Handbook of
Thermal
Process
Modeling
of Steels
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page i 6.11.2008 6:02pm Compositor Name: VBalamugundan
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page ii 6.11.2008 6:02pm Compositor Name: VBalamugundan
Handbook of
Thermal
Process
Modeling
of Steels
Edited by
Cemil Hakan Gür
Jiansheng Pan
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page iii 6.11.2008 6:02pm Compositor Name: VBalamugundan
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 334872742
© 2009 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed in the United States of America on acidfree paper
10 9 8 7 6 5 4 3 2 1
International Standard Book Number13: 9780849350191 (Hardcover)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
9787508400. CCC is a notforprofit organization that provides licenses and registration for a variety of users. For orga
nizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page iv 6.11.2008 6:02pm Compositor Name: VBalamugundan
Contents
Preface............................................................................................................................................. vii
Editors .............................................................................................................................................. ix
Contributors ..................................................................................................................................... xi
Chapter 1 Mathematical Fundamentals of Thermal Process Modeling of Steels...................... 1
Jiansheng Pan and Jianfeng Gu
Chapter 2 Thermodynamics of Thermal Processing................................................................ 63
Sivaraman Guruswamy
Chapter 3 Physical Metallurgy of Thermal Processing ........................................................... 89
Wei Shi
Chapter 4 Mechanical Metallurgy of Thermal Processing .................................................... 121
Božo Smoljan
Chapter 5 Modeling Approaches and Fundamental Considerations ..................................... 185
Bernardo HernandezMorales
Chapter 6 Modeling of Hot and Warm Working of Steels ................................................... 225
Peter Hodgson, John J. Jonas, and Chris H.J. Davies
Chapter 7 Modeling of Casting.............................................................................................. 265
Mario Rosso
Chapter 8 Modeling of Industrial Heat Treatment Operations.............................................. 313
Satyam Suraj Sahay
Chapter 9 Simulation of Quenching ...................................................................................... 341
Caner S ¸ ims ¸ ir and C. Hakan Gür
Chapter 10 Modeling of Induction Hardening Processes ........................................................ 427
Valentin Nemkov
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page v 6.11.2008 6:02pm Compositor Name: VBalamugundan
v
Chapter 11 Modeling of Laser Surface Hardening.................................................................. 499
Janez Grum
Chapter 12 Modeling of Case Hardening................................................................................ 627
Gustavo Sánchez Sarmiento and María Victoria Bongiovanni
Chapter 13 Industrial Applications of Computer Simulation of Heat
Treatment and Chemical Heat Treatment ............................................................. 673
Jiansheng Pan, Jianfeng Gu, and Weimin Zhang
Chapter 14 Prospects of Thermal Process Modeling of Steels................................................ 703
Jiansheng Pan and Jianfeng Gu
Index............................................................................................................................................. 727
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page vi 6.11.2008 6:02pm Compositor Name: VBalamugundan
vi
Preface
The whole range of steel thermal processing technology, from casting and plastic forming to
welding and heat treatment, not only produces workpieces of the required shape but also optimizes
the endproduct microstructure. Thermal processing thus plays a central role in quality control,
service life, and the ultimate reliability of engineering components, and now represents a funda
mental element of any company’s competitive capability.
Substantial advances in research, toward increasingly accurate prediction of the microstructure
and properties of workpieces produced by thermal processing, were based on solutions of partial
differential equations (PDEs) for temperature, concentration, electromagnetic properties, and stress
and strain phenomena. Until the widespread use of highperformance computers, analytical solution
of PDEs was the only approach to describe these parameters, and this placed severe limitations in
terms of prediction for engineering applications so that thermal process developments themselves
relied on empiricism and traditional practice. The level of inaccuracy inherent in computational
predictions hindered both materials performance improvements and process cost reduction.
Since the 1970s, the pace of development of computer technology has made possible effective
solution of PDEs in complicated calculations for boundary and initial conditions, as well as non
linear and multiple variables. Mathematical models and computer simulation technology have
developed rapidly; currently wellestablished mathematical models integrate fundamental theories
of materials science and engineering including heat transfer, thermoelastoplastic mechanics, ﬂuid
mechanics, and chemistry to describe physical phenomena occurring during thermal processing.
Further, evolution of transient temperature, stress–strain, concentration, microstructure, and ﬂow
can now be vividly displayed through the latest visual technology, which can show the effects of
individual process parameters. Computation=simulation thus provides an additional decision
making tool for both the process optimization and the design of plant and equipment; it accelerates
thermal processing technology development on a scientiﬁcally sound computational basis.
The basic mathematical models for thermal processing simulation gradually introduced to date
have yielded enormous advantages for some engineering applications. Continued research in this
direction attracts increasing attention now that the cuttingedge potential of future developments is
evident. Increasingly profound investigations are now in train globally. The number of important
research papers in the ﬁeld has risen sharply over the last three decades. Even so, the existing
models are regarded as highly simpliﬁed by comparison with real commercial thermal processes.
This has meant that the application of computer simulation has thus far been relatively limited
precisely because of these simplifying assumptions, and their consequent limited computational
accuracy. Extensive and continuing research is still needed.
This book is now offered as both a contribution to work on the limitations described above and
as an encouragement to increase the understanding and use of thermal process models and
simulation techniques.
The main objectives of this book are, therefore, to provide a useful resource for thermal
processing of steels by drawing together
.
An approach to a fundamental understanding of thermal process modeling
.
A guide to process optimization
.
An aid to understand realtime process control
.
Some insights into the physical origin of some aspects of materials behavior
.
What is involved in predicting material response under real industrial conditions not easily
reproduced in the laboratory
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page vii 6.11.2008 6:02pm Compositor Name: VBalamugundan
vii
Linked objectives are to provide
.
A summary of the current state of the art by introducing mathematical modeling method
ology actually used in thermal processing
.
A practical reference (industrial examples and necessary precautionary measures are
included)
It is hoped that this book will
.
Increase the potential use of computer simulation by engineers and technicians engaged in
thermal processing currently and in the future
.
Highlight problems requiring further research and be helpful in promoting thermal process
research and applications
This project was realized due to the hard work of many people. We express our warm appreciation
to the authors of the respective chapters for their diligence and contribution. The editors are truly
indebted to everyone for their contribution, assistance, encouragement, and constructive criticism
throughout the preparation of this book.
Here, we also extend our sincere gratitude to Dr. George E. Totten (Totten Associates and a
former president of the International Federation for Heat Treatment and Surface Engineering
[IFHTSE]) and Robert Wood (secretary general, IFHTSE), whose initial encouragement made
this book possible, and to the staff of CRC Press and Taylor & Francis for their patience and
assistance throughout the production process.
C. Hakan Gür
Middle East Technical University
Jiansheng Pan
Shanghai, Jiao Tong University
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page viii 6.11.2008 6:02pm Compositor Name: VBalamugundan
viii
Editors
C. Hakan Gür is a professor in the Department of Metallurgical
and Materials Engineering at Middle East Technical University,
Ankara, Turkey. He is also the director of the Welding Tech
nology and Nondestructive Testing Research and Application
Center at the same university. Professor Gür has published
numerous papers on a wide range of topics in materials science
and engineering and serves on the editorial boards of national
and international journals. His current research includes simula
tion of tempering and severe plastic deformation processes,
nondestructive evaluation of residual stresses, and microstruc
tures obtained by various manufacturing processes.
Jiansheng Pan is a professor in the School of Materials Science
and Engineering at Shanghai Jiao Tong University, Shanghai,
China. He was an elected member of the Chinese Academy of
Engineering in 2001. Professor Pan’s expertise is in chemical
and thermal processing of steels (including nitriding, carburiz
ing, and quenching) and their computer modeling and simula
tion. He has established mathematical models of these processes
integrating heat and mass transfer, continuum mechanics, ﬂuid
mechanics, numerical analysis, and software engineering. These
models have been used for computational simulation to design
and optimize thermal processes for parts with complicated shape.
Pan and his coworkers have published extensively in these areas
and have been awarded over 40 Chinese patents. In addition to a
number of awards for scientiﬁc and technological achievements,
Professor Pan was the president of the Chinese Heat Treatment Society (2003–2007) and is the
chairman of the Mathematical Modeling and Computer Simulation Activity Group of the Inter
national Federation for Heat Treatment and Surface Engineering.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page ix 6.11.2008 6:02pm Compositor Name: VBalamugundan
ix
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page x 6.11.2008 6:02pm Compositor Name: VBalamugundan
Contributors
María Victoria Bongiovanni
Facultad de Ingeniería
Universidad Austral
Buenos Aires, Argentina
and
Facultad de Ciencias Exactas y Naturales
Universidad de Buenos Aires
Buenos Aires, Argentina
Chris H.J. Davies
Department of Materials Engineering
Monash University
Melbourne, Victoria, Australia
Janez Grum
Faculty of Mechanical Engineering
University of Ljubljana
Ljubljana, Slovenia
Jianfeng Gu
School of Materials Science and Engineering
Shanghai Jiao Tong University
Shanghai, China
C. Hakan Gür
Department of Metallurgical and Materials
Engineering
Middle East Technical University
Ankara, Turkey
Sivaraman Guruswamy
Department of Metallurgical Engineering
University of Utah
Salt Lake City, Utah
Bernardo HernandezMorales
Departamento de Ingeniería Metalúrgica
Universidad Nacional Autónoma de México
Mexico
Peter Hodgson
Centre for Material and Fibre Innovation
Institute for Technology Research and
Innovation
Deakin University
Geelong, Victoria, Australia
John J. Jonas
Department of Materials Engineering
McGill University
Montreal, Quebec, Canada
Valentin Nemkov
Fluxtrol, Inc.
Auburn Hills, Michigan
and
Centre for Induction Technology
Auburn Hills, Michigan
Jiansheng Pan
School of Materials Science and Engineering
Shanghai Jiao Tong University
Shanghai, China
Mario Rosso
R&D Materials and Technologies
Politecnico di Torino
Dipartimento di Scienza dei Materiali e
Ingegneria Chimica
Torino, Italy
and
Politecnico di Torino
Sede di Alessandria
Alessandria, Italy
Satyam Suraj Sahay
Tata Research Development and Design Centre
Tata Consultancy Services Limited
Pune, Maharashtra, India
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page xi 6.11.2008 6:02pm Compositor Name: VBalamugundan
xi
Gustavo Sánchez Sarmiento
Facultad de Ingeniería
Universidad de Buenos Aires
Buenos Aires, Argentina
and
Facultad de Ingeniería
Universidad Austral
Buenos Aires, Argentina
Wei Shi
Department of Mechanical Engineering
Tsinghua University
Beijing, China
Caner S¸ims¸ir
Stiftung Institüt für Werkstofftechnik (IWT)
Bremen, Germany
Božo Smoljan
Department of Materials Science and
Engineering
University of Rijeka
Rijeka, Croatia
Weimin Zhang
School of Materials Science
and Engineering
Shanghai Jiao Tong University
Shanghai, China
Gur/Handbook of Thermal Process Modeling of Steels 190X_C000 Final Proof page xii 6.11.2008 6:02pm Compositor Name: VBalamugundan
xii
1
Mathematical Fundamentals
of Thermal Process
Modeling of Steels
Jiansheng Pan and Jianfeng Gu
CONTENTS
1.1 Thermal Process PDEs and Their Solutions.......................................................................... 2
1.1.1 PDEs for Heat Conduction and Diffusion .................................................................. 2
1.1.2 Solving Methods for PDEs ......................................................................................... 5
1.2 FiniteDifference Method....................................................................................................... 6
1.2.1 Introduction of FDM Principle ................................................................................... 6
1.2.2 FDM for OneDimensional Heat Conduction and Diffusion ..................................... 6
1.2.3 Brief Summary.......................................................................................................... 12
1.3 FiniteElement Method ........................................................................................................ 12
1.3.1 Brief Introduction...................................................................................................... 12
1.3.1.1 Stage 1: Preprocessing ................................................................................ 13
1.3.1.2 Stage 2: Solution ......................................................................................... 13
1.3.1.3 Stage 3: Postprocessing............................................................................... 13
1.3.2 Galerkin FEM for TwoDimensional Unsteady Heat Conduction........................... 14
1.3.3 FEM for ThreeDimensional Unsteady Heat Conduction ........................................ 19
1.4 Calculation of Transformation Volume Fraction................................................................. 21
1.4.1 Interactions between Phase Transformation and Temperature ................................. 21
1.4.2 Diffusion Phase Transformation ............................................................................... 21
1.4.2.1 Modiﬁcation of Additivity Rule for Incubation Period .............................. 23
1.4.2.2 Modiﬁcation of Avrami Equation ............................................................... 25
1.4.2.3 Calculation of Proeutectoid Ferrite and Pearlite Fraction........................... 26
1.4.3 Martensitic Transformation....................................................................................... 28
1.4.4 Effect of Stress State on Phase Transformation Kinetics ......................................... 30
1.4.4.1 Diffusion Transformation............................................................................ 30
1.4.4.2 Martensitic Transformation ......................................................................... 30
1.5 Constitutive Equation of Solids ........................................................................................... 31
1.5.1 Elastic Constitutive Equation.................................................................................... 31
1.5.1.1 Linear Elastic Constitutive Equation........................................................... 31
1.5.1.2 Hyperelastic Constitutive Equation............................................................. 33
1.5.2 Elastoplastic Constitutive Equation .......................................................................... 36
1.5.2.1 Introduction ................................................................................................. 36
1.5.2.2 Yield Criterion............................................................................................. 36
1.5.2.3 Flow Rule .................................................................................................... 37
1.5.2.4 Hardening Law............................................................................................ 38
1.5.2.5 Commonly Used Plastic Constitutive Equations ........................................ 39
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 1 3.11.2008 3:17pm Compositor Name: BMani
1
1.5.2.6 Elastoplastic Constitutive Equation............................................................. 44
1.5.2.7 Thermal Elastoplastic Constitutive Equation .............................................. 45
1.5.3 Viscoplastic Constitutive Equation........................................................................... 47
1.5.3.1 OneDimensional Viscoplastic Model ........................................................ 47
1.5.3.2 Viscoplastic Constitutive Equation for General Stress State ...................... 49
1.5.3.3 Commonly Used Viscoplastic Models........................................................ 49
1.5.3.4 Creep ........................................................................................................... 50
1.6 Basics of Computational Fluid Dynamics in Thermal Processing...................................... 53
1.6.1 Introduction............................................................................................................... 53
1.6.2 Governing Differential Equations for Fluid.............................................................. 53
1.6.2.1 Generalized Newton’s Law......................................................................... 53
1.6.2.2 Continuity Equation (Mass Conservation Equation) .................................. 54
1.6.2.3 Momentum Conservation Equation............................................................. 55
1.6.2.4 Energy Conservation Equation.................................................................... 55
1.6.3 General Form of Governing Equations..................................................................... 56
1.6.4 Simpliﬁed and Special Equations in Thermal Processing........................................ 56
1.6.4.1 Continuity Equation for Incompressible SourceFree Flow ....................... 57
1.6.4.2 Euler Equations for Ideal Flow................................................................... 57
1.6.4.3 Volume Function Equation ......................................................................... 58
1.6.5 Numerical Solution of Governing PDEs .................................................................. 58
References ....................................................................................................................................... 59
Steels are usually under the action of multiple physical variable ﬁelds, such as temperature ﬁeld,
ﬂuid ﬁeld, electric ﬁeld, magnetic ﬁeld, plasm ﬁeld, and so on during thermal processing. Thus, heat
conduction, diffusion, phase transformation, evolution of microstructure, and mechanical deform
ation are simultaneously taken place inside. This chapter includes the mathematical fundamentals of
the most widely used numerical analysis methods for the solution of partial differential equations
(PDEs), and the basic knowledge of continuum mechanics, ﬂuid mechanics, phase transformation
kinetics, etc. All these are indispensable for the establishment of the coupled mathematical models
and realization of numerical simulation of thermal processing.
1.1 THERMAL PROCESS PDEs AND THEIR SOLUTIONS
1.1.1 PDES FOR HEAT CONDUCTION AND DIFFUSION
The ﬁrst step of computer simulation of thermal processing is to establish an accurate mathematical
model, i.e., the PDEs and boundary conditions that can quantiﬁcationally describe the related
phenomena.
The PDE describing the temperature ﬁeld inside a solid is usually expressed as follows:
@
@x
l
@T
@x
_ _
þ
@
@y
l
@T
@y
_ _
þ
@
@z
l
@T
@z
_ _
þQ ¼ rc
p
@T
@t
(1:1)
where
T is the temperature
t is the time
x, y, z are the coordinates
l is the thermal conduction coefﬁcient
r is the density
c
p
is the heat capacity
Q is the intensity of the internal heat resource
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 2 3.11.2008 3:17pm Compositor Name: BMani
2 Handbook of Thermal Process Modeling of Steels
Equation 1.1 has a very clear physical concept, and can be illustrated as in Figure 1.1. The ﬁrst
item on the lefthand side of the equation is the net heat ﬂux input to the inﬁnitesimally small
element along axis x, i.e., the difference between the heat ﬂux entering dQ
xin
and the heat ﬂux
effusing dQ
xout
. The second and third items are the net heat ﬂux along axes y and z, respectively
(Figure 1.1). The intensity of the internal heat source Q may be caused by different factors, such as
phase transformation, plastic work, electricity current, etc. The righthand side of the equation
stands for the change in heat accumulating in the inﬁnitesimal element per time unit due to the
temperature change. Equation 1.1 shows that the sum of the heat input and heat generated by the
internal heat source is equal to the change in heat accumulating for an inﬁnitesimal element in each
time unit, so it functions in accordance with the energy conservative law. The heat conduction
coefﬁcient l, density r, heat capacity c
p
, and the intensity of the internal heat source are usually the
functions of temperature, making Equation 1.1 a nonlinear PDE.
