The RDHT model uses the radiation
equation given in Equation 2.22 from the MISG investigation, and view factors given in
Equations 2.23 to 2.26. This radiation equation considers radiation between the strip and
walls only in concurrent calculation grids i.e. at grid point x, the radiative heat transfer
is only considered between wall and strip both at grid point x, and not x − 1 or x + 1
mesh points, which are approximately 1m away in either direction.
This assumption is reasonable, as the view factors between offset grid points are much
lower than between concurrent grid points in this furnace geometry. This is tested using
an equation for view factors between two finite rectangles on a parallel plane given by
Howell [23]. The equation used to calculate these view factors is given in Equations 4.22
and 4.23, and the geometry showing the location of the measurements required in Figure
4.30.
2 2 2 2
1 XXXX
F1−2 = (−1)(i+j+k+l) G(xi , yi , ηk , ξl ) (4.22)
(x2 − x1 )(y2 − y1 ) l=1 k=1 j=1 i=1
125
!
1 1/2 y−η
(y − η) (x − ξ)2 + z 2 tan−1

G=
2π [(x − ξ)2 + z 2 ]1/2
!
1/2 x−ξ
+ (x − ξ) (y − η)2 + z 2 tan−1

[(y − η)2 + z 2 ]1/2
!
z2
− ln (x − ξ)2 + (y − η)2 + z 2

(4.23)
2
Figure 4.30: Geometry used to calculate view factors between finite rectangles on parallel
planes [23]
Figure 4.31: Representation of parallel rectangles approximation used to evaluate view

factors between furnace and strip
The view factor between strip and furnace is approximated as parallel rectangles 1.8m
wide × 1.0m long for the furnace roof or floor, with a rectangle 1m wide × 1m long for
126
the strip, separated by 0.4m (half the furnace cavity height). View factors are calculated
using equations 4.22 and 4.23 between the strip and floor in concurrent zones, as well as
in offset zones. This scheme is shown in Figure 4.31.
Strip Location Floor Location View Factor Fsf

x x 0.35
x x+1 0.067
x x+2 0.0037
x x+3 0.0007
x x+4 0.0002
SUM 0.42
Table 4.8: View factors between furnace strip and furnace floor at increasing grid position
offsets
The view factor between the strip at location x and the furnace floor at the same
location x in this approximation is Fsf = 0.35. This rapidly reduces when considering
furnace floor sections offset from location x, as shown in Table 4.8. In this calculation,
the view factor between the strip and a grid section of furnace floor offset by only 2m
(x + 2) is 100× less than to the floor in the concurrent grid section.
Because the view factors are so low, there will be little radiative heat exchanged
between offset grid sections i.e. most of the radiative heat exchange occurs between the
strip and furnace walls in the same grid sections. This shows that simplifying the radiative
heat exchange to only occur between the strip and walls within the same 1 metre long grid
sections is reasonable. This may be improved slightly by including grid sections offset by
one position (x ± 1), but the view factors are negligible to sections offset further (x ± 2 or
more). While the view factors to grid sections offset by ±1 are relatively large compared
to that for concurrent grid sections, especially when including both x + 1 and x − 1, the
temperatures at these wall grid points are very close to the temperature at position x and
so including the offset points will not strongly effect the overall radiative heat transfer
calculation. This is shown by the furnace temperature profiles in Figure 4.13, where there
is a gradual temperature change along the furnace, excluding the large step change at the
start of the heating zones, noting that each zone is divided into 6-10 grid points each.
The radiative heat transfer calculated at each grid position in the RDHT model is
calculated in MATLAB using Equation 4.24. The heat flux calculation uses the strip
temperature and furnace inner wall temperature (y = 0) at time t to calculate the radiative
exchange at time t+1. The heat flux Q in Joules is the heat applied to the strip in Equation
127
4.31, and is also used to calculate the heat flux to furnace walls in Equation 4.25.
2ws σ
× T W (x, 0)4 − ST (x)4 × δ(x)

Q(x) = s (1−w ) w
(4.24)
1+ w p
4.6.6 Heat Balance Calculation
The RDHT model uses calculations given in the MISG investigation [39] for the heat
balance in each calculation grid point. The heat input at each grid point is defined by the
power input to the heating elements in each zone, divided by the number of grid points
falling inside each zone. The radiative heat transfer equation gives the amount of this
input heat transferred to the strip, and the remainder is the heat input to the furnace
walls, φ(x) for each grid position x, calculated in MATLAB using Equation 4.25. This
heat balance is shown schematically in Figure 4.29.
P OW ER(x) − Q(x)
P HI(x) = (4.25)
2p × δ(x)
Where:
PHI = Heat flux to walls, W/m2
POWER = Power applied to furnace heaters per grid section, W
Q = Radiative heat transfer to strip per grid section, W
p = Furnace cavity height + width, m
4.6.7 Strip Heat Equation Approximation
The strip temperature is calculated by a finite difference approximation to the heat equa-
tion. The heat equation for conduction in a 3D solid with advection given in Equation
4.5 is simplified by ignoring heat conduction in the x − direction (down the length of the
furnace) and the y − direction (across the width of the strip). The rate of conduction in
these directions is shown to be negligible compared to the rate of heat advection by the
movement of the strip by McGuinness and Taylor [39]. The heat equation used is given
in Equation 4.26.

cp ρ ∂T ∂T q
+v = (4.26)
k ∂t ∂x whδ
128
The difference equation is written by finding approximations to the ∂T
∂t
and ∂T
∂x
terms.
Because all of the heat is travelling in the same direction i.e. in the direction of strip
movement, a first-order upwind approximation to the partial differential equation is used.
This uses a forward difference approximation for ∂T ∂t
(Equation 4.27), and a backward
difference approximation for ∂T
∂x
(Equation 4.28).
∂T Tx,t+1 − Tx,t
= (4.27)
∂t k
∂T Tx,t − Tx−1,t
= (4.28)
∂x δ
Where k = step size in time (∆t)

δ = step size in x-direction (∆x)
The PDE now is written as the difference equation:
Tx,t+1 − Tx,t Tx,t − Tx−1,t q

+v = (4.29)
k δ ρs Cp,steel whδ
The difference equation is solved iteratively by rearranging to find an expression for

Tx,i+1 . This gives the solution for all x-values of temperature at time i + 1 based on the
current values at time i. This is the explicit solution to the PDE.
kv kv qk
Tx,t+1 − Tx,t + Tx,t − Tx−1,t = (4.30)
δ δ ρs Cs wh

kv kv qk
Tx,t+1 = 1 − Tx,t + Tx−1,t + (4.31)
δ δ ρs Cs wh
This is written in the MATLAB code using the ST and ST 2 matrices as in Equation
4.32 for the steady state solution. Equation 4.33 is used instead for a transient solution,
where the strip temperatures are stored at all time steps in an expanded ST matrix, and
the ST 2 matrix is not required.

kv kv Q(x)k
ST 2(x) = 1 − ST (x) + ST (x − 1) + (4.32)
δ δ Cp,steel ρs whδ

kv kv Q(x)k
ST (x, t + 1) = 1 − ST (x, t) + ST (x − 1, t) + (4.33)
δ δ Cp,steel ρs whδ
129
The only boundary condition defined for the strip difference equation is the temper-
ature at entry to the furnace (i.e. at x = 0), as measured by the contact thermocouple
described in Section 4.2.3.3. This is represented in the MATLAB code with Equation
4.34. In the case of a transient simulation, the initial temperature Tinit may change as a
function of time if the power to the induction heaters changes, or the strip size or speed
changes as per equation 4.35.
ST (0) = Tinit (4.34)

ST (0, t) = Tinit (t) (4.35)
4.6.8 Wall Finite Difference Approximation
The conduction in the furnace walls is solved by a finite difference approximation to

the heat equation PDE. As with the F3DC model, an irregular calculation mesh is used
to discretise the space, where finer elements are used adjacent to the inner walls where
radiative heat transfer is occurring.
Figure 4.32: Furnace walls represented as a 2D calculation mesh, including notation for
calculation point spacing.
The mesh design for the walls is shown in Figure 4.32, with notation indicating the
direction of adjacent mesh grid points. For each calculation point Tx,y , the notation sn
refers to the multiplier of a mesh spacing distance ‘h’, to the north, or the distance from
y = n to y = n + 1. Similarly, the length ss is the multiplier for the south distance, and
se and sw are the east and west multipliers.
130
The heat equation for conduction in three dimensions as given in Equation 4.1 is
simplified to two dimensions in Equation 4.36 for use in the RDHT model.
∂ 2T ∂ 2T

∂T
cp ρ =k + (4.36)
∂t ∂x2 ∂y 2
The Laplace operator ∇2 T is discretised using a general formula for a non-equidistant

grid in Equation 4.37 given by Grossman, Roos and Stynes [18].
"
2 1 1
∇2 u(x, y) = 2 u(x + se h, y) + u(x − sw h, y)
δ se (se + sw ) sw (se + sw )
1 1
+ u(x, y + sn h) + u(x, y − ss h)
sn (sn + ss ) ss (sn + ss )
#
1 1
− + u(x, y) (4.37)
sw se sn ss
This is calculated by setting the grid spacing distance h = 1m. The multipliers se and
sw depend on the x-grid spacing, and the multipliers sn and ss vary depending on the
y-coordinate spacing, and each factor is calculated individually at each calculation step
in the RDHT model.
For grid notation (xi , yk )

with h = 1m:
se (x, y) = xi+1 − xi (4.38)
sw (x, y) = xi − xi−1 (4.39)
sn (x, y) = yk+1 − yk (4.40)
ss (x, y) = yk − yk−1 (4.41)
The difference equation is constructed using a first-order upwind difference to the

wall temperature at time t, and approximation to the Laplace operator in Equation 4.37
with mesh spacing factors in Equations 4.38 to 4.41. The difference equation is given by
Equation 4.42.

Cp ρ Tx,y,t+1 − Tx,y,t
= ∇2 T (x, y) (4.42)
kwall ∆t
131
The finite difference approximation to the 2D heat equation that is solved in the
RDHT model MATLAB code is shown in Equation 4.43. Rather than storing the wall
temperature result for all 5 y grid points at each time step, only two matrices are used.
Matrix T W (x, y) represents the wall temperatures at each grid point at time t, and matrix
T W 2(x, y) represents the wall temperatures at time t + 1. At each time step, the T W
matrix is overwritten by the T W 2 matrix, and a new T W 2 matrix is calculated.
"
kwall T W (x + 1, y) T W (x − 1, y)
2∆t
T W 2(x, y) = T W (x, y) + +
ρCp δ2 se (se + sw ) sw (se + sw )
#
T W (x, y + 1) T W (x, y − 1) 1
+ + − T W (x, y) 1 + (4.43)
sn (sn + ss ) ss (sn + ss ) sn ss
In the case of transient temperature solutions, a new matrix is created to store the
furnace inner radiating face temperature only, as this can be roughly correlated with
plant measurements. This matrix T W Y 0(x, j) represents the furnace wall temperature
at position y = 0 for each time step j, as in Equation 4.44.
T W Y 0(x, j) = T W 2(x, 0) (4.44)
Equation 4.43 is solved for all mesh coordinates that do not lie on the mesh boundaries
i.e. where x 6= 0 or 158 and y 6= 0 or 4. There are three boundary conditions defined for
the walls, which are the insulated boundaries at x = 0 and x = 158, radiative heat flux
at y = 0 and convective heat loss at y = 4. These boundary conditions are described in
Sections 4.6.8.1 to 4.6.8.3.
4.6.8.1 Insulated Boundary Condition
The ends of the furnace walls at x = 0 and x = 158 are modelled as insulated boundaries,
which prevent heat flow across the boundary. This condition is a simplifying approxi-
mation, however the surface area on the furnace ends is very small relative to along its
length, and heat losses in this direction will be minor.
132
To have an insulated boundary with no heat transfer, the heat flux approaching the
boundary must be equal to the heat flux leaving the boundary as in equation 4.46.
∂T
=0 (4.45)
∂x
Tx+1,y − Tx,y
=0 (4.46)
∆x
This is done in the numerical scheme by specifying the temperature at the boundary
to be equal to the bulk wall temperature one grid point in from the boundary, so that
there is no temperature gradient at the boundary. This is represented in MATLAB by
Equations 4.47 and 4.48.
T W 2(0, y) = T W (1, y) (4.47)

T W 2(158, y) = T W (157, y) (4.48)
4.6.8.2 Radiation Boundary Condition
The radiative flux Φ(x) reaching the wall inner boundary is calculated by Equation 4.25. A
heat balance on the boundary shows that the heat flux approaching one side by radiation
must be balanced by the same heat flux propagating through the furnace walls, as in
Equation 4.49 and the approximation in Equation 4.50.

∂T
k = Φ(x) (4.49)
∂y y=0

T (x, 0) − T (x, 1)
k = Φ(x) (4.50)
∆y
The temperature is calculated in MATLAB using Equation 4.51.
P HI(x)sn (x, 0)
T W 2(x, 0) = T W (x, 1) + (4.51)
kwall
133
4.6.8.3 Convection Boundary Condition
The heat flux leaving the wall outer boundary at y = 4 is calculated using Newton’s Law
of Cooling, as in Equation 2.19 for natural convection. This is approximated by Equation
4.52, and rearranged and shown using the RDHT model matrices in Equation 4.53 for
y-coordinate y = 4.

Tx,y−1 − Tx,y
k = hA(Tx,y − T∞ ) (4.52)
∆y

hAss hAss
T W 2(x, 4) = T W (x, 3) − Tinf / 1 − (4.53)
kwall kwall
4.6.9 RDHT Model Outputs
As per the F3DC model outputs described in Section 4.5.5, the RDHT model may output
data for steady state or transient solutions. Unlike the F3DC model, there is no ambiguity
in selecting a location on the strip or furnace walls to output temperature results from.
The one dimensional strip temperature or two dimensional wall temperature results may
be output from the final time step only for the steady state simulation. A transient
simulation allows output of either the temperature profile down the furnace or through
the furnace walls at any chosen time step, or from any temperature points changing over
time during the transient simulation. The steady state results are described in Section
4.6.9.1, and the transient temperature results are described in Section 4.6.9.2.
4.6.9.1 RDHT Model Steady State Results
Due to the potentially large number of time steps required to approach the steady state
solution, the RDHT model does not store temperature results at every time step when
calculating the steady state solution. The only outputs available from the RDHT model
steady state solution are the temperatures at any calculation mesh location at the final
time step (at steady state).
The steady state result allows plotting of the strip temperature profile down the length
of the furnace. The furnace wall temperature profile may be plotted at any y-coordinate
depth in the wall mesh, but the most useful wall temperature is at the inner wall of the
radiating cavity at y = 0. This wall temperature result does not refer to any specific
position on the wall, but is a ‘general’ wall temperature obtained by approximating the
134
Figure 4.33: Sketch of furnace showing approximate location of steady state temperature
results for furnace wall or strip temperature profiles
3D wall shape as a 2D section. The difference between this and the F3DC model results
are shown by Figures 4.25 and 4.33, where the RDHT model result represents the whole
body temperature on average, rather than a specific point or location on the body.
4.6.9.2 RDHT Model Transient Results
Similarly to the F3DC model transient results in Section 4.5.5.2, the best way to visualise
the transient temperature results of the RDHT model is by selecting a single point to plot
temperature change over time. Figure 4.34 shows this graphically, where a single coor-
dinate along the furnace length is tracked over time during changing furnace conditions,
such as a step change in power input or other furnace variable.
Figure 4.34: Sketch of furnace showing transient temperature results for furnace wall or
strip temperature at specified location
A transient temperature model can be constructed by changing any furnace input or

control variable at the start of the period of transient operations, as detailed in Section
135
4.6.2, or these variables may be entered into the RDHT model as a function of simulation
time. For example, the furnace input power can be changed as a function of time to match
the power applied by the furnace PID controllers, or may be changed in steps at specified
points in the simulation i.e. the disturbance variables that are different from the steady
state settings may be applied at any time or as a function of time during the transient
simulation.
Figure 4.35: Example of transient simulation results with power input changes applied at
different times
Figure 4.35 shows an example of this, where the power applied to furnace Zone 7 is
reduced immediately at the start of the transient simulation, and power to Zone 8 is in-
creased after a time delay. This simulation shows the strip temperature initially declining,
and then rapidly recovering after power to Zone 8 has been increased to compensate for
the loss of power in Zone 7.
136
4.7 Metallurgical Modelling
4.7.1 Introduction
Integration of a metallurgical model into the furnace model is vital to enable decision mak-
ing on furnace control using the model results. A metallurgical model is used to convert
the model temperature results into a measure of the metallurgical changes experienced
by the steel. In this way, the effectiveness of any heat treatment scheme can be analysed,
by understanding whether the calculated temperatures are sufficient to recrystallise the
steel fully in the case of ‘soft iron’ recrystallised steels, or to ensure that recrystallisation
does not occur in ‘hard iron’ recovered steels.
The metallurgical model described here is coupled to the temperature calculations of

the RDHT furnace temperature model. The coupled temperature-metallurgical model is
hereafter referred to as the ‘Reduced Dimensionality Heat Transfer - Coupled Metallur-
gical Model’, or RDHT-CM Model.
The process of fitting metallurgical model equations to NZ Steel recrystallisation ex-

periments on MCL steel grades is shown in Section 4.7.2. The model results are compared
against the recrystallisation ‘start’ and ‘complete’ temperatures chosen from the recrys-
tallisation trials in Section 4.7.3. Section 4.7.4 shows how the metallurgical model is
integrated into the RDHT model, and used to predict strip recrystallisation caused by
the calculated strip temperature profile.
4.7.2 Equation Fitting
The metallurgical model equations were fitted to selected strip sizes and types by aligning
the measured recrystallisation softening curves with the calculated recrystallisation curves
using the ‘improved model’ Equations 2.35 and 2.36. The following parameters from the
‘improved model’ [44] were also used in these equations to suit the steel types in the MCL
as developed in Section 2.7.2.
137
Strip properties :
= Cold reduction of strip
˙ = 1000s−1
D = 10µm (hot rolled)
or 20µm (warm rolled)
Equation Parameters :
p = −4.3D−0.169 (4.54)
q = −0.53 (4.55)
s=1 (4.56)
A = 3.754 × 10−4 exp(−7.869 × 10−5 Qrex ) (4.57)
The formula used for activation energy is modified from the ‘improved model’ equation
proposed by Medina et al [44] as discussed in Section 2.7.2.2. The factors for Nb and Ti
concentration were removed due to the inapplicable range of the Medina models, but it is
understood that the small percentages of Nb and Ti present may have a noticeable effect
on the activation energy. These factors are removed from the equation and replaced with
a constant value fitted to the recrystallisation temperature experiments conducted by NZ
Steel as follows.
Using equations 2.35 and 2.36, a recrystallisation curve is calculated for 25 seconds
hold time at each target temperature, as per the recrystallisation experiments. This
results in the curve shape in Figure 4.36, showing the expected increasing amount of
recrystallisation as the temperature is increased.
This curve can be fitted to the recrystallisation experiments by plotting 1 − Xv against

temperature, and comparing against the experimental yield stress curve of the steel strip,
where the fractional reduction in yield stress is assumed to be equivalent to the fractional
recrystallisation. The yield stress curve is scaled to match the limits (0-1) of the 1 − Xv
curve, as shown in figure 4.37.
This assumes that steel at the start of the yield stress curve (highest yield stress)
is fully unrecrystallised, and the point at the end of the recrystallisation curve where
138
Figure 4.36: Fractional recrystallisation (Xv ) against temperature for 25s hold time
the slope has flattened is equal to 100% recrystallisation (lowest yield stress). Scaling the
1−Xv curve allows the effect of changing the empirical constants to be examined visually.
Changing the activation energy term, Qrex , moves the calculated recrystallisation curve
left and right (recrystallisation starts at higher or lower temperatures), and changing the
Avrami constant, n, changes the slope of the curve.
These results obtained using the metallurgical model show good agreement with the
experimental results for this 0.4mm 100 grade steel sample. A comparison of the predicted
recrystallisation and the fractional softening of the steel for this trial is shown in Table 4.9.
This shows that the average difference between the predicted fraction of recrystallisation
and the measured fraction of reduction in upper yield stress (UYS) for this steel type is
only 1.6%. This comparison is also shown graphically in Figure 4.38.
The calculated recrystallisation results also match well with other gauges and chem-
istry grades examined, such as the range of model result curves plotted against recrys-
tallisation experiment curves shown in figures 4.39 to 4.40. These charts show the scaled
comparison of yield stress against 1 − Xv , as this shows the value of the yield stress rather
than the fraction of yield stress change, showing that the charts apply to different strength
grades of steel.
In order to match these model calculations with the experimental testing results, the
values of the activation energy, Qrex , and avrami exponent, n, were manually adjusted. In
accordance with other literature [4, 16, 62], an avrami exponent of n = 2 best matched the
139
Figure 4.37: Calculated recrystallisation (1−Xv ) compared with experimental yield stress
softening determined in Section 2.7.3 for 0.4mm 100 grade strip
Figure 4.38: Calculated recrystallisation (Xv ) compared with fractional reduction in mea-
sured upper yield stress (UYS) determined in Section 2.7.3 for 0.4mm 100 grade strip
140
Temp UYS Fractional Calculated Difference
◦
C MPa UYS Reduction Xv
537 683 0.000 0.005 0.005
550 664 0.055 0.014 -0.041
563 663 0.058 0.036 -0.022
570 651 0.093 0.059 -0.035
585 640 0.125 0.156 0.031
595 612 0.207 0.283 0.076
605 510 0.504 0.472 -0.033
612 468 0.627 0.632 0.005
622 428 0.743 0.846 0.103
630 346 0.983 0.953 -0.029
645 372 0.907 0.999 0.093
655 357 0.950 1.000 0.050
665 357 0.950 1.000 0.050
680 346 0.983 1.000 0.017
685 334 1.017 1.000 -0.017
700 340 1.000 1.000 0.000
AVERAGE 0.016
Table 4.9: Comparison of fractional reduction in yield strength against fractional recrys-
tallisation calculated by metallurgical model
slope of the experimental curves, and was used here. The linear function of temperature
found by Haff and Schulson [21] also agreed well at most temperatures, but was not useful
at the lowest recrystallisation start temperatures such as with 0.3mm gauge hot rolled
strip, so the Avrami Exponent was set at n = 2.
The value of the activation energy, Qrex , was increased beyond that calculated in Equa-
tion 2.39, as the recrystallisation curves otherwise started around 30◦ C too early compared
with the curves determined in the laboratory recrystallisation experiments shown in Sec-
tion 2.7.3. It was previously noted that this formula may under-predict Qrex due to the
presence of small amounts of other alloying elements that can cause noticeable retarda-
tion of recrystallisation, such as Ti, Al and N, that were not included in Equation 2.39.
Using the addition for Ti mentioned in an earlier version of the Medina models [42],
+76830[T i]0.123 , the activation energy was increased by approximately 15,000 J/mol. For
NZ Steel grade 100 steel composition, this was an increase of around 8% in Qrex , from
192,000 J/mol to 207,000 J/mol.
This addition corresponding to a nominal Ti content allows the calculated recrystalli-

sation curve to match well with experimental results. This addition is included in the
141
(a) 0.3mm 100 grade (b) 1.75mm 100 grade
Figure 4.39: Comparison of model results against steel recrystallisation experiments for
two different gauge sizes (a) 0.3mm and (b) 1.75mm of the same chemistry (grade 100)
(a) 1.15mm 210 grade (b) 1.15mm 310 grade
Figure 4.40: Comparison of model results against steel recrystallisation experiments for
two different chemistry grades (a) 210 and (b) 310, of the same gauge size (1.15mm)
formula for Qrex , but not as a function of Ti concentration, as actual Ti is difficult to

measure. It is instead used as an empirical adjustment to the Qrex formula, which was
empirically fit originally. The formula now used for the activation energy Qrex is given by
Equation 4.58:
Qrex = 162476 − 71981[C] + 56537[Si] + 21180[M n]

+ 121243[M o] + 64469[V ] + 109731[N b]0.15 (4.58)
Given that the Niobium content, similar to Titanium, is very low (less than 0.001 wt%),
and below the sensitivity of the computer recording system at NZ Steel, the variation in
this element across steel production batches is not measurable and cannot meaningfully
be considered in the metallurgical model other than as an empirical constant. The con-
tribution of Nb to Qrex can be included in the first term, and the formula simplified to
the final version used in this model given by Equation 4.59. This Nb term is equivalent to
142
approximately 37,000 J/mol, or 18% of the total activation energy. Note that this value
has not been added beyond that calculated in Equation 4.58, but has been included in
the 205,000 J/mol ‘base’ value of Qrex , as the typical variation of Nb is not known or
easily measurable, and is not sensible to include in the final Qrex calculation.
Qrex = 205000 − 71981[C] + 56537[Si] + 21180[M n]

+ 121243[M o] + 64469[V ] (4.59)
This leads to the values of activation energy for the four chemistry grades used on the
MCL in Table 4.10. The mean value and variation of these is calculated using 11 months
of chemistry analysis data from Jan - Nov 2007. These are within the range expected for
Al-killed steels as in table 2.3, and show some of the expected variation between grades
as seen in practice, where grade 210 has higher recrystallisation temperatures than grades
100 and 110. This should also be seen for grade 310, but has not been found using the
Qrex Equation 4.59.
Chemistry Grade Qrex (J/mol) st.dev

100 206986 370
110 206278 425
210 213175 630
310 206530 790
Table 4.10: Calculated average activation energy and standard deviation for MCL chem-
istry grades
The Qrex values calculated in Table 4.10 do not differ much from the 205,000 J/mol
’base’ value in Equation 4.59, as the concentration of alloying elements in these compo-
sitions is low. The addition to activation energy provided by each element is small, with
carbon and manganese being by far the most significant contributions. These elements
have offsetting effects however, as carbon decreases activation energy as discussed in Sec-
tion 2.7.2.2. While grades 210 and 310 have much higher manganese content than 100
and 110, grade 310 also has high carbon content, reducing Qrex . Table 4.11 shows the
mean concentration of each element used in Equation 4.59 for each of the four steel grades
as measured over the period Jan - Nov 2007.
The difference between the highest and lowest calculated Qrex values is only 7,000
J/mol, or 3% of the maximum value. The range of Qrex values due to composition
variation in the steel are also low in each case, with the standard deviation of each Qrex
143
calculation being only 0.4% of the total. The variation in actual measured concentrations
of each element is much higher, but this variation has only small effect on the calculated
activation energy. For example, the measured variation of carbon content in grade 100
steel was between 0.026 and 0.05wt%, but this difference only corresponds to 1700 J/mol
between highest and lowest concentration, less than 1% of the total Qrex .
Equation 4.59 has allowed an estimate for the mean Qrex for each grade, but individual
composition differences amongst steel batches of the same grade do not provide significant
changes to Qrex . Therefore it is unnecessary to calculate Qrex for any individual products,
and generic values of 206,500 J/mol for grades 100, 110 and 310, and 213,000 J/mol for
grade 210 will be used.
Element Grade Grade Grade Grade

wt % 100 110 210 310
C 0.038 0.054 0.082 0.177
Si 0.006 0.006 0.010 0.010
Mn 0.179 0.199 0.600 0.607
Mo 0.001 0.001 0.001 0.001
V 0.007 0.007 0.01 0.011
Table 4.11: Measured average composition of each element used in Qrex calculation for
MCL steel grades over Jan - Nov 2007
144
4.7.3 Recrystallisation Temperature Comparison
The results from the metallurgical model across a wide range of gauges and chemistries are
compared against the NZ Steel experimental results described in Section 2.7.3. The yield
strength softening curves produced from the recrystallisation experiments are used to find
‘recrystallisation start’ and ‘recrystallisation stop’ temperatures, which correspond to the
points of inflection on the calculated recrystallisation curves such as shown in Figure 4.41.
This curve shows the inflection point for the start of recrystallisation to occur at 10% of
the total recrystallisation, and the inflection point at the end of recrystallisation to occur
at 95% recrystallisation. The equations 2.36 and 2.35 give this characteristic shape to the
recrystallisation curve for all steel types and compositions.
Figure 4.41: Comparison of model results of recrystallisation start and finish temperatures
for grade 110 hot rolled steel against experimental data
The recrystallisation start temperature is not taken as the temperature when recrys-
tallisation first appears on the model curves, as this point cannot be matched against the
yield stress curve. The point where recrystallisation first occurs cannot be distinguished
on experimental curves from the slow reduction in yield stress due to the recovery process
occurring immediately prior to recrystallisation.
The selection of recrystallisation start and stop temperatures on an experimental yield

stress curve is shown in Figure 4.42, where the yield stress curve has been scaled using the
maximum and minimum yield stresses measured from the experiment. The highest yield
stress (lowest temperature) is taken as 0% softening, and the lowest yield stress (highest
145
temperature) taken as 100% softening. The temperatures corresponding to the 10% and
95% relative softening positions are measured and compared against the temperatures
predicted by the metallurgical model.
Figure 4.42: Scaled yield stress reduction against temperature for recrystallisation trial
of 0.55mm hot rolled grade 100 steel showing selection of recrystallisation start and stop
temperatures
Selection of single temperatures for start and end points of the recrystallisation curve
allows the measured and model predicted temperatures to be compared across the whole
range of gauge sizes at once, showing the suitability of the model to the full strip size
range. This is shown in Figure 4.43 for the 110 grade hot-rolled steel experiments and
model results.
Figure 4.43 shows good agreement between the model calculated temperatures for
10% and 95% recrystallisation and the temperatures in the recrystallisation experiments
causing 10% and 95% of the softening of the upper yield stress. One of the major sources
of error in this comparison is the determination of the 10% and 95% temperatures on
the experimental curves, as there are many instances of outliers and discontinuities in
the experimental data. For example, in Figure 4.42 the softening curve shows a rough
plateau around the 95% position, and the 95% temperature could be selected between
635 and 670◦ C. The temperature selected for Figure 4.43 in this case is 640◦ C, as this is
most representative of the inflection point on the softening curve.
146
for all gauges of grade 110 hot rolled steel against experimental data
Improvement of the experimental data by increasing the number of samples taken at

each temperature position could result in a better comparison with the model recrystalli-
sation temperature results, as the trials conducted by NZ Steel use only one steel sample
at each temperature. Individual sample and experimental variations may strongly impact
the yield stress curves due to the small sample size. It would also be useful to test samples
at a greater number of temperature divisions to improve the determination of the position
of the inflection points on the softening curves. For example, the sample frequency around
the 10% position on the softening curve in Figure 4.42 is very coarse. The space between
sample positions around this position is approximately 25◦ C, which makes selection of an
accurate temperature in the 10% position difficult.
Tables 4.12 and 4.13 show the results for the model and experimental results for 10%
and 95% recrystallisation respectively, for each of the gauge sizes used in the trial. The
model results show good agreement with the experimental temperatures, as the average
difference between model and experimental temperature is +8◦ C for the 5% recrystalli-
sation temperature, and only +3◦ C for the 95% recrystallisation temperature.
147
Gauge Expt. 10% Model 10% Difference
◦ ◦ ◦
mm C C C
0.3 570 587 17
0.4 590 589 -1
0.55 580 593 13
0.67 600 599 -1
0.75 595 600 5
0.95 595 606 11
1.15 600 612 12
1.4 610 616 6
1.45 610 615 5
1.55 600 616 16
1.75 610 621 11
1.85 620 620 0
AVERAGE 8
Table 4.12: Comparison of experimental and model calculated recrystallisation start (10%
softening) temperatures
Gauge Expt. 95% Model 95% Difference

◦ ◦ ◦
mm C C C
0.3 635 639 4
0.4 640 642 2
0.55 640 647 7
0.67 650 653 3
0.75 650 654 4
0.95 650 661 11
1.15 660 668 8
1.4 675 673 -2
1.45 670 672 2
1.55 670 672 2
1.75 680 678 -2
1.85 685 677 8
AVERAGE 3
Table 4.13: Comparison of experimental and model calculated recrystallisation stop (95%
softening) temperatures
148
The comparison of model and experimentally derived temperatures is extended to

grade 210 steel chemistry in Figure 4.44, which has a higher activation energy of recrys-
tallisation as shown in Section 4.7.2. A very limited number of gauge sizes were tested
in the recrystallisation trials for this product, but the model results appear to agree well
for the three smaller gauge sizes tested - 0.75, 0.85 and 0.95mm. The model results agree
less well for the heavier gauge of 1.15mm. This is also seen in model results for grade
310 steel chemistry in Figure 4.45, which was only tested at heavy gauge sizes 1.15 and
1.55mm.
The yield stress softening curves for gauge sizes 1.15 and 1.55mm are markedly different
from the smaller gauge sizes, as they both exhibit extended softening beyond the inflection
points on the curves i.e. softening continues for an extended temperature range. Figure
4.46 shows the experimental yield stress softening curves for grade 210 1.15mm and grade
310 1.55mm strips.
If the temperatures for 95% softening are taken at the position of the inflection points
on these curves, rather than the absolute position of 95% of the total softening, the
temperatures match much closer with the model predictions as shown in Figures 4.47
and 4.48. The differences between model predictions and experimental temperatures
for the heavy gauge (1.15 and 1.55mm) steels is due to the shape of the experimental
recrystallisation curve, which becomes flatter and the inflection points less pronounced.
This may be due to the experimental setup, which may not allow enough time for the
149
temperature to propagate through the heavier gauge strip.
Using these comparisons, it appears that the metallurgical model static recrystalli-
sation calculations generally agree very well with the measured temperatures from NZ
Steel’s recrystallisation experiments, and are suitable to couple to a furnace heat transfer
model. Improving the metallurgical model to better fit the experimental recrystallisation
trial data is difficult due to the limited sample sizes used in these trials. The model re-
sults shown in Figures 4.43 matches the data to within around 5 − 10◦ C, which is closer
than the average temperature spacing of experimental sample points. It is not practical
to fit the model more accurately without increasing the sample size and frequency of the
experimental data.
Inclusion of the metallurgical model in the RDHT furnace and strip temperature model
will allow calculation of predicted recrystallisation arising from the heat treatment expe-
rienced by any combination of strip gauge, chemistry and rolling type that is processed
on the metal coating line.
150
(a) 1.15mm Grade 210 (b) 1.55mm Grade 310
Figure 4.46: Yield stress softening curves (fractional) for 1.15mm grade 210 steel and
1.55mm grade 310 steel
Figure 4.47: Comparison of model results for grade 210 hot rolled steel against experi-
mental data with alternative 95% temperature selection for 1.15mm gauge
151
Figure 4.48: Comparison of model results for grade 310 hot rolled steel against experi-
mental data with alternative 95% temperature selections for 1.15 and 1.55mm gauge
152
4.7.4 Integration into RDHT Model
The strip temperature profile calculated by the RDHT model as shown in Section 4.6.9 can
be used to calculate the amount of recrystallisation of the strip at any time for both steady
state and transient cases. Equation 2.35 is used to calculate the 50% recrystallisation time,
t0.5 , based on the strip properties and the temperature profile, and Equation 2.36 is used
to convert this to a percentage of strip recrystallisation, Xv , achieved during the time of
heat treatment.
For a steady state temperature solution, the recrystallisation calculation is carried out
once using the steady state temperatures. For transient temperature calculations, the
recrystallisation is calculated every time step so it can be followed over time as the strip
temperature changes.
The strip temperature profile is discretised into 157 segments (158 grid points) by
the calculation grid in the x-direction down the furnace, and the time spent at each
temperature by the strip is given by the grid spacing divided by the linespeed. Hence a
linespeed of 2m/s gives a duration of approximately 0.5s spent in each grid section, but
this varies by the length of the grid spacing in each zone.
Calculation of Equations 2.35 and 2.36 requires the strip gauge and width, linespeed,
and strip temperature output from the RDHT Model, as well as the following inputs for
strip physical properties:
Steel Chemistry = One of grade 100, 110, 210, 310

Strain, = Cold Reduction
Grain Size, D = 10µm (hot rolled)
= 20µm (warm rolled)
At each x-coordinate, the current strip temperature is used to calculate t0.5 , and
t
the time duration, t, to cross the previous x-grid distance used in the fraction t0.5 in
Equation 2.36 to calculate the amount of recrystallisation in that particular grid space.
The recrystallisation amount, Xv , is cumulatively summed at each x-coordinate from

x = 0 to give the amount of recrystallisation from furnace entry to x. This shows the
strip recrystallisation occurring down the length of the furnace as the heat treatment
progresses, as described by Equation 4.60. The cumulative recrystallisation at x = 158
153
represents the total recrystallisation occurring during the whole heat treatment i.e. the
amount of recrystallisation where the strip exits the furnace, and will be referred to as
ΣXv .
x
X
Xv (x) = Cumulative recrystallisation from entry to x coordinate (4.60)
0
The amount of recrystallisation depends on the values assigned for Qrex and D, which
are difficult to measure or define exactly. To give an indication of the amount of un-
certainty this creates in the recrystallisation calculations, an uncertainty band is created
by showing the recrystallisation calculations for likely highest and lowest values of Qrex .
The variation in the calculated values of Qrex is a maximum of ±3% from the average
values given in Table 4.10, however the variation in grain size can only be estimated to
be around ±20%.
As increasing grain size and increasing activation energy both decelerate recrystalli-
sation (Equation 2.35), the low recrystallisation calculation, ΣXv,low is calculated using
Equations 4.61 and 4.62 for Qrex and D.
Qrex,low = 1.03 × Qrex (4.61)

Dlow = 1.2 × D (4.62)
Conversely, ΣXv,high is calculated using Equations 4.63 and 4.64 for Qrex and D.
Qrex,high = 0.97 × Qrex (4.63)

Dhigh = 0.8 × D (4.64)
Figure 4.49 shows the recrystallisation calculation for a soft iron (fully recrystallised)
strip during heat treatment against the x-coordinate furnace length. Note that ΣXv is
capped at 1, as recrystallisation cannot be greater than 100%. This shows full recrystalli-
sation of the strip after approximately 80m travel through the furnace for the nominal
recrystallisation calculation. This occurs even faster for the ‘high’ recrystallisation cal-
culation, however the ‘low’ recrystallisation calculation shows that furnace temperatures
are only just sufficient to give 100% recrystallisation by furnace exit.
Recrystallisation calculations can also be used in transient simulations to ensure that

strip recrystallisation remains acceptable during times of changing strip temperature.
154
Figure 4.49: Cumulative recrystallisation calculation of steel strip against x-coordinate,
showing highest, nominal and lowest likely recrystallisation
Figure 4.50 shows a transient simulation including recrystallisation, where the integrated
metallurgical model calculations show the amount of predicted recrystallisation (ΣXv )
changing in response to strip temperature changes.
Figure 4.50: Example of transient result from RDHT-CM model
In this example, heating power in furnace Zone 6 is suddenly cut for a period of 30min
before being reinstated. This results in a gradual decrease in furnace temperature in
this zone, shown by plotting furnace temperature at an x-coordinate inside zone 6 such
as at x = 33m. The peak strip temperature reduces, which causes a decrease in the
calculated final recrystallisation of the strip, shown initially by the low recrystallisation
calculation dropping below 100%, followed by the nominal recrystallisation. Once heating
power is restored, and furnace and strip temperatures start recovering, the calculcated
recrystallisation amounts also start recovering towards 100%.
155
Chapter 5
Model Results and Validation
5.1 Introduction
In this Chapter, the accuracy of the furnace and metallurgical models is demonstrated by
comparison with real furnace operational data, and against prior mathematical investiga-
tions of the furnace system. The available data and method used to select data suitable
for model validation is explained in Section 5.2.
Section 5.3 shows the validation of the F3DC furnace model against both furnace
measurements and previous mathematical investigations. The RDHT model is compared
against furnace measurements in Section 5.4, in terms of the temperature results only
to validate the mathematical approximations. The results from the F3DC and RDHT
models are then compared against each other in Section 5.5.
The RDHT-CM model is then validated in Section 5.6, demonstrating the accuracy of
metallurgical predictions from the model, which is based on the temperature calculation
results.
5.2 Validation Data Selection
The furnace model is compared against ‘coil summary data’ gathered during furnace op-
erations, which contains the measured power settings and recorded temperatures for each
furnace zone, as well as the other required input variables of line speed, strip dimensions,
156
induction heating power, and strip pyrometer temperatures. This provides maximum,
minimum and average values of each variable indexed against an identification number
for each coil. The duration each coil spends in the furnace depends on the strip gauge
and linespeed, but is typically between 5-15 minutes per coil.
Because the steady state solutions of the models are being examined, it is necessary
to find coil summary data from situations when the furnace is operating in a completely
steady state. The large thermal mass of the furnace and the length of time required for it to
achieve a steady state complicates the selection of furnace measurement for comparison, as
there is a high likelihood that any selection of coil data will contain transient temperature
effects. The transient situations are not only caused by product type changes, but also
by the manual adjustment of linespeed and induction preheaters by operators.
Suitable values for zone temperatures and power inputs are found by selection of
individual steel coils that have been processed during very steady conditions. Data from a
number of these ‘steady state coils’ is used to find average values for zone power inputs and
the other model inputs including the amount of preheating, the linespeed and measured
strip gauge. The temperature solution of the model is compared against the average zone
temperatures measured for these steady state coils.
Figure 5.1: Variation in applied power to two heating zones after a cycle change from
550MPa 0.4x940mm strip to 300MPa 0.4x905mm strip
Steady state coils are located by finding coils of the desired product type far away
from cycle changes, once the furnace temperatures have had sufficient time to steady out.
This is demonstrated in Figure 5.1, showing for example the variation in applied power to
heating zones 6 and 12 after a cycle change from recovered 550MPa strip to recrystallised
157
300MPa strip with similar dimensions. The power applied to these heating zones changes
slowly until a new steady state is reached approximately 2 hours after the time of cycle
change. Examination of the coil summary data at this steady state shows the zone power
for all 12 heating zones to have stabilised to within ±1% during the processing time of
the coil. This data is difficult to collect for uncommon product types which typically run
for short periods of time only, and may not be of long enough duration to reach a steady
state.
Furnace data extracted for these steady state coils may still exhibit strong variation
in several factors. This variation includes the heating power applied to each zone, as well
as the exact measured dimensions of the strip, the furnace linespeed etc. This variation
occurs both in the short term, such as required zone power changing between steady state
coils which may be only several hours or days apart. This short term variation may be
due to the variation in strip thickness or surface properties (affecting the strip emissivity)
from the cold rolling mill, changes in ambient temperature, or failures of heating elements
or other mechanical equipment. Figure 5.2 shows the variation in zone input power for
several coils of the same specification processed at steady state within a period of three
weeks. This shows that furnace heating zone inputs even at steady state may typically
vary by up to ±5% in the short term with no special cause.
Figure 5.2: Variation in power applied to furnace heating zones 1-12 for several steady
state coils of the same specification over a three week period
Longer term variation also occurs when examining steady state data from coils pro-
cessed months apart or longer. This may also be due to the failure of heating elements or
other equipment, or the slow degradation of furnace condition or materials between pe-
158
riods of major maintenance. Figure 5.3 shows much stronger variation when considering
coils processed at longer time intervals. This indicates that major changes have happened
in the furnace system over this time period. In this case, heating power has been reduced
in several zones due to the failure of heating elements, resulting in low power to some
zones, and high power to adjacent zones to compensate. This can be a major issue when
selecting furnace data for model input, as heating elements that become damaged are only
replaced during major maintenance at 2-yearly intervals, as discussed in Section 4.2.1.
Figure 5.3: Variation in power applied to furnace heating zones 1-12 for steady state coils
of the same specification processed two years apart
5.3 F3DC Model Results and Validation
Simulation of different strip dimensions (width and gauge) in the F3DC model is time
consuming due to the requirement to manually create a new computational mesh with
each change. For this reason, most effort on validation of the F3DC model is spent on the
most common strip size only, being 0.4mm gauge × 940mm width. As shown in Section
6.2, this size comprises over 20% of the total production by weight of the MCL furnace
and is twice as common as the next most popular size.
Validation against plant measurements is only considered here for the heating zones
of the furnace, as this is main area of interest for NZ Steel as peak strip temperature is
reached at the end of the heating zone. Use of the complete three-model system for strip
temperature predictions along the entire furnace are not practical for use by NZ Steel due
159
to the computation power and time required. Validation of the heating zones model will
confirm the accuracy of the mathematical and geometric simplifications made in Section
4.5.2, and confirm the physical properties assumed for the modelled bodies. Once these
assumptions are confirmed, they can be used to inform the design of the simple model.
Validation of the F3DC model requires comparison of the available measurements of

the real furnace against predicted temperatures in the model. As described in Section
4.2.3, the furnace pyrometers are highly inaccurate and unusable other than the ‘L1’
pyrometer at the start of Zone 13. Gold cup pyrometer measurements have shown the
L1 pyrometer to commonly read 10-50◦ C high. This is the only measurement of strip
temperature available for validation of model results against historical plant data, and
will be compared with the strip temperature exiting Zone 12 of the heating zones model.
The temperature recorded on furnace thermocouples is compared directly against the