There are three kinds of boundary conditions for heat exchange in all kinds of thermal
processing technologies.
The ﬁrst boundary condition S
1
: The temperature of the boundary (usually certain surfaces) is
known; it is a constant or function of time.
T
s
¼ C(t) (1:2)
The second boundary condition S
2
: The heat ﬂux of the boundary is known.
l
@T
@n
¼ q (1:3)
where
@T=@n is the temperature gradient on the boundary along the external normal direction
q is the heat ﬂux through the boundary surface
The third boundary condition S
3
: The heat transfer coefﬁcient between the workpiece and
environment is known.
z
dQ
z out
dQ
z in
dQ
y out
dQ
z in
dQ
y in
dQ
x out
x
y
FIGURE 1.1 Heat ﬂux along coordinates subjected to an inﬁnitesimal element.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 3 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 3
Àl
@T
@n
_ _
¼ h(T
a
ÀT
s
) (1:4)
where
T
a
is the environment temperature
T
s
is the surface temperature of the workpiece
h is the overall heat transfer coefﬁcient, representing the heat quantity exchanged between
the workpiece surface and the environment per unit area and unit time when their tempera
ture difference is 18C
It is worth mentioning that only convective heat transfer occurs in some cases; however, radiation
heat transfer should also be considered in other complicated ones, such as gas quenching and
heating under protective atmosphere. Hence, the overall heat transfer coefﬁcient h should be the sum
of the convective heat transfer coefﬁcient h
c
and the radiation heat transfer coefﬁcient h
r
. Therefore,
we have
h ¼ h
c
þh
r
(1:5)
The radiation heat transfer coefﬁcient h
r
can be obtained as follows:
h
r
¼ «s(T
2
a
þT
2
s
)(T
a
þT
s
) (1:6)
where
« is the radiation emissivity of the workpiece
s is the Stefan–Boltzmann constant
The boundary condition can be set according to the speciﬁc thermal process, and the tempera
ture ﬁeld inside the workpiece at different times, the socalled unsteady temperature ﬁeld, can be
obtained by solving Equation 1.1. When the temperature ﬁeld inside the workpiece does not change
with time any more, it arrives at the steady temperature ﬁeld, and the lefthand side of Equation 1.1
becomes zero.
The unsteady concentration ﬁeld inside the workpiece subjected to carburizing or nitriding is
usually governed by the following PDE.
@
@x
D
@C
@x
_ _
þ
@
@y
D
@C
@y
_ _
þ
@
@z
D
@C
@z
_ _
¼
@C
@t
(1:7)
where
C is the concentration of the element being penetrated (carbon or nitrogen)
D is the diffusion coefﬁcient
The boundary conditions can also be classiﬁed into the following three kinds.
Boundary s
1
: The surface concentration is known.
C
s
¼ C (1:8)
Boundary s
2
: The mass ﬂux through the surface is known.
D
@C
@n
_ _
¼ q (1:9)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 4 3.11.2008 3:17pm Compositor Name: BMani
4 Handbook of Thermal Process Modeling of Steels
Boundary s
3
: The mass transfer coefﬁcient between the workpiece surface and environment
(ambient media) is known.
ÀD
@C
@n
_ _
¼ b(C
g
ÀC
s
) (1:10)
where
D is the diffusion coefﬁcient
b is the mass transfer coefﬁcient
C
g
is the atmosphere potential of carbon (or nitrogen)
C
s
is the surface concentration of carbon (or nitrogen)
Although the diffusion and heat conduction PDEs describe different physical phenomena, their
mathematical expression and solving method are exactly the same.
1.1.2 SOLVING METHODS FOR PDES
Usually, there are two methods to solve the PDEs, analytical method and numerical method. The
analytical method, taking speciﬁc boundary conditions and initial conditions, can obtain the analytical
solution by deduction (for example, variables separation method), which is a type of mathematical
representation clearly describing certain ﬁeld variables under space coordinates and time.
The analytical solution has the advantage of concision and accuracy, so it is also called exact
solution. Although it plays an important role in fundamental research, it is only applicable to very
few cases with relatively simple boundary and initial conditions. Therefore, the analytical solution
cannot cope with massive problems under practical manufacture environment, which are featured
with complicated boundary conditions and a high degree of nonlinearity.
The numerical solution, also named approximate solution, is applicable for different kinds
of boundary conditions and can cope with nonlinear problems. It is the most basic simulation
method in engineering. Up to now, the ﬁniteelement method (FEM) and ﬁnitedifference method
(FDM) are the most widely used methods in simulation of the process, and their common
characteristic is discretization of continuous functions, thus transforming the PDEs into large
systems of simultaneous algebraic equations and solving the large algebraic equation group ﬁnally
(Figure 1.2).
y
f
1
f
0
f
−1
x
−1
x
0
x
1
y = f (x)
x
FIGURE 1.2 Discretization of the continuous function.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 5 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 5
1.2 FINITEDIFFERENCE METHOD
1.2.1 INTRODUCTION OF FDM PRINCIPLE
First, for a continuous function of x, namely f(x), f
À1
, f
0
, and f
1
are retained as the values of f at x
À1
,
x
0
, and x
1
, respectively (Figure 1.2). When the function has all its derivatives deﬁned at x
0
and f
1
, f
À1
can be expressed by a Taylor series as follows:
f
1
¼ f
0
þDx Á f
0
0
þ
(Dx)
2
2!
f
00
0
þ
(Dx)
3
3!
f
000
0
þ
(Dx)
iV
4!
f
iV
0
þÁ Á Á (1:11)
f
À1
¼ f
0
ÀDx Á f
0
0
þ
(Dx)
2
2!
f
00
0
À
(Dx)
3
3!
f
000
0
þ
(Dx)
iV
4!
f
iV
0
ÀÁ Á Á (1:12)
Truncating the items after (Dx)
2
, Equation 1.11 can be written as
@f
@x
¸
¸
¸
¸
x¼x
0
¼ f
0
0
¼
f
1
Àf
0
Dx
À
Dx
2
f
00
0
%
f
1
Àf
0
Dx
(1:13)
Equation 1.13 is the ﬁrstorder forward difference with its truncation error of V(Dx). Here V(Dx) is a
formal mathematical notation, which represents terms of order Dx.
In the same way, another difference scheme from Equation 1.12 can be obtained as follows:
@f
@x
¸
¸
¸
¸
x¼x
0
¼ f
0
0
¼
f
0
Àf
À1
Dx
þ
Dx
2
f
00
0
%
f
0
Àf
À1
Dx
(1:14)
This is the ﬁrstorder backward difference with its truncation error of V(Dx).
Subtracting Equation 1.12 from Equation 1.11 yields
@f
@x
¼ f
0
0
¼
f
1
Àf
À1
2
þ2
(Dx)
2
3!
f
000
0
%
f
1
Àf
À1
2
(1:15)
Equation 1.15 is the secondorder central difference with its truncation error of V(Dx
2
).
Summing Equations 1.11 and 1.12, and solving for @
2
f=@x
2
, we have
@
2
f
@x
2
¼ f
00
0
¼
f
1
À2f
0
þf
À1
(Dx)
2
þ2
(Dx)
2
4!
f
iV
0
%
f
1
À2 f
0
þf
À1
(Dx)
2
(1:16)
Equation 1.16 is the secondorder central second difference with its truncation error of V(Dx
2
).
It can be observed that the truncation error, originating from the replacement of the partial
derivatives by ﬁnitedifference quotients, makes the FDM solution an approximate one; however,
the accuracy can be improved by reducing the step size.
1.2.2 FDM FOR ONEDIMENSIONAL HEAT CONDUCTION AND DIFFUSION
In this section, two simple cases are taken to elucidate the FDM to solve the PDEs in engineering.
The ﬁrst case is the unsteady, onedimensional heat conduction PDE without an internal heat
resource item, and the second one is the onedimensional diffusion PDE.
The governing PDE for the unsteady, onedimensional heat conduction without an internal heat
resource item has the following concise form:
a
@
2
T
@x
2
¼
@T
@t
(1:17)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 6 3.11.2008 3:17pm Compositor Name: BMani
6 Handbook of Thermal Process Modeling of Steels
where a¼l= rc
p
, x and t are independent variables. Equation 1.17 has two independent variables,
x and t.
Replacing the partial derivatives in Equation 1.17 with ﬁnitedifference quotients, the difference
equation can be obtained as follows:
T
n
iÀ1
À2T
n
i
þT
n
iþ1
(Dx)
2
¼
1
a
T
nþ1
i
ÀT
n
i
Dt
(1:18)
where
i is the running index in the x direction
n is the running index in the t direction
When one of the independent variables is a marching variable, such as t in Equation 1.17, it is
conventional to denote the running index for this marching variable by n and to display this index as
a superscript in the ﬁnitedifference quotient (Figure 1.3), T
n
iÀ1
, T
n
i
, and T
n
iþ1
are the temperatures on
node i À1, i, and i þ1 at time level n, respectively, and T
nþ1
i
is the temperature on node i þ1 at time
level n þ1.
With some rearrangement, this equation can be written as
T
nþ1
i
¼ F
0
T
n
iþ1
þF
0
T
n
iÀ1
þ(1 À2F
0
)T
n
i
(1:19)
where
F
0
¼
aDt
(Dx)
2
¼
l Á Dt
rc
p
(Dx)
2
Equation 1.19 is written with temperatures at time level n on the righthand side and temperatures
at time level n þ1 on the lefthand side. Within the timemarching philosophy, all temperatures at
level n are known and those at level n þ1 are to be calculated. Of particular signiﬁcance is that only
one unknown T
nþ1
i
appears in Equation 1.19. Hence, Equation 1.19 allows for the immediate
solution of T
nþ1
i
from the known temperatures at time level n.
Equation 1.19 is one of the socalled explicit ﬁnitedifference approaches, which provide a
straightforward mechanism to accomplish this time marching (Figure 1.3). However, this approach
T
Δx
i – 1 i + 1 i
Δx
n + 1
n
x
FIGURE 1.3 Illustration of discretization and time marching.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 7 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 7
has the limitation that only when the stability criterion F
0
1=2 is met, the converged solution can be
obtained. The truncation error can be estimated as V(Dx
2
)(Dt).
Writing the spatial difference on the righthand side of Equation 1.17 in terms of temperatures at
time level n þ1, we can obtain
a Á
T
nþ1
iÀ1
À2T
nþ1
i
þT
nþ1
iþ1
(Dx)
2
¼
T
nþ1
i
ÀT
n
i
Dt
(1:20)
With some rearrangement, Equation 1.21 can be written as
F
0
T
nþ1
iÀ1
À(2F
0
þ1)T
nþ1
i
þF
0
T
nþ1
iÀ1
¼ T
n
i
(1:21)
Observing Equation 1.21, the unknown T
nþ1
i
is not only expressed in terms of the known temper
atures at time level n, namely T
n
i
, but also in terms of other unknown temperatures at time level
n þ1, namely, T
nþ1
iÀ1
and T
nþ1
iþ1
. In other words, Equation 1.21 represents one equation with three
unknowns, namely T
nþ1
iÀ1
, T
nþ1
i
, and T
nþ1
iþ1
. Hence, Equation 1.21 applied at a given grid point i does
not stand alone; it cannot by itself result in a solution for T
nþ1
i
. Rather, Equation 1.21 must be
written at all interior grid points, resulting in a system of algebraic equations from which the
unknowns T
nþ1
i
for all i can be solved simultaneously.
Equation 1.21 is one of the socalled implicit ﬁnitedifference approaches, in which the unknown
must be obtained by means of simultaneous solution of the difference equations applied at all grid
points arrayed at a given time level. Because of this need to solve large systems of simultaneous
algebraic equations, implicit methods are usually involved with the manipulations of large matrices.
One advantage of these methods lies in the fact that they are unconditionally stable, i.e., they can
always get the converged solution. Their truncation error can also be estimated as V(Dx
2
) Á (Dt).
There are different difference equations that can represent Equation 1.17 except Equations 1.21
and 1.22, which are the only two of many difference representations of the original PDE. For
example, writing the spatial difference on the righthand side in terms of average temperatures
between time level n and n þ1, Equation 1.17 can be represented by
a
T
n
iÀ1
À2T
n
i
þT
n
iþ1
2(Dx)
2
þ
T
nþ1
iÀ1
À2T
nþ1
i
þT
nþ1
iþ1
2(Dx)
2
_ _
¼
T
nþ1
i
ÀT
n
i
Dt
ÀF
0
T
nþ1
iÀ1
þ2(1 þF
0
)T
nþ1
i
ÀF
0
T
nþ1
iþ1
¼ F
0
T
n
iÀ1
þ2(1 ÀF
0
)T
n
i
þF
0
T
n
iþ1
(1:22)
This special type of differencing employed in Equation 1.22 is called the Crank–Nicolson form,
which is also unconditionally stable and has a small truncation error of V(Dx)
2
(Dt)
2
.
As a typical case of onedimensional heat conduction without an internal heat resource, an inﬁnite
plate with a thickness of d is subjected to the boundary condition that can be expressed as follows:
h(T
a
ÀT
s
) ¼ Àl
@T
@n
¸
¸
¸
¸
x¼0
(1:23)
The spatial discretization is shown in Figure 1.4, generating mþ1 nodes from surface (node 0) to
center (node m). Here, because of the symmetry half of the slab can only be considered, and the
symmetry axis can be set as the adiabatical boundary.
For the surface node (i ¼0), the boundary condition is introduced and the Crank–Nicolson form
can be obtained as follows:
h T
a
À
1
2
T
nþ1
0
þT
n
0
_ _
_ _
À
l
2Dx
T
nþ1
0
ÀT
nþ1
1
þT
n
0
ÀT
n
1
_ _
¼
Dxrc
p
2Dt
T
nþ1
0
ÀT
n
0
_ _
(1:24)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 8 3.11.2008 3:17pm Compositor Name: BMani
8 Handbook of Thermal Process Modeling of Steels
With some rearrangement, Equation 1.24 can be written as
(1 þF
0
þBi)T
nþ1
0
ÀF
0
T
nþ1
1
¼ (1 ÀF
0
ÀBi)T
n
0
þF
0
T
n
1
þ2BiT
a
(1:25)
where Bi ¼hDt=Dx.
For the central node (i ¼m), the adiabatical boundary @T=@x ¼0 is input, and the difference
equation is
(1 þF
0
)T
nþ1
mÀ1
ÀF
0
T
nþ1
mÀ1
¼ (1 ÀF
0
)T
n
m
þF
0
T
n
mÀ1
(1:26)
The ﬁnitedifference form for transient heat conduction in the inﬁnite plate is composed by
Equations 1.22, 1.25, and 1.26, providing mþ1 algebraic equations for mþ1 unknown T
i
for all
nodes. The unique solution for the temperature ﬁeld can usually be obtained by a mature algorithm.
The onedimensional diffusion PDE (Equation 1.27) and its difference equations have the same
form as that of the unsteady, onedimensional heat conduction without an internal heat resource.
Hence, the corresponding equations are brieﬂy repeated and then entered into solving of the
algebraic equation group.
D
@
2
C
@X
2
¼
@C
@t
(1:27)
The boundary conditions at the surface node and the center node are the third type of boundary
condition and the adiabatical condition, respectively, and are listed as follows:
ÀD
@C
@X
¸
¸
¸
¸
x¼0
¼ b(C
g
ÀC
s
)
D
@C
@X
¸
¸
¸
¸
x¼m
¼ 0
_
¸
¸
¸
_
¸
¸
¸
_
(1:28)
d
2
T
0 1 2 i m x
/
FIGURE 1.4 Spatial difference scheme of nodes for an inﬁnite plate.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 9 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 9
For the inner nodes (i ¼1 Á Á Á mÀ1), the Crank–Nicolson difference form can be written as
ÀF
0
C
nþ1
iÀ1
þ2(1 þF
0
)C
nþ1
i
ÀF
0
C
nþ1
iþ1
¼ F
0
C
n
iÀ1
þ2(1 ÀF
0
)C
n
i
þF
0
C
n
iþ1
(1:29)
where F
0
¼
(D Á Dt)
(Dx)
2
.
For the surface node (i ¼0), the Crank–Nicolson form is
(1 þF
0
þBi)C
nþ1
0
ÀF
0
C
nþ1
1
¼ (1 ÀF
0
ÀBi)C
n
0
þF
0
C
n
1
þ2BiC
g
(1:30)
For the central node (i ¼m), the difference equation is
(1 þF
0
)C
nþ1
m
ÀF
0
C
nþ1
mÀ1
¼ (1 ÀF
0
)C
n
m
þF
0
C
n
mÀ1
(1:31)
The algebraic equation group constituted by Equations 1.29 through 1.31 can be expanded and
written in the following matrix form:
d
0
a
0
b
1
d
1
a
1
.