temperature of the geometry blocks in the F3DC model representing the thermocouple
units. The thermocouple units are included in the F3DC model for this purpose, as
described in Section 4.5.2.3.
5.3.1 Validation of Cycle 24 0.4mm x 940mm G550 Strip Model
This is the most common product specification as shown in Section 6.2, and the furnace
data used in the F3DC model are collected from a large number of coils of this type pro-
cessed at various times under steady state conditions. The power input data used is as for
Figure 5.2, which covers a time period where all furnace heating elements were operating
at full capacity, and there were no other major problems with furnace equipment.
The designation ‘cycle 24’ refers to the temperature setpoints of the furnace, where
many product types and size ranges are grouped into a small number of cycles, as shown
in Figure 4.12. Note that while the temperature setpoints may be the same for many
product variations under the same cycle, the actual power applied by the PID controllers
will vary strongly depending on the strip dimensions, linespeed and pre heating etc. G550
refers to the target yield strength of the strip after heat treatment as described in Section
4.3.2. In this case, the strip should have over 550MPa yield stress, which falls into the
‘hard iron’ category undergoing a recovery heat treatment only, without recrystallisation.
Figure 5.4 shows the thermocouple temperature results from the F3DC model for each
heating zone using either the maximum, minimum or average power inputs for the group
160
Figure 5.4: F3DC model results for thermocouple temperatures using minimum, maxi-
mum and average power input values as shown in Figure 5.2
of steady state coils in Figure 5.2. The variation of the input data for a small group of
steady state coils of the same product type is shown to give around ±20◦ C temperature
variation in each of the heating zones, with the zone temperatures ranging from 670 to
805◦ C. It is obvious that significant variation can be produced in the results due to the
variation of power input for individual coil data selections.
The variation of the results may be due to furnace operational parameters or equipment
behaviours that are not captured in the numerical operating data and hence are not
specifically included in the model. The furnace heater PID control covers up this variation
by modulating heating power to achieve a fixed zone temperature, and so any single
measurement of zone power may not provide accurate results in the 3D model. Further
model results shall be calculated using averaged input data from a range of coils processed
in the furnace at different times.
Figure 5.5 shows the combined F3DC model result using averaged steady state input
data. The combined result shows the actual recorded furnace wall temperatures, compared
with model results for furnace wall temperature and thermocouple temperature, as well as
strip edge and centre temperatures compared with the range measured for L1 Pyrometer
strip temperature. Note that the thermocouple temperature results are discrete, and
only exist at the positions along the furnace length corresponding to the thermocouple
positions at the end of each heating zone. The strip centre and edge temperatures are
sampled from the model at 1m intervals for the whole distance along the furnace heating
zones, as shown in Figure 5.5.
161
Figure 5.5: F3DC model result for strip and furnace temperatures for cycle 24 0.4x940mm
product using average steady state input data from Figure 5.2
The F3DC heating zone model results show good agreement with the measured furnace
temperatures in zones 1-12. The model result of the thermocouple temperatures matches
well with the measured temperature in most cases, especially in the centre of the furnace,
but appears less accurate at the ends of the furnace where there may be other effects
not modelled, such as heat transfer to the subsequent cooling zones. It was anticipated
that wall temperatures in the F3DC model should be higher than that measured on the
furnace thermocouples due to radiative cooling of the thermocouple tubes, as described
in Section 4.2.3.1.
The strip temperature also agrees well with that measured by the L1 Pyrometer at
position x = 70m for the same furnace input data. The readings made by the L1 pyrom-
eter for this data range between 468 − 490◦ C, with an average value across the coil data
of 480◦ C. The model predicts a temperature of 460◦ C in the centre of the strip and 490◦ C
at the strip edge. The L1 pyrometer aims at the centre of the strip, so this value of 460◦ C
is compared against the measured average 480◦ C. As described in Section 4.2.3.2, the L1
pyrometer typically measures 10 − 50◦ C offset above the strip temperature measured by
the gold cup pyrometer. The model strip centre temperature is 20◦ C below the averaged
162
measured L1 temperature, which is well inside the expected range in this case.
Figure 5.6: Detail of F3DC model result for wall and thermocouple temperatures for cycle
24 0.4×940mm product using averaged steady state input data
Figure 5.6 shows the F3DC model wall and thermocouple temperature results only for
the same averaged input data from Figures 5.4 and 5.5, compared against the temperatures
measured by furnace zone thermocouples for these coils. Note that there is very little
variation in the measured temperature of the thermocouples in the steady state coils,
typically ±1◦ C for any coil, as the zone heaters are PID controlled to maintain these
temperatures i.e. the zone temperatures are constant while there is much variation in the
zone power inputs as in Figure 5.2.
The F3DC model results agree well with the measured temperature data, as the cal-
culated thermocouple temperatures are within 5◦ C of the measured temperatures for 6 of
the 12 heating zones, and within 10◦ for 8 of 12 zones. Four of the zones have differences
between measured and modelled thermocouple temperatures between 15 and 30◦ C - Zones
1 and 3 are both lower than expected, and Zones 9 and 10 are both higher than expected.
The lower than expected temperatures calculated in Zones 1 and 3 appear to be radia-
tive effects only on the thermocouple blocks in the model, as the furnace wall temperatures
in the model are in line with temperatures in adjacent zones. The calculated average wall
temperatures in Zones 1 - 4 are all within the range 798 − 802◦ C, but the thermocouple
temperatures vary between 742 − 775◦ C. High radiative cooling of the thermocouples in
Zones 1 and 3 may be due to their proximity to the end wall of the furnace, and the high
rate of heat transfer in the early heating zones where the strip is coldest.
163
The higher than expected temperatures calculated in Zones 9 and 10 suggests that
actual heat loss in these furnace zones is higher than allowed for in the model. This
is assumed because both furnace wall temperatures and thermocouple temperatures are
higher than expected i.e. it is not a radiative effect on the thermocouples only. This is
likely due to a ‘dam wall’ which is present in the furnace in this exact location, between
zones 9 and 10, which has not been represented in the F3DC model. This dam wall
separates these two zones, and will increase heat loss by conduction through the furnace
walls.
Figure 5.7: F3DC model result for wall and thermocouple temperatures for cycle 24
0.4×940mm product after introducing 10kW heat loss in position of dam wall between
zones 9 and 10
Figure 5.7 shows the wall and thermocouple temperature results once a heat loss
has been introduced between Zones 9 and 10 in the F3DC model. Trial and error has
shown the required heat loss in this zone to be approximately 10kW to match the recorded
thermocouple temperatures. Comparison with Figure 5.6 shows the temperatures in Zones
9 and 10 to be significantly closer to that recorded in the real furnace. Including the dam
wall in the 3D model would be preferable, but was not practical due to geometry and
meshing restrictions of the software.
The additional heat loss required in the model shows how simplification of the geom-
etry has lead to some errors in the model solution. The error introduced by removing
the dam wall between Zones 9 and 10 is easily diagnosed when comparing the simplified
model geometry with the actual furnace, and is compensated for by introducing this heat
loss. Other effects such as the low model thermocouple temperatures in Zones 1 and 3
cannot be easily compensated for, as there is no obvious cause for this variation.
164
5.3.2 Validation of Cycle 15 0.39mm x 905mm G300 Strip Model
Further model results are briefly summarised for 0.39mm x 905mm product of 300MPa
tensile strength, which is the second most popular processed on the MCL furnace, and is
significantly different from the product in Section 5.3.1, as it is a ‘soft iron’ recrystallised
product that is heat treated at a much higher strip temperature.
Figure 5.8: F3DC model result for strip and furnace temperatures for cycle 15
0.39mm×905mm product using averaged steady state input data
Figures 5.8 and 5.9 show the F3DC model results for this product, showing both
the strip and furnace temperatures, and then compared against recorded furnace zone
temperatures. No additional changes to the F3DC model are required to simulate this
different product type, other than the power inputs to each zone, strip dimensions, entry
temperature and linespeed. Note that the additional heat loss due to the dam wall between
zones 9 and 10 is retained in this and all further models.
As in Section 5.3.1, the modelled thermocouple temperatures generally agree with

the measured thermocouple temperatures during steady state operation. The average
absolute difference between model and measured thermocouple temperature is ±15◦ C,
but this occurs in both positive and negative directions so that the average model and
165
Figure 5.9: F3DC model result for wall and thermocouple temperatures for cycle 15
0.39mm×905mm product compared against recorded furnace zone temperature
measured thermocouple temperatures match exactly. In this model, the average wall
temperature is around 30◦ C higher than the average thermocouple temperature. The
exact difference is higher at the start of the furnace and reduces as the strip heats up i.e.
as the strip-furnace temperature difference reduces.
The strip centre-temperature agrees well with the range of temperatures measured by
the L1 pyrometer, as the strip centre temperature is 755◦ C, compared with the average
L1 pyrometer temperature of 765◦ C. This gives a temperature difference for the strip
temperature 10◦ C below the average L1 pyrometer temperature, similar to the 20◦ C dif-
ference calculated by the model for the product in Section 5.3.1, and within the range of
temperature offsets measured for this pyrometer as in Section 4.2.3.2.
166
5.3.3 Thermocouple Block Temperature Variation
The temperatures of the thermocouple blocks in the F3DC model show strong variation
along their lengths, and do not generally agree with the expected temperature profile as
calculated for a simplified system [39, 78], as shown in Section 2.5.4. Figure 5.10 shows
the temperatures measured along each thermocouple block in heating zones 1-6 for the
product type as modelled in Section 5.3.1. Figure 5.11 shows the thermocouple block
temperature for the remaining heating zones 7-12.
Figure 5.10: F3DC model result showing temperature variation in the thermocouple
blocks for heating zones 1-6
The thermocouple block temperature results in Zones 1-6 of the model as shown in
Figure 5.10 have the opposite temperature variation to that expected. As discussed in
Section 2.5.4, it is expected that the temperature will be the lowest at the bottom of the
thermocouple block, where it is radiatively cooled by the strip passing beneath. The model
results actually show the opposite trend in most cases, although the average temperature
of the block is below the average temperature of the furnace walls, which also agrees well
with the actual recorded thermocouple temperatures as shown in Section 5.3.1.
The modelled thermocouple blocks in zones 1 and 3 show much stronger variation
in temperature, although these two zones give results that do not agree with measured
temperatures unlike the other zones. Complex variation in the temperature profiles of
the thermocouple blocks is also present in the remaining heating zones as in Figure 5.11,
with the Zone 8 thermocouple in particular having very high temperature variation.
167
Figure 5.11: F3DC model result showing temperature variation in the thermocouple
blocks for heating zones 7-12
Across all heating zones excluding Zones 1, 3 and 8, the average variation along the
length of the thermocouple block is around 15◦ C hotter at the base than at the top end
adjacent to the furnace ceiling. The pattern of variation along these blocks however is
complex, and not consistent between zones despite having identical furnace geometry in
each zone.
It appears that the 3D furnace geometry in the F3DC model has a strong effect on
the calculation of temperatures along the length of the thermocouple blocks, as the model
predictions have much higher complexity than predicted for a simple 2D furnace cavity
in Section 2.5.4. One major factor affecting the temperature of the blocks is likely to
be the geometry of the heating elements, as they are in reality much more complex than
represented in the F3DC model, and not present at all in Wang’s calculations.
The thermocouple sheaths are very close to the heating elements, and simplifications
of the element shape are likely to have a stronger effect on the thermocouple blocks than
on the furnace heat transfer as a whole. The blocks in the model were shaped to provide
equivalent heat flux to the furnace floor and roof parallel to the elements only, as described
in Section 4.5.2.2, but are much closer to the base of the thermocouple sheaths than in
the actual furnace and may provide too much direct radiative heat flux. The different
temperature patterns seen between thermocouple blocks may also be due to numerical
artifacts caused by different alignments with the mesh points along the furnace walls and
heater elements. As discussed in Section 7.2 further investigation using a significantly
more accurate geometry and mesh may be required to fully explain these differences.
168
The thermocouple representations in the F3DC model appear sufficient to predict
furnace temperatures for each zone, as the average temperature of the blocks matches
closely with the recorded thermocouple temperatures from the real furnace. The detailed
variation in temperature along these blocks does not agree with previous works [39, 78],
although the F3DC model uses a much more accurate representation of the furnace geom-
etry. It is likely that a model using even more detailed geometry is required to accurately
gauge the temperature variation along these blocks, considering the exact shape of the
heating elements, the hollow Inconel sheaths, and the connection of the sheaths to the
furnace walls, all of which are beyond the capability of the F3DC model with current
software and hardware limitations.
As shown in the product-specific models in Sections 5.3.1 and 5.3.2, the average tem-
perature of the thermocouple blocks is consistently lower than the average furnace wall
temperature in the models. This agrees with the general understanding from previous
investigations of the MCL Furnace that the thermocouples are cooled by the strip passing
below [39, 73, 78], even if the pattern of temperature variation on the sheaths is unclear.
Discounting the thermocouples in Zones 1 and 3, which display irregular temperatures

when compared against other thermocouples, the average thermocouple temperature in
the Cycle 24 model (Section 5.3.1) is 26◦ C below the average furnace wall temperature in
each zone. This is consistent with the Cycle 15 model (Section 5.3.2), where the average
thermocouple temperature is 27◦ C below the average furnace wall temperature. Note that
the first case is for a recovered product with low furnace temperatures around 780◦ C, and
the second case is a recrystallised product with furnace temperatures around 950◦ C.
This result is useful when examining results from the RDHT model in Section 5.4,
which does not separately calculate thermocouple temperatures. Because there are no
model thermocouple temperatures to compare against measured thermocouple temper-
atures, the RDHT model wall temperatures must be used. When examining results in
further models, it is assumed that model wall temperatures should be 25 − 30◦ C hotter
than thermocouple measurements in the heating zones.
5.3.4 F3DC Model Strip Edge Heating
Heating of the edges of the strip as previously investigated by Taylor and Wang [73] and
described in Section 2.5.4 is clearly visible in F3DC model results presented in Figures
5.5 and 5.8. Figure 5.12 shows that strip edge temperature initially increases much faster
169
than the strip centre temperature, and stabilises to a constant difference approximately
half way along the heating section.
Figure 5.12: F3DC model result showing difference between strip edge and centre tem-
perature for cycle 24 0.4mm×940mm product
For the cycle 24 product investigated in Section 5.3.1, the strip edge temperature
at the end of the heating zones is 472◦ C, and the strip centre temperature is 454◦ C, a
difference of +18◦ C between edge and centre. The cycle 15 product simulated in Section
5.3.2 calculates a strip centre temperature of 758◦ C, and an edge temperature of 776◦ C,
which is a difference of +18◦ C between edge and centre.
(a) 0.4×940mm G550 Strip (b) 0.39×905mm G300 Strip
Figure 5.13: F3DC model result showing strip temperature variation across the strip
width at the end of the heating zones model for products simulated in Sections 5.3.1 and
5.3.2
The strip temperature profile calculated across the strip width for the two products
simulated in Sections 5.3.1 and 5.3.2 is shown in Figure 5.13, which appears to be similar
shape to that predicted by Wang in Figure 2.11, but of higher magnitude. Wang predicted
170
only +7◦ C at the strip edges for a simplified system with 2D square geometry and constant
wall and strip centre temperature. The F3DC model has significantly more accurate 3D
geometry, and is likely to provide a better calculation of strip edge temperature for actual
operating conditions. Unfortunately this cannot be measured in the furnace, as the gold
cup pyrometer can only be used to measure temperature at the strip centre.
While the temperature increase at the strip edges predicted here is greater than pre-
dicted by Wang in Section 2.5.4, the F3DC result appears realistic as the product seen in
Figure 5.13(a) has previously experienced severe ‘wavy edge’ problems, despite the strip
centre temperature read on the L1 pyrometer being safely below temperatures usually
causing recrystallisation. To prevent wavy edges in this particular product, strip target
temperature had to be reduced by MCL operators by as much as 50◦ C to completely
eliminate this problem.
(a) 0.4×940mm G550 Strip (b) 0.39×905mm G300 Strip
Figure 5.14: F3DC model result showing strip temperature variation across the strip
width at the end of the heating zones model for products simulated in Sections 5.3.1 and
5.3.2 with doubled strip emissivity along edge
Even higher levels of increased edge heating than calculated here may be due to the
irregular shape of the strip edges formed by the cold rolling mill. Cold rolling creates strip
edges that are rougher than the strip faces, and additionally the strip is physically thinner
at the edges, and hence likely to heat up faster than the strip bulk. The first factor can
be included in the 3D model by increasing the emissivity of the strip edge. F3DC Model
results found by doubling the strip emissivity at the edge of the strip only are shown in
Figure 5.14. This increases the temperature difference at the strip edges by a further 5◦ C
for the cycle 24 product, and 9◦ C for the cycle 15 product, giving total edge temperature
increases of +23◦ C and +27◦ C respectively.
Reduced strip edge thickness is difficult to model due to the very large aspect ratio of
171
the strip i.e. very thin edge length relative to strip width. This causes difficulties forming
the calculation mesh, which is substantially more difficult for an irregular shape as occurs
when the strip edges are thinner. This effect has not been included in the F3DC model,
but it is worthwhile to consider that F3DC model results may still underestimate the edge
temperature increase.
5.4 RDHT Model Temperature Results Validation
Figures 5.15 and 5.16 show a graphical comparison of the temperature results of the
RDHT model developed in Section 4.6, against the measured temperatures of the furnace
thermocouples and the L1 strip pyrometer recorded in the CITEC computer system.
The results displayed in these figures represent some of the common products of the
MCL, as defined in Section 6.2. The RDHT model easily calculates the furnace and strip
temperatures for the full length of the furnace, including both heating and cooling zones;
This contrasts with the F3DC model, for which the heating zones model results only were
examined in Section 5.3.
(a) Thin gauge (b) Thick gauge
Figure 5.15: Wall and strip temperature results from RDHT furnace model compared
against furnace thermocouple and L1 pyrometer readings for two example soft-steel re-
crystallised products
The RDHT model appears to give accurate temperature results for the furnace wall
and strip temperature predictions in the case of either high temperature (recrystallised
strip) and lower temperature (recovered strip) operating conditions, across the range of
sizes and widths of strip processed in the MCL furnace.
172
(a) Thin gauge (b) Thick gauge
Figure 5.16: Wall and strip temperature results from RDHT furnace model compared
against furnace thermocouple and L1 pyrometer readings for two example hard-steel re-
covered products
5.4.1 Model Result Interpretation
Because the RDHT model does not contain 3D representations of the furnace body, the
strip, or the zone thermocouples, results from the F3DC model can be used to interpret
RDHT model results when comparing against measured values.
As described in Sections 2.5.2, 2.5.4 and 4.2.3.1, the temperature measured by fur-
nace thermocouples and recorded in the CITEC system is affected by radiation in the
furnace cavity, and the thermocouples are expected to read cooler than the actual wall
temperatures. The results in Section 5.3.3 show that this effect is likely to be larger
than previously estimated, with the F3DC model predicting an average difference around
25 − 30◦ C. For the furnace walls, the model temperature result will be compared against
the recorded thermocouple temperatures, with the expectation that the average model
wall temperature will be around 25−30◦ C higher than the average recorded thermocouple
temperature.
The RDHT model strip temperature is compared against the furnace L1 pyrometer
measurements, which are also strongly affected by reflected radiation as described in
Section 4.2.3.2. The L1 pyrometer is in Zone 13, the position where strip temperature
can be measured with the least error, however these readings are typically still 10 − 50◦ C
too high, which represents up to 12.5% measurement error of the strip temperature. There
is currently no way to improve the accuracy of this measurement, and so the RDHT model
strip temperature result must be compared against this. Using the position in Zone 13
of x = 70m, the predicted strip temperature is expected to be in the range of 10 − 50◦ C
below the measured pyrometer reading.
173
5.4.2 RDHT Model Temperature Results
To examine the results of the RDHT model, a selection of furnace operating data is made
to find ‘steady state coils’ as discussed in Section 5.2. A number of these coils are chosen
to model a range of combinations of strip dimensions and heat treatment types to cover
most of the plant product types. The results from these models are used to compare the
average predicted wall temperatures against the average thermocouple temperatures, and
the difference between the Zone 13 strip temperature and the L1 pyrometer, as discussed
in Section 5.4.1.
Coil No. Grade Dimensions Avg Wall Temp L1 Strip Temp

MPa mm Difference ◦ C Difference ◦ C
G574447 G550 0.4 x 940 3 -26
G576583 G500 1.15 x 1200 77 -16.5
G575678 G250 0.54 x 1135 7 -31
G573537 G300 0.55 x 670 59 -11
G572536 G250 1.55 x 1220 64 -20
Table 5.1: Difference between measured and model average temperature for heating zones,
and zone 13 vs. L1 pyro strip temperature for a selection of coils without preheating
Table 5.1 gives the difference between the model results and measured temperatures
for a selection of simulations of individual steady-state coils that do not use induction
preheating. The wall temperatures are compared by subtracting the average thermocouple
temperature of the heating zones from the average model result for wall temperature from
these zones. The RDHT model results are shown by the ‘Model Walls’ data series in
Figures 5.15 and 5.16. As described in Section 5.3.3, it is expected that the model result
for wall temperature should be around 25 − 30◦ C higher than the recorded thermocouple
temperature.
The strip temperature results are compared by subtracting the recorded L1 pyrometer
temperature from the model predicted strip temperature at x = 70m. The strip tem-
perature result is given by the data series ‘Model Strip’ in Figures 5.15 and 5.16. As in
Section 4.2.3.2, the model results are expected to be in the range 10 − 50◦ C lower than
the measured temperature i.e. -10 to -50◦ C in Table 5.1. The results show that for all
cases, the predicted strip temperature in the location of the L1 pyrometer is within the
expected 10 − 50◦ C range. The strong variation in the measured L1 pyrometer tempera-
tures described in Section 4.2.3.2 means that model predicted strip temperatures cannot
be validated more accurately for any given production data.
174
The model predicted wall temperatures in Table 5.1 have much higher variation than
the predicted strip temperature. The model results fall on both sides of the expected
25 − 30◦ C temperature difference, and the variation in model result is from 3 − 77◦ C
higher than the average recorded heating zone temperature. In each case there is a
positive difference i.e. the average model predicted wall temperatures are higher than the
recorded thermocouple temperatures as expected.
Table 5.2 gives further comparison of model temperatures against measured temper-
atures for coils utilising the induction preheaters to significantly raise the temperature
of the strip at entry to the furnace. These results also show good agreement with the
10−50◦ C expected difference to the measured L1 pyrometer temperature. The differences
between model wall and recorded wall temperatures are also high for these coils, but it
can be seen that they have the same behaviour as for the coils without preheating. The
model equations appear robust enough to work for any strip thickness, width or entry
temperature that can be expected in the plant.
Coil No. Grade Dimensions Avg Wall Temp L1 Strip Temp Preheating
MPa mm Difference ◦ C Difference ◦ C ◦
C
G576634 G300 0.39 x 1095 35 -28 +20
G573690 G250 0.54 x 1135 48 -8 +200
G580515 G250 0.945 x 1210 57 -9 +225
G576478 G300 1.15 x 1190 63 -22 +275
Table 5.2: Difference between measured and model average temperature for heating zones,
and zone 13 vs. L1 pyro strip temperature for a selection of coils with preheating
Improvement of the prediction of the wall temperature results is very difficult, as

even the detailed F3DC model does not give a clear indication of the difference between
wall and thermocouple temperatures, as discussed in Section 5.3.3. While the average
thermocouple temperature was found to be 25 − 30◦ C below the average adjacent wall
temperature, the model predicted up to 50◦ C variation along any thermocouple block as
shown in Figures 5.10 and 5.11. The wall temperatures are related to the strip temperature
by radiative heat transfer, and as the predicted strip temperatures are within the expected
range as shown in Tables 5.1 and 5.2, the wall temperature predictions appear sufficient
for the RDHT model.
Exact validation of the temperature results of the RDHT model is difficult, as the
accuracy of temperature measurements in the furnace is low. The reduction of the furnace
geometry for the RDHT model does not allow modelling of individual thermocouples or
175
the radiative effects on their measured temperature as with the 3D model. The strip
temperature predictions shown in Tables 5.1 and 5.2 are within the expected limits of
accuracy for strip temperature, and are suitable for a rapid online-control and optimisation
model. The predicted strip temperatures agree well with the expected range of 10 − 50◦ C
below the L1 pyrometer measurements in all cases.
176
5.5 Comparison of RDHT and F3DC Model Results
The accuracy of the RDHT model is further examined by comparison with temperature
results from the F3DC model for the 12 heating zones of the furnace. This is done for the
two most common products in the MCL simulated in the F3DC model in sections 5.3.1
and 5.3.2.
(a) Furnace Temperatures (b) Strip Temperatures
Figure 5.17: Comparison of results from the F3DC model and the RDHT model for
0.4×940mm G550 strip
The comparison of F3DC and RDHT model temperature results for the 0.4mm ×
940mm strip in Figures 5.17 and 5.18 show remarkably good agreement considering the
high level of simplification required to construct the RDHT model. The reduction in
model geometry from a complex 3D shape to a simple 1D/2D approximation does not
appear to have strongly affected the calculation of heat transfer from the furnace to the
strip.
The average furnace wall temperature from the F3DC model in Figure 5.17 differs
from the average RDHT model wall temperature by only 3◦ C, and the difference for the
0.39mm × 905mm product in Figure 5.18 is even lower at only 1◦ C. The temperature
distribution in the furnace walls is slightly different in the RDHT model, probably due to
the simplification of radiative transfer only occurring between concurrent grid sections,
but the average temperatures are very similar in both models.
This is further shown by the strip temperature predictions from both models. The
RDHT model strip temperature in Figure 5.17 is virtually identical to the F3DC model
temperature at the centre of the strip, for the whole length of the heating zones model. The
temperatures are also very similar in Figure 5.18, although the RDHT model temperature
is somewhere between the centre and edge temperature of the strip in the F3DC model.
177
(a) Furnace Temperatures (b) Strip Temperatures
Figure 5.18: Comparison of results from the F3DC model and the RDHT model for
0.39×905mm G300 strip
It is apparent that the RDHT wall temperature results agree closely with the wall
temperatures in the F3DC model. This means that RDHT model wall temperature results
for a given steady state coil input should be expected to be 25 − 30◦ C higher than the
recorded thermocouple temperatures, as was found for the F3DC model in Section 5.3.3.
The strip temperature calculated by the RDHT model is most likely to represent the
temperature in the centre of the strip, when compared against F3DC model results. As
shown in section 5.3.4, the edges of the strip can be significantly hotter than the centre
temperature. When considering the strip temperature result of the RDHT model, it
should be understood that the strip edges may be 20 − 30◦ C hotter than the strip centre,
as shown in Section 5.3.4. This is of most importance when examining model results
for recovered hard-iron steels, where the elevated edge temperature may cause localised
recrystallisation and the formation of ‘wavy-edge’ defects.
5.6 RDHT-CM Model Recrystallisation Results Val-

idation
The exact amount of recrystallisation occurring in steel strip produced on the MCL plant
is not measured by NZ Steel, except in special cases where it may be examined with low
accuracy by visual examination of the steel microstructure. This means it is impossible to
validate exact predictions of the recrystallisation percentage from the RDHT-CM model.
178
As exact recrystallisation percentages are not explicitly comparable, coils where re-
crystallisation is expected to be either 100% complete, or to not occur at all, are used for
validation. It is expected that the RDHT-CM model will result in predictions of 100%
recrystallisation for all soft-iron recrystallised products when provided with high quality
steady-state input data, and will similarly result in 0% recrystallisation for all hard-iron
recovered products. This is demonstrated in Section 5.6.2 for hard iron recovered products
and in Section 5.6.1 for soft iron recrystallised products.
This is further examined in Section 5.6.3 by using the RDHT-CM model to simulate
cases from MCL operations where the heat treatment has not occurred to meet product
specifications. This includes hard products that have been overheated and rejected as too
soft by tensile testing, and soft products that have been under heated and tested to be
too hard. The RDHT-CM model results show the undesirable recrystallisation occurring
or not occurring respectively, when given the same input data as existed during these
production problems.
In this way, the RDHT-CM model is shown to be suitable for predicting whether a
given furnace setup and strip combination will deliver a fully recrystallised product, a
completely unrecrystallised product, or if the given parameters are likely to cause only
partial recrystallisation. It is shown that a combination exhibiting partial recrystallisation
is in a region of rapid metallurgical change in response to temperature change, and not
possible to predict the exact outcomes within the limits of error of the temperature and
metallurgical models.
5.6.1 Soft Iron Recrystallised Products
The RDHT-CM model was used to simulate soft iron recrystallised products using steady
state input data. Table 5.3 shows the model results for the calculated strip temperature,
recrystallisation amount, and the ‘lowest calculated’ recrystallisation amount, RXmin .
For recrystallised products, the ‘lowest’ calculation is important to examine, to ensure
recrystallisation will take place. If the nominal and lowest likely tolerance are both equal
to 1, this indicates that full recrystallisation is very likely.
For fully recrystallised products, the RDHT-CM model predicted full recrystallisation
in every case (RX = 1), covering the range of gauge sizes from thin (0.39mm) to thick
(1.55mm) gauge. For some of the strip models, the minimum likely calculation shows less
than complete recrystallisation (RXmin < 1) however. This may be due to a number of
179
Grade Dimensions L1 Temp Predicted Predicted Predicted
◦
MPa mm C Z13 Strip ◦ C RX RXmin
300 0.39 x 905 750 705 1.0 1.0
250 0.55 x 1135 721 678 1.0 0.6
300 0.59 x 1260 708 690 1.0 0.8
250 0.75 x 900 708 621 1.0 1.0
250 1.55 x 1220 731 702 1.0 1.0
Table 5.3: Nominal (RX) and Minimum (RXmin ) expected recrystallisation results from
the RDHT-CM model for a selection of soft iron products
factors, such as furnace operating data that was not at steady state as believed, which
affects the temperature results from the simple model used to calculate recrystallisation
in the metallurgical model.
This result of RXmin < 1 does not necessarily mean that less that 100% recrystallisa-
tion is actually occurring, but that the amount occurring could be less than 100% within
the assumed limits of uncertainty of the metallurgical model. These furnace setups have
proven acceptable in practice however, indicating that the ‘nominal’ recrystallisation cal-
culation is correct.
(a) Temperature (b) Recrystallisation
Figure 5.19: Results from the RDHT-CM model, showing the predicted furnace and strip
temperatures and the resulting cumulative recrystallisation with maximum and minimum
likely tolerance values for a soft iron recrystallised product
The range of temperatures causing partial recrystallisation is narrow for any given
product, as shown on the softening curve from recrystallisation trials in Section 2.7.3 and
the metallurgical model equation fit to these curves in Section 4.7.2. Figure 4.43 shows
the calculated and measured temperatures for the ‘recrystallisation start’ and ‘recrystalli-
sation end’ temperatures for Grade 110 hot rolled products across the gauge range. It can
be seen that the average difference between start and end temperatures is close to 55◦ C
180
in all cases i.e. the band in which partial recrystallisation occurs is 55◦ C wide.
For strip temperatures near the partial recrystallisation band of temperatures, the
metallurgical model predicted recrystallisation amount is sensitive to small changes in
temperature. In this region, a small change in calculated strip temperature can have
a large effect on the calculated recrystallisation, as shown in Table 5.4. Increasing the
furnace heating to give a small 4◦ C rise in Zone 13 strip temperature causes the ‘minimum
likely’ recrystallisation calculation to rise from RXmin = 0.8 to RXmin = 1.0.
Grade Dimensions Predicted Predicted Predicted

◦
MPa mm Z13 Strip C RX RXmin
300 0.59 x 1260 690 1.0 0.8
300 0.59 x 1260 694 1.0 1.0
Table 5.4: Recrystallisation results from the RDHT-CM model showing high sensitivity
to temperature change when in the region of partial recrystallisation
(a) 690◦ C Peak Strip Temperature (b) 694◦ C Peak Strip Temperature
Figure 5.20: Example of model sensitivity to temperature. The lowest likely recrystalli-
sation increases from 80% to 100% by increasing peak strip temperature by only 4◦ C
The recrystallisation calculations for these two cases are shown in Figure 5.20. The
increase in peak (Zone 13) strip temperature of 4◦ C required to increase lowest predicted
calculation to 100% is well below the limits of validation of the RDHT temperature model,
as the predicted strip tempeature can only be validated within a range of 10 − 50◦ C
below the L1 pyrometer temperature as shown in Section 4.2.3.2. It is not possible to
distinguish the cause of the model calculating less than 100% recrystallisation for these
soft iron recrystallised products, as the amount of temperature error required to cause
this is below the limit of model accuracy.
This means that for any model result that predicts partial recrystallisation, the exact
amount of recrystallisation calculated is not meaningful, as the uncertainty in the model
181
temperature can have a large effect on the recrystallisation amount. Rather than using
exact recrystallisation amounts from the metallurgical model of the range 0 ≤ RX ≤ 1,
a more meaningful recrystallisation result is RX = 0, 1, or partial .
5.6.2 Hard Iron Recovered Products
The RDHT-CM model was also used to simulate hard iron recovered products using
steady state input data. The results from a range of coil dimensions is given in Table 5.5,
showing that a very low amount of recrystallisation was predicted for each coil type as
expected. The calculated recrystallisation amount is negligible in each case, supporting
the results of the RDHT-CM model for predicting metallurgical outcomes for hard iron
recovered products. This covers a range of strip types, of both thin gauge (0.4mm) and
thick gauge (1.45mm), which require significantly different amounts of power input to the
furnace.
Grade Dimensions L1 Temp Calculated Calculated Calculated

◦
MPa mm C Z13 Strip ◦ C RX RXmax
−10
G550 0.4 x 940 465 434 4e 4e−9
−7
G550 0.55 x 1221 495 479 2e 3e−6
G500 1.45 x 1174 550 506 2e−7 2e−6
G500 1.145 x 992 539 499 1e−7 1e−6
G500 1.219 x 995 525 503 6e−8 5e−7
Table 5.5: Nominal (RX) and Maximum (RXmax ) expected recrystallisation results from
the RDHT-CM model for a selection of hard iron products
A typical graphical result from the RDHT-CM model is shown in Figure 5.21, where
the cumulative amount of recrystallisation never rises above 0%.
While the model would also give apparently correct results for hard iron products
by simply predicted no recrystallisation for any situation, it was shown in Section 5.6.1
that the RDHT-CM model is also accurate in predicting when 100% recrystallisation
occurs in the strip. The RDHT-CM model is capable of predicting the boundaries of the
temperature zone causing recrystallisation in the strip, and it is possible to show that
the model also correctly predicts the region of temperature where the transition from
0% to 100% recrystallisation occurs i.e. when there is partial recrystallisation. This is
demonstrated in Section 5.6.3, by simulating coils which have experienced temperatures
outside of the region usually used for that product.
182
(a) Temperature (b) Recrystallisation
Figure 5.21: Results from the RDHT-CM model, showing the predicted furnace and strip
temperatures and the resulting cumulative recrystallisation with maximum and minimum
likely tolerance values for a hard iron recovered product
5.6.3 Production Coils With Incorrect Heat Treatment
The accuracy of the metallurgical model was examined further using the RDHT-CM
model to simulate actual furnace operational situations which caused the product coils
to fail mechanical property tests. These failures involve either hard iron products being
too soft, which indicates some recrystallisation took place, or soft iron products being too
hard, indicating incomplete recrystallisation.
Section 5.6.3.1 shows an example where strip hardness is higher than expected due to
the failure of strip preheaters, and section 5.6.3.2 demonstrates model results when early
control actions have caused softening in a hard iron strip due to the onset of recrystalli-
sation.
5.6.3.1 Failure of Preheater
The example in Figure 5.22 for G250 0.55x1200mm product shows a sudden decrease
in strip temperature caused by the tripping of the induction preheaters, which affects
coil numbers G611781 and G611782 over a period of 15 minutes. The measured strip
temperature drops from 710◦ C to 630◦ C, causing an increase in hardness above product
specifications for these two coils as shown in Table 5.6, where the average hardness in-
creases from around 56.6 HR30T to 60.4 HR30T. Note that ‘G250’ refers to the desired
tensile strength grade of the coil (250MPa) whereas ‘G6117xx’ refers to individual coil
designation numbers from the plant.
183
Figure 5.22: Example of incorrect heat treatment in two coils caused by tripping of the
induction preheaters
Coil L1 Temperature Hardness Hardness Average

◦
No. C Quarter Position Edge
G611779 710 59 54.2 56.6
G611782 630 62.4 58.4 60.4
Table 5.6: Hardness test results (HR30T) from 2 measurement positions on coils before
and during reduction in strip temperature shown in Figure 5.22
The RDHT-CM model can be tested by simulating this coil type both for normal
operating conditions, and the abnormal condition without induction preheating. Figure
5.23 shows the simple model results for the case of normal operation giving acceptable
recrystallisation such as for coils G611771 and G611779 in Table 5.6.
Figure 5.24 shows the RDHT-CM model simulation of the case where coil number
G611782 is processed at lower strip temperature, due to the failure of induction preheaters.
Note the significantly lower strip temperature at furnace entry (x = 0) on Figure 5.24
compared with the normally processed coil number G611779 in Figure 5.23.
The RDHT-CM model has therefore given the expected result for this case of incorrect
heat treatment of a coil in production. In practice, the strip hardness was measured to
be harder than desired due to the failure of the induction preheaters causing a low strip
temperature in the furnace. The RDHT-CM model has predicted full recrystallisation
under normal operations, as expected for a soft iron recrystallised strip, and when given
the lower strip entry temperature as a new model input has predicted much lower levels
of recrystallisation (partial recrystallisation).
184
(a) Temperature Results (b) Metallurgical Results
Figure 5.23: RDHT-CM model results for G250 0.55x1200mm coil number G611779 under
normal operating conditions showing full recrystallisation
Figure 5.24: RDHT-CM model results for G250 0.55x1200mm coil number G611782 with-
out preheating showing incomplete recrystallisation
The model shows recrystallisation in the case without preheating to be in the region
which is highly sensitive to changes in temperature, as described in Section 5.6.1. This
predicts that recrystallisation is likely to be incomplete (partial recrystallisation), but
the actual amount of recrystallisation occurring cannot be accurately gauged due to the
limits of accuracy of the temperature model. This cannot be compared with exact amount
of recrystallisation actually occurring, as NZS do not routinely conduct microstructural
examination of MCL coils, however the increase in average hardness measurements for
coil G611782 over coil G611779 in Table 5.6 indicate that the amount of recrystallisation
certainly reduced without preheating (hardness increased).
185
5.6.3.2 Early Cycle Change
The RDHT-CM model is used to examine a case where a hard iron recovered product
has experienced higher temperatures at the end of the coil, due to early control actions
by furnace operators preceding a cycle change to ‘soft iron’ recrystallised steel. It is seen
in Figure 5.25 that the normal strip temperature for this coil type is around 550◦ C, but
this increases to 600◦ C near the end of the coil as the operators have reduced linespeed
around 3 minutes before the cycle change when the ‘soft iron’ strip enters the furnace.
Figure 5.25: Example showing high temperature at end of coil G609702 due to early
change of furnace settings before cycle change
The specification for G500 grade steel states a minimum tested tensile strength of
500Mpa. In this case, the tensile tests from the end of this coil show only 368Mpa yield
strength, indicating that partial recrystallisation has occurred in this strip. The RDHT-
CM model is used to simulate two cases - the normal operating case as in Figure 5.26
where the strip temperature is below 550◦ C, and the ‘error’ case in Figure 5.27, where the
strip temperature has been allowed to increase to 600◦ C by the reduction in linespeed.
The normal case in Figure 5.26 predicts no recrystallisation of the strip as expected,
whereas Figure 5.27 shows that partial recrystallisation is likely to start occurring for the
case of high strip temperatures, with the calculation results ranging from 2-9% recrys-
tallisation. As discussed in Section 5.6.1, a partial recrystallisation prediction means the
strip is in the region of high sensitivy to temperature, and the actual amount of recrys-
tallisation could be significantly higher than 9%, due to the combined limits of accuracy
of the temperature and metallurgical models.
186
Figure 5.26: RDHT-CM model results for G500 1.45 x 1174mm coil under normal oper-
ating conditions showing no recrystallisation
Figure 5.27: RDHT-CM model results for G500 1.45 x 1174mm coil under incorrect
operating conditions showing partial recrystallisation
This demonstrates that the RDHT-CM model has predicted the onset of recrystalli-
sation in this strip, where temperature has increased above the normal operating tem-
perature. The onset of recrystallisation is confirmed by the measured tensile strength of
this coil reducing to only 368Mpa, from the minimum required strength of 500MPa under
normal operations.
187
5.7 RDHT-CM Transient Temperature Results
Simulation of transient effects in the furnace requires detailed input of the furnace param-
eters changing over time, such as the power applied to each furnace zone. These inputs
may be single step changes, such as the change in strip dimensions at the time of cycle
change, or may be continuously varying as the power changes under PID control. Any
possible combination of changes can be modelled using the RDHT-CM model, but the
only way to validate the transient behaviour of the model is to programme it to match
actual transient conditions occurring in the furnace.
Many occurrences of cycle change in the furnace are for small magnitude changes in
strip dimensions, speed or temperatures. The main cycle changes of interest however are
the cases where the furnace and strip setup change from hard iron to soft iron, or vice
versa, as they have a large magnitude temperature change, and the greatest opportunity
to produce off-specification product.
Ideally, the MCL schedulers will arrange for the highest temperature of the hard iron
cycles to transition to the coolest temperature of the soft iron cycles, so there is the
smallest temperature change required at cycle change, but this is not always possible
due to production requirements. In cases where the temperature change is too great,
the operators have the option of running a ‘stringer’ of non-production coil through the
furnace until the temperature reaches the desired range. This is preferential to causing
prime material to be incorrectly heat treated, but is not always necessary or possible due
to the availability of stringers.
At the time of cycle change, the furnace zone setpoints are immediately changed to
the new cycle. As the furnace zones take a long time to change, the linespeed and the
induction furnace are manually controlled by the operators to maintain the correct strip
temperature as measured at the L1 pyrometer.
For example, when changing from a hard iron (cool furnace) to soft iron (hot furnace),
the furnace takes a long time to heat up to temperature, so the linespeed is reduced and
induction heating manually adjusted. These are controlled in small steps by the operators,
and could be changed between 5-20 times before the furnace temperature stabilises, which
typically takes between 30-60 minutes. These settings are adjusted to maintain the strip
temperature as close as possible to the target temperature for the new cycle, but there
is no automatic warning when the temperature is outside of the allowable temperature
range.
188
The main difficulty in modelling cycle change is approximating the changes in zone
power to the radiant furnace. The furnace is controlled to temperature set points using
PID control on the applied power, but the RDHT-CM model requires input of the per-
centage power applied. To model the power changes, it is necessary to approximate the
applied power over time as recorded in the CITECT system, either as a series of step
changes or using linear equations.
Figure 5.28: Example of power change in a single furnace zone during two consecutive
cycle changes and a linear equation approximation for the RDHT-CM model
Figure 5.28 shows how the power applied to a single furnace zone can change over time
during cycle changes, and how this can be approximated using a series of linear equations.
The accuracy of this method is dependent on the number of separate equations used to
fit the changes in power, and the complexity of the shape of the power changes.
This is very time-consuming, as it commonly requires manually fitting a series of 5-