.
.
.
.
.
.
.
.
b
i
d
i
a
i
.
.
.
.
.
.
.
.
.
b
mÀ1
d
mÀ1
a
mÀ1
b
m
d
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
C
0
C
1
.
.
.
C
i
.
.
.
C
mÀ1
C
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¼
d
0
0
a
0
0
b
0
1
d
0
1
a
0
1
.
.
.
.
.
.
.
.
.
b
0
i
d
0
i
a
0
i
.
.
.
.
.
.
.
.
.
b
0
mÀ1
d
0
mÀ1
a
0
mÀ1
b
0
m
d
0
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
Â
C
0
C
1
.
.
.
C
i
.
.
.
C
mÀ1
C
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
n
þ
2lC
0
0
.
.
.
0
.
.
.
0
0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:32)
where
d
0
¼ 1 þF
0
þBi
d
1
¼ d
2
¼ Á Á Á ¼ d
mÀ1
¼ 2(1 þF
0
)
d
m
¼ 1 þF
0
b
1
¼ b
2
¼ Á Á Á ¼ b
m
¼ ÀF
0
a
0
¼ a
1
¼ Á Á Á ¼ a
mÀ1
¼ ÀF
0
d
0
0
¼ 1 ÀF
0
ÀBi
d
0
1
¼ d
0
2
¼ Á Á Á ¼ d
0
mÀ1
¼ 2(1 ÀF
0
)
d
0
m
¼ 1 ÀF
0
b
0
1
¼ b
0
2
¼ Á Á Á ¼ b
0
m
¼ F
0
a
0
0
¼ a
0
1
¼ Á Á Á ¼ a
0
mÀ1
¼ ÀF
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 10 3.11.2008 3:17pm Compositor Name: BMani
10 Handbook of Thermal Process Modeling of Steels
When the concentration of the nodes at time level n is known, the righthand side of Equation 1.32
can be simpliﬁed as a column matrix. Thus, we obtain
d
0
a
0
b
1
d
1
a
1
.
.
.
.
.
.
.
.
.
b
i
d
i
a
i
.
.
.
.
.
.
.
.
.
b
mÀ1
d
mÀ1
a
mÀ1
b
m
d
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
C
0
C
1
.
.
.
C
i
.
.
.
C
mÀ1
C
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
nþ1
¼
F
0
F
1
.
.
.
F
i
.
.
.
F
mÀ1
F
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:33)
The coefﬁcient matrix on the lefthand side of Equation 1.33 is a tridiagonal matrix, deﬁned as
having nonzero elements only along the three diagonals, which are marked with three dashed lines.
The solution of the system of equations denoted by Equation 1.33 involves the manipulation of the
tridiagonal arrangement; such solutions are usually obtained using Thomas’ algorithm, which has
become almost standard for the treatment of tridiagonal systems of equations. A description of this
algorithm is given as follows.
First, Equation 1.33 is changed into an upper bidiagonal form by dropping the ﬁrst term of each
equation, replacing the coefﬁcient of the maindiagonal term by Equation 1.34, and replacing the
righthand side with Equation 1.35.
d
i
*
¼ d
i
À
b
i
d
iÀ1
*
a
iÀ1
(i ¼ 1, 2, 3, . . . , m À1, m) (1:34)
F
i
*
¼ F
i
À
b
i
d
iÀ1
*
F
iÀ1
*
(i ¼ 1, 2, 3, . . . , m À1, m) (1:35)
Then Equation 1.33 transforms into an upper bidiagonal form as follows:
d
0
a
0
d
1
*
a
1
.
.
.
.
.
.
.
.
.
d
i
*
a
i
.
.
.
.
.
.
.
.
.
d
mÀ1
*
a
mÀ1
d
m
*
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
C
0
C
1
.
.
.
C
i
.
.
.
C
mÀ1
C
m
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
nþ1
¼
F
0
F
1
*
.
.
.
F
i
*
.
.
.
F
mÀ1
*
F
m
*
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:36)
The elements with the superscript asterisk are those subjected to Gaussian elimination. The last
equation in Equation 1.36 has only one unknown. Solving C
nþ1
m
from Equation 1.37, all other
unknowns are found in sequence from Equation 1.38, starting from C
nþ1
mÀ1
to C
nþ1
1
.
C
nþ1
m
¼
F
m
*
d
m
*
(1:37)
C
nþ1
i
¼
F
i
*
Àa
i
C
nþ1
iþ1
_ _
d
i
*
(i ¼ m À1, m À2, . . . , 2, 1) (1:38)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 11 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 11
The concentration of the surface node C
nþ1
0
can be obtained by
C
nþ1
0
¼
F
0
Àa
0
C
nþ1
1
_ _
d
0
(1:39)
So far, the concentration of all nodes at time level n þ1 has been calculated. Sometimes, the activity
is used instead of the concentration in diffusion problems such as the nitriding process.
1.2.3 BRIEF SUMMARY
The main advantage of FDM lies in its rigorous mathematical derivation, and it is very simple when
applied in onedimensional problems. For the two and threedimensional problems, FDM can also
be applied but for objects with relatively simple shapes due to its limitation in coping with
complicatedshape boundaries. Hence, the FEM method is mainly used in the simulation of
temperature ﬁelds and concentration ﬁelds with threedimensional complicatedshape parts.
1.3 FINITEELEMENT METHOD
The FEM, sometimes referred to as ﬁniteelement analysis (FEA), is a computational technique used
to obtain approximate solutions of boundary value problems in engineering. Simply stated, a
boundary value problem is a mathematical problem in which one or more dependent variables
must satisfy a differential equation everywhere within a known domain of independent variables
and satisfy speciﬁc conditions on the boundary of the domain. Usually, FEM divides the deﬁnition
domain into reasonably deﬁned subdomains (element) by hypothesis and supposes the unknown
state variable function in each element approximately deﬁned, so that the approximate solution of
boundary value and initial value problems is thus obtained. Since the respectively deﬁned functions
can be harmonized at element nodes or certain joint points, the unknown function can approxi
mately be expressed in the whole deﬁnition domain.
Because of the extraordinary ﬂexibility of element division, the FEM elements can ﬁt well to
objects with complex shape and the boundaries with complex curved surfaces. For example,
complex threedimensional regions can be effectively ﬁlled by tetrahedral elements, similar to
triangular elements ﬁlling a twodimensional region. Therefore, the FEM is the most widely used
method in heat treatment numerical simulation so far.
FEM has been dissertated in detail in related monographs [1–3]. Hence, a brief introduction is
presented in this section.
1.3.1 BRIEF INTRODUCTION
No matter what the physical nature of the problem, the standard FEM is always performed with a
sequential series of steps. Certain steps in formulating an FEA of a physical problem are common to
all such analyses, whether structural, heat transfer, ﬂuid ﬂow, or some other problems. The steps are
described as follows:
1. Deﬁnition of problem and its deﬁnition domain
2. Discretization of the deﬁnition domain
3. Determination of all kinds of state variables
4. Formulations of the problem
5. Establishing of coordinate system
6. Construction of the approximate function for elements
7. Derivation of element matrix and equation
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 12 3.11.2008 3:17pm Compositor Name: BMani
12 Handbook of Thermal Process Modeling of Steels
8. Coordinate transformation
9. Assembly of element equation
10. Introduction of the boundary condition
11. Solution to the ﬁnal algebraic equation
12. Explanation of the results
When FEM for a speciﬁc engineering problem is performed by computer (these steps are embodied
in commercial ﬁniteelement software packages), it usually involves three stages of activity:
preprocessing, solution, and postprocessing.
1.3.1.1 Stage 1: Preprocessing
The preprocessing stage involves the preparation of data, such as nodal coordinates, connectivity,
boundary conditions, loading, and materials information. It is generally described as deﬁning the
model and includes the following:
.
Deﬁne the geometric domain of the problem
.
Deﬁne the element type(s) to be used
.
Deﬁne the material properties of the elements
.
Deﬁne the geometric properties of the elements (length, area, and the like)
.
Deﬁne the element connectivities (mesh of the model)
.
Deﬁne the physical constraints (boundary conditions)
.
Deﬁne the loadings
1.3.1.2 Stage 2: Solution
The solution stage involves stiffness generation, stiffness modiﬁcation, and solution of equations,
resulting in the evaluation of nodal variables. Other derived quantities, such as gradients for stresses,
may be evaluated at this stage. In other words, the ﬁniteelement software assembles the governing
algebraic equations in matrix form and computes the unknown values of the primary ﬁeld variable(s).
The computed values are then used by back substitution to compute additional derived variables, such
as reaction forces, element stresses, and heat ﬂow.
1.3.1.3 Stage 3: Postprocessing
Analysis and evaluation of the solution results is referred to as postprocessing, so the postprocessing
stage deals with the presentation of results. Typically, the deformed conﬁguration, mode shapes,
temperature, and stress distribution are computed and displayed at this stage. Postprocessor software
contains sophisticated routines used for sorting, printing, and plotting selected results from a ﬁnite
element solution. Examples of operations that can be accomplished include
.
Sort element stresses in order of magnitude
.
Check equilibrium
.
Calculate factors of safety
.
Plot deformed structural shape
.
Animate dynamic model behavior
.
Produce colorcoded temperature plots
While solution data can be manipulated in many ways in postprocessing, the most important
objective is to apply sound engineering judgment in determining whether the solution results are
physically reasonable.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 13 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 13
1.3.2 GALERKIN FEM FOR TWODIMENSIONAL UNSTEADY HEAT CONDUCTION
The method of weighted residuals, especially the embodiment of the Galerkin FEM, is a powerful
mathematical tool that provides a technique for formulating a ﬁniteelement solution approach to
practically any problem for which the governing differential equation and boundary conditions can
be written.
Here, twodimensional unsteady heat conduction is taken as an example to derive the FEM
formulations by Galerkin’s weighted residual method.
The governing PDE for unsteady, twodimensional heat conduction with an internal heat
resource is
l
@
2
T
@x
2
þ
@
2
T
@y
2
_ _
þQ Àrc
p
@T
@t
¼ 0 (1:40)
The initial condition, supposing the temperature ﬁeld is known, can be written as
t ¼ 0: T ¼ T
0
(1:41)
The three types of boundary conditions have been listed by Equations 1.2 through 1.4.
The righthand side of Equation 1.40 equals zero when the column vector T, the exact solution
of the temperature ﬁeld, is substituted. On the contrary, the approximate solution T makes the
residual error:
R ¼ l
@
2
T
@x
2
þ
@
2
T
@y
2
_ _
þQ Àrc
p
@T
@t
(1:42)
The basic idea of the weighted residual method is to construct a suitable weight function so that the
integration of products by residual error and weight function equals zero; the approximate solution
on the whole domain can thus be obtained. Therefore, we have
__
D
W
i
l
@
2
T
@x
2
þ
@
2
T
@y
2
_ _
Àrc
p
@T
@t
þQ
_ _
dx dy ¼ 0 (1:43)
where W
i
is the weight function.
Several variations of the weighted residual method exist and the techniques vary primarily in
how the weight functions are determined or selected. The most common techniques are point
collocation, subdomain collocation, least squares, and Galerkin’s method [4].
The secondorder differential item in Equation 1.43 can be transformed into a ﬁrstorder item by
integration in parts. Therefore, the secondorder differential item can be expressed as
__
D
W
i
l
@
2
T
@x
2
þ
@
2
T
@y
2
_ _ _ _
dx dy ¼ À
__
D
l
@W
i
@x
@T
@x
þ
@W
i
@y
@T
@y
_ _
dx dy þ
_
S
W
i
l
@T
@n
ds (1:44)
and
_
S
W
i
l
@T
@n
ds ¼
_
S
1
W
i
l
@T
@n
ds þ
_
S
2
W
i
l
@T
@n
ds þ
_
S
3
W
i
l
@T
@n
ds (1:45)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 14 3.11.2008 3:17pm Compositor Name: BMani
14 Handbook of Thermal Process Modeling of Steels
Since the temperature on boundary S
1
is known, we have
_
S
1
W
i
l(@T=@n)ds ¼ 0. Substituting
Equations 1.44 and 1.45 into Equation 1.43, we can get
__
D
l
@W
i
@x
@T
@x
þ
@W
i
@y
@T
@y
_ _
dx dy þW
i
rc
p
@T
@t
ÀQ
_ _
dx dy ¼
_
S
2
W
i
l
@T
@n
ds þ
_
S
3
W
i
l
@T
@n
ds (1:46)
The element analysis is to establish the formulations by which the continuous function in the
subdomain DD can be expressed by the node values, i.e., the function of the node values. Supposing
there are n elements in the solution domain, and there are m nodes in each element with the
temperature T
i
(i ¼1, 2, . . . , m), the unknown temperature T
e
(x, y, z)
deﬁned in the element can be
expressed as the interpolation function of each node. Therefore, we have
T
e
(x, y, z)
¼ N
1
T
1
þN
2
T
2
þÁ Á Á þN
m
T
m
¼ [N
i
]{T
i
} (1:47)
where N
i
is the shape function, which is a function of the components of each element nodes (x
i
, y
i
, z
i
)
and those of the location (x, y, z).
For example, the shape functions of a triangle element with three nodes can be simply derived
and expressed as
N
i
(x, y) ¼
1
2A
(a
i
þb
i
x þc
i
y)
N
j
(x, y) ¼
1
2A
(a
j
þb
j
x þc
j
y)
N
m
(x, y) ¼
1
2A
(a
m
þb
m
x þc
m
y)
(1:48)
where
A ¼
1
2
(b
i
c
j
Àb
j
c
i
)
a
i
¼ x
j
y
m
Àx
m
y
j
; a
j
¼ x
m
y
i
Àx
i
y
m
; a
m
¼ x
i
y
j
Àx
j
y
i
b
i
¼ y
j
Ày
m
; b
j
¼ y
m
Ày
i
; b
m
¼ y
i
Ày
j
c
i
¼ x
m
Àx
j
; c
j
¼ x
i
Àx
m
; c
m
¼ x
j
Àx
i
Although the shape functions for different types of elements have different forms, they have
common characteristics and can be obtained by the geometrical method. Especially the coefﬁcients
in the shape functions are the functions of the coordinates of each node, that is,
N
i
(x, y, z) ¼ F(x
i
, y
i
, z
i
, x, y, z), i ¼ 1, 2, . . . , m (1:49)
Obviously, the shape functions are determined only by the coordinates of each node and the type of
element. Hence, the temperature on certain points in an element can be expressed as the function of
node temperature, namely, the column vector {T
i
} (see Equation 1.47).
Equation 1.46 is applicable in the whole solution domain D, so for its subdomains, i.e., each
element DD, we have
__
DD
l
@W
i
@x
@T
@x
þ
@W
i
@y
@T
@y
_ _
dx dy þW
i
rc
p
@T
@t
ÀQ
_ _
dx dy
¼
_
DS
2
W
i
l
@T
@n
ds þ
_
DS
3
W
i
l
@T
@n
ds (1:50)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 15 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 15
In Galerkin’s weighted residual method, the weight functions are chosen to be identical to the trial
function. Here, the shape functions are taken as the weight functions, for example, W
i
¼N
i
(x, y), and
substitute into Equation 1.50. Hence, we have
__
DD
l
@N
i
@x
@T
@x
þ
@N
i
@y
@T
@y
_ _
dx dy þN
i
rc
p
@T
@t
ÀQ
_ _
dx dy
¼
_
DS
2
N
i
l
@T
@n
ds þ
_
DS
3
N
i
l
@T
@n
ds (1:51)
Substituting Equation 1.47 into Equation 1.51, and introducing the boundary conditions from
Equations 1.2 through 1.4, we have
__
DD
l
@N
i
@x
@N
i
@x
T
i
þ
@N
j
@x
T
j
þ
@N
m
@x
T
m
_ _
þl
@N
i
@y
@N
i
@y
T
i
þ
@N
j
@y
T
j
þ
@N
m
@y
T
m
_ _ _
þ N
i
rc
p
Á N
i
@T
i
@t
þN
j
@T
j
@t
þN
m
@T
m
@t
_ _
ÀN
i
Q
_
dx dy
¼
_
DS
2
qN
i
ds À
_
DS
3
hN
i
(N
i
T
i
þN
j
T
j
ÀTa)ds (1:52)
Taking W
j
¼N
j
(x, y) and W
m
¼N
m
(x, y), and deriving in exactly the same way, can be obtained
as follows:
__
DD
l
@N
j
@x
@N
i
@x
T
i
þ
@N
j
@x
T
j
þ
@N
m
@x
T
m
þl
@N
j
@y
_ _
@N
i
@y
T
i
þ
@N
j
@y
T
j
þ
@N
m
@y
T
m
_ _ _
þ N
j
rc
p
N
i
@T
i
@t
þN
j
@T
j
@t
þN
m
@T
m
@t
_ _
ÀN
i
Q
_
dx dy
¼
_
DS
2
qN
j
ds À
_
DS
3
hN
j
(N
i
T
i
þN
j
T
j
ÀTa)ds (1:53)
__
DD
l
@N
m
@x
@N
i
@x
T
i
þ
@N
j
@x
T
j
þ
@N
m
@x
T
m
_ _
þl
@N
m
@y
@N
i
@y
T
i
þ
@N
j
@y
T
j
þ
@N
m
@y
T
m
_ _ _
þ N
m
rc
p
N
i
@T
i
@t
þN
j
@T
j
@t
þN
m
@T
m
@t
_ _
ÀN
i
Qdx dy
_
¼
_
DS
2
qN
m
ds À
_
DS
3
hN
m
(N
i
T
i
þN
j
T
j
ÀTa)ds (1:54)
For the interior elements, the righthand side of Equations 1.52 through 1.54 equals zero respect
ively; while for the boundary element, only the righthand side of Equation 1.54 equals zero with the
assumption that only node i and j are on the boundary.