10 linear equations to the power change, individually for each furnace zone. As well as
the zone power, it is also necessary to approximate the changes made to the linespeed
and induction furnace. As these are manually adjusted by operators, they are more easily
approximated using a series of step changes. Modelling the changes that occur in a typical
cycle change can require over 50-100 separate linear equations for ramps and step changes
in factors over time.
Section 5.7.1 shows the example of a transient model for a cycle change from a hard
iron, low temperature cycle to a soft iron, high temperature cycle.
189
5.7.1 Cycle Change from Hard to Soft Iron
This example shows the transition from a hard iron (0.4 x 902mm G550) to a soft iron
(0.55 x 1200mm G300). The recorded furnace data, including the zone power inputs,
strip dimensions, linespeed and preheating amount for several coils immediately before
the cycle change time is averaged and used as input for the steady state solution to predict
the initial values before the cycle change. This particular cycle change was selected as the
furnace operating parameters were very steady before the change, meaning the starting
point calculations should be reasonably accurate.
Figure 5.29: RDHT-CM model results for G550 0.4 x 902mm coil using steady state input
data before a cycle change occurs
Figure 5.29 shows the RDHT-CM model result for the G550 0.4×902mm coil before
the cycle change. The strip and wall temperature results from this model are used as
the initial temperature values for the transient model, as described by Section 4.6.2. The
initial model solution is rather poor, as there is a large power imbalance in the heating
zones - Zone 1 has 95% power, and the remaining heating Zones 2-12 use only 38% power
on average. The conduction of heat along the furnace body in the model does not work
well for this large imbalance, with the result that predicted Zone 1 wall temperature is
too high, and remaining heating zone temperatures are too low. The overall heat balance
is still maintained however, as the predicted Zone 13 strip temperature agrees well with
the L1 pyrometer reading.
An examination of the time-series data from the furnace shows an immediate rise in
induction furnace power, and drop in linespeed at the time of cycle change, causing a
rapid rise in strip temperature measured by the L1 pyrometer as shown in Figure 5.30.
The induction heaters shown in Figure 5.31(a) are then slowly reduced in power, and the
linespeed shown in Figure 5.31(b) increased as the furnace zones heat up and approach
190
their new setpoints.
The induction heaters and linespeed are manually controlled to maintain the L1 tem-
perature above 700◦ C, and below approximately 725◦ C. It can be seen from Figures 5.30
and 5.31 that manual control allows the L1 pyrometer temperature to be maintained at a
steady level after less than 5 minutes, while the preheaters (strip entry temperature) are
slowly changed over a 30 minute period as the furnace zones change temperature.
Figure 5.30: Strip temperature measured by L1 pyrometer changing over the cycle change
period
(a) Induction Furnace Power (b) Linespeed
Figure 5.31: Manual changes applied to furnace induction heater power and linespeed
during cycle change
These particular changes in induction heating and linespeed are each approximated
using several step changes and linear equations to roughly fit the setpoint changes made by
operators. The changes in power applied to each heating zone are more difficult to model,
as there are changes occurring continuously in all furnace zones. The power changes
applied by the PID controllers to Zones 1-12 are shown in Figure 5.32 and the power
applied in Zones 13-19 is shown in Figure 5.33.
191
(a) Heating Zones 1-6 (b) Heating Zones 7-12
Figure 5.32: Changes to power input in heating zones 1-12 during cycle change due to
PID control actions
Figure 5.33: Changes to power input in cooling zones 13-19 during cycle change due to
PID control actions
192
This group of power inputs for the heating zones 1-12 is simple to model, as the power
inputs all immediately rise to 100% at the time of cycle change, as the furnace is being
heated up. Zones 1-6 stay at 100% for the duration of this simulation, and the other
heating zones can be approximated using one equation for each zone, to reduce the power
input linearly (from 100%) over the course of the simulation. This is significantly easier
to approximate than many other cycle changes that were investigated. The power inputs
for the cooling zones are more difficult to model, but have little effect on the peak strip
temperature, and so are heavily simplified to reduce the number of equations required.
Once the variation of power inputs, linespeed and induction heating have been approx-
imated over time, and the pre-cycle change steady state has been calculated, the transient
solution of the RDHT-CM model can be used to simulate the cycle change.
For the purpose of evaluating the results of the transient solution, it is useful to choose
data points which can be compared against actual furnace measurements. The main data
point of interest is usually the strip temperature at x = 70m, corresponding approxi-
mately to the position of the L1 pyrometer, and in this case a selection of furnace wall
temperatures are also compared against temperatures measured by the furnace thermo-
couples in these zones. The zones selected for comparison in this case are heating zones 2,
5, 10 and cooling zone 15. The comparison of model predicted strip temperature against
L1 pyro temperature is shown in Figure 5.34, and the comparison of model furnace wall
temperatures with recorded thermocouple temperatures is shown in Figure 5.35 for zones
2 and 5, and Figure 5.36 for zones 10 and 15.
Figure 5.34: Transient solution strip temperature result at x = 70m compared with actual
L1 pyrometer measurement
193
(a) Heating Zone 2 (b) Heating Zone 5
Figure 5.35: Transient temperature solutions from Zones 2 and 5 compared with zone
thermocouple measurements
(a) Heating Zone 10 (b) Cooling Zone 15
Figure 5.36: Transient temperature solutions from Zones 10 and 15 compared with zone
thermocouple measurements
194
The comparison of model predicted strip temperature at x = 70m against the L1
pyrometer temperature, as in Figure 5.34 shows a good agreement with the rate and
magnitude of temperature increase but is too high in the period 0-500s after cycle change,
recalling that the model strip temperature should always be 10 − 50◦ C below the L1
Pyrometer temperature, as shown in Section 4.2.3.2. This brief overshoot in model pre-
dicted temperature may be caused by the large number of approximations made in fitting
equations to the changes in heating power, induction preheating and linespeed.
Comparison of the model wall temperatures in four zones with the measured ther-
mocouple temperatures shows a good agreement with the measured temperature rises.
The rate of temperature change is comparable in each case, although the magnitude of
increase appears higher than for the measured temperatures. As discussed in Section 5.5,
the RDHT model agrees with the F3DC model, and predicted wall temperatures should
usually be 25 − 30◦ C higher than measured thermocouple temperatures.
The predicted wall temperatures following the cycle change are higher than thermo-
couple measurements as expected. The lower than expected initial temperatures before
the cycle change are shown in the initial steady state solution in Figure 5.29, caused by
the large imbalance in power applied to the heating zones described earlier. The predicted
wall temperatures reach the expected offset above the measured temperatures after the
cycle change as the zone power inputs are changed to a more even pattern.
The transient solution of the RDHT-CM model has predicted furnace and strip tem-
peratures which agree well with the recorded temperatures for this particular cycle change,
but these results are dependent on the accuracy of the approximation of changes in the
zone applied power. The PID control system of the zone heaters adds a layer of complexity
and hinders the ability of the model to predict the transient response of the furnace. This
particular situation was modelled acceptably because the change in products resulted in
zone powers post cycle change that were constant or easily approximated.
5.7.2 Other Cycle Change Simulations
Further to the cycle change situation modelled in Section 5.7, several other cycle changes
have been modelled to demonstrate the transient solution of the RDHT-CM model.
The strip temperature results from each of these cycle change simulations is compared
against the measured L1 pyrometer strip temperatures and shown in Figures 5.37 to 5.39.
195
The cycle changes simulated are:
1. 0.442 x 940mm G550 to 0.39 x 905mm G300 (hard to soft iron)
2. 0.39 x 905mm G300 to 0.4 x 940mm G550 (soft to hard iron)
3. 0.39 x 905mm G300 to 0.55 x 940mm G550 (soft to hard iron)
Figure 5.37: Transient strip temperature result and L1 pyrometer measurement for cycle
change from 0.442x940mm G550 to 0.39x905mm G300
Each of the 3 cycle change models in Figures 5.37 to 5.39 show reasonable agreement
with the measured L1 pyrometer temperatures, exhibiting the same shape and rate of
change in temperature, although there is some undershoot/overshoot in each case.
196
The second transient result in Figure 5.38 is a good example of a case where a ‘stringer’
coil would be required, as the time taken to change from hot (750◦ C) to cold (450◦ C) strip
temperatures is very long, around 10-15 minutes, and would cause a significant amount
of product to be incorrectly heat treated. At 140m/min linespeed for this product, 10
minutes of off-specification temperature could cause 1400m of wasted product strip. In
this case, it would be the G550 coil immediately after the cycle change experiencing high
temperatures and possible recrystallisation that would be wasted.
The third transient result in Figure 5.39 is an example of poor control of the induc-
tion preheaters during the cycle change. The spike in strip temperature shown is due to a
sudden power increase in the preheaters, which are manually controlled by the operators.
For this hard iron product, the preheaters should not be used at all, and the strip tem-
perature was nearly at the target level when the manual control action caused a sudden
75◦ C increase in the strip temperature.
197
5.8 Summary of Model Results
The modelling of the MCL furnace consisted of two separate approaches, being the three-
dimensional F3DC model discussed in Section 4.5, and the simplified RDHT model using
1D/2D geometry representations shown in Section 4.6. The RDHT model was further
enhanced by the integration of a metallurgical model, as described in Section 4.7 to create
the final RDHT-CM model. The results generated by the F3DC and RDHT-CM models
presented in Sections 5.3 to 5.7 each have distinct uses, due to their varying complexity,
type of outputs available, and speed of solution.
The typical time required to calculate the F3DC model solution approaches 2-3 hours
for each of the three separate COMSOL models (Zones 1-2, 13-14 and 15-19) using a
dedicated modelling computer (64-Bit, Quad-Core 2.6Ghz, 8GB Ram). To solve the
F3DC model for all three sections may therefore take up to 9 hours of computing time,
and is obviously not practical for simulating many alternative options or making decisions
based on model results in the production environment. The model is also difficult to
modify and operate, and uses proprietary software.
The strength of the F3DC model is that it does not involve extensive simplification
of the furnace geometry, although some compromises have to be made, especially for the
shape of the heating elements. The relatively accurate geometry therefore allows the
F3DC model to be used for investigation of the spatial effects of radiation inside the
furnace cavity, which has a strong effect on the temperature of the thermocouples used to
measure furnace temperatures, and the variation of temperature across the width of the
strip.
To predict the strip and furnace temperatures with a fast solution time required a high
degree of simplification of the furnace system in the construction of the RDHT model,
as described in Section 4.6.2, but this enables solution of the model equations in around
10 seconds on a typical desktop computer. The RDHT model is programmed using the
MATLAB language, but is easily translated into other codes for use without requiring
proprietary software, such as for integration into the NZ Steel computer system.
The purpose of the RDHT-CM model is to enable the prediction of strip and wall
temperature and strip metallurgical change inside the furnace in a way that is flexible and
fast to solve. This enables it to be used to rapidly investigate operational alternatives,
and optimise the control of the furnace to improve quality, production rate and energy
efficiency. Demonstration of this will be shown in Chapter 6. The RDHT model results
198
do not allow investigation of spatial temperature effects such as strip edge heating or
thermocouple temperature changes, but understanding of these behaviours can be gained
from the F3DC model.
Validation of the model temperature results for both the RDHT and F3DC models is
difficult due to the very limited accuracy of the temperature measurement devices in use
in the MCL annealing furnace. The L1 pyrometer in furnace zone 13 is used in all cases to
validate strip temperature results, as it has been shown to be accurate within 10 − 50◦ C
in Section 4.2.3.2. The furnace zone thermocouples are used to validate model furnace
temperatures, and understanding of their readings has been enhanced by mathematical
investigations and results from the F3DC model in Section 5.3.3.
In almost all cases, the RDHT model has predicted strip temperatures that lie within
the expected band of 10 − 50◦ C below the temperature recorded on the L1 pyrometer, for
all combinations of thick and thin gauge, long and narrow width, and hard and soft iron
products. This shows that the model is robust, and useful across the range of product
types used in the furnace. While the RDHT model uses great simplification of the furnace
geometry, the radiation and furnace wall equations used are capable of dealing with the
wide range of strip dimensions and furnace temperatures likely to occur.
Comparison of the RDHT and F3DC model results in Section 5.5 shows that the
simplifications used in the RDHT model gave results that agree very closely with the F3DC
model. This demonstrates that the RDHT model is an effective method of calculating
furnace and strip temperatures, when specific results that can be provided by the F3DC
model such as temperature distribution and increased heating of the strip edges are not
of interest.
The RDHT-CM model is capable of predicting the softening of the strip within the
limitations of available plant data. The limits of validation of the temperature calcu-
lation and the low accuracy of recrystallisation measurements in practice mean that an
exact percentage recrystallisation occurring cannot realistically be calculated, however the
model has been shown to correctly identify cases where full recrystallisation occurs, and
when recrystallisation will not occur. Further to this, the RDHT-CM model has been able
to predict cases where partial recrystallisation has occurred due to poor furnace control.
While the exact percentage recrystallisation calculated by the model is not meaningful,
it can be used to predict when the strip is in danger of either unwanted recrystallisation
for hard iron recovered products, or when temperature is unlikely to be sufficient for full
199
recrystallisation of soft iron recrystallised products.
The RDHT-CM model was also used to define a region during recrystallisation with
high sensitivity to temperature, where small changes in the strip temperature profile have
large effects on the calculated recrystallisation amount. This is an area where control
of recrystallisation of the strip is very difficult due to the combination of poor temper-
ature measurement and high sensitivity to temperature. This area should be avoided
during production, as it is likely to result in incorrect heat treatment of the strip for both
recrystallised and recovered products.
The transient temperature solution of the RDHT-CM model was demonstrated by

simulation of cycle changes in the MCL furnace. It was shown that good agreement
can be found with the changing wall and strip temperatures, however the model results
are dependent on a large number of approximate linear or step change equations fitted
to the changes in furnace parameters over time. This is done manually, and a different
method for model data input or approximation would be required for practical, real-time
simulation of furnace and strip transient temperature.
200
Chapter 6
MCL Furnace Optimisation
6.1 Introduction
In this chapter, the use of the RDHT-CM model for furnace control optimisation is demon-
strated by simulation of existing and alternative operating conditions for some of the
most popular strip products in the NZS MCL Annealing Furnace. Section 6.2 identifies
the most common product types processed in the plant, and Section 6.3 demonstrates
how the RDHT-CM model can be applied to some of these products to optimise furnace
setups, or find furnace settings for alternative conditions.
6.2 Selection of MCL Products
There is a very large range of product sizes and specifications processed on the NZ Steel
Metal Coating Line. Some combinations (product types) are very popular, and are often
run on the MCL, whereas some types may only run for special orders and occur very
rarely. When applying the RDHT-CM model to analyse furnace setups, it is obviously
most useful to start by investigating the most common products, as these maximise the
amount of gain relative to modelling effort.
Table 6.1 shows the factors describing the product types processed on the MCL, and
the range in these factors occurring in the 12 months preceding May 2008. These factors
can be combined into a single ‘tag’ describing the product type. For example, the tag
0.4 940 G550 A 24A indicates a strip 0.4 × 940mm dimensions, heat treated to G550
201
Factor Range
Gauge 0.3 - 2.25mm
Width 638 - 1260mm
Tensile Strength 250 - 550Mpa
Coating Type Galvanised (Z) or Zincalume (A)
Furnace Cycle 4-30 (A/B)
Table 6.1: Factors describing the range of product types processed on the MCL
grade (550Mpa tensile strength), with Zincalume (A) coating, using furnace cycle setup
24A. This single tag format allows each coil processed to be indexed with a tag, and the
weight of coil processed with each tag summed to give the total weight production of each
product type.
Rank Product Tag Tonnage Wt. Percent

1 0.4 940 G550 A 24A 45309 20.8%
2 0.39 905 G300 A 15B 22875 10.5%
3 0.55 938 G300 A 14B 11069 5.1%
4 0.55 1200 G300 A 14B 7358 3.4%
5 0.55 1221 G550 A 29A 5987 2.7%
6 0.55 940 G550 A 29A 5484 2.5%
7 0.59 1260 G300 Z 14B 4100 1.9%
8 0.42 940 G550 A 24A 3964 1.8%
9 0.442 940 G550 A 25A 3810 1.7%
10 0.4 940 G550 Z 24A 2520 1.2%
Table 6.2: The 10 most common products by weight processed on the MCL in the 12
months preceding May-2008
Table 6.2 gives the product specifications, product tonnage and percentage of total
tonnage for the 10 most common products processed on the MCL in the 12 months
preceding May 2008. Each product type also has further variation regarding the coating
weight (thickness) applied to the strip, but this factor has no effect on furnace operation.
After discounting coating weight variation, there were still 750 distinct products processed
on the MCL during these 12 months.
The top 10 products shown in Table 6.2 represent over 51% of the total production,
although it is seen that the top 2 products contribute over 31% of the total themselves.
Below the top 11 products, each type represents under 1% of annual production, and
hence there are rapidly diminishing returns to investigatie furnace setups for any of these
products individually. Over 600 of the 751 products represent less than 0.1% of annual
production each, although many of these products are likely to be minor variations of
202
other products.
The RDHT-CM model was used to simulate these top 10 products, first using the
standard steady state furnace settings to create a ‘base’ case. Various options for furnace
setup were also modelled, to look for any opportunities to either increase production by
increasing linespeed while maintaining acceptable heat treatment, or to decrease power
usage by reducing either zone or induction heating. The products used for this analysis
comprised both hard and soft irons, in both Zincalume and Galvsteel coatings.
6.3 Simulation of Selected Products
Simulations of the top 3 products are shown here, including examination of alternative
operating setups for the MCL furnace. As per Table 6.2, Products 1, 2 and 3 are examined
in detail in Sections 6.3.1 to 6.3.3.
6.3.1 Product 1. 0.4 940 G550 A 24A
The total yearly tonnage of the first ranked product is approximately double the tonnage
of the second ranked, and four times the third ranked, so the most modelling effort and
analysis was focussed on this product, being 0.4mm x 940mm, G550 steel with Zincalume
(A) coating on furnace cycle 24A. This is the same product used for validation of the
F3DC model in Section 5.3.1.
Table 6.3 shows the applied power input to the RDHT-CM model, as extracted from
steady state coil summary data for this product, as well as the average measured tem-
peratures for each zone, and the condition of each of the heating elements. A factor of
1 for heating elements indicates that the elements in that zone are undamaged and run-
ning at full capacity. This is reduced to 0.66, 0.5 or 0 if the elements are damaged, as
described in Section 4.2.1. Further data input required is given in Table 6.4, containing
the strip dimensions and physical properties, and plant control factors of linespeed and
strip temperature post-preheating.
The condition of the elements is not recorded in coil summary data, and must be
found from plant maintenance technicians. The CITECT computer system records only
the percentage of power applied by the PID controllers, and so will show 100% power for
203
cases when damaged elements are running at their full, but reduced capacity. Failure to
include the element condition in cases where elements are damaged will result in higher
power levels applied to the model than in reality, and model temperature results will be
higher than expected.
Zone Power Input % Temperature (◦ C) Elements

1 46.5 785 1
2 48.3 785 1
3 46.8 780 1
4 47.7 775 1
5 40.7 765 1
6 38.7 755 1
7 36.7 750 1
8 33 740 1
9 37.5 735 1
10 36.3 735 1
11 25.8 725 1
12 40.7 720 1
13 100 497 1
14 100 291 1
15 98.4 530 1
16 85.5 525 1
17 78.7 510 1
18 92.8 510 1
19 75.5 500 1
20 57 520 1
Table 6.3: Applied power to each furnace zone, zone target temperature and element
condition applied to the RDHT-CM model for Product No.1
The steady state solution of the RDHT-CM model for the data inputs in Tables 6.3
and 6.4 are shown in Figure 6.1. These results show the model predicted temperatures to
match well against the temperatures recorded in the furnace, although as in Section 5.3.3,
the calculated temperatures could be expected to be higher, typically around 25 − 30◦ C
higher than the recorded thermocouple temperatures. The predicted wall temperatures
in this model are on average around 10◦ C higher than the thermocouple temperatures.
The strip temperature prediction appears sufficient also, as the predicted strip temper-
ature at the position x = 70m is 455◦ C, which is 10◦ C lower than the average recorded L1
Pyrometer temperature of 465◦ C, although there is strong variation in the recorded tem-
perature from 443 - 489◦ C. This is just within the range of the 10 − 50◦ C offset expected
below the L1 pyrometer temperature, as shown in Section 4.2.3.2.
204
Strip Property Value
Gauge 0.4mm
Width 940mm
Linespeed 130m/min
Grade G550
L1 Pyro Min Reading 443◦ C
L1 Pyro Avg Reading 465◦ C
L1 Pyro Max Reading 489◦ C
Chemistry Grade 100
Cold Reduction 0.826
Entry Temperature 50◦ C
Table 6.4: Strip properties and plant operating data applied to the RDHT-CM model for
Product No.1
(a) Temperature Results (b) Recrystallisation Results
Figure 6.1: Temperature and recrystallisation results for RDHT-CM model simulation of
Product No.1 - 0.4 940 G550 A 24A
Figure 6.1(b) shows that zero recrystallisation is calculated for this product, which is
expected for G550 grade ‘hard iron’ recovered steel. The recrystallisation temperature for
0.4mm Chemistry Grade 100 steel found from NZ Steel recrystallisation tests as shown in
Section 4.7.2 was 570◦ C, and so no recrystallisation should occur in this strip with peak
temperature around 460◦ C.
At no point does the recrystallisation calculation rise above 0%, which should be
maintained for all alternative options investigated for this product, as a calculation show-
ing partial recrystallisation indicates that the strip is in the region of high sensitivity to
temperature on the recrystallisation curve, as described in Section 5.6, and the strip is
likely to be unsuitable to meet the 550MPa desired tensile strength. Several alternative
options for processing this strip type are now investigated to demonstrate the use of the
RDHT-CM model.
205
6.3.1.1 Increased Line Speed
During 2008, MCL Technologists undertook a trial to increase the maximum allowable
furnace linespeed from 130m/min to 140m/min. This trial was successful, and most
products running at 130m/min were subsequently increased to run at 140m/min with an
increase in heating power to maintain strip temperatures. For this particular product,
increasing the linespeed required an increase in heating power only to Zone 1. Steady
state coil data shows the average power requirement in this zone increased from 46.5% to
95.1%.
Product No.1 after increasing linespeed to 140m/min
Figure 6.2 shows the RDHT-CM model results for the case of 140m/min linespeed,
with increased heating power to Zone 1 only. This model also shows very good agreement
with all measured values, other than in heating Zone 1, which appears significantly hotter
than measured. It is likely that this is due to limited conduction along the x-direction
in the RDHT-CM model, and is not considered important in these results, as it does not
affect the other zone temperatures or the strip temperature. The strip temperature in
this case also stays below the approximate temperature required to start recrystallisation,
shown by the model recrystallisation results equal to zero.
Measurement Base 130m/min 140m/min

Zone 13 Strip Temp, ◦ C 455 453
Total Electric Heating, kW 2755 2940
Production Rate, T/hr 23.0 24.8
Specific Energy, kWhr/T 119.8 118.3
Table 6.5: Comparison of Product No.1 base case of 130m/min, and with increased line-
speed of 140m/min
206
Table 6.5 shows the comparison of some key measurements for the base case of
130m/min, and the alternative 140m/min. This shows that as well as increasing the
absolute rate of production, the case with faster linespeed also has lower specific energy
consumption, representing a decrease in production cost when compared against the orig-
inal case. This is because the increase in power required for the faster strip is relatively
lower than the increase in production.
The production rate is directly proportional to the MCL linespeed, which increases
by 8% from 130m/min to 140m/min. Given the limits of accuracy of the RDHT-CM
temperature calculation, the 1% decrease in specific energy consumption may not be
significant, however there is certainly an increase in production rate without incurring a
penalty in energy efficiency.
This alternative is a good choice when running the furnace in steady state conditions,
although it does not consider the other parts of the MCL plant shown in Figure 1.1. The
effect of increased linespeed on their power consumption and operability should first be
investigated to determine if there is a net benefit in running at the higher linespeed. When
restricting evaluation to the furnace system, it does appear that the case of 140m/min
has both efficiency and production rate advantages over the base 130m/min situation.
6.3.1.2 Alternative Heating Using Induction Preheater
This alternative investigates how the induction heater could be used at 140m/min line-
speed, instead of increasing the amount of power to the first heating zone. To accomplish
this, the lower zone heating power required for 130m/min was used, and the strip entry
temperature increased until the peak strip temperature model result matched with the
base case temperature. The power required in the induction preheaters to achieve this
entry temperature was found from NZ Steel induction heater data, and compared with
the increase in zone 1 heating power required to achieve the same peak strip temperature.
The induction preheaters are not commonly used at all for this product type, so the
induction power required was calculated by interpolation between similar sizes of soft iron
products that do use the induction furnace. The results summarised in Table 6.6 show the
power increase using the induction heater is significantly more than the power required
using the electric radiant furnace only, meaning this option is much less efficient than the
current standard setup using increased Zone 1 heating for 140m/min production.
207
Measurement Base 130m/min 140m/min Preheated
Strip Entry Temp, ◦ C 45 45 88
Zone 13 Strip Temp, ◦ C 455 453 456
Total Electric Heating, kW 2755 2940 2755
Induction Heating, kW 0 0 520
Production Rate, T/hr 23 24.8 24.8
Specific Energy, kWhr/T 119.8 118.3 132.1
Table 6.6: Comparison of Product No.1 existing 140m/min case, and alternative using
preheating instead of increased Zone 1 power
The energy cost of using the preheaters instead of radiant heating is over 14kWhr/T.
It appears that use of the preheaters is not power efficient for small increases in entry
temperature, and the preheaters should only be utilised in cases where zone heating is
already at maximum capacity i.e. for heavy gauge, recrystallised products.
6.3.1.3 Reduced Linespeed and Heating Power
This section deals with an alternative option of reducing linespeed below the previous
standard 130m/min. This simulates the case where overall power consumption may be-
come more important to New Zealand Steel in the future. While this directly reduces the
amount of production from the MCL, it is also useful to look for options where greater
overall heating efficiency may be found.
To simulate this case using the RDHT-CM model, the linespeed value was reduced
from 130 m/min to 120 m/min, and the heating zone power inputs were reduced evenly
until the model predicted peak strip temperature was reduced to the same value as for
the base case simulation of 130m/min.
Measurement Base 130m/min 120m/min

Zone 13 Strip Temp, ◦ C 454 456
Total Electric Heating, kW 2755 2650
Production Rate, T/hr 23.0 21.3
Specific Energy, kWhr/T 119.8 124.4
Table 6.7: Comparison of Product No.1 base case of 130m/min, and with reduced line-
speed of 120m/min
The results in Table 6.7 show that the reduced linespeed case requires 105kW less
heating power than the base case, but this reduction is smaller than the reduction in
production rate, so that the specific heating power required is higher. As per Section
208
6.3.1.1, this does not include power consumption of other MCL plant equipment such as
the drive motors, which will also decrease with decreased linespeed and possibly make
this situation more favourable.
This result is consistent with the case of increased linespeed in Section 6.3.1.1, as the
simulation of 140m/min was found to be more power efficient than 130m/min. Similarly,
130m/min was found to be more efficient than 120m/min when using only electric zone
heating power. This reduced linespeed option does not appear to be a good option in
terms of power efficiency, but may be useful in the case where overall power consumption
needs to be reduced for cost savings etc.
6.3.1.4 Maximum Allowable Temperature
This simulation was used to find the maximum strip temperature before recrystallisation
occurs in this hard iron product. This information can be useful during cycle changes
and during cases of off-gauge strip ends, which are both causes of sudden strip temper-
ature spikes. This is also done to assist with validation of the RDHT-CM model, as it
was commonly understood by furnace operators and technologists that the highest strip
temperature for hard-iron products should not exceed around 550 − 560◦ C.
This test was conducted using base case furnace setup for this product type at 130m/min,
and gradually reducing the linespeed until the RDHT-CM model predicted 1% recrystalli-
sation at the highest likely calculation (RXmax ).
Measurement Reduced 88m/min

Zone 13 Strip Temp 560◦ C
Nominal Recrystallisation (RX) 0.19%
Max Recrystallisation (RXmax ) 0.99%
Table 6.8: Results for reduced linespeed case to find maximum allowable strip temperature
As shown in Table 6.8, recrystallisation calculations rose above negligible levels once
the strip temperature reached 560◦ C. This result matches well with the knowledge of
experienced operators and technologists, and also agree with NZ Steel recrystallisation
temperature experiments shown in Sections 2.7.3 and 4.7.2, where a temperature of 570◦ C
was found to cause 10% softening on the experimental curve for this strip gauge.
There are many cases where the strip must be run considerably colder however, espe-
cially in the case of very thin gauge product. The 0.4mm gauge simulated here is a good
209
example, as it has a high occurrence of wavy-edge defects. These are likely to occur due to
the elevated strip temperatures at the edges of the strip, which can cause localised soften-
ing and recrystallisation as shown in Section 5.3.4. This product is run with a maximum
allowable L1 pyrometer strip temperature of only 460◦ C as measured at the start of Zone
13 to prevent this wavy-edge defect. From a selection of coils operating at steady state,
it was seen that the hard iron products ran with typical L1 pyrometer temperatures of
420 − 480◦ C for thin gauge of less that 0.5mm, and 500 − 560◦ C for heavy gauge products
of 1-2mm.
6.3.2 Product 2. 0.39 905 G300 A 15B
Similar trials were also run for the second ranked product type in Table 6.2, being product
specification 0.39 905 G300 A 15B. The main difference between these two products is
that product 2 is a soft iron rather than hard, and instead of avoiding recrystallisation
in each case it is necessary to maintain full recrystallisation. The standard setup for this
product has also been increased from 130 to 140m/min linespeed during 2008 using both
increased electric radiant heating and increased induction preheating.
Product No.2 with 140m/min linespeed
The steady state temperature and recrystallisation results for this product shown in
Figure 6.3 again agree well with the measured furnace and strip temperatures. The
average difference between model and thermocouple temperatures in the heating zones is
27◦ C, which is close to the difference expected as per Section 5.3.3. The zone 13 model
strip temperature at x = 70m is 9◦ C below the recorded L1 pyrometer strip temperature
measurement, which is at the edge of the range of expected temperature difference as per
Section 4.2.3.2.
210
The strip temperature predicted at the position of the L1 pyrometer is 754◦ C, which
is well above the ‘recrystallisation stop’ temperature of 640◦ C found from the NZ Steel
recrystallisation trials in Section 2.7.3. The recrystallisation result curves are as expected
for this situation of a ‘soft iron’ recrystallised steel, with all three curves reaching 100%
recrystallisation.
Measurement 130m/min 140m/min 125m/min

Strip Entry Temperature, ◦ C 92 230 45
Preheating Power, kW 500 1200 0
Production Rate, T/hr 21.7 23.3 20.8
Table 6.9: Comparison of alternative options for Product 2 - 0.39 905 G300 A 15B
The RDHT-CM model is used to calculate furnace and strip temperatures for the
existing cases of the old 130m/min setting, and the new furnace setup with 140m/min
strip speed in Table 6.9, with the induction heater power requirement found from plant
measurements. The RDHT-CM model is then used to predict the highest linespeed that
will maintain required recrystallisation without preheating of the strip (125m/min).
As shown in Table 6.9, increasing the amount of induction preheating of the strip
strongly reduces the energy efficiency. There is a significant reduction in energy consump-
tion when the induction furnace is not used, with linespeed reduced to 125m/min to main-
tain the same strip temperature, saving 16kWhr/T against the base case of 130m/min.
This represents 12.5% saving in energy requirements, and costs only 3.9% reduction in
production rate. This option appears highly desirable in the case where energy consump-
tion becomes most important to plant managers.
This situation was also found for the 0.4×940mm product investigated in Section
6.3.1.2, where using induction heating instead of electric radiant heating has a strong
penalty in energy consumption. The induction heater units are much less power efficient
than the electric radiant furnace, and should not be used preferentially to radiant heating.
Induction preheating may be best used in cases where the radiant furnace is running at
100% power already, or when short runs of heavy gauge product are required, as the
induction heaters can be used to compensate for the change in strip dimensions without
the time delay for the radiant furnace to change temperature.
211
6.3.3 Product 3. 0.55 938 G300 A 14B
The third ranked product type in Table 6.2 is product specification 0.55 938 G300 A 14B.
The steady state results from the RDHT-CM model running at the pre-2008 standard
linespeed of 130m/min are shown in Figure 6.4, and two alternative options of running at
140m/min with increased electric heating to compensate, and running at lower 105m/min
linespeed without requiring preheating are shown in Table 6.9.
Product No.3 with 130m/min linespeed
Measurement 130m/min 140m/min 105m/min

Induction Preheating Power, kW 1171 1217 0
Strip Entry Temperature, ◦ C 200 200 45
Production Rate, T/hr 31.6 34.0 25.5
Table 6.10: Comparison of all alternative options for Product 3 - 0.55 938 G300 A 14B
The RDHT-CM model temperature results in Figure 6.4 agree reasonably well with
the measured results. The average predicted wall temperature in the heating zones is
50◦ C below the average recorded thermocouple temperatures from the CITEC system.
The zone 13 strip temperature at x = 70m is 21◦ C below the measured L1 pyrometer
strip temperature at this location, well inside the expected range of temperature difference
expected as in Section 4.2.3.2.
The recrystallisation results show the minimum calculated recrystallisation (RXmin ) to

be well below 100%, but the nominal recrystallisation easily reaches full recrystallisation.
The actual product runs at this strip temperature without problem, indicating the nominal
212
calculation of 100% recrystallisation is correct. The predicted L1 strip temperature is
699◦ C, which is well above the ‘recrystallisation stop’ temperature of 645◦ C.
The alternatives in Table 6.10 show that increasing the linespeed to 140m/min gives
a benefit in specific energy consumption of kWhr/T produced, which is similar to that
seen in Product 1 in Section 6.3.1. This is simulated using the RDHT-CM model by
increasing the electric heating power to maintain zone 13 strip temperature, with only a
small increase in induction heating to maintain the same strip entry temperature as for
the 130m/min base setup.
The best energy efficiency was found with a further alternative, where induction pre-
heating is not used at all, which is in agreement with the cases considered previously.
This was simulated by disabling preheating, reducing the strip entry temperature, and
also reducing the linespeed until the required zone 13 strip temperature was achieved as
for the base case. This setup gives an overall increase in energy efficiency of 7.7 kWhr/T
over the base case, at the expense of 6.1 T/hr production rate.
213
6.4 Furnace Optimisation Summary
It was found in all three strip products examined in Sections 6.3.1 to 6.3.3 that increasing
the linespeed from the previous standard speed of 130m/min to 140m/min gave higher
production rates and lower specific energy requirements for heating the steel. This is true
for cases where the radiant furnace electric heating power only is increased to maintain
the strip temperature. Use of the induction preheater to maintain strip temperature
appears to have a strong negative impact on the specific energy requirements of the strip,
showing that in these situations the induction heaters are much less energy-efficient than
the radiant furnace.
As the standard linespeed used for all three of these products was increased from
130m/min to 140m/min during 2008, the alternative furnace setups for these two line-
speeds were investigated using recorded plant data, as the actual power applied to the
radiant furnace and induction heaters was known in each case. The RDHT-CM model was
also used to investigate alternative options for which no recorded plant data was available.
The alternative setups were adjusted until the predicted strip temperature matched the
temperatures found in the base cases that used actual plant data. This allows predicted
heating energy to be compared to find methods of furnace operation with the lowest
specific energy consumption.
This is shown by the example of Product 1 in Table 6.6, where two options are exam-
ined for increasing the furnace linespeed from 130m/min to 140m/min. The strip temper-
ature is maintained by either increasing the radiant zone heating only, or by maintaining
zone heating and using the induction preheater to increase strip entry temperature. The
case of using radiant heating only is now evidenced by plant data, where the increase
in production rate of 1.8 T/hr (7.8%) requires a 185kW (6.7%) increase in zone heating
power. The alternative using preheating was simulated, where it was found that the strip
entry temperature must be increased from 45◦ C to 88◦ C to maintain peak strip tempera-
ture, which requires an additional 520kW (18.8%) power consumption in the preheaters.
This shows that the increased linespeed using only radiant heating is more energy effi-
cient than the base 130m/min linespeed, but that using induction preheating makes it
signficantly less energy efficient than the base linespeed case.
The inefficiency of the induction preheaters is further shown when preheating is turned
off in cases in which it is normally used, and furnace linespeed reduced to maintain strip
temperature. In each of the cases in Sections 6.3.1 to 6.3.3, there is a significant benefit
214
in specific energy consumption by turning off the preheaters. This could be important to
NZ Steel operations in the case where energy use and expenditure needs to be optimised
or reduced.
While using the induction preheaters are inefficient compared to the radiant furnace,
they allow significant increases in production rate, particularly for recrystallised coils
which run with full radiant furnace heating capacity. For example, Product 2 in Section
6.3.3 is commonly run at 140m/min using the induction heaters, but can only run at a
maximum of 125m/min if the preheaters are not used. Despite the power inefficiency, the
induction preheaters are an important tool to maximise production rate, as the MCL is
a bottleneck in the production chain at New Zealand Steel.
215
Chapter 7
Conclusions and Further Work
7.1 Conclusions
This thesis describes a method to predict temperature evolution and metallurgical changes
inside a continuous annealing furnace under both steady state, and strongly unsteady state
thermal conditions. The models developed here allow a priori prediction of furnace tem-
perature and metallurgical changes in a practical way - where predictions are achieved
in real time before furnace control actions occur. These models can be used to optimise
furnace settings and control to prevent product wastage, and to increase the production
rate and energy efficiency. Existing temperature measurement methods are strongly af-
fected by thermal radiation in the furnace, causing poor control during changes in furnace
temperature, strip dimensions or plant linespeed.
7.1.1 F3DC Model Results
A set of three 3D finite-element models comprising the F3DC model was constructed
to simulate a continuous annealing furnace using COMSOL Multiphysics software as
described in Section 4.5. The 3D modelling approach allows an accurate representation
of the full furnace geometry of the NZS MCL furnace to be constructed, which includes
all of the main components of the furnace such as heating elements, hearth rolls, cooling
tubes and thermocouples, as well as complete representations of the refractory walls and
the steel strip.
216
It was found that the finite-element modelling approach was suitable to accurately
predict furnace temperature evolution and distribution in three dimensions, under both
steady state and changing temperature conditions, where the temperatures and bound-
ary conditions change over several time scales. This research shows that the internal
components and geometry of the furnace have strong effects on the heat transfer rates,
temperature distribution and temperature measurement within the system.
Modelling individual furnace thermocouple probes was shown to suitably predict the
temperature recorded by the actual furnace thermocouples, but also showed complex
temperature variation along the length of the thermocouple units which was not predicted
by simpler mathematical calculations previously and would benefit from further modelling
work. The prediction of the temperature difference between thermocouple probes and the
adjacent furnace wall is an important finding, as it allows comparison of modelled wall
temperatures against recorded thermocouple temperatures. This is required create a
simplified furnace model which does not explicitly model the thermocouple probes.
The 3D finite-element modelling approach was shown to be suitable to predict tem-

perature distribution across the moving strip, which exhibits a strong edge-heating effect
due to the radiative heat transfer and furnace geometry. The strip edge temperature was
found to increase fastest in the early stages of heating, where the radiative heat transfer
is strongest, and eventually decrease to the same heating rate as the bulk strip i.e. the
temperature rise is constant by the end of the heating zones of the furnace. It is predicted
that the strip temperature increases progressively from the strip centre towards the edges,
with a strong edge heating effect on the outside 5mm of strip width only.
Given understandings developed of the differences between actual furnace and strip
temperatures and the radiative effects on the recorded measurements, the F3DC model
predictions compare favourably with measured temperatures, although validation is diffi-
cult due to the large variations in the furnace applied power and the range of temperature
measurements from the thermocouples and pyrometers. For the product types investi-
gated, the predicted thermocouple temperatures in the heating zones were commonly
within 5 − 10◦ C of the actual furnace thermocouple measurements. It is concluded that
the average model thermocouple temperature matches closely to the average measured
furnace thermocouple temperature.
The strip temperature exiting the heating zones as predicted by the F3DC model agrees
well with strip temperature measured by the furnace ‘L1’ Pyrometer in this location. This
pyrometer has been shown to read higher than the actual strip temperature measured by
217
gold cup pyrometer, as shown in Section 4.2.3.2. The predicted strip temperature for
all product types investigated is within the band of expected actual temperatures, offset
below the L1 pyrometer reading.
It has been shown that the 3D finite-element modelling approach of the F3DC model
is sufficient to predict temperature evolution inside the complex heat transfer system of
a continuous annealing furnace. Solution of the F3DC model however is computationally
expensive and time consuming and cannot therefore be used to improve process control,
which requires the simulation of a large number of alternative operational settings and
real time simulation of transient temperature effects due to control actions and product
changes.
7.1.2 RDHT Model Results
The RDHT model is a simplification of the complex annealing furnace system, developed
by approximating the 3D furnace geometry as a 2D wall domain and a 1D strip domain
linked by simple radiative heat transfer calculations. This model was written using MAT-
LAB software, which can be easily translated into other languages for use by NZ Steel
and can be easily modified and maintained in future.
The RDHT model does not contain the accurate furnace geometry of the F3DC model,
and does not contain individual representations of the furnace thermocouples, heating
elements, cooling tubes or hearth rolls. The simplification of the furnace in this manner
allowed a set of equations to be developed that can be solved rapidly without specialised
computer hardware, and so can be used to predict furnace and strip temperatures in a
practical manner, before changes occur in the real furnace. The F3DC model is used
to enhance understanding of the results from the simplified RDHT model, which cannot
otherwise be related to the thermocouple temperature measurements.
The RDHT model was shown to agree very closely with the furnace and strip tem-
perature results from the F3DC model in Section 5.5. This shows that the important
physics of the furnace system have been retained in the RDHT model, and that the re-
duction of furnace geometry to 1D and 2D shapes is a reasonable method to simplify the
furnace model. The wall and strip temperatures in the RDHT model are shown to agree
closely with the F3DC model predictions, and so the differences between the recorded and
predicted temperatures can be expected to be the same in both models. This includes
the thermocouple temperature variation, difference between pyrometer and actual strip
218
temperatures, and temperature rise at the strip edges.
The RDHT model temperature results were examined over a wider range of product
types than possible with the computationally expensive F3DC model, and it was found
that predicted strip temperature agrees very well with the expected temperature, being
in all cases within the expected temperature offset range below the L1 pyrometer reading.
This is a critical result from the RDHT model, as the strip temperature is used to calculate
the metallurgical changes in the strip when coupled with the metallurgical model. The
strip temperature prediction appears to be as accurate as can be expected given the errors
in actual temperature measurements used for model fitting and validation.
The predicted wall temperatures from the RDHT model are slightly higher than ex-
pected, but improvement of wall temperature prediction is difficult due to the strong
variation in furnace input power seen even for products running under steady state con-
ditions. Understanding of the wall temperature prediction could be improved by further
modelling of thermocouple temperatures. This would require a completely new model
containing very detailed representations of the thermocouples and heater elements, which
is not possible for a full furnace model as discussed in Section 4.5.2.
It has been shown that a model containing reduced complexity of furnace geometry
and heat transfer calculations is suitable for predicting furnace and strip temperatures
inside a continuous annealing furnace, and can provide predictions in real-time that are
suitable for process control and optimisation. Simplification in this manner was enabled
by investigation of furnace temperature evolution and distribution using a complex three
dimensional model, which revealed effects important to strip heating and furnace control
that could not be calculated using a simplified model only.
7.1.3 RDHT-CM Model Results
A metallurgical model was developed using established calculations for the recrystalli-
sation process, and fitted to the steel compositions used by NZ Steel. The prediction
of recrystallisation ‘start’ and ‘stop’ temperatures using the model was shown to agree
closely with recrystallisation temperature trials conducted by NZ Steel across the gauge
range. This metallurgical model was then coupled to the RDHT temperature model,
which allowed prediction of the fractional recrystallisation occurring in the steel.
219
The RDHT-CM model with temperature-metallurgical coupling was shown to predict
100% recrystallisation for all simulations of ‘soft iron’ recrystallised steels, and also to
predict 0% recrystallisation for all simulations of ‘hard iron’ recovered steels. The cou-
pled model was used to simulate specific cases where errors in furnace operation caused
incorrect steel heat treatments. In these simulations, the RDHT-CM model was able to
predict partial recrystallisation both for cases where ‘soft iron’ steels were under-heated,
and where ‘hard iron’ steels were over-heated.
A prediction of partial recrystallisation using the RDHT-CM model cannot be vali-

dated exactly, as microstructural examinations of recrystallised fraction are not conducted
by NZ Steel for these coils with incorrect heat treatment. Further to this, the recrystalli-
sation curve of these steels shows recrystallisation to be highly sensitive to temperature
in the range of incomplete recrystallisation. The range found between recrystallisation
‘start’ and ‘stop’ temperatures was around 55◦ C for all steel gauges as shown in Section
4.7.3.
Given the limit of validation of the strip temperature prediction from the RDHT
model, and uncertainty in the metallurgical model calculations detailed in Section 4.7.4,
an exact amount of recrystallisation predicted between 0 - 100% is not meaningful beyond
a measure of Partial Recrystallisation. The model was proven accurate in predicting all
cases of partial recrystsallisation occurring, as well as all cases were either zero or complete
recrystallisation occurred.
7.1.4 Furnace Optimisation
The RDHT-CM model was used to simulate furnace operating conditions for a number of
the most popular products processed on the NZS MCL plant. Several common factors were
found by simulating the existing situations, plus a combination of alternative operating
conditions using either faster or slower linespeed, and different heating methods utilising
the furnace radiant heating and induction preheaters.
It was found that increasing the linespeed to the maximum available speed gave not
only higher production rate, but also lower specific energy requirements in all cases where
the radiant electric furnace only was used. It was found, however, that heating from
the induction preheaters was very inefficient, and strongly increased the specific energy
requirements. While most thin-gauge ‘hard iron’ recovered products may run at full
speed using radiant heating only, thick-gauge or ‘soft iron’ recrystallised products often
220
required induction preheating as well as full power applied to the radiant furnace. For
these products, it was found that using radiant heating only (no preheating) and reducing
the linespeed to the maximum which still maintained the required strip temperature, gave
significant improvement to the specific energy. In this case, a decision may be made by
NZ Steel as to whether absolute production rate is more important than specific energy
consumption. As the MCL plant is a bottleneck in the NZ Steel production chain, this
may be directly related to product demand.
7.1.5 Project Aims
Development of the F3DC and RDHT-CM models achieved the aims of this research as
given in Chapter 1:
• The F3DC model allows simulation of a continuous annealing furnace system with
minimal simplification of furnace geometry and equipment. It allows transient sim-
ulations for any combination of operating conditions, and provides insight into the
temperature distribution and heating behaviour inside this complex non-linear sys-
tem with time-varying boundary conditions. This understanding enables the use of
a simplified model with fast solution time for process control and optimisation.
• The RDHT model allows rapid solution of the furnace and strip temperatures for any
combination of operating conditions, which agrees well with the F3DC model results
and can be related to real furnace temperatures using understandings developed
from the F3DC model.
• When coupled with the metallurgical model, the RDHT-CM model can predict the
recrystallisation temperatures of the steel types processed by NZ Steel. The model
accurately predicts cases where the strip temperatures will cause zero recrystallisa-
tion, full recrystallisation, and can identify situations likely to produce incomplete
recrystallisation.
• The RDHT-CM model is suitable for a priori prediction of furnace and strip tem-
peratures and strip recrystallisation, and is suitable for use in the optimisation of
furnace operating conditions. This model can be used to investigate alternative fur-
nace setups to improve product quality, production rate or energy efficiency, and is
capable of simulating furnace and strip temperatures during transient states, which
is a common cause of poor quality product.
221
7.2 Recommended Further Work
There are several areas in which further development could improve the results and appli-
cation of the furnace models developed as reported in this thesis. However, the investment
required to improve these areas may outweigh the gains possible beyond the accuracy of
the existing models. An extensive series of plant measurements of furnace and strip tem-
peratures and metallurgical properties would provide a complete set of data suitable for
refining the accuracy of the computer models.
To better understand the behaviour of the furnace thermocouple temperatures, it ap-