An equation group constructed by Equations 1.52 through 1.54 contains only the three
unknowns, T
i
, T
j
, and T
m
. After rearrangement, it can be written in matrix form as follows:
[K]
e
{T
e
} þ[C]
e
@
@t
{T}
e
¼ {r}
e
(1:55)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 16 3.11.2008 3:17pm Compositor Name: BMani
16 Handbook of Thermal Process Modeling of Steels
where
[K]
e
is the element stiffness matrix
{T}
e
is the column vector of temperature on element nodes (unknown)
[C]
e
is the element heat capacity matrix
{r}
e
is the element constant vector item
The element stiffness matrix assembled from Equations 1.52 through 1.54 is
[K]
e
¼
__
DD
l
@N
i
@x
Á
@N
i
@x
þ
@N
i
@y
Á
@N
i
@y
@N
j
@x
Á
@N
i
@x
þ
@N
j
@y
Á
@N
i
@y
@N
m
@x
Á
@N
i
@x
þ
@N
m
@y
Á
@N
i
@y
@N
i
@x
Á
@N
j
@x
þ
@N
i
@y
Á
@N
j
@y
@N
j
@x
Á
@N
j
@x
þ
@N
j
@y
Á
@N
j
@y
@N
m
@x
Á
@N
j
@x
þ
@N
m
@y
Á
@N
j
@y
@N
i
@x
Á
@N
m
@x
þ
@N
i
@y
Á
@N
m
@y
@N
j
@x
Á
@N
m
@x
þ
@N
j
@y
Á
@N
m
@y
@N
m
@x
Á
@N
m
@x
þ
@N
m
@y
Á
@N
m
@y
_
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
_
dx dy
þ
_
DS
2
q
N
i
N
j
0
_
¸
_
_
¸
_ds þ
_
DS
3
h
N
i
N
i
N
j
N
i
0
N
i
N
j
N
j
N
j
0
0 0 0
_
¸
_
_
¸
_ds (1:56)
Substituting Equation 1.48 into Equation 1.56, and integrating, we obtain
[K]
e
¼
l
4A
b
2
i
þc
2
i
b
i
b
j
þc
i
c
j
b
i
b
m
þc
i
c
m
b
j
b
i
þc
j
c
i
b
2
j
þc
2
j
b
j
b
m
þc
j
c
m
b
m
b
i
þc
m
c
i
b
m
b
j
þc
m
c
j
b
2
m
þc
2
m
_
_
_
_
þ
hl
ij
6
2 1 0
1 2 0
0 0 0
_
_
_
_
(1:57)
where
A is the area of element DD
l
ij
is the length of exterior boundary; l
ij
¼0 for the interior element
The coefﬁcients b
i
, b
j
, b
m
, c
i
, c
j
, c
m
are determined by the coordinates of nodes. Obviously, every
element in [K]
e
is determined.
For the element heat capacity matrix, we have
[C]
e
¼
__
DD
rC
p
N
i
N
i
N
j
N
i
N
m
N
i
N
i
N
j
N
j
N
j
N
m
N
j
N
i
N
m
N
j
N
m
N
m
N
m
_
_
_
_
dx dy (1:58)
Substituting Equation 1.48 into Equation 1.58, and integrating, we obtain
__
DD
N
i
N
j
dx dy ¼
__
DD
N
i
N
m
dx dy ¼
__
DD
N
j
N
m
dx dy ¼
A
12
__
DD
N
i
N
i
dx dy ¼
__
DD
N
j
N
j
dx dy ¼
__
DD
N
m
N
m
dx dy ¼
A
6
(1:59)
Thus, Equation 1.58 can be rewritten as
[C]
e
¼
rc
p
A
12
2 1 1
1 2 1
1 1 2
_
_
_
_
(1:60)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 17 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 17
The element constant vector item {r}
e
in Equation 1.55 can be expanded as
{p}
e
¼ {p
Q
}
e
þ{p
q
}
e
þ{p
h
}
e
(1:61)
where {p
Q
}
e
, {p
q
}
e
, {p
h
}
e
originate from the internal heat resource, heat ﬂux in the second kind of
boundary condition, and the heat transfer coefﬁcient in the third kind of boundary condition,
respectively.
The item {p
Q
}
e
can be obtained by
{p
Q
}
e
¼
_
DD
QN
i
dx dy þ
_
DD
QN
j
dx dy þ
_
DD
QN
m
dx dy (1:62)
When the internal heat resource intensity Q is a constant, we have
{P
Q
}
e
¼
AQ
3
1
1
1
_
_
_
_
_
_
(1:63)
When the internal heat resource intensity Q is a linear function, we have
{P
Q
}
e
¼
A
12
2Q
i
Q
j
Q
m
Q
i
2q
j
Q
m
Q
i
Q
j
2Q
m
_
_
_
_
(1:64)
where Q
i
, Q
j
, and Q
m
are the internal heat resource intensity on node i, j, and m, respectively.
The item {p
q
}
e
due to the heat ﬂux on the boundary can be obtained by
{p
q
}
e
¼
_
DS
2
qN
i
ds þ
_
DS
2
qN
j
ds þ
_
DS
2
qN
m
ds (1:65)
When the heat ﬂux through the boundary q is a constant, we have
{p
q
}
e
¼
l
ij
2
q
1
1
0
_
_
_
_
_
_
(1:66)
When the heat ﬂux through the boundary q is a linear function, we have
{p
q
}
e
¼
l
ij
6
q
2q
i
þq
j
q
i
þ2q
j
0 þ0
_
_
_
_
_
_
(1:67)
The item {p
h
}
e
due to the heat transfer coefﬁcient on the boundary can be obtained by
{p
h
}
e
¼
_
DC
3
hT
a
N
i
ds þ
_
DC
3
hT
a
N
j
ds þ
_
DC
3
hT
a
N
m
ds ¼
l
ij
2
hT
a
1
1
0
_
_
_
_
_
_
(1:68)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 18 3.11.2008 3:17pm Compositor Name: BMani
18 Handbook of Thermal Process Modeling of Steels
Therefore, the continuous function T
e
(x, y, t)
on the subdomains, i.e., the element DD, has been
expressed by the node temperature, and the algebraic equation group with the unknown variables
of node temperature has been established.
For the whole solution domain, the algebraic equation groups on each element are assembled
together so that a largescale equation group can be obtained as follows:
[K]{T} þ[C]
@
@t
{T} À{P} ¼ 0 (1:69)
where
General stiffness matrix: [K] ¼
[K]
e
(1:69a)
Heat capacity matrix: [C] ¼
[C]
e
(1:69b)
Heat flux matrix: {P} ¼
{P}
e
(1:69c)
The solving of Equation 1.69 brings the solution of the temperature ﬁeld for the unsteady two
dimensional heat conduction. The evolution of temperature can be observed by recording the node
temperature at each time level, and the heating (or cooling) curves at special points can also be
extracted from the result ﬁles.
1.3.3 FEM FOR THREEDIMENSIONAL UNSTEADY HEAT CONDUCTION
The principles and procedure in the FEM for threedimensional unsteady heat conduction are
identical to those in the twodimensional one, as presented in the previous section. The derivation
of its FEM formulations is more complex and can be found in related monographs [5–7]. Here, only
the derived results are brieﬂy outlined.
The governing PDE of threedimensional unsteady heat conduction with an internal heat
resource is repeated here.
l
@
2
T
@x
2
þ
@
2
T
@y
2
þ
@
2
T
@z
2
_ _
þQ ¼ rc
p
@T
@t
(1:70)
Supposing that there are total n elements (m nodes in each element) and p nodes after the
discretization of the whole solution domain, Equation 1.70 can be transformed into an algebraic
equation group as follows:
[K]{T} þ[C]
@T
@t
_ _
À{P} ¼ 0 (1:71)
[K] ¼
[K]
e
(1:71a)
[C] ¼
[C]
e
(1:71b)
{P} ¼
{P}
e
(1:71c)
[K]
e
¼
___
V
e
[B]
T
[D] [B]dv þ
__
S
e
h[N]
T
[N]ds (1:71d)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 19 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 19
[C]
e
¼
___
V
e
rc
p
[N]
T
[N]dv (1:71e)
[p]
e
¼
___
V
e
Q[N]
T
dv þ
__
S
e
hT
a
[N]
T
ds (1:71f)
[D] ¼
l 0 0
0 l 0
0 0 l
_
_
_
_
(1:71g)
[B] ¼
@N
1
@x
@N
2
@x
. . .
@N
P
@x
@N
1
@y
@N
2
@y
. . .
@N
P
@y
@N
1
@z
@N
2
@z
. . .
@N
P
@z
_
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
_
(1:71h)
[N] ¼ [N
1
N
2
Á Á Á N
P
] (1:71i)
where
V
e
is the element volume
S
e
is the exterior boundaries
In the unsteady temperature ﬁeld, the variable temperature T
(x,y,z,t)
is a function of spatial
location and time. Equation 1.71 is the spatial discretized formulation with the time differential
item @T=@t, which can be discretized by the difference method in a uniform format as follows:
u
@T
@t
_ _
t
þ(1 Àu)
@T
@t
_ _
tÀDt
¼
1
Dt
(T
t
ÀT
tÀDt
) (1:72)
If u ¼1, the backward difference format of time discretization is
@T
@t
_ _
t
¼
1
Dt
(T
t
ÀT
tÀDt
) (1:73)
Substituting Equation 1.73 into Equation 1.71, we get
[K] þ
1
Dt
[C]
_ _
{T
t
} ¼
1
Dt
[C]{T
tÀDt
} þ{P} (1:74)
If u ¼1=2, the Crank–Nicolson difference format (also named central difference format) of time
discretization can be obtained as
1
2
@T
@t
_ _
t
þ
@T
@t
_ _
tÀDt
_ _
¼
1
Dt
(T
t
ÀT
tÀDt
) (1:75)
Substituting Equation 1.75 into Equation 1.71, the corresponding FEM algebraic equation can be
obtained as follows:
[K] þ
2
Dt
[C]
_ _
{T
t
} ¼
2
Dt
[C] À[K]
_ _
{T
tÀDt
} þ
1
Dt
{P}
t
þ{P}
tÀDt
ð Þ (1:76)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 20 3.11.2008 3:17pm Compositor Name: BMani
20 Handbook of Thermal Process Modeling of Steels
The Crank–Nicolson difference format is, in general, more accurate than the backward difference
method, in that it does not give preference to either temperature at time level t ÀDt or time level t
but, rather, gives equal credence to both.
In FDM of time, the key parameter governing solution accuracy is the selected time step Dt. In a
fashion similar to the FEM, in which the smaller the elements are, physically, the better is the
solution, the FDM converges more rapidly to the true solution as the time step is decreased.
When the temperature ﬁeld at time level t ÀDt is known, and input into Equation 1.74 or
Equation 1.76, the temperature {T
t
} on all nodes at time level t can thus be solved, and the
marching solution can progress in time until a steady state is reached. The nodal temperatures,
recorded step by step at different time levels, constitute the evolution history of the threedimen
sional temperature ﬁeld.
1.4 CALCULATION OF TRANSFORMATION VOLUME FRACTION
1.4.1 INTERACTIONS BETWEEN PHASE TRANSFORMATION AND TEMPERATURE
The latent heat releases as an internal heat source inside the solid when phase transformation occurs.
Thus, the kinetics of phase transformation strongly depends on the temperature history of steel parts,
but also strongly affects the temperature ﬁeld inside. The interactions between phase transformation
and temperature are not oneway, but bilateral, increasing the complexity in the accurate numerical
simulation of the thermal processing.
The latent heat per time unit, expressed as the internal heat source in the heat conduction
equation, i.e., Equation 1.1, is usually calculated by the following Equation 1.77:
Q ¼ DH
DV
Dt
(1:77)
where
DH is the enthalpy difference when a new phase of unit volume forms from the parent phase
DV is the change of transformation fraction in the time step Dt
The calculation of the volume fraction is the key to predict the evolution of temperature and
microstructure during the thermal processing, as well as the ﬁnal microstructure constituents and
rough mechanical property. Since phase transformations are usually classiﬁed into two categories
according to their mechanisms, diffusion transformation (for example, pearlite transformation),
and nondiffusion transformation, i.e., martensitic transformation, the mathematical models of the
transformation kinetics are very different.
In this section, the numerical method to calculate the transformed volume fraction of a new
phase is mainly introduced.
1.4.2 DIFFUSION PHASE TRANSFORMATION
A time–temperature–transformation (TTT) diagram describes the relationship between the trans
formation starting, ending, and the transformed volume fraction during the isothermal process at
different temperatures. The isothermal kinetics equation, namely Johnson–Mehl equation [8],
provides a solid base for numerical simulation of thermal process although it cannot be directly
applied to calculate the volume fraction due to the nonisothermal process of the practical heating
or cooling process. Until now, the method proposed by Fernandes et al. [9] has been widely
accepted, in which the practical nonisothermal process is considered as many isothermal stages
with a tiny time duration as shown in Figure 1.5, and the effect of these stages can be summed
together according to Scheil’s additivity rule.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 21 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 21
Due to the strict limits of Johnson–Mehl equation, Avrami [10] proposed an empirical equation,
which has a simpler form and has been widely used, as follows:
f ¼ 1 Àexp(Àbt
n
) (1:78)
where
f is the volume fraction of a new phase
t is the isothermal time duration
b is a constant dependent on the temperature, composition of parent phase, and the grain size
n is also a constant dependent on the type of phase transformation and ranges from 1 to 4
Coefﬁcient b and n at different temperatures are usually calculated from the experimentally
obtained TTT diagrams by the following equations:
n
(T)
¼
ln[ ln(1 Àf
1
) Àln(1 Àf
2
)]
ln t
1
Àln t
2
(1:79)
b
(T)
¼
ln(1 Àf
1
)
t
n(T)
1
(1:80)
Therefore, the relationship between the volume fraction and time under certain isothermal temper
atures can be calculated by Equations 1.78 through 1.80. Most experimental data of phase trans
formation agree well with the Avrami equation.
Only with the help of Scheil’s additivity rule can the transformation starting time, namely, the
incubation period, for a nonisothermal cooling or heating process be determined. The general form
of Scheil’s additivity rule can be written as
_
t
s
0
dt
t
TTT
i
¼ 1 or
n
i¼1
Dt
i
t
TTT
i
¼ 1 (1:81)
T
e
m
p
e
r
a
t
u
r
e
Time time
t
2
,
Fictitious
Δt
1
Δt
1
Δt
2
t
1
, Fictitious
V
o
l
u
m
e
f
r
a
c
t
i
o
n
,
f
f
2
f
1
T
2
f
1
f
0
Δt
2
T
1
T
1
T
2
FIGURE 1.5 Schematic model for fraction calculation of diffusion transformation during continuous cooling
process.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 22 3.11.2008 3:17pm Compositor Name: BMani
22 Handbook of Thermal Process Modeling of Steels
where
t
s
is the incubation period to be determined for the nonthermal process
t
TTT
i
is the incubation period under a certain isothermal temperature
Dt
i
is the time step
It can be understood that the transformation in the nonisothermal process occurs when the
summation of the relative fractions Dt
i
t
TTT
i
_
reaches unity.
Scheil’s additivity rule has also been applied in volume fraction calculation. For example, to
ﬁnd the time (t
m
) needed to attain a certain fraction transformed (j
m
) in a nonisothermal transform
ation with time–temperature path T
t
, Scheil’s additivity rule is expressed as
_
t
m
0
dt
t
j
m
(T)
¼ 1 (1:82)
where t
j
m
(T) is the time to transform the fraction j
m
isothermally at the current isotherm T
t
.
According to Equation 1.82, the total time required for completion of the transformation is obtained
by adding the absolute durations of time Dt spent at each temperature T, until the sum of the relative
durations Dt=t
j
m
(T) becomes unity. The time t
j
m
(T), to reach transformed fraction j
m
, can be obtained
from the kinetics equation, i.e., the Avrami equation (Equation 1.78).
To explicitly demonstrate its calculation method, the concept of ﬁctitious time t* and
ﬁctitious transformed volume fraction f
i
*
are commonly developed [9]. These two variables can
be calculated by
t
i
*
¼
Àln(Àf
iÀ1
)
b
i
_ _
1=n
i
(1:83)
f
i
*
¼ 1 Àexp b
i
t
i
*
þDt ð Þ
n
i
½ (1:84)
Hence, the practical transformed fraction is
f ¼ f
i
*
f
g
iÀ1
þf
iÀ1
_ _
f
max
(1:85)
where
f
g
iÀ1
and f
i À1
are the austenite fraction and transformed fraction at the end of the previous time
step
f
max
is the maximum possible transformed fraction for this type of transformation
Equations 1.78 through 1.85 constitute the basic frame of volume fraction calculation for the
diffusion transformation, whereas it should be carefully applied in speciﬁc case studies and
necessary modiﬁcations be made.
1.4.2.1 Modiﬁcation of Additivity Rule for Incubation Period
Based on theoretical analysis and experimental results, Hsu [11] points out that the additivity rule
is not always accurate enough to be applicable in incubation period prediction. Hawbolt et al. [12]
and Reti and Felde [13] held the same viewpoint that the additivity rule sometimes seriously
overestimates the incubation time.
Taking the eutectoid steel as the example, the practical incubation period t
CCT
s
under certain
cooling rate, as well as the incubation periods for different isothermal stages t
TTT
s
, can be obtained
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 23 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 23
from CCT and TTT diagrams, respectively. Hence, the relative durations of each tiny isothermal
stage Dt t
TTT
s
_
can be summed stage by stage from zero to t
CCT
s
by
x ¼
t
CCT
s
0
Dt
t
TTT
s
(1:86)
Beyond expectation, the summation x, listed in Table 1.1, is much less than unity.
Another approach has been proposed to modify the additivity rule for prediction of incubation
period using TTT and CCT diagrams simultaneously [14].
The starting curves of TTT and CCT diagram are plotted together as Figure 1.6. The average
cooling rate during arbitrary time step Dt
i
on the practical cooling curve, that is, the slope of the
tangent line dd drawn at the middle point of time step Dt
i
, can be calculated by
V
i
¼
T
iÀ1
ÀT
i
Dt
i
(1:87)
TABLE 1.1
Summation of Relative Durations in Nonisothermal
Process of Eutectoid Steel
Average Cooling
Rate, 8C=s
Summation of
Relative Durations, x Source
7.5 0.2 [12]
2.0 0.23
38.5 0.24
5.3 0.43 [14]
21.2 0.31
47.6 0.28
Time
0
Practical cooling curve
d
dЈ
dЈ
d
V
i
A
cl
Starting curve of CCT
Starting curve of TTT
T
e
m
p
e
r
a
t
u
r
e
T
i −1
t
s
T
i
Δt
i
t
CCT
s(i)
t
TTT
s(i)
t
CCT
s
TTT
FIGURE 1.6 Schematic sketch of the modiﬁcation of additivity rule for incubation period.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 24 3.11.2008 3:17pm Compositor Name: BMani
24 Handbook of Thermal Process Modeling of Steels
Cooling from the critical point A
c1
with the constant rate of V
i
, the cross point with the starting curve
t
CCT
s
in the CCT diagram gives the incubation period t
CCT
s(i)
, during which the modiﬁcation coefﬁcient
w
i
of relative durations in this time step for the TTT diagram can be calculated by
w
i
¼
_
t
CCT
s(i)
0
dt
t
TTT
s
(1:88)
The modiﬁcation coefﬁcient w
i
reﬂects the difference in incubation period between the CCT and
TTT diagram. When w
i
equals unity, there is no difference. Usually, it is bigger than unity since it is
very common that the starting curve in CCT lags behind that in the TTT diagram. Therefore, the
relative time duration Dt
i
=t
TTT
s(i)
for the time step Dt
i
can be modiﬁed as (Dt
i
=t
TTT
s(i)
)(1=f
i
).
Hence, for the transformation starting time, namely, the incubation period for the isothermal
process, the criterion is modiﬁed as follows:
Dt
i
t
TTT
s(i)
Â
1
w
i
¼ 1 (1:89)
The theoretical derivation of this modiﬁcation is still in the process of consummation, and the
application and the accuracy also need to be further validated.
1.4.2.2 Modiﬁcation of Avrami Equation
It is very clear that the time in the Avrami Equation 1.78 counts since the transformation occurs, that
is, the incubation period is not included. Therefore, Hawbolt et al. [12] suggested that the modiﬁed
form of Avrami equation should be
f ¼ 1 Àexp Àb(t Àt
s
)
n
½ (1:90)
The calculation of coefﬁcients n and b is also changed as
n(T) ¼
ln ln(1 Àf
1
) Àln(1 Àf
2
) ½
ln(t
1
Àt
s
) Àln(t
2
Àt
s
)
(1:91)
b(T) ¼
ln(1 Àf
1
)
(t
1
Àt
s
)
n(T)
(1:92)
where t
s
is the incubation period of the isothermal process at a certain temperature.
To demonstrate the difference between these two methods, again, the eutectoid steel (T8 steel) is
taken as an example. In the algorithm 1 based on the Avrami equation, the transformation starting
time t
s
(approximate as t
0.01
) and ending time t
e
(approximate as t
0.99
) read from the TTT diagram
are input to Equations 1.78 through 1.80 to calculate the coefﬁcients n, b, and t
0.25
(the time needed
to get a fraction of 0.25 of the new phase). In the algorithm 2 based on the modiﬁed Avrami
equation, t
0.5
and t
0.75
are the time to transform the new phase of the volume fraction of 0.5 and
0.75, respectively, both of which include the incubation period t
s
and can be got from the TTT
diagram. With the help of Equations 1.90 through 1.92, the coefﬁcients n, b, and t
0.25
are also
calculated. All these data are listed in Table 1.2. It is obvious that the modiﬁed Avrami equation
seems more reasonable and ﬁts better with the TTT diagram than the unmodiﬁed one.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 25 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 25
1.4.2.3 Calculation of Proeutectoid Ferrite and Pearlite Fraction
The transformations in the hypoeutectoid steel can be explained by Figure 1.7. When the steel is
cooled rapidly until the temperature is below line SE
0
, the stable ferrite a and cementite Fe
3
C
precipitate simultaneously and form the quasieutectoid. However, if the temperature is above
point a, for example, at point a
0
, the nucleus of ferrite appears ﬁrst at the austenite grain boundary,
and the ferrite grows up as the time goes on. At the same time, carbon diffuses into the neighboring
austenite grain due to its low dissolvability in ferrite, increasing the carbon concentration in
austenite. When the concentration reaches point b located on the boundary of twophase area
aþFe
3
C, the retained austenite starts to decompose into pearlite.
The mechanism of the transformation process has been investigated intensively and has almost
reached common understanding, while the fraction calculation of proeutectoid ferrite and pearlite is
an unsettled dispute.
The nucleation sites and the mode of the proeutectoid ferrite are theoretically different from
those of eutectoid pearlite. Although their growth is controlled by diffusion, the growth mode and
diffusion route of carbon are different. These factors support the viewpoint that the fraction of
proeutectoid ferrite and pearlite should be calculated separately, that is, two sets of independent
TABLE 1.2
Coefﬁcients n, b, and t
0.25
Calculated by Different Methods
Temperature
(8C)
Data from TTT Diagram of
Eutectoid Steel (T8 Steel)
Algorithm 1: Avrami
Equation
Algorithm 2: Modiﬁed
Avrami Equation
t
s
(s) t
0.25
(s) t
0.5
(s) t
0.75
(s) t
e
(s) n(T) b(T) Â10
À5
t
0.25
(s) n(T) b(T) Â10
À5
t
0.25
(s)
700 12.5 39 63 90 120 2.71 1.07 43.1 1.62 121 41.9
650 2.7 6.5 8.8 11.5 18.5 3.18 42.5 7.7 1.89 2271 6.5
450 1.4 5.5 8.0 11 20 2.13 463 6 2.05 1460 5.7
400 3 30 4.3 65 100 1.75 147 20.5 1.58 200 26
G
P
α
α + γ
γ
α + Fe
3
C
γ + Fe
3
C
C
S
PЈ
EЈ
GЈ
E
aЈ bЈ
a
C (%)
T
(
Њ
C
)
T
i
b
FIGURE 1.7 Schematic sketch of the quasiequilibrium diagram of FeC.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 26 3.11.2008 3:17pm Compositor Name: BMani
26 Handbook of Thermal Process Modeling of Steels
Avrami equations should be used. Therefore, Wang et al. [15] proposed that the volume fraction of
proeutectoid ferrite and pearlite in steel 1025 be calculated by two Avrami equations as
f
F
¼ 1 Àexp Àb
F
(t Àt
SF
)
n
F
½ (1:93)
f
P
¼ 1 Àexp Àb
P
(t Àt
SP
)
n
P
½ (1:94)
On the other hand, the proeutectoid ferrite precipitates along the austenite grain boundary, and there
is enough space for growth at the initial stage. However, with the growth of the new phase, the
volume fraction increases, and the transformation rate decreases because of the encounter of the new
phase. Usually, the pearlite transformation occurs when the proeutectoid ferrite precipitation is still
in process. Hence, the proposal [12,15] that the proeutectoid ferrite precipitation should be
described by one dependent Avrami equation seems to lack convincing evidence.
In addition, the starting temperature of pearlite transformation during continuous cooling is hard
to determine because the carbon content of austenite varies with the precipitation of proeutectoid
ferrite. The pearlite transformed is usually the quasieutectoid in which the carbon content decreases
with the decrease in temperature. These are not in agreement with the hypotheses of the additivity rule.
Based on the above analysis, Pan et al. [16] developed a set of new approaches that uses one
combined Avrami equation to calculate the total volume fraction of proeutectoid ferrite and pearlite.
Assuming that the steel is isothermally kept for enough long time to precipitate the proeutectoid
ferrite only and the pearlite transformation does not occur, the ﬁnal volume fraction of ferrite f
F
end
under certain temperature T
i
can be calculated by lever rule as
f
F
end
¼
a
0
b
0
cb
0
(1:95)
Therefore, the deﬁnition domain of the Avrami equation describing the proeutectoid ferrite is
[0, f
F
end
]. However, the practical situation is that the pearlite transformation usually occurs before
the volume fraction of proeutectoid ferrite reaches f
F
end
. Thus, there exists a maximum volume
fraction of proeutectoid ferrite f
F
max
at a certain temperature, which can be calculated as
f
F
max
¼
ba
0
cb
(1:96)
When the volume fraction of proeutectoid ferrite reaches f
F
max
, the carbon concentration in the
retained austenite reaches the composition range of quasieutectoid. The decomposition of austenite
enters the stage of pearlite transformation. The accurate kinetics of austenite decomposition, i.e., the
solid line in Figure 1.8, can be split into two parts: the ﬁrst part is the transformation of proeutectoid
ferrite from [0, f
F
max
], and the second part is that of pearlite from [f
F
max
,1]. The whole curve is still
S type and can be united into one Avrami equation.
The total volume fraction of proeutectoid ferrite f
F
i
and pearlite f
P
i
can be calculated referring to
previous sections, and their separation is performed:
If f
i
< f
F
max
, then
f
F
i
¼ f
i
f
P
i
¼ 0
_
(1:97)
If f
i
! f
F
max
, then
f
F
i
¼ f
F
max
f
P
i
¼ f
i
Àf
F
max
_
(1:98)
In order to test this approach, the TTT and CCT diagram of 45# steel was used to calculate the volume
fraction of proeutectoid ferrite f
F
i
and pearlite f
P
i
, which were then input into the additivity rule as
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 27 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 27
x
i
¼
t
CCT
e
t
CCT
s
Dt
i
t
TTT
ei
Àt
TTT
si
(1:99)
The results listed in Table 1.3 indicate that the summation of relative durations almost reaches unity
under different cooling rates, demonstrating the feasibility of the approach discussed here.
Related experimental and simulation results [17] also demonstrated that the whole decompos
ition process of austenite in hypoeutectoid steel can be reasonably described by one Avrami
equation.
1.4.3 MARTENSITIC TRANSFORMATION
In most cases, the transformed volume fraction of martensite is independent of the cooling rate, but a
function of temperature. The Koistinen–Marburger [18] equation is always adopted in thermal
process simulation, and it is
f
M
¼ 1 Àexp Àa(M
s
ÀT) ½ (1:100)
Ferrite
Pearlite
Ferrite
V
o
l
u
m
e
f
r
a
c
t
i
o
n
t
sF
t
sP
Time
Pearlite
1
0
f
max
F
f
end
F
FIGURE 1.8 Schematic sketch of isothermal transformation in hypoeutectoid steel.
TABLE 1.3
Results of Check Computation of the Additivity Rule in 45# Steel
Average Cooling Rate (8C=s) t
CCT
s
(s) t
CCT
s
(8C) t
TTT
e
(s) x
i
0.6 113.3 717 288.3 0.98
8.1 9.5 689 635 0.97
25 2.7 670 625 0.94
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 28 3.11.2008 3:17pm Compositor Name: BMani
28 Handbook of Thermal Process Modeling of Steels
where
f
M
is the transformed volume fraction of martensite
T is the temperature
M
s
is the start temperature of martensitic transformation
a is the constant reﬂecting the transformation rate and varying with the steel composition
For carbon steel with the carbon content lower than 1.1%, we have a¼0.011.
Assuming that the average volume of martensite lamella is constant, Magee [19] theoretically
derived the constant a as
a ¼ Vw
@DG
g!M
v
@T
(1:101)
where
V is the average volume of martensite lamella
w is the ratio of the martensite number newly formed to the driving force in the austenite of
unit volume
DG
g!M
v
is the free energy difference between austenite and martensite
Hsu [20] inferred that the Koistinen–Marburger equation is only applicable in high and medium
carbon content steels in which the carbon does not diffuse when martensite transform occurs. While
in the low carbon steel and the medium carbon steel with strip austenite the alloy element affects the
carbon diffusion. Therefore, the Koistinen–Marburger equation should be modiﬁed for the low
carbon steel, and it becomes
f ¼ 1 Àexp b(C
1
ÀC
0
) Àa(M
s
ÀT
q
)
_ ¸
(1:102)
where
C
0
is the carbon concentration in austenite
C
1
is the carbon concentration in martensite
a and b are constants dependent on the materials; the former can be obtained according to Equation
1.101, while the latter can be calculated as follows:
b ¼ Vw
@DG
g!M
v
@C
(1:103)
where C is the carbon concentration.
Generally, the volume of martensite in the carbon steel and alloy steel is calculated using
Equations 1.100 and 1.102, and the M
s
point can be obtained by experiment and the following
empirical equations:
M
s
(
C) ¼ 520 À(C%) Â320 (1:104)
M
s
(
C) ¼ 512 À453C À16:9Ni þ15Cr À9:5Mo þ217(C)
2
À71:5(C)(Mn) À67:6(C)(Cr) (1:105)
The M
s
point is considered as a constant in Koistinen–Marburger equation, while it varies if other
transformations, for example, pearlite transformation, bainite transformation, and so on, occur
heretofore, changing the composition of the parent phase or consuming the embryos. It can be
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 29 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 29
observed from the CCT curves that the more the austenite decomposes, the more the M
s
point
decreases. The phenomenon has not been generally reﬂected in the additivity rule in calculation of
volume fractions using TTT curves. Hence, the calculation model, i.e., the Koistinen–Marburger
equation, for martensite is in great need of further amendment based on transformation mechanisms
to improve its universality and accuracy.
1.4.4 EFFECT OF STRESS STATE ON PHASE TRANSFORMATION KINETICS
The stress state has considerable effects on the phase transformation kinetics irrespective of the
diffusion transformation or that of nondiffusion. Numerous investigations [21–25] have been carried
out to clarify the mechanism and to establish effective models for accurate simulation. Up to now,
the research in this ﬁeld is still far from satisfactory. Some mathematical models proposed by
different groups are only applicable for some speciﬁc steels, and some need to be further validated.
In this section, representative models are brieﬂy introduced.
1.4.4.1 Diffusion Transformation
For the pearlite transformation, Inoue and Wang [26] developed the model of transformation
kinetics under stress. In his work, the Johnson–Mehl equation in the stressfree state is expressed as
f ¼ 1 Àexp À
_
t
0
F(T) t Àt ð Þ
3
dt
_
_
_
_
(1:106)
The Johnson–Mehl equation under stress state was modiﬁed as
f ¼ 1 Àexp À
_
t
0
exp (cs
m
)F(T) t Àt ð Þ
3
dt
_
_
_
_
(1:107)
where
F(T) is the temperature function
s
m
is the mean stress
c is a constant
Denis et al. [27] introduced the effect of stress into the coefﬁcients (n and b) in the Avrami
equation (Equation 1.78) as follows:
n
s
¼ n (1:108)
b
s
¼
b
(1 ÀCs
e
)
n
(1:109)
where
s
e
is the equivalent stress
C is a constant
1.4.4.2 Martensitic Transformation
For martensitic transformation, the stress state in the steel has much greater effects on the trans
formation process. The coefﬁcient a in the Koistinen–Marburger equation (Equation 1.100), as well
as the M
s
point, is strongly dependent on the stress state.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 30 3.11.2008 3:17pm Compositor Name: BMani
30 Handbook of Thermal Process Modeling of Steels
Observing Equation 1.101, the coefﬁcient a can be obtained from the driving force in which the
stress plays an important role. Thus, Hsu proposed an empirical equation as follows, indicating the
relationship between a and equivalent stress s
i
.
a ¼ a
0
þa
1
s
i
(1:110)
where the coefﬁcients are about a
0
¼1.2430 Â10
À2
and a
1
¼6.9752.
Inoue and Wang [26] expressed the Koistinen–Marburger equation without stress as
f ¼ 1 Àexp f(T) ð Þ (1:111)
while the kinetics equation under stress was modiﬁed as
f ¼ 1 Àexp As
m
þBJ
1
2
2
_ _
þf(T)
_ _
(1:112)
where
s
m
is the mean stress, i.e., hydrostatic pressure
J
2
is the second stress invariant
coefﬁcients A and B are the material constants
Similarly, Denis et al. [27] developed the model of martensitic transformation under stress, and
her equation for volume fraction of martensite is
f ¼ 1 Àexp Àa(M
s
þAs
m
þBs
i
ÀT) þ As
m
þBs
1=2
i
_ _ _ _
(1:113)
Compared to Equation 1.112, this equation additionally considers the effect of stress on the M
s
point, which can be clearly expressed as an extra item DM
s
[28]. Therefore, we have
M
s
s
¼ M
s
þDM
s
(1:114)
DM
s
¼ As
m
þBs
i
(1:115)
1.5 CONSTITUTIVE EQUATION OF SOLIDS
1.5.1 ELASTIC CONSTITUTIVE EQUATION
1.5.1.1 Linear Elastic Constitutive Equation
In a solid under deformation, the stress depends on the current strain state, and is a single value
function of the strain. The elastic deformation will be recovered, when the load is released, i.e., the
elastic deformation is reversible. Usually elastic constitutive relation is taken as linear.
1.5.1.1.1 Isotropic Elasticity
In the isothermal, small deformation process of solids, the stress components are a linear function of
the components of the small strain tensor, as follows:
s
ij
¼ C
e
ijkl
«
e
kl
(1:116)
where C
e
ijkl
is the elastic tensor, the superscript ‘‘e’’ means elastic
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 31 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 31
Such a material model is called linear elastic constitutive equation. The incremental form of the
above equation is
ds
ij
¼ C
e
ijkl
d«
e
kl
(1:117)
Usually metals can be taken as isotropic, linear elastic materials or ideal elastic materials. For such
materials, the elastic tensor is a fourthorder isotropic tensor, including only two independent
parameters, as follows:
C
e
ijkl
¼ 2G d
ik
d
jl
þ
v
1 À2v
d
ij
d
kl
_ _
(1:118)
where
G is the shear modulus
v is the Poisson’s ratio
G is related with Young’s modulus E by the following equation:
G ¼
E
2(1 þv)
In ﬁnite deformation, to meet the requirement of objectivity, the Jaumann rate of Cauchy stress
tensor and the rate of deformation tensor, which possess objectivity, are adopted, and the elastic
constitutive equation is rewritten as
^ s
ij
¼ C
e
ijkl
d
e
kl
(1:119)
For the convenience of ﬁniteelement formulation, the stress and strain are usually rewritten from
secondorder tensors into vectors, e.g.,
s ¼ [s
11
s
22
s
33
s
12
s
23
s
31
]
T
« ¼ [«
11
«
22
«
33
2«
12
2«
23
2«
31
]
T
Accordingly, the fourthorder tensor C
e
is rewritten into a matrix:
C
e
¼
2G
1 À2n
1 Àn n n 0 0 0
n 1 Àn n 0 0 0
n n 1 Àn 0 0 0
0 0 0 (1 À2n)=2 0 0
0 0 0 0 (1 À2n)=2 0
0 0 0 0 0 (1 À2n)=2
_
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
_
(1:120)
Then the elastic constitutive equation can be written in the matrix form:
s ¼ C
e
«
In this chapter, the symbol s is used to represent the stress both in tensorial form and in vectorial
form. Similar usage is also applied for « and C
e
. The exact meaning of the symbol can be deduced
from the context.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 32 3.11.2008 3:17pm Compositor Name: BMani
32 Handbook of Thermal Process Modeling of Steels
Elastic deformation can be decomposed into a volumetric component,
«
e
m
¼ «
e
11
þ«
e
22
þ«
e
33
_ _
=3
«
e
0
ij
¼ «
e
ij
À«
e
m
d
ij
«
e
0
ij
¼
1
2G
s
0
ij
«
e
m
¼
1 À2n
E
s
m
1.5.1.1.2 Orthotropic Elasticity
Some materials, such as a single crystal, show different elastic behavior along different directions. If
the principal axes are mutually orthogonal, it is called orthotropic elasticity. If along each principal
axis, the elastic properties are identical, then the elastic matrix is
C
e
¼
C
11
C
12
C
12
0 0 0
C
12
C
11
C
12
0 0 0
C
12
C
12
C
11
0 0 0
0 0 0 C
44
0 0
0 0 0 0 C
44
0
0 0 0 0 0 C
44
_
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:121)
1.5.1.2 Hyperelastic Constitutive Equation
Some materials such as rubbers, have highly nonlinear stress–train relationships; they remain
elastic even when they go through a huge deformation. Such materials are described by hyperelastic
constitutive equation.
If a material possesses the strain energy density function, w, which is an analytic function of the
strain tensor, and the increment of which is the work done by the stress, then it is called hyperelastic
material. For hyperelastic material
w ¼
1
r
0
S
ij
_
E
ij
(1:122)
where
r
0
is the mass density in initial conﬁguration
S
ij
is the second Piola–Kirchhoff stress tensor
_
E
ij
is the Green strain tensor
On the other hand, the strain energy function is the analytic function of strain tensor; then
_ w ¼
@w
@E
ij
_
E
ij
(1:123)
1
r
0
S
ij
À
@w
@E
ij
_ _
_
E
ij
¼ 0 (1:124)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 33 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 33
If the components of
_
E
ij
are mutually independent, from the arbitrariness of
_
E
ij
, we have
S
ij
¼ r
0
@w
@E
ij
(1:125)
Equation 1.125 is the hyperelastic constitutive equation.
If the volume change in deformation process is negligible, then the components of
_
E
ij
are not
independent, but should meet the incompressible condition:
«
ii
¼
@X
m
@x
i
@X
n
@x
i
_
E
mn
¼ 0 (1:126)
Comparing Equations 1.124 and 1.126, it is known that
1
r
0
S
ij
À
@w
@E
ij
¼
@X
i
@x
m
@X
j
@x
m
That is,
S
ij
¼
1
r
0
@w
@E
ij
þh
@X
m
@x
i
@X
n
@x
i
(1:127)
where h is the hydrostatic pressure, determined by the boundary conditions. Equation 1.127 is the
constitutive equation of incompressible hyperelastic materials.
For isotropic hyperelastic materials, the deformation energy density function could be expressed
as a function of the invariables of Green deformation tensor:
w ¼ w(I
1
, I
2
, I
3
)
where
I
1
¼ C
ii
¼ 2E
ii
þ3
I
2
¼ C
ii
C
jj
ÀC
ij
C
ji
_ _
=2 ¼ 2E
ii
E
jj
þ4E
ii
À2E
ij
E
ij
þ3
I
3
¼ det C ¼ det
@x
m
@X
i
@x
m
@X
j
_ _
_
¸
¸
¸
_
¸
¸
¸
_
(1:128)
So
S
ij
¼ r
0
@w
@I
1
@I
1
@E
ij
þ
@w
@I
2
@I
2
@E
ij
þ
@w
@I
3
@I
3
@E
ij
_ _
(1:129)
where
@I
1
@E
ij
¼ 2d
ij
@I
2
@E
ij
¼ 2(d
ij
d
rs
Àd
ir
d
js
)(2E
rs
þd
rs
)
@I
3
@E
ij
¼ 2
@X
i
@x
m
@X
j
@x
m
I
3
_
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
_
(1:130)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 34 3.11.2008 3:17pm Compositor Name: BMani
34 Handbook of Thermal Process Modeling of Steels
For an incompressible hyperelastic material, I
3
need not be considered. Then,
S
ij
¼ r
0
@w
@I
1
@I
1
@E
ij
þ
@w
@I
2
@I
2
@E
ij
_ _
þh
@X
i
@x
m
@X
j
@x
m
(1:131)
By deﬁning different deformation energy density functions, different hyperelastic material models
could by deﬁned. Some commonly used models are given below:
1. Neo Hooke model [3,29]
w ¼ C
1
(I
1
À3) (1:132)
2. Moony model [30–32]
W ¼ C
1
(I
1
À3) þC
2
(I
2
À3) (1:133)
where C
1
and C
2
are the material constants.
The deformation energy density functions of the above models are deﬁned by the
invariables of Green deformation tensor. Because C
ii
¼ l
2
i
, l
i
is the principal stretch; the
deformation energy density functions of the models are functions of even power of l
i
, so
they belong to the Rivilin materials. For the Rivilin materials, the deformation energy
density functions can be chosen as functions of higherorder power of I
1
and I
2
. With an
increase in the order, or the number of the items, the agreement of the constitutive equation
with the behavior of real materials will be improved, but the parameters to be determined
increase, and the computation becomes more complicated. It is found that I
3
is not included
in the deformation energy density functions of the above models, so they are incompress
ible hyperelastic materials.
3. Ogden model [33,34]
Volume change in deformation is taken into account in this model. Its deformation energy
density function is
w ¼
N
n¼1
m
n
a
n
J
Àan
3
l
a
n
1
þl
a
n
2
þl
a
n
3
_ _
À3
_ ¸
þ4:5K J
1
3
À1
_ _
2
(1:134)
where
m
n
and a
n
are the material constants
K is the voluminal modulus
J is the ratio of volume change expressed as J ¼ l
1
l
2
l
3
Equation 1.134 should be applied in case the ratio of volume change is in the order of 0.01. If
volume change is large, the following generalized compressible Ogden model could be used:
w ¼
N
n¼1
m
n
a
n
l
a
n
1
þl
a
n
2
þl
a
n
3
À3
_ ¸
þ
N
n¼1
m
n
b
n
(1 ÀJ
b
n
) (1:135)
where m
n
, a
n
, and b
n
are the material constants. Usually N¼2 or N¼3 is appropriate.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 35 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 35
1.5.2 ELASTOPLASTIC CONSTITUTIVE EQUATION
1.5.2.1 Introduction
The plastic behavior of the metal at normal temperature depends on deformation history of the
material, hence the plastic ﬂow theory, that establishes the relationship between stress rate and strain
rate, should be normally used.
In establishing the constitutive equation, ﬁrst a small deformation problem is considered, and
the relationship between Cauchy stress and small strain is established. Then the constitutive
equation is extended to ﬁnitedeformation problem; here Cauchy stress and small strain should be
substituted by corresponding appropriate stress rate and strain rate. The appropriate stress rate could
be chosen as Jaumann rate of Cauchy stress, ^ s, and the objective strain rate could be chosen as the
rate of deformation tensor, d.
The elastoplastic deformation is decomposed into a recoverable elastic deformation and an un
recoverable plastic deformation. In small deformation, the strain increment d« is decomposed as
follows:
d« ¼ d«
e
þad«
p
Correspondingly, in ﬁnite deformation, the strain rate, d, is decomposed as
d ¼ d
e
þad
p
where
superscripts ‘‘e’’ and ‘‘p’’ mean the elastic and plastic components, respectively
a is the loading factor
In pure elastic deformation, including unloading, a¼0; in elastoplastic deformation, a¼1.
1.5.2.2 Yield Criterion
When a metal sample is deformed under uniaxial loading, if the deformation is small, the material is
in elastic state, and the relation between stress and strain is linear. When the stress reaches the yield
point, s
s
, the material yields and deforms plastically, and the relation between stress and strain
becomes nonlinear.
In a deforming body, the stress distribution is generally not uniform. For a particle in the
material, when the stress components meet a certain condition, the particle will yield. This condition
is called yield criterion. The yield criterion can be written in the following general mathematical
expression:
F ¼ f (s
ij
) Às
s
¼ 0 (1:136)
where
f(s
ij
) is the function of the stress components
s
s
is the initial yield stress
The yield criterion may be geometrically represented by a surface in the stress space, called
yield surface. If the stress, s
ij
, locates inside the surface, i.e., F (s
ij
) <0, then the particle is in elastic
state. If the stress, s
ij
, locates on the surface, i.e., F(s
ij
) ¼0, then the particle yields and turns into
the plastic state. But it is impossible for the stress, s
ij
, to get out the surface, i.e., F(s
ij
) >0.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 36 3.11.2008 3:17pm Compositor Name: BMani
36 Handbook of Thermal Process Modeling of Steels
1.5.2.3 Flow Rule
1.5.2.3.1 Associated Flow Rule
For ordinary engineering materials, the ﬂow stress is an increasing function of the plastic strain
in plastic deformation; such material is called a stable material, which meets the condition
Ds
ij
D«
ij
>0 during plastic deformation. Consequently, successive yield surface, or loading surface,
F(s
ij
, Y) is convex. Let the stress space coincide with the strain space; then the plastic strain
increment must point in the outward normal direction of the yield surface F(s
ij
, Y), i.e.,
_ «
ij
¼
_
l
@F
@s
ij
(1:137)
where
@F=@s
ij
, the geometric meaning of the partial derivative, is the component of the unit outward
normal vector of the yield surface F, in the stress space
_
l is the scalar factor
Adopting the concept of plastic potential theory, F is taken as the plastic potential function. When
the yield criterion is taken as the plastic potential function, associated ﬂow rule can be obtained.
Using a different yield function, different plastic stress–strain relationships can be derived from
associated ﬂow rule. If the plastic potential function is not chosen as the yield criterion, the resulting
plastic stress–strain relationship is called nonassociated ﬂow rule. For metals, associated ﬂow rule is
normally used.
1.5.2.3.2 Loading–Unloading Condition
Work hardening usually occurs in plastic deformation of metals at normal temperature, which
causes an increase in successive yield stress or ﬂow stress. Successive yield criterion can normally
be written as
F ¼ f (s) ÀY(«
P
) ¼ 0
where Y is the ﬂow stress, and is usually taken as a function of the accumulated plastic strain, «
P
. In
small elastoplastic deformation, the loading–unloading condition of work hardening material is as
follows.
Loading:
F ¼ 0, dF ¼ 0 and
@F
@s
ij
ds
ij
> 0
Unloading:
F ¼ 0, dF < 0 (1:138)
Neutral variation:
F ¼ 0, dF ¼ 0 and
@F
@s
ij
ds
ij
¼ 0
where, the geometric meaning of (@F=@s
ij
)ds
ij
is the projection of stress increment, ds, in the
outward unit normal vector of the yield surface, @F=@s. In loading state, the value of the projection
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 37 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 37
is positive. During loading, F¼0 and dF¼0 must be satisﬁed; they are called consistency
condition. From the consistency condition, we obtain
@F
@s
ij
ds
ij
ÀdY ¼ 0
Since s ¼Y and ds¼dY during loading, the above equation can be written as
@F
@s
ij
ds
ij
Àds ¼ 0
1.5.2.4 Hardening Law
1.5.2.4.1 Two Kinds of Hardening Assumptions
For hardened materials, it is assumed that the successive yield behavior still obeys the initial yield
criterion, but the initial yield stress must be replaced by ﬂow stress.
The change in successive yield surface, and yield locus, which is the intersection of the yield
surface with a plane in the stress space, is very complicated. For simplicity, two kinds of assump
tions are commonly adopted. One of them is the isotropic hardening assumption. Its key points are
(1) the material remains isotropic after hardening; (2) the center and the shape of the yield locus do
not change after hardening, but its size grows continuously during deformation. The other is the
kinematic hardening assumption. Its key point is that the size and shape of the yield locus is ﬁxed
after hardening, and the yield locus only moves rigidly in the stress space. Sometimes both
assumptions are combined in application.
1.5.2.4.2 Equivalent Stress–Equivalent Strain Relation
Flow stress is the function of accumulated plastic strain or plastic work; their relationship can be
determined by the single curve assumption. It is assumed that in the equivalent stress–equivalent
strain relationship in general the stress state is identical with the true stress–true strain relationship in
uniaxial stress state.
The commonly used formulations describing the relationship between equivalent stress and
equivalent strain are given below. These formulations only describe the equivalent stress and
equivalent strain relationship in plastic state; the consistency between elastic and plastic stress–
strain relationships can be realized by the compatibility of initial yield stress s
s
0
.
1. Ideal plasticity
s ¼ s
s
where, work hardening is neglected.
2. Linear hardening
s ¼ s
s
þK«
where K is the tangent modulus.
3. Power law
s ¼ K«
n
where
K is the strength factor
n the hardening exponent.
4. Power law with initial strain
s ¼ K(«
0
þ«)
n
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 38 3.11.2008 3:17pm Compositor Name: BMani
38 Handbook of Thermal Process Modeling of Steels
1.5.2.4.3 Bauschinger Effect
For metals, if loading direction is reversed after plastic deformation, the yield stress in reverse
loading direction is lower than the initial yield stress, as shown in Figure 1.9. This phenomenon is
called Bauschinger effect and can be described by the kinematic hardening assumption.
1.5.2.5 Commonly Used Plastic Constitutive Equations
1.5.2.5.1 J
2
Flow Rule
Supposing the material is plastically isotropic, as the yield criterion is an appropriate physical law, it
can be expressed as the function of stress invariants. Since hydrostatic pressure does not affect
yielding, the yield criterion is related only with the deviatoric stress. Consequently, the yield
condition can be written as
F(J
0
2
, J
0
3
) ¼ 0
where J
0
2
and J
0
3
are the second and third invariants of the deviatoric stress, s
0
ij
. It should be noted
that J
0
1
¼ 0.
For the isotropic hardening material, the von Mises yield criterion can be written as
F ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
3
2
s
0
ij
s
0
ij
_
ÀY ¼
ﬃﬃﬃﬃﬃﬃﬃ
3J
0
2
_
ÀY ¼ 0 (1:139)
Since the yield criterion is the function of the second invariant of deviatoric stress, J
0
2
, the associated
ﬂow rule derived by it is called the J
2
ﬂow rule.
VonMises yield criterion can be represented geometrically by a cylinder in the principal stress
space; the axis of the cylinder is along the isoclinal direction, and its radius is
ﬃﬃﬃﬃﬃﬃﬃﬃ
2=3
_
Y, as shown in
Figure 1.10. In principal stress space, the plane passing through the origin and normal to the
isoclinal direction is called p plane. The Mises yield locus in p plane is a circle.
Substituting the Mises yield criterion into the associated ﬂow rule, we get
_ «
p
ij
¼
_
ls
0
ij
(1:140)
s
s
s
e
s
s1
O
FIGURE 1.9 True stress–true strain curve.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 39 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 39
If both sides of the above equation are multiplied by themselves and then summed,
_
l is obtained
_
l ¼
3
2
_
«
p
s
where s and
_
« are the equivalent stress and equivalent strain rates, respectively, which are deﬁned as
s
2
¼
3
2
s
0
ij
s
0
ij
_
«
2
¼
2
3
_ «
ij
_ «
ij
Substituting the above equations into Equation 1.140, the SaintVenant ﬂow equation is obtained.
_ «
p
ij
¼
3
2
_
«
p
s
s
0
ij
Its incremental form is
d«
p
ij
¼
3d«
p
2s
s
0
ij
(1:141)
1.5.2.5.2 Kinematic Hardening Rule
Supposing the initial yield criterion is
F ¼
ﬃﬃﬃﬃﬃﬃﬃ
3J
0
2
_
Às
s
0
¼ 0
During plastic deformation, s
s
0
in the above equation remains constant. This model is called the J
2
kinematic hardening model. Supposing the center of the yield surface moves to u
ij
, the successive
yield criterion can be written as
F ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
3
2
(s
0
ij
Àu
0
ij
)(s
0
ij
Àu
0
ij
)
_
Às
s
0
¼ 0 (1:142)
s
1
s
2
s
3
FIGURE 1.10 Mises yielding surface in the principal stress space.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 40 3.11.2008 3:17pm Compositor Name: BMani
40 Handbook of Thermal Process Modeling of Steels
where u
0
ij
is the deviator of u
ij
; u
ij
is called backing stress.
@F
@s
ij
¼
3(s
0
ij
Àu
0
ij
)
2s
s
0
The equivalent stress is deﬁned as
s
02
¼
3
2
(s
0
ij
Àu
0
ij
)(s
0
ij
Àu
0
ij
)
When (s
0
ij
Àu
0
ij
) ^ s
ij
> 0 and s
0
¼ s
s
0
, a¼1; when (s
0
ij
Àu
0
ij
) ^ s
ij
< 0 or s
0
< s
s
0
, a¼0.
The movement of the center of the yield surface, u, can be determined by Ziegler rule. It is
assumed that during plastic deformation, u moves along the direction of vector sÀu; then
^
u
ij
¼ ^ m(s
ij
Àu
ij
)
where
^
u
ij
is the Jaumann rate of u
ij
.
From the consistency condition, we obtain
^ m ¼
3(s
0
ij
Àu
0
ij
)
2s
2
s
0
^ s
ij
It is found that if the kinematic hardening rule with Jaumann rate is used in analyzing the simple
shear deformation, after shear strain exceeds 1, shear stress will oscillate, which is unreasonable. To
prevent such a problem, it is suggested that Jaumann rate be substituted by Green–Naghdi rate in the
constitutive equation, i.e., the rotational speed v be substituted by W
R
¼
_
R Á R
T
.
1.5.2.5.3 Orthotropic Material
Hill proposed the yield criterion for orthotropic materials in 1948:
F(s
22
Às
33
) þG(s
33
Às
11
)
2
þH(s
11
Às
22
)
2
þ2Ls
2
23
þ2Ms
2
31
þ2Ns
2
12
¼ 1 (1:143)
where
F, G, H, L, M, and N are the anisotropic parameters
x
1
, x
2
, and x
3
are the material anisotropic principal axes
Let s denote the equivalent stress and Y the ﬂow stress:
s ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
3
2
F(s
22
Às
33
)
2
þG(s
33
Às
11
)
2
þH(s
11
Às
22
)
2
þ2Ls
2
23
þ2Ms
2
31
þ2Ns
2
12
F þG þH
¸
Y ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
3
2(F þG þH)
_
Then Equation 1.143 can be rewritten in the following form:
F ¼ s ÀY ¼ 0
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 41 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 41
According to the associated ﬂow law,
d«
p
ij
¼ dl
@F
@s
ij
¼ dl
Y
2
s
A
ij
where
A
11
¼ G(s
11
Às
33
) þH(s
11
Às
22
)
A
22
¼ F(s
22
Às
33
) þH(s
22
Às
11
)
A
33
¼ G(s
33
Às
11
) þF(s
33
Às
22
)
A
12
¼ A
21
þNs
12
A
23
¼ A
32
¼ Ls
23
A
31
¼ A
13
¼ Ms
31
dl ¼ d«
p
¼
YC
e
mnpq
A
mn
d«
pq
E
t
þY
2
C
e
mnpq
A
mn
A
pq
For a plane stress problem, e.g., the sheet metal forming problem, the yield criterion is
simpliﬁed as
F ¼
3
2(F þG þH)
(G þH)s
2
11
þ(F þH)s
2
22
À2Hs
11
s
22
þ2Ns
2
12
_ ¸
ÀY
2
¼ 0
The coefﬁcients in the above equation can be determined by the strain ratio in transverse and
thickness directions during simple tension, r ¼«
t
=«
z
. Let r
0
, r
45
, and r
90
denote the strain ratio along
the direction 08, 458, and 908 from the rolling direction, respectively, s
s
1
, s
s
45
, and s
s
2
the yield
stresses in simple tension along and above directions respectively; then
G ¼
1
(1 þr
0
)s
2
s
1
F ¼
1
(1 þr
90
)s
2
s
2
H ¼
r
0
(1 þr
0
)s
2
s
1
¼
r
90
(1 þr
90
)s
2
s
2
N ¼ (G þH)
1
2
þr
45
_ _
Supposing r ¼r
0
¼r
45
¼r
90
, (i.e., the sheet is anisotropic along the thickness direction, but isotropic
in the horizontal plane where x, y directions coincide with the principal stress directions), then, the
yield criterion can be further simpliﬁed as
F ¼
3
2(2 þr)
(1 þr)(s
2
1
þs
2
2
) À2rs
1
s
2
_ ¸
ÀY
2
¼ 0
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 42 3.11.2008 3:17pm Compositor Name: BMani
42 Handbook of Thermal Process Modeling of Steels
In this condition, the following relationship among the anisotropic parameters exists. Let
s
s
1
¼s
s
2
¼s
s
, then
F ¼ G ¼
1
(1 þr)s
2
s
H ¼ rF
N ¼ (1 þ2r)F
L ¼ M ¼
(5 þr)
2
F
Y
2
¼
3(1 þr)s
2
s
2(2 þr)
when r is more than unity, i.e., r >1, this yield criterion cannot describe the abnormal phenomenon
of the ratio of s
b
and s
u
less than unity, i.e., s
b
=s
u
<1, where s
b
means the yield stress in biaxial
tension, and s
u
the yield stress in simple tension. Therefore, Hill, Bassani, et al. proposed
nonquadratic yield criteria. However, the original form of the orthotropic yield criterion is more
convenient and commonly used.
1.5.2.5.4 Compressible Material
Some materials, such as metal powder, have volume change during plastic deformation; besides
deviatoric and hydrostatic stress also affect their yielding. The yield criterion can be written as
F ¼ AJ
0
2
þBJ
2
1
ÀCY
2
¼ 0 (1:144)
where
J
0
2
is the second invariant of the deviatoric stress tensor
J
1
is the ﬁrst invariant of the stress tensor
Y is the ﬂow stress of matrix material
A, B, and C are the function of void volume fraction, f ¼V
cavity
=(V
matrix
þV
cavity
), or the mass
density, r
They should be determined by experiments.
From the associated ﬂow rule, we obtain
_ «
p
ij
¼
3
_
«
2s
As
0
ij
þ2BJ
1
d
ij
_ _
_ «
kk
¼
3
_
«
s
BJ
1
The equivalent stress and equivalent strain are deﬁned respectively as
s
2
¼ 3(AJ
0
2
þBJ
2
1
)
_
«
2
¼
2
3
1
A
_ «
0
ij
_ «
0
ij
þ
1
18B
_ «
kk
ð Þ
2
_ _
If the material compressibility is small enough, then B approaches zero, i.e., B !0. If A !1 and
C !1=3 are also assumed, then the yield criterion, the equivalent stress, and the equivalent strain of
voided material coincide with those deﬁned by Mises criterion. For A and B, the relation Aþ3B¼1
should be satisﬁed.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 43 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 43
1.5.2.6 Elastoplastic Constitutive Equation
The formulation of elastic constitutive equation and plastic ﬂow rule were presented here separately.
They are now combined to get the elastoplastic constitutive equation. First, the elastoplastic
constitutive equation in small deformation is derived. If the yield criterion is written in such a
form, in which the yield stress, Y, is in the ﬁrst power, then in the associated ﬂow rule dl ¼d«
p
; i.e.,
the associated ﬂow rule can be expressed as d«
p
kl
¼ d«
p
(@F=@s
kl
). Substituting this equation into
Equation 1.117, we get
ds
ij
¼ C
e
ijkl
(d«
kl
Àd«
p
kl
) ¼ C
e
ijkl
d«
kl
Àd«
p
@F
@s
kl
_ _
If both sides of the above equation are multiplied by @F=@s
ij
and summed up, we get
@F
@s
ij
ds
ij
¼
@F
@s
ij
C
e
ijkl
d«
kl
Àd«
p
@F
@s
kl
_ _
Applying the consistency condition, (@F=@s
ij
) ds
ij
¼ds, and supposing H¼ds=d«
p
, the above
equation can be written as
Hd«
p
¼
@F
@s
ij
C
e
ijkl
d«
kl
Àd«
p
@F
@s
kl
_ _
d«
p
can be solved from the above equation:
d«
p
¼
@F
@s
ij
C
e
ijkl
d«
kl
H þ
@F
@s
ij
C
e
ijkl
@F
@s
kl
Then, the general form of the elastoplastic constitutive equation referring to the stress increment in
small deformation condition is obtained:
ds
ij
¼ C
e
ijkl
Àa
C
e
ijmn
@F
@s
mn
@F
@s
rs
C
e
rskl
H þ
@F
@s
mn
C
e
mnrs
@F
@s
rs
_ _
d«
kl
¼ C
e
ijkl
ÀaC
p
ijkl
_ _
d«
kl
¼ C
ep
ijkl
d«
kl
(1:145)
where a is the loading factor. For example, if Mises yield criterion is substituted into the above
equation, the elastoplastic constitutive equation of J
2
ﬂow rule in small deformation condition can
be obtained.
ds
ij
¼ C
e
ijkl
À
2Ga
g
s
0
ij
s
0
kl
_ _
d«
kl
¼ C
ep
ijkl
d«
kl
where, G is the shear modulus, g ¼2s
2
[1 þh=(2G)]=3, and h ¼2H=3. Let E denote the slope of the
stress–strain curve of simple tension, assuming that the material is incompressible; then
1
H
¼ d«
p
=ds ¼ (d« Àd«
e
)=ds ¼ 1=E
t
À1=E
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 44 3.11.2008 3:17pm Compositor Name: BMani
44 Handbook of Thermal Process Modeling of Steels
In ﬁnite strain, ds
ij
and d«
p
ij
in the loading–unloading criterion, ﬂow rule and the elastoplastic
constitutive equation should be substituted by ^ s
ij
and d
p
ij
, respectively.
1.5.2.7 Thermal Elastoplastic Constitutive Equation
1.5.2.7.1 Thermal Elastic Constitutive Equation
In the elastic zone, the strain, expressed by a vector, is decomposed as follows:
d« ¼ d«
e
þd«
T
¼ d«
e
þadT (1:146)
where
d«
T
is the thermal strain increment caused by thermal expansion; only the normal strain
components are nonzero
a is the vector of the thermal expansion coefﬁcients, a¼a [1 1 1 0 0 0]
T
From the Hooke’s law
«
e
¼ (C
e
)
À1
s
Since the elastic matrix, C
e
, depends on the temperature, T, differentiation of the above equation
leads to
d«
e
¼
d(C
e
)
À1
dT
sdT þ(C
e
)
À1
ds
Substituting the above equation into Equation 1.146, ds can be solved:
ds ¼ C
e
d« À(a þ
d(C
e
)
À1
dT
s)dT
_ _
If the variation of the elastic properties with changing temperature can be neglected, we obtain
ds ¼ C
e
(d« Àd«
T
)
1.5.2.7.2 Thermal Elastoplastic Constitutive Equation
In the plastic zone, since the ﬂow stress, Y, depends on the temperature, T, the successive Mises
yield criterion can be written as
F ¼ s ÀY
_
d«
p
, T
_ _
¼ 0
The differentiation of the above equation is
@F
@s
ds ¼ Hd«
p
þ
@Y
@T
dT
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 45 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 45
where
H ¼
dY
d«
p
@F
@s
¼
@F
@s
11
@F
@s
22
@F
@s
33
@F
@s
12
@F
@s
23
@F
@s
33
_ _
T
In the plastic zone
d« ¼ d«
e
þd«
p
þd«
T
Substituting Equation 1.146 and the ﬂow rule into the above equation, ds can be solved.
ds ¼ C
e
d« À
@F
@s
d«
p
À a þ
d C
e
ð Þ
À1
dT
s
_ _
dT
_ _
If both sides of the above equation are multiplied by @F=@s, and then summed up, d«
p
can be
solved.
d«
p
¼
@F
@s
_ _
T
C
e
d« À
@F
@s
_ _
T
C
e
a þ
d C
e
ð Þ
À1
dT
s
_ _
dT À
@Y
@T
dT
H þ
@F
@s
_ _
T
C
e @F
@s
Then, the incremental stress–strain relationship in the plastic zone is obtained.
ds ¼ C
ep
d« À a þ
d(C
e
)
À1
dT
s
_ _
dT
_ _
þds
T
% C
ep
(d« Àd«
T
) þds
T
(1:147)
where ds
T
denotes the additional stress arising from the effect of temperature on the plastic
modulus:
ds
T
¼
C
e @F
@s
@Y
@T
dT
H þ
@F
@s
_ _
T
C
e @F
@s
1.5.2.7.3 Phase Transformation
When multiple phases exist in the material, each material parameter, such as the Young’s modulus,
E, the ﬂow stress, Y, etc., should be taken as weighted averaging values:
A ¼
n
i¼1
m
i
A
i
where
A is the parameter value after weighted averaging
n is the number of phases
m
i
is the percentage of phase i at the moment
A
i
is the value of the parameter for phase i
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 46 3.11.2008 3:17pm Compositor Name: BMani
46 Handbook of Thermal Process Modeling of Steels
Phase transformation causes a change in the speciﬁc volume of the material; such an effect is
similar to thermal expansion and may be computed by an equivalent thermal expansion coefﬁcient.
Let b
j
denote the thermal expansion coefﬁcient corresponding to the speciﬁc volume change caused
by transformation of phase j; then
b ¼
n
j¼1
m
j
b
j
where
b is the average thermal expansion coefﬁcient; it is the function of temperature, T
n is the number of phases
m
j
is the increment of phase j in the computation step
Then the strain increment is obtained:
D«
e
tr
¼ b[1 1 1 0 0 0]
T
DT
1.5.2.7.4 Loading–Unloading Condition
Considering the effects of temperature on the ﬂow stress and the equivalent stress–equivalent strain
curve, the loading–unloading condition is modiﬁed as follows:
Loading:
@F
@s
_ _
T
ds þ
@F
@T
dT > 0
Unloading:
@F
@s
_ _
T
ds þ
@F
@T
dT < 0
Neutral variation:
@F
@s
_ _
T
ds þ
@F
@T
dT ¼ 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:148)
1.5.3 VISCOPLASTIC CONSTITUTIVE EQUATION
For materials deforming under impact loading and at high temperature, the yield stress and plastic
modulus increase with an increase in the strain rate, as shown in Figure 1.11. Such behavior is called
viscosity. The permanent deformation of viscoplastic material depends on time. For viscoplastic
material, the strain rate can be decomposed into the elastic part and viscoplastic part. Here only its
viscoplastic part, is consider denoted by superscript vp. A description of the elastic part can be found
in the previous section. The viscoelastic response of the material is neglected.
1.5.3.1 OneDimensional Viscoplastic Model
The mechanical model of viscoplastic material can be expressed by the combination of an elastic
element, a plastic element, and a viscous element, as shown in Figure 1.12. The superscripts ‘‘e,’’
‘‘p,’’ and ‘‘vp’’ in the ﬁgure mean elasticity, plasticity, and viscoplasticity, respectively. This kind
of model takes into account viscosity only in the plastic part, not in elastic part. It is named
elastic=viscoplastic model.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 47 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 47
The constitutive relation of the elastic element is s¼E_ «. Let Y denote the static yield stress of
the plastic element. The plastic element does not deform when s<Y, and the stress applied to the
plastic element keeps Y constant when s>Y. The constitutive relation of the viscous element is
s
vp
¼m_ «
vp
; here m is the viscosity coefﬁcient.
The total stress and strain rates meet the following relation:
s ¼ s
p
þs
vp
_ « ¼ _ «
e
þ _ «
vp
e =e
3
e =e
2
e =e
1
e
1
< < e
2
e
3
s
e O
FIGURE 1.11 Stress–strain curve of viscoplastic material.
s
vp
= s −Y
e
v
p
e
e
Y
s
FIGURE 1.12 Onedimensional elastic=viscoplastic model.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 48 3.11.2008 3:17pm Compositor Name: BMani
48 Handbook of Thermal Process Modeling of Steels
Then the constitutive equation of elastic=viscoplastic material is
_ « ¼
1
E
_ s if s < Y
_ « ¼
1
E
_ s þ
1
m
(s ÀY) if s > Y
_
¸
¸
_
¸
¸
_
The stress exerted to the viscous element is s
vp
¼sÀY, which is called overstress. If overstress
exists, then _ « cannot be zero, and the strain « will increase with time.
1.5.3.2 Viscoplastic Constitutive Equation for General Stress State
Perzyna proposed the viscoplastic constitutive equation for general stress state by associated ﬂow
law of plastic potential theory.
_ «
vp
ij
¼ ghf(F)i
@F
@s
ij
(1:149)
where
g is the viscous coefﬁcient of the material
F is the static yield function
The meaning of the symbol h i is
hf(F)i ¼ 0 if F < 0
hf(F)i ¼ f(F) if F > 0
_
f(F) is a function of overstress; it should be determined by experiment. For example, w(F) ¼(F=Y)
n
can be adopted. If Mises yield criterion is adopted as F in the equation, we can ﬁnally obtain
_ «
vp
ij
¼
3
_
«
vp
2s
s
0
ij
1.5.3.3 Commonly Used Viscoplastic Models
The ﬂow stress of viscoplastic material is the function of strain, strain rate, and temperature, and the
following expressions of equivalent stress are commonly used in engineering computations.
Backofen model [35–37]
s ¼ c
_
«
m
where c and m are material constants.
Rosserd model [38–40]
s ¼ k«
m
_
«
n
where k, m, and n are material constants. This model takes into account the effect of both strain and
strain rate on ﬂow stress.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 49 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 49
The overstress model [41–43]
s ¼ Y(«) 1 þ
_
«
vp
r
_ _
n
_ _
where
Y(«) is the static yield stress
n and r are the material constants
Powerexponent model [44,45]
s ¼ g
_
«
vp
_ a
_ _
m
where
g is the reference stress
_ a is the reference strain rate
m is the material constant
1.5.3.4 Creep
If the material is loaded for a long time, especially at elevated temperature, then even the stress does
not reach the yield stress, and it will deform permanently with time. If the stress is originated
from the elastic deformation of the material, e.g., the prestress or residual stress, the stress will relax
with time. This phenomenon is called creep. Creep is similar to viscoplasticity, and they differ
in whether there exits the yield point and the time scale. A typical uniaxial creep curve is shown in
Figure 1.13.
The duration of the initial creep and the third stage creep are short. Therefore, attention is
usually concentrated on the second stage creep.
When analyzing creep problems, viscosity must be considered in both elastic and plastic
constitutive responses. Such a constitutive model is called a viscoelastoplastic model.
Initial creep
Second stage creep
e
Rupture
Third stage creep
t
O
FIGURE 1.13 A typical onedimensional creep curve.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 50 3.11.2008 3:17pm Compositor Name: BMani
50 Handbook of Thermal Process Modeling of Steels
The strain increment is decomposed as follows:
D« ¼ D«
e
þD«
p
þD«
T
þD«
c
(1:150)
where the superscripts ‘‘e,’’ ‘‘p,’’ ‘‘T,’’ and ‘‘c’’ denote the elastic, plastic, thermal, and creep strain,
respectively.
Creep strain D«
c
can be expressed by creep strain rate _ «
c
. In the ith computation step
D«
c
i
¼ _ «
c
i
Dt
i
(1 Àu) þu_ «
c
iþ1
Dt
i
where
Dt
i
is the time step size of step i
_ «
c
i
and _ «
c
iþ1
are the creep rate at time i and time i þ1, respectively
u is the parameter of ﬁnite difference, 0 u 1
Usually u ¼1=2 $ 2=3 is adopted.
Suppose
_
«
c
is the function of s, and taking the two anterior items of the Taylor expansion of
_ «
c
(sþds), we get
_ «
c
iþ1
¼ _ «
c
i
þ
@ _ «
c
@s
¸
¸
¸
¸
i
Ds
i
¼ _ «
c
i
þH
i
Ds
i
where
H
i
¼
@ _ «
c
@s
¸
¸
¸
¸
i
This matrix can be computed from viscoplastic constitutive relation:
_ «
c
¼
_
«
c
(@F=@s) ¼
_
«
c
(@s=@s)
Then
D«
c
i
¼ _ «
c
i
Dt
i
þuDt
i
H
i
s
i
(1:151)
1. The viscoelastic constitutive equation
Substituting Equations 1.150 and 1.151 into the elastic constitutive equation and supposing
D«
p
¼ 0
we obtain
Ds
i
¼
^
C
e
D«
i
ÀD«
T
i
À _ «
e
i
Dt
i
_ _
where
^
C
e
¼ (C
e
)
À1
þuH
i
Dt
i
_ ¸
À1
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 51 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 51
2. The viscoplastic constitutive equation
Expressing D«
p
by the ﬂow rule and adding it to the viscoelastic constitutive equation,
we get
Ds
i
¼
^
C
ep
(D«
i
ÀD«
T
i
À _ «
e
i
Dt
i
) þDs
T
i
(1:152)
where
^
C
ep
¼
^
C
e
À
^
C
p
^
C
p
¼
^
C
e @F
@s
@F
@s
_ _
T
^
C
e
H þ
@F
@s
_ _
T
^
C
e @F
@s
Ds
T
i
can be referred to in Equation 1.147.
Different types of creep equations are determined by experiment, which vary with different
materials and deformation conditions. Some of them, commonly used in engineering computations,
are given below.
1. A creep rule in the uniaxial loading [46]
«
c
¼ As
n
t
m
i.e.,
_ «
c
¼ mAs
n
t
mÀ1
where
m, n, and A are the material constants [46]
t is the time
2. Anand model [47]
_
« ¼ A
0
(shBs)
n
e
V
Tþ273
where
s is the equivalent stress
T is the temperature (8C)
A
0
, B, V, and n are the material constants
3. Power law for the second stage creep [48]
_
«
e
¼ k
1
s
k
2
e
À
k
3
T
where k
1
, k
2
, and k
3
are the material constants.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 52 3.11.2008 3:17pm Compositor Name: BMani
52 Handbook of Thermal Process Modeling of Steels
1.6 BASICS OF COMPUTATIONAL FLUID DYNAMICS IN THERMAL
PROCESSING
1.6.1 INTRODUCTION
The ﬂow phenomena are very common during thermal processing and have various forms. For
example, the liquid materials fulﬁll the ﬁlling through ﬂowing in the casting and injection
forming; the gas ﬂow in the heat treatment furnaces are helpful in ensuring the uniformity of
furnace wall and atmosphere; and the quenching media ﬂow is usually critical to improve
quenching ability and product quality. In all these engineering problems, the continuum model
has been widely adopted, which takes a series of state variables, such as pressure, velocity,
density, temperature, and so on, describing the ﬂow as continuous functions of time and spatial
coordinates. Hence, the daedal forms of ﬂow during different thermal processes are dominated by
the most fundamental physical laws, namely, the mass conservation law, the momentum conser
vation law, and the energy conservation law.
In this section, the mathematical expressions of different laws, namely, the governing PDEs,
in ﬂuid ﬂow and the monodromy conditions (i.e., initial and boundary conditions) making
one process different from others, are introduced from the view of ﬂuid mechanics. The
fundamental ideas and commonly used numerical methods to solve the governing PDEs are
also brieﬂy discussed.
1.6.2 GOVERNING DIFFERENTIAL EQUATIONS FOR FLUID
All of computational ﬂuid dynamics (CFD), in one form or another, is based on the fundamental
governing equations of ﬂuid dynamics: the continuity, momentum, and energy equations. They are
the mathematical statement of three fundamental physical principles upon which all of ﬂuid
dynamics is based:
1. Mass is conserved
2. Newton’s second law, F¼ma
3. Energy is conserved
1.6.2.1 Generalized Newton’s Law
The relationship between stress tensor and strain tensor for the viscous ﬂuid can be described by
generalized Newton’s law. The constitutive equation is
[t] ¼ 2h[«] À(p ÀlrV)[I] (1:153)
where
[t] is the stress tensor
h is the dynamic viscosity coefﬁcient
p is the static pressure of ﬂuid
V is the velocity vector
[I] is the unit vector
[«] is the strain tensor
l (l¼À2=3 h) is the second viscosity coefﬁcient
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 53 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 53
In the Cartesian space, the strain tensor can be expressed as
[«] ¼
«
x
«
xy
«
xz
«
yx
«
y
«
yz
«
zx
«
zy
«
z
_
_
_
_
¼
@u
@x
1
2
@v
@x
þ
@u
@y
_ _
1
2
@u
@z
þ
@w
@x
_ _
1
2
@v
@x
þ
@u
@y
_ _
@v
@y
1
2
@w
@y
þ
@v
@z
_ _
1
2
@u
@z
þ
@w
@x
_ _
1
2
@w
@y
þ
@v
@z
_ _
@w
@z
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
(1:154)
where u, v, and w are the components of V along the coordinates x, y, and z.
The constitutive equation in the Cartesian space can thus be expressed as
s
x
¼ Àp þlrv þ2h
@u
@x
s
y
¼ Àp þlrv þ2h
@v
@y
s
z
¼ Àp þlrv þ2h
@w
@z
t
xy
¼ t
yx
¼ h
@u
@y
þ
@v
@x
_ _
t
zx
¼ t
xz
¼ h
@u
@z
þ
@w
@x
_ _
t
yz
¼ t
zy
¼ h
@v
@z
þ
@w
@y
_ _
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:155)
where
s
x
, s
y
, and s
z
are the normal stress
t
x
, t
y
, and t
z
are the shear stress
1.6.2.2 Continuity Equation (Mass Conservation Equation)
The fundamental physical principle that mass is conserved means that net mass ﬂow out of the
inﬁnitesimal ﬂuid element through surface equals the time rate of decrease in mass inside the
inﬁnitesimal ﬂuid element. Hence, the continuity equation can be derived as
@r
@t
þr(rv) ¼ 0 (1:156)
In the Cartesian space, it becomes
@r
@t
þ
@(ru)
@x
þ
@(rv)
@y
þ
@(rw)
@z
¼ 0 (1:157)
where
r is the ﬂuid
t is the time
@r=@t is the time rate of increase in mass inside the element volume
((@(ru)=@x) þ(@(rv)=@y) þ(@(rw)=@z)) is the time rate of net mass ﬂow out of the element
volume
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 54 3.11.2008 3:17pm Compositor Name: BMani
54 Handbook of Thermal Process Modeling of Steels
1.6.2.3 Momentum Conservation Equation
Applying another fundamental physical principle to a model of the ﬂow, namely,
Physical principle: F¼ma (Newton’s second law)
The conclusion that the time rate of ﬂow momentum equals the sum of external force acting
on it can be arrived at. Therefore, we have the momentum equation:
rdV
dt
¼ rF þr[t] (1:158)
where F is the body force of ﬂow of unit volume.
Substituting Equation 1.153 into Equation 1.158, and writing in the Cartesian space, the
expanded form of the momentum equation can be obtained.
r
@u
@t
þu
@u
@x
þv
@u
@y
þw
@u
@z
_ _
¼rF
x
À
@(pÀlrV)
@x
þ
@
@x
2h
@u
@x
_ _
þ
@
@y
h
@v
@x
þ
@u
@y
_ _ _ _
þ
@
@z
h
@u
@z
þ
@w
@x
_ _ _ _
r
@v
@t
þu
@v
@x
þv
@v
@y
þw
@v
@z
_ _
¼rF
y
À
@(pÀlrV)
@y
þ
@
@y
2h
@v
@y
_ _
þ
@
@x
h
@v
@x
þ
@u
@y
_ _ _ _
þ
@
@z
h
@v
@z
þ
@w
@y
_ _ _ _
r
@w
@t
þu
@w
@x
þv
@w
@y
þw
@w
@z
_ _
¼rF
z
À
@(pÀlrV)
@z
þ
@
@z
2h
@w
@z
_ _
þ
@
@x
h
@u
@z
þ
@w
@x
_ _ _ _
þ
@
@y
h
@v
@z
þ
@w
@y
_ _ _ _
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:159)
Equation 1.159 is also called Navier–Stokes equation. The lefthand side of Equation 1.159 is the
inertia force, while on the righthand side of Equation 1.159, the ﬁrst item is the mass force, the
second item is the pressure difference, the third item is the viscous expansion force, and the fourth
and ﬁfth items are the viscous deformation force. The third to ﬁfth items are only related to the
viscous coefﬁcient and strain tensor.
1.6.2.4 Energy Conservation Equation
According to the ﬁrst law of thermodynamics, the energy equation can be written as
rc
p
dT
dt
¼ Àp(rV) þr(krT) þrq þF (1:160)
where
c
p
is the speciﬁc heat at constant pressure
T is the ﬂow temperature
q is the heat ﬂux in the ﬂow with unit mass
k is the thermal conductivity of the ﬂow
The lefthand side of Equation 1.160 is the increase in system internal energy. On the righthand
side of Equation 1.160, the ﬁrst item is the work done by the volume change of ﬂow (it equals zero
for the incompressible ﬂow), the second item is the energy input by the thermal conduction, the third
item is the energy generated by the internal heat source, and the fourth item is the viscous
dissipation work, which can be expressed as
F ¼ h 2
@u
@x
_ _
2
þ2
@v
@y
_ _
2
þ2
@w
@z
_ _
2
þ
@v
@x
þ
@u
@y
_ _
2
þ
@w
@y
þ
@v
@z
_ _
2
þ
@w
@x
þ
@u
@z
_ _
2
_ _
þl(rÁ V)
2
(1:161)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 55 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 55
Then,
rc
p
@T
@t
þu
@T
@x
þv
@T
@y
þw
@T
@z
_ _
¼ (Àp þlrÁ V)
@u
@x
þ
@v
@y
þ
@w
@z
_ _
þ h 2
@u
@x
_ _
2
þ2
@v
@y
_ _
2
þ2
@w
@z
_ _
2
þ
@v
@x
þ
@u
@y
_ _
2
þ
@w
@y
þ
@v
@z
_ _
2
þ
@w
@x
þ
@u
@z
_ _
2
_ _
þ
@
@x
k
@T
@x
_ _
þ
@
@y
k
@T
@y
_ _
þ
@
@z
k
@T
@z
_ _
þrq (1:162)
1.6.3 GENERAL FORM OF GOVERNING EQUATIONS
For the viscous Newtonian ﬂuid, the governing equations of the main state variables that need to be
solved in ﬂuid ﬂow problems can be expressed in a generalized form as [49]
@
@t
(rF) þ
@
@x
j
(ru
j
F) ¼
@
@x
j
G
F
@F
@x
j
_ _
þS
F
(1:163)
where
j ¼1, 2, 3, x
j
is the coordinate components
u
j
is the velocity components along x
j
F is the general variable
G
F
is the transmission coefﬁcient
S
F
is the source item
For the incompressible turbulent ﬂow, the coefﬁcient and source in the equations of CFD model
are listed in Table 1.4.
Since there are usually three forms of heat transfer, conduction, convection, and radiation, it is
worth mentioning that the control equation discussed here can describe the conduction and
convection, but for the radiative ﬂow media, the radiation heat transfer with the furnace wall or
among themselves should be considered additionally. The numerical calculation of radiation heat
transfer can be found in the related literature [50,51].
1.6.4 SIMPLIFIED AND SPECIAL EQUATIONS IN THERMAL PROCESSING
The basic equations described here are applicable in an unsteady viscous Newtonian ﬂuid, while
their solving is extraordinarily complicated. In the modeling of thermal processing, reasonable
hypotheses are usually adopted to simplify the governing