pears that an additional 3D model is required, including highly accurate representations
of the thermocouple units and adjacent heating elements. The level of detail required to
model the shape of the elements precludes this for a full-length furnace model as described
in Section 5.3.3. It is recommended that a single zone furnace model using exact repre-
sentations of the heating elements and the thermocouple probes would be suitable. This
should also include further complexity beyond the existing model by enabling conductive
heat transfer from the thermocouple sheaths and heater element support structures to
the furnace walls. This model would be highly detailed and time consuming to develop
and solve, and so may provide questionable value compared to the modelling effort and
time investment required.
7.2.1 RDHT Model Improvements and Applications
Further minor improvements to the RDHT-CM model are easy to implement due to the
transparent nature of the model code and the ability to translate it into other program-
ming languages. Possible improvements to this model could be either to the equation
system, or to the method of application of the model. Some areas which may be of
interest to investigate further in the equation system include:
• Improved representation of the gas jet cooler units, using a correlation for heat
removal from the strip and furnace zone based on the mass throughput of strip and
the gas velocity / cooler fan power.
• Improved equation for cooling tubes to show the thermal resistance to conduction
through the tube body, combined with measurements of gas flow speed inside the
tubes.
222
• Adjustments to heat capacity and conductivity of the furnace walls to simulate the
effect of hearth rolls on the furnace thermal mass, or by inclusion as lumped thermal
objects at the relevant furnace grid points.
• Improved fit of metallurgical recrystallisation curves to actual furnace products

rather than using limited results obtained from existing trial specimens. This re-
quires extensive metallurgical examinations of furnace products as per the following
discussion.
There are some areas of application of the model which may be improved further,
including automatic optimisation and transient simulations.
Present optimisation methods using the RDHT-CM model require the manual simu-
lation of a variety of alternative options to find the best operating conditions. This could
be strongly improved by the development of an automatic system for running the model
to find operating points which fulfil a set of constraints defined by the product type. For
example, given the constraint of 100% recrystallisation for a ‘soft iron’ product, an auto-
matic procedure could run the model iteratively to find the lowest energy furnace setup
to achieve this, or a setup that enables using the fastest linespeed, or an optimum point
such as the fastest linespeed not requiring the use of the inefficient induction preheaters.
This would greatly improve the application of the model to optimising the use of the
NZS MCL furnace, as any individual product type not covered in the products investigated
in Section 6.3 could be rapidly optimised without any manual effort. The large number
of possible product types used in the MCL as shown in Section 6.2 makes it impractical
to conduct manual optimisation procedures for any but the most common products.
The RDHT-CM model has been integrated into a software interface by NZ Steel for
easy simulation using furnace measurements from actual product coils, but this does not
allow transient simulations. This requires the development of a new software interface
capable of solving a steady state solution to define an initial condition, plus a selection
of transient conditions or control actions, such as changes in heating setpoints, linespeed,
induction preheating or strip dimensions that will then be simulated following the initial
condition.
223
7.2.2 Improved Transient Solutions
Simulation of an unknown transient condition is difficult, as each of the models require

the input of furnace zone power over time, which changes via PID control to temperature
setpoints. To model any arbitrary change in furnace zone setpoints, it is required that
the PID control is also modelled to give the furnace model inputs in terms of the applied
power. This requires construction and fitting of an entirely separate type of model which
has not been considered here.
Simulation of an existing transient condition using furnace data is currently achieved

by linear approximation to these PID controlled changes in input power, but could be
made more accurate by input of the actual recorded zone power over time during these
transient periods. A software interface to the RDHT-CM model could collect recorded
power input data over the desired transient period, and feed this changing data to the
model as step changes, at a frequency which allows the model to solve the furnace and
strip temperatures between updates. For example, the model could run in real-time to
shadow the furnace operations by receiving updates on furnace power and strip properties
every 10 seconds, and solving a 10 second transient solution between each update.
7.2.3 Improved Temperature Model Fitting
The final area in which further work is recommended is to more accurately measure all
furnace and strip temperatures and metallurgical properties to better enable model fitting
to measurements. It was a recurring problem throughout comparison of both F3DC and
RDHT model results in Chapter 5 that models must be fitted using measurements that
exhibit wide ranges even during steady state operations. Examples of such measurements
are the L1 pyrometer temperatures and the zone power inputs.
The strip temperature recorded by the L1 pyrometer is an important measurement for

fitting model equations, but refining this measurement is difficult and time consuming,
due to the need to use a gold cup pyrometer [48]. The existing measurements show that
actual strip temperatures recorded by the gold cup pyrometer are in all cases 10 − 50◦ C
below the temperature recorded by the L1 pyrometer. This could be improved if the
temperature error could be correlated to the strip dimensions and furnace temperatures,
i.e. to reduce the expected temperature difference to a smaller range for all cases.
224
To accurately develop this correlation, a large number of temperature measurements
across many product types would be required, and given the variation in furnace power
inputs found even at steady state, this must also be conducted across a sufficient time
frame encompassing multiple measurements of the same product type. Developing an
accurate correlation may be expensive both in time and manpower, and also equipment
usage due to gold cup corrosion and wear in the furnace environment.
7.2.4 Improved Metallurgical Model Fitting
The final recommendation for further plant measurements is the metallurgical examina-
tion of strip from the NZS MCL furnace. At present, metallurgical examination is not
conducted by NZ Steel, but this would be required to improve the fit of the metallurgi-
cal model to actual process conditions and steel types used by NZS. This examination
is required both for strips prior to cold rolling to better gauge variation of model input
factors such as the grain size, and post MCL heat treatment for product strips that have
experienced incorrect heat treatment such as the cases investigated in Section 5.6.3.
Measurements of actual recrystallisation amount are usually conducted visually, and

hence are of low accuracy, but estimates to the level of ±10% for a sufficient number of
cases of incorrect heat treatment could assist in refining the fit of the metallurgical model
to predict the situations where partial recrystallisation occurs.
225
Appendix A
Nomenclature
A.1 General Nomenclature
Symbol Description Units

A Area m2
Cp Heat Capacity kJ/kg.◦ C
Eb Emissivity of equivalent black body -
Fi−j View factor from area i to area j -
G Irradiation W/m2
Gr Grashof Number -
h Heat transfer coefficient W/m2 K
J Radiosity W/m2
k Thermal Conductivity W/m.K
L Characteristic Length m
Nu Nusselt Number -
Pr Prandtl Number -
q Heat flux W/m2
qc Heat flux by convection W/m2
qr Heat flux by radiation W/m2
Ra Rayleigh Number -
Re Reynolds Number -
t Time s
continued following page
226
◦
T Temperature C
◦
T∞ Ambient Temperature C
◦
Tf Fluid (Air) Temperature C
◦
Ts Strip Temperature C
◦
Tw Wall (Boundary) Temperature C
U∞ Fluid bulk velocity m/s
A.2 Greek Symbols

α Fluid thermal diffusivity W/m.K
α Absorbtivity -
β Fluid coefficient of thermal expansion 1/K
Emissivity -
φ Heat flux W/m2
µ Fluid dynamic viscosity Pa.s
σ Stefan-Boltzmann Constant W/m2 K4
ρ Density kg/m3
ρ Reflectivity -
2
υ Fluid kinematic viscosity m /s
A.3 Physical Constants
Symbol Description Value

R Gas constant 8.314 J/mol.K
σ Stefan-Boltzmann Constant 5.67 × 10−8 W/m2 K4
227
A.4 RDHT-CM Temperature Model Specific

h Strip thickness (gauge) m
k Duration of each time step s
N Number of steps in time domain -
P Power supplied to heating elements W/m
p Furnace inside perimeter m
sn,s,e,w Grid spacing multiplier in n,s,e,w directions -
v Strip movement speed m/s
w Strip Width m
δ Grid spacing in x-direction m
Φ Heat flux to walls only W/m
A.5 RDHT-CM Metallurgical Model Specific

B Kinetic parameter -
D0 Initial grain size µm
n Avrami Exponent -
Qrex Activation energy for recrystallisation J/mol
RX Nominal calculation of fractional recrystallisation %
RXmin Minimum expected fractional recrystallisation %
RXmax Maximum expected fractional recrystallisation %
tx Time required for x% recrystallisation s
Xv Volume fraction recrystallised %
Strain (cold reduction) -
˙ Strain rate 1/s
A, p, q, s Empirical Constants -
228
Appendix B
RDHT-CM Model Variable List
Variable Description
A coolingtubes Area cooling tubes per metre furnace length
COOLING(x) Additional heat removal at position x
COOLING LOSS(x) Total heat removal by cooling at position x
cp steel Function for steel heat capacity
cpw Heat capacity of furnace walls
CR Cold reduction of strip
del(x) Length grid spacing from x to x − 1
ELEMENTS(Z) Condition of heating elements in Zones 1-20
es Emissivity of strip
ew Emissivity of walls
FF(Z) ’Fudge Factor’ for heating power to Zones 1-20
h coolingtubes Heat transfer coefficient for cooling tubes
hconvect Outer wall heat transfer coefficient
k Duration of each time step
kwall Conductivity of furnace walls
kwall 14 Conductivity of walls in Zone 14 only
M Total number of x-divisions
M endheating Number x-divisions to end of zone 12
M endzone13 Number x-divisions to end of zone 13
M endzone14 Number x-divisions to end of zone 14
MZP(Z) Maximum available power in each zone
229
N Number of time divisions
outerA Wall outer surface area per metre length
p cool Wall inner perimeter in cooling zones
p heat Wall inner perimeter in heating zones
PHI(x) Heat flux to wall at grid point x
POWER(x) Heating power applied to grid point x
POWER percent(Z) Percentage power applied in each zone
Q(x) Heat flux to strip at point x
Q GJC Heat loss from strip by GJC per metre
Q Z20 LOSS Extra heat loss from Zone 20
Q Z9 LOSS Extra heat loss from Zone 9
qconst cool Constant part of radiation equation in cooling zones
qconst heat Constant part of radiation equation in heating zones
rex A Empirical constant A for RX calculation
rex D Initial grain size
rex D H Initial grain size high estimate
rex D L Initial grain size low estimate
rex n Avrami exponent
rex p Empirical constant p for RX calculation
rex q Empirical constant q for RX calculation
rex Q Recrystallisation activation energy
rex Q H Recrystallisation activation energy high estimate
rex Q L Recrystallisation activation energy low estimate
rex s Empirical constant s for RX calculation
rex t50(x) Time for 50% recrystallisation at point x
rex xv(x) Volume recrystallised at point x
rex xvsum(x) Cumulative volume recrystallised by point x
rex xvsum H(x) Cumulative volume recrystallised high estimate
rex xvsum L(x) Cumulative volume recrystallised low estimate
rho Strip density assumed constant
rhow Furnace wall density
sigma Stefan-Boltzmann constant
230
se(x) Strip and wall grid spacing East at point x
sn(x,y) Wall grid spacing North at point x, y
ss(x,y) Wall grid spacing South at point x, y
sw(x) Strip and wall grid spacing West at point x
ST(x) Strip temperature at point x time t
ST2(x) Strip temperature at point x time t + 1
steel chem Steel chemistry grade number
strain rate Strain rate in cold rolling
T Duration of whole time domain
T init Strip entry temperature post-preheaters
th Strip thickness (gauge)
Tinf Ambient temperature
TW(x,y) Furnace wall temperature at point x, y at time t
TW2(x,y) Furnace wall temperature at point x, y at time t + 1
v Strip velocity (linespeed)
w Strip width
wtp Strip mass per metre length
xgrid Position of each x point
ygrid Position of each y point
ZONE COOL(Z) Fixed part of cooling per furnace zone
zone length(Z) Length of each furnace zone
zone sections Number of x grid points per zone
ZONE REX(Z) Recrystallisation result at end of each zone
ZONE REX H(Z) Recrystallisation result high estimate
ZONE REX L(Z) Recrystallisation result low estimate
ZONE STRIP TEMP(Z) Strip temperature result at end of each zone
ZONE WALL TEMP(Z) Inner wall temperature result at end of each zone
Note:
Variables(x,y) are per grid point x, y
Variables(Z) are per furnace Zone 1-20
231
Appendix C
RDHT-CM Model Example Code
% Final thesis code type using variable x-grid spacing

% Variables used for 0.4x940mm Cycle 24 Product
start = cputime; % start time needed to display calculation time duration
%%%%%%%%% Adjustable Values for Different Coils %%%%%%%%%%%

% Furnace / Strip Parameters
th = 0.4e-3; % strip thickness, m
w = 0.94; % strip width, m
v = 120 / 60; % Speed of strip/advection
T_init = 45 +273; % Strip entry temp (K)
CR = 0.788; % Cold Reduction of Strip (Engineering Strain)
steel_chem = 100; % Steel chemistry recipe grade - 100/110/210/310
% Input zone power %’s

POWER_percent = [40
40
40
38.7
38.7
38.7
36.7
33
37.5
36.3
25.8
40.7
100
100
232
98.4
85.5
78.7
92.8
75.5
34] / 100;
% Amount of cooling in each zone with Cooling Tubes or GJCs

% Replace later with correlation
h_coolingtubes = 5; % W/m2.K coefficient for cooling tube surface
A_coolingtubes = 2.6; % approx m2 cooling tube surface per metre furnace length
Q_Z9_LOSS = 10000; % fudge for zone 9 heat loss through dam wall
Q_Z20_LOSS = -100000; % fudge for zone 20 heat gain from hot metal / gas
Q_GJC = 0; % Heat loss directly from strip PER METRE to GJC
kwall_14 = 1.0; % Higher conductivity of GJC zone walls
% Power available modifiers to account for damaged elements

ELEMENTS = [1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1];
%%%%%%%%% End of Adjustments for Different Coils %%%%%%%%%%%

plot1=1; % = 1 to plot temp profiles and rex curve separate axis
T = 100000; % Length t-domain (s)
N = (v*1.05)*T; % No. t-divisions
Tinf = 30+273; % Ambient Temp (K)
233
es = 0.2; % strip emissivity
ew = 0.9; % wall emissivity
sigma = 5.67e-8; % Stefan-Boltzmann Constant
rho = 7854; % steel density kg/m3 (assumed constant)
p_heat = 2.671; % half furnace cavity perimeter in heating zones only
p_cool = 3.153; % half furnace cavity perimeter in cooling zones only
rhow = 497; % density refractive bricks
cpw = 500; % heat capacity refractive bricks
kwall = 0.25; % k refractory assumed constant
hconvect = 5; % convective HTC for outer wall (natural convection)
outerA = 8.5; % area of outer wall per grid length
wtp = w * th* rho ; % strip width x thickness x density = strip mass/m
% qconst = constant part of equation for Q

qconst_heat = (2 * w * es * sigma) / (1 + es*(1-ew)*w/(ew*p_heat));
qconst_cool = (2 * w * es * sigma) / (1 + es*(1-ew)*w/(ew*p_cool));
%%%%%%%%%%%% Recrystallisation Constants %%%%%%%%%%%%%

strain_rate = 1000; % 1/s
rex_D = 10; % um typical hot-rolled grain size
rex_s = 1; % grain size exponent
rex_p = -3.7 * (rex_D^-0.137); % strain exponent
rex_q = -0.53; % strain rate exponent
if steel_chem == 100
rex_Q = 201826; % activation energy, J/mol
elseif steel_chem == 110
rex_Q = 201251;
rex_Q = 211352;
rex_Q = 204028;
end
rex_A = 0.0003869* exp(-0.00007921*rex_Q); % Q exponent
rex_D_H = rex_D * 0.8;
rex_Q_H = rex_Q * 0.98;
rex_D_L = rex_D * 1.2;
rex_Q_L = rex_Q * 1.03;
% Calculate k and convert time steps to time in seconds for plotting results
k = T/N; % Duration of each t-step
timescale = transpose(0:k:T);
% Zone length and number of division data

zone_length = zeros(20,1);
234
zone_sections = zeros(20,1);
zone_length(1) = 5.395;
zone_sections(1) = 5;
235
M = sum(zone_sections); % number grid sections in total
M_endheating = sum(zone_sections(1:12))+1; % grid sections to end heating zones
M_endzone13 = sum(zone_sections(1:13))+1;
M_endzone14 = sum(zone_sections(1:14))+1;
% Initialise matrices :
Q = zeros(M+1,1); % Q = Heat transfer to strip at each grid position
PHI = zeros(M+1,1); % Power flux into wall at each time step
rex_xv = zeros(M+1,1); % Strip Recrystallisation at each point
rex_xvsum = zeros(M+1,1); % Cumulative Recrystallisation
rex_n = zeros(M+1,1); % Rex ’n’ factor at each point
rex_t50 = zeros(M+1,1); % time for 50% rxn at each point
rex_xvsum_H = zeros(M+1,1);
rex_xvsum_L = zeros(M+1,1);
ST = zeros(M+1,1); % ST = Strip Temps at x,t
ST2 = zeros(M+1,1); % ST = Strip Temps at x,t+1
TW = zeros(M+1,5); % Stored wall temps for 5 y-pos
TW2 = zeros(M+1,5);
% Create x- and y-position grids and s factors

ygrid = zeros(M+1,5); % Position grid of mesh points
ygrid(1:M_endheating,2) = 0.004; % ygrid(1,:) = heating zones grid spacing
ygrid(1:M_endheating,3) = 0.02;
ygrid(M_endheating+1:M+1,2) = 0.0027; % ygrid(2,:) = cooling zones grid spacing

ygrid(M_endheating+1:M+1,3) = 0.0134;
ss = zeros(M+1,5);
sn = zeros(M+1,5);
for ypos = 2:5
ss(:,ypos) = ygrid(:,ypos) - ygrid(:,ypos-1);
end
for ypos = 1:4
sn(:,ypos) = ygrid(:,ypos+1) - ygrid(:,ypos);
end
ss(:,1) = ss(:,2);
sn(:,5) = sn(:,4);
x = zeros(M+1,1);
counter=2;
236
for i = 1:20
for j = 1:zone_sections(i)
x(counter) = x(counter-1) + zone_length(i) / zone_sections(i);
counter = counter+1;
end
end
se = zeros(M+1,1);
sw = zeros(M+1,1);
se(1:M) = x(2:M+1) - x(1:M);
sw(2:M+1) = x(2:M+1) - x(1:M);
se(M+1) = se(M);
sw(1) = sw(2);
% stability check for CFL condition

% with varying del -> del for CFL is the shortest grid spacing in x
del = min(se(1:M));
if (k*v/del) > 1
disp(’Error - Unstable - CFL condition not met. Required 0 < k*v/del < 1’);
k*v/del
end
% Rough initial temperature guess for walls to speed up steady state solution
TW(:,1) = 800+273;
TW(:,2) = 700+273;
TW(:,3) = 500+273;
TW(:,4) = 200+273;
TW(:,5) = 50+273;
% New Max zone powers to fit exact zone lengths

MZP = zeros(20,1);
MZP(1) = 346000;
MZP(2) = 346000;
MZP(3) = 346000;
MZP(4) = 346000;
MZP(5) = 346000;
MZP(6) = 346000;
MZP(7) = 346000;
MZP(8) = 346000;
MZP(9) = 346000;
MZP(10) = 346000;
MZP(11) = 346000;
MZP(12) = 346000;
MZP(13) = 115000;
237
MZP(14) = 70000;
MZP(15) = 115000;
MZP(16) = 200000;
MZP(17) = 200000;
MZP(18) = 200000;
MZP(19) = 200000;
MZP(20) = 225000;
% ’Fudge Factors’ for zone power inputs to match furnace heat losses
FF = ones(20,1);
FF(12) = 0.75;
% Zone heat losses - additional specific W heat loss rather than power%
ZONE_COOL = zeros(20,1);
ZONE_COOL(9) = Q_Z9_LOSS;
ZONE_COOL(20) = Q_Z20_LOSS;
COOLING=zeros(M+1,1);
% Populate zone power and heat losses into each grid section
counter = 2;
for i = 1:20
POWER(counter) = MZP(i)/zone_sections(i) * POWER_percent(i) *ELEMENTS(i) *FF(i);
COOLING(counter) = ZONE_COOL(i)/zone_sections(i);
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%% Calculate results for all time steps %%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
ST(:) = T_init;
for j=1:N % start j-loop
%xpos = 1
ST2(1) = T_init;
TW2(1,1) = TW(2,1);
TW2(1,2) = TW(2,2);
TW2(1,3) = TW(2,3);
TW2(1,4) = TW(2,4);
TW2(1,5) = TW(2,5);
for xpos = 2:M_endheating % x-pos loop for heating zones exc. end points
del = sw(xpos);
238
Q(xpos) = qconst_heat * ( (TW(xpos,1).^4 - ST(xpos).^4) )*del;
cp_steel = 0.0003275*ST(xpos)*ST(xpos) + 0.101*ST(xpos) + 396.4;
ST2(xpos) = (1-v*k/del) * ST(xpos) + (v*k/del) * ST(xpos-1) + (Q(xpos)*k /
(wtp * cp_steel * del));
COOLING_LOSS = COOLING(xpos);
PHI(xpos) = (POWER(xpos) - Q(xpos) - COOLING_LOSS)/(2*p_heat*del);
TW2(xpos,1) = (0.0040/kwall)*PHI(xpos) + TW(xpos,2);
for ypos = 2:4
TW2(xpos,ypos) = TW(xpos,ypos) + ((kwall * 2 * k)/(rhow*cpw))*
( TW(xpos+1,ypos)/(se(xpos)*(se(xpos)+sw(xpos)))
+ TW(xpos-1,ypos)/(sw(xpos)*(se(xpos)+sw(xpos)))
+ TW(xpos,ypos+1)/(sn(xpos,ypos)*(sn(xpos,ypos)+ss(xpos,ypos)))
+ TW(xpos,ypos-1)/(ss(xpos,ypos)*(sn(xpos,ypos)+ss(xpos,ypos)))
- TW(xpos,ypos)*(1/(sw(xpos)*se(xpos)) + 1/(sn(xpos,ypos)*ss(xpos,ypos))));
end
TW2(xpos,5) = (TW(xpos,4) + (0.2579*hconvect*outerA*del/kwall)*Tinf)/
(1 + 0.2579*hconvect*outerA*del/kwall);
end % end x-pos loop for heating zones
% ZONE 13 ONLY
for xpos = M_endheating+1:M_endzone13 % x-pos loop for cooling zones ex. end points
del = sw(xpos);
Q(xpos) = qconst_heat * ( (TW(xpos,1).^4 - ST(xpos).^4) );
ST2(xpos) = (1-v*k/del) * ST(xpos) + (v*k/del) * ST(xpos-1) + (Q(xpos)*k /
(wtp * cp_steel * del));
COOLING_LOSS = COOLING(xpos) +
(h_coolingtubes * A_coolingtubes * del * (TW(xpos,1) - Tinf));
PHI(xpos) = (POWER(xpos) - Q(xpos) - COOLING_LOSS)/(2*p_cool*del);
for ypos = 2:4
end
end % end x-pos loop for cooling zones
% ZONE 14 - GJC ZONE ONLY

for xpos = M_endzone13+1:M_endzone14 % x-pos loop for cooling zones ex. end points
239
del = sw(xpos);
ST2(xpos) = (1-v*k/del) * ST(xpos) + (v*k/del) * ST(xpos-1) +
((Q(xpos)-Q_GJC*del)*k / (wtp * cp_steel * del));
COOLING_LOSS = COOLING(xpos);
TW2(xpos,1) = (0.0027/kwall_14)*PHI(xpos) + TW(xpos,2);
for ypos = 2:4
end
TW2(xpos,5) = (TW(xpos,4) + (0.1721*hconvect*outerA*del/kwall_14)*Tinf)/
(1 + 0.1721*hconvect*outerA*del/kwall_14);
% COOLING ZONES 15-20 ONLY

for xpos = M_endzone14+1:M % x-pos loop for cooling zones ex. end points
del = sw(xpos);
(Q(xpos)*k / (wtp * cp_steel * del));
COOLING_LOSS = COOLING(xpos) +
(h_coolingtubes * A_coolingtubes * del * (TW(xpos,1) - Tinf));
for ypos = 2:4
end
% END POSITION ONLY xpos=M+1
240
xpos = M+1;
del = sw(xpos);
(Q(xpos)*k / (wtp * cp_steel * del));
TW2(M+1,1) = TW2(M,1);
TW2(M+1,2) = TW2(M,2);
TW2(M+1,3) = TW2(M,3);
TW2(M+1,4) = TW2(M,4);
TW2(M+1,5) = TW2(M,5);
% Copy temperatures at time t+1 to time t matrix for next time step
ST(:,1) = ST2(:,1);
TW(:,:) = TW2(:,:);
end %end j loop

% end of time stepping equations
%%%%%%%%%%%%%% Recrystallisation Calculations %%%%%%%%%%%%%%

% Recrystallisation calculations based only on strip temperatures at final time step
for xpos = 1:M+1
% Using linear interpolation formula for ’n’ from Haff & Schulson 1982
rex_n(xpos) = 2;
% rex_n(xpos) = 1 + 0.8*((998-ST(xpos,1))/115);
rex_t50(xpos) = rex_A * CR^rex_p * strain_rate^rex_q * rex_D^rex_s *
exp( rex_Q / ( 8.314*ST(xpos,1)));
rex_xv(xpos) = 1 - exp( -0.693*( (del/v) / rex_t50(xpos) )^ rex_n(xpos));
end
% sum each fraction recrystallised at each grid point to get total rxn in
% strip. Maximum is 1.0 recrystallised.
for xpos = 1:M+1
rex_xvsum(xpos) = sum(rex_xv(1:xpos));
if rex_xvsum(xpos) > 1
rex_xvsum(xpos) = 1;
end
end
% End of Recrystallisation Calculations
% Repeat for Low Recrystallisation Tolerance

for xpos = 1:M+1
rex_t50(xpos) = rex_A * CR^rex_p * strain_rate^rex_q * rex_D_L^rex_s *
exp( rex_Q_L / ( 8.314*ST(xpos,1)));
241
end
for xpos = 1:M+1
rex_xvsum_L(xpos) = sum(rex_xv(1:xpos));
if rex_xvsum_L(xpos) > 1
rex_xvsum_L(xpos) = 1;
end
end
% Repeat for High Recrystallisation Tolerance

for xpos = 1:M+1
rex_t50(xpos) = rex_A * CR^rex_p * strain_rate^rex_q * rex_D_H^rex_s *
exp( rex_Q_H / ( 8.314*ST(xpos,1)));
end
for xpos = 1:M+1
rex_xvsum_H(xpos) = sum(rex_xv(1:xpos));
if rex_xvsum_H(xpos) > 1
rex_xvsum_H(xpos) = 1;
end
end
%%%%%%%%%%%%%%%%%%%%%%
%%%% Output Stage %%%%
%%%%%%%%%%%%%%%%%%%%%%
ZONE_STRIP_TEMP = ones(20,1);
ZONE_WALL_TEMP = ones(20,1);
ZONE_REX = ones(20,1);
ZONE_REX_L = ones(20,1);
ZONE_REX_H = ones(20,1);
counter = 1;
for i = 1:20
counterzonestart = counter;
end
ZONE_STRIP_TEMP(i) = ST(counter) - 273.15;
ZONE_REX(i) = rex_xvsum(counter);
ZONE_REX_L(i) = rex_xvsum_L(counter);
ZONE_REX_H(i) = rex_xvsum_H(counter);
ZONE_WALL_TEMP(i) = MEAN(TW(counterzonestart:counter,1)) - 273.15;
end
if ZONE_REX_H(:) < 0.001
ZONE_REX(:) = 0
ZONE_REX_H(:) = 0
242
ZONE_REX_L(:) = 0
end
if plot1 == 1 % plot temps and recrystallisation curves per zone
subplot(2,1,1); % Temperature Plots

hold on
xlabel(’X-Position (m)’);
ylabel(’Temperature (deg C)’);
hline1 = plot(x,ST(:,1)-273,’b’);
hline2 = plot(x,TW(:,1)-273,’r’);
xlim([0 164.5]);
ylim([0 1200]);
set(hline1,’Linewidth’,2);
legend(’Strip Temperature’,’Wall Temperature’);
subplot(2,1,2); % Recrystallisation Plots

hold on
hline3 = plot(x,rex_xvsum,’b’);
hline4 = plot(x,rex_xvsum_H, ’g’);
hline5 = plot(x,rex_xvsum_L, ’r’);
xlim([0 164.5]);
ylim([0 1]);
set(hline4,’Linewidth’,2,’LineStyle’,’--’);
set(hline5,’Linewidth’,2,’LineStyle’,’--’);
ylabel(’Fraction Recrystallised, Xv’);
xlabel(’X-Position (m)’);
legend(’Centre’,’High’,’Low’,2);
end
ZONE_STRIP_TEMP % Display strip temp for end of each zone

ZONE_WALL_TEMP % Display average wall temp of each zone
ZONE_REX % Display Recrystallisation at end of each zone
ZONE_REX_L
ZONE_REX_H
ST(67)-273.15
finish = cputime;
duration = finish - start;
243
Bibliography
[1] Anonymous. New zealand steel website - glenbrook steel site [online]. Available:
http://nzsteel.co.nz/go/about-new-zealand-steel/history, [July 29, 2008].
[2] Anonymous. New zealand steel website - history [online]. Available:

http://nzsteel.co.nz/go/about-new-zealand-steel/history, [July 29, 2008].
[3] ASM. American society for metals: Metals handbook. volume 4: Heat treating.
[4] Avrami, M. Kinetics of phase change i. J. Chem. Phys. 7 (1939).
[5] Bagshaw, P., and Kimber, R. Advances in warm rolling technology at bhp new
zealand steel. In Proceedings of the IPENZ Technical Conference (1999).
[6] Barraclough, D., and Sellars, C. Static recrystallization and restoration after
hot deformation of type 304 stainless steel. Metal Science, 13 (1979).
[7] Bejan, A., and Kraus, A. D. Heat Transfer Handbook. John Wiley and Sons,
2003.
[8] Beynon, J., and Sellars, C. Modelling microstructure and its effects during
multipass hot rolling. ISIJ International 32, 3 (1992).
[9] Bhadeshia, H. H., and Honeycombe, R. Steels - Microstructures and Properties,

3rd ed. Butterworth-Heinemann, 2006.
[10] Burke, J., and Turnbull, D. Recrystallization and grain growth. Prog. Met.
Phys. 3 (1952).
[11] Cahn, R., and Haasent, P. Physical Metallurgy: Fourth, Revised and Enhanced
Edition. Elsevier Science BV, 1997.
[12] Capdevila, C., Garcia-Mateo, C., Caballero, F., and Garcia de An-
dres, C. Neural network analysis of the influence of processing on strength and
244
ductility of automotive low carbon sheet steels. Computational Materials Science 38
(2006).
[13] Cengel, Y. An Introduction to Thermodynamics and Heat Transfer. McGraw-Hill,

1997.
[14] Churchill, S., and Chu, H. H. S. Correlating equations for laminar and turbu-
lent free convection from a vertical plate. Int J. Heat Mass Transfer 18 (1975).
[15] Fausett, L. V. Applied Numerical Analysis Using Matlab. Prentice Hall, 1999.
[16] Fernandez, A., Uranga, P., Lopez, B., and Rodriguez-Ibabe, J. Static
recrystallization behaviour of a wide range of austenite grain sizes in microalloyed
steels. ISIJ International 40, 9 (2000).
[17] Gale, W., and Totemeier, T. C. Smithells Metals Reference Book, Eighth
Edition. Elsevier Butterworth-Heinemann, 2004.
[18] Grossman, C., Roos, H., and Stynes, M. Numerical Treatment of Partial
Differential Equations, 3rd ed. Springer, 2007.
[19] Haberman, R. Elementary Applied Partial Differential Equations, 3rd Ed. Prentice
Hall, 1998.
[20] Haberman, W., and James, J. Engineering Thermodynamics with Heat Transfer,
2nd Ed. Allyn and Bacon, 1992.
[21] Haff, G., and Schulson, E. Recrystallisation and grain growth in NiAl. Metal-
lurgical Transactions A 13a (1982).
[22] Hottel, H., and Sarofim, A. Radiative Transfer. McGraw-Hill, 1967.
[23] Howell, J. A catalog of radiation heat transfer configuration factors. Website

http://www.me.texas.edu/h̃owell/intro.html, 2006.
[24] Humphreys, F., and Hatherly, M. Recrystallization and Related Annealing

Phenomena. Elsevier, 2004.
[25] Incropera, F., and DeWitt, D. Introduction to heat transfer, 4th ed. John
Wiley and Sons, 2002.
[26] Jinks, D. Pyrometer temperature measurement for horizontal annealing furnaces.

the experience at NZ Steel. BlueScope Steel TechNote BSR/N/2004/045 (2004).
245
[27] Jinks, D. A steady state model of the NZ Steel MCL radiant electric furnace.
BlueScope Steel TechNote BSR/N/2004/049 (2004).
[28] Johnson, W., and Mehl, R. Reaction kinetics in processes of nucleation and
growth. Trans. Metall. Soc. A.I.M.E. 135 (1939).
[29] Joveljic, N. Continuous annealing of cold rolled steel strip at the metal coating
plant. NZ Steel Intranet Page, 2006.
[30] Joveljic, N. Skinpassing. NZ Steel Intranet Page, 2006.
[31] Joveljic, N. Skinpassing and levelling. NZ Steel Intranet Page, 2006.
[32] Kolmogorov, A. On statistical theory of metal crystallisation. Izv. Akad. Nauk.

USSR-Ser-Matemat. 3, 1 (1937).
[33] Kreith, F., and Bohn, M. Principles of Heat Transfer, sixth ed. Brooks/Cole,
2001.
[34] Lenard, J. Primer on Flat Rolling. Elsevier Science & Technology Books, 2007.
[35] Major Engineering Pty Ltd. Equipment Specification and Operations Man-
ual Major/Selas for MCP Upgrade Gas Jet Coolers for NZ Steel Ltd, manual no.
20030600/3 ed., November 2003.
[36] Marlow, D. Modelling direct-fired annealing furnaces for transient operations.

Applied Math Modelling 20 (January 1996).
[37] Martinez-de-Guerenu, A., Arizti, F., Diaz-Fuentes, M., and Gutierrez,

I. Recovery during annealing in a cold rolled low carbon steel. part i: Kinetics and
microstructural characterization. Acta Materialia, 52 (2004).
[38] Massardier, V., Guetaz, V., Merlin, J., and Soler, M. Study of the role
played by nitrogen on the deep-drawing properties of aluminium killed steel sheets
obtained after a continuous annealing. Materials Science Forum 426-432 (2003).
[39] McGuinness, M., and Taylor, S. Strip temperature in a metal coating line
annealing furnace. Australian and New Zealand Mathematics in Industry Study
Group, 2004.
[40] Medina, S., and Lopez, V. Static recrystallization in austenite and its influence
on microstructural changes in c-mn steel and vanadium microalloyed steel at the hot
strip mill. ISIJ International 33, 5 (1993).
246
[41] Medina, S., and Mancilla, J. Influence of alloying elements in solution on static
recrystallization kinetics of hot deformed steels. ISIJ International 36, 8 (1996),
1063–1069.
[42] Medina, S., and Mancilla, J. Static recrystallisation modelling of hot deformed
steels containing several alloying elements. ISIJ International 36, 8 (1996), 1070–
1076.
[43] Medina, S., Mancilla, J., and Hernandez, C. Influence of vanadium on

the static recrystallization of austenite in microalloyed steels. Journal of Materials
Science, 28 (1993).
[44] Medina, S., and Quispe, A. Improved model for static recrystallisation kinetics of
hot deformed austenite in low alloy and nb/v microalloyed steels. ISIJ International
41, 7 (2001).
[45] Mehl, R. Recrystallization. American Society of Metals, 1948.
[46] Michael F. Ashby, D. R. J. Engineering Materials 2. An Introduction to Mi-

crostructures, Processing and Design. Butterworth Heineman, 1998.
[47] Modest, M. Radiative Heat Transfer, second ed. Academic Press, 2003.
[48] Morrison, B., Jinks, D., and Edmonds, I. NZ Steel MCL furnace - gold cup
temperature measurements. BlueScope Steel TechNote SRL/N/2003/038 (2003).
[49] New Zealand Steel. Prediction of strip temperature in a metal coating line
annealing furnace. Problem Presentation for MISG from NZ Steel personnel, January
2004. File Reference MISG04.ppt.
[50] Ohring, M. Engineering Materials Science. Academic Press, 1995.
[51] Ordieres Mere, J., Gonzalez Marcos, A., Gonzalez, J., and Lobato
Rubio, V. Estimation of mechanical properties of steel strip in hot dip galvanising
lines. Ironmaking and Steelmaking 31, 1 (2004).
[52] Packer, J. Chemical Processes in New Zealand, 2nd Ed. New Zealand Institute of
Chemistry, 1999.
[53] Pernia-Espinoza, A., Castejon-Limas, M., Gonzalez-Marcos, A., and

Lobato-Rubio, V. Steel annealing furnace robust neural network model. Ironmak-
ing and Steelmaking 32, 5 (2005).
247
[54] Perry, R., and Green, D. Perry’s Chemical Engineer’s Handbook, 7th ed. Mc-
Graw Hill, 1997.
[55] Prieto, M. M., Fernandez, F. J., and Rendueles, J. L. Development of step-

wise thermal model for annealing line heating furnace. Ironmaking and Steelmaking
32, 2 (2005).
[56] Ramamurthy, H., Ramadhyani, S., and R.Viskanta. A thermal system model
for a radiant-tube continuous reheating furnace. Journal of materials engineering and
performace 4, 5 (1995).
[57] Renshaw, W. A recrystallisation model for continuous heat treatment of low carbon
aluminium killed steels. Tech. Rep. BHPR/SF/R/011, The Broken Hill Proprietary
Company Ltd, 1996.
[58] Roberts, W. Cold Rolling of Steel. Marcel Dekker, 1978.
[59] Rohsenow, W. M., Hartnett, J. P., and Cho, Y. I. Handbook of Heat

Transfer (3rd Edition). McGraw-Hill, 1998.
[60] Sahay, S., and Kapur, P. Model based scheduling of a continuous annealing
furnace. Ironmaking and Steelmaking 34, 3 (2007).
[61] Sample, V., Lalli, L., and Richmond, O. Experimental aspects and phe-
nomenology of internal state variable constitutive models. Proc. 3rd Int. Conf. on
Al Alloys, Trondheim, Norway (1992).
[62] Sellars, C., and Whiteman, J. Recrystallisation and grain growth in hot rolling.
Metal Science, 13 (1979).
[63] Shercliff, H., and Lovatt, A. Modelling of microstructure evolution in hot

deformation. Phil. Trans. R. Soc. London. A, 357 (1999).
[64] Sinha, A. K. Physical Metallurgy Handbook. McGraw-Hill, 2003.
[65] Smallman, R., and Bishop, R. Modern Physical Metallurgy and Materials Engi-
neering, 6th Edition. Butterworth-Heinemann, 1999.
[66] Smith, W. Foundations of Materials Science and Engineering, second ed. McGraw-
Hill, 1993.
[67] Somani, M., and Karjalainen, L. Modelling the deformation and annealing
processes: Physical and regression approaches. Materials Science Forum 550 (2007).
248
[68] Sydac Engineering Ltd. Bluescope steel annealing furnace model. Report from
Sydac Engineering Ltd to NZ Steel, 2004. Commercial in confidence.
[69] Sydac Engineering Ltd. Mcl annealing furnace model specification. phase 2 -
steady state model. Report from Sydac Engineering Ltd to NZ Steel no. AD05-002-
SP-01, 2004. Commercial in confidence.
[70] Sydac Engineering Ltd. Modelling of the strip temperature in a metal coating
line annealing furnace. Report from Sydac Engineering Ltd to NZ Steel no. AD05-
003-DR-001, March 2005. Commercial in confidence.
[71] Tang, K.-T. Mathematical Methods for Engineers and Scientists 3. Springer-Verlag
Berlin Heidelberg, 2007.
[72] Tang Guang-Bo, Liu Zheng-dong, D. G. K. L. M. Numerical simulation of

austenite recrystallization in csp hot rolled c-mn steel strip. Journal of Iron and Steel
Research, International 14, 4 (2007).
[73] Taylor, S., and Wang, S. Modelling steel strip heating within an annealing
furnace. In Press, 2009.
[74] Ueda, I., Hosada, M., and Taya, K. Strip temperature control for a heating
section in cal. In Proc. IECON ’91 (1991).
[75] Wang, Z., Shao, C., and Chai, T. Application of multivariable techniques in
temperature control of reheating furnaces. In Proceedings of the 1999 International
Conference on Control Applications (1999).
[76] Willis, D. Developments in hot dipped metallic coated steels processing. Materials
Forum 29 (2005). Institute of Materials Engineering Australia Ltd.
[77] Yahiro, K., Shigemori, H., Hirohata, K., Ooi, T., Haruna, M., and
Nakanishi, K. Development of strip temperature control system for a continu-
ous annealing line. In Proc. IECON ’93 (1993).
[78] Yicheng, L. The new zealand steel problem. Student project for University of
Auckland B.Tech programme, supervised by Prof. S.Wang, 2005.
[79] Yoshitani, N., and Hasegawa, A. Model-based control of strip temperature for
the heating furnace in continuous annealing. IEEE Transactions on Control Systems
Technology 6, 2 (1998).
249

2009 - Annealing Furnace Model - Effect of Warm Rolled Steel On Recrystallization PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2009 - Annealing Furnace Model - Effect of Warm Rolled Steel On Recrystallization PDF

Uploaded by

Copyright:

Available Formats

http://researchspace.auckland.ac.

• Any use you make of these documents or images must be for

To request permissions please use the Feedback form on our webpage.

General copyright and disclaimer

Note : Masters Theses

The digital copy of a masters thesis is as submitted for examination and

Nicholas Brian Depree

3 Practical Temperature Calculation to Improve Control 72

4 Model Methodology and Development 77

5 Model Results and Validation 156

6 MCL Furnace Optimisation 201

7 Conclusions and Further Work 216

B RDHT-CM Model Variable List 229

C RDHT-CM Model Example Code 232

1.1 Schematic diagram of the Metal Coating Line process . . . . . . . . . . . . 2

2.1 Comparison of grain structure distortion in cold rolled low-carbon steel . . 11

4.1 Typical cross section of MCL annealing furnace heating zone . . . . . . . . 78

6.1 Temperature and recrystallisation results for RDHT-CM model simulation

4.1 Material properties of furnace bodies to be used in conductive heat transfer

GJC Gas Jet Cooler

F3DC Full 3D Comsol

Recrystallisation of the strip is dependent on both strip temperature and duration of

Ensuring the correct metallurgical outcome is complicated significantly by the diffi-

In cases of large magnitude changes in strip dimensions or required temperature, called

1.2 Annealing Furnace Characteristics

1.3 Study Aims

1.5 Thesis Outline

2.1 The Continuous Galvanising Process

• Cleaning and pickling

• Heat treatment furnace

• Hot dip coating with thickness control

• Continuous cooling or quenching

• Tension leveling and skinpassing

• Chemical coating and passivation

• Coiling, cutting and packaging

2.1.1 Cold Rolling

(a) 30% Cold Reduction (b) 70% Cold Reduction

2.1.1.1 Sheet Metal Forming

Figure 2.2: The iron-carbon phase diagram [9]

• Amount of deformation (cold work)

• Deformation rate during processing

• Initial grain size

• Composition (alloy chemistry) of metal

The rate of recrystallisation is strongly dependent on temperature - at elevated tem-

Recrystallisation of the deformed microstructure is called primary recrystallisation,

2.2.2.1 Laws of Recrystallisation

1. A minimum amount of deformation is required to initiate recrystallisation.

3. The required temperature decreases as the strain increases

6. Higher deformation temperatures increase the required recrystallisation temperature

2.2.3 Grain Growth

Further exposure to temperatures above the recrystallisation temperature after recrys-

The calculation of temperatures inside furnaces is a common industrial problem. As

Statistical modelling of furnace systems is a completely different approach to furnace

Further to the development of models to calculate furnace temperature or product

(a) Strip Temperature (b) Convective and radiative transfer

A model of an indirectly fired furnace was created by Ramamurthy et al [56], for a

An alternative approach to creating a physical equation-based mathematical model is to

2.3.3 Model Based Control Systems

(a) Yoshitani et al [79] (b) Ueda et al [74]

Figure 2.8: Schematic diagrams of multi-layer predictive control schemes implemented on

Another approach to model-based control is the ‘expert system’ used by Yahiro et al

2.3.4 Model Type Selection

As covered in Section 3.3, a 3D mathematical computer model is first developed, which

2ws σ(Tw4 − Ts4 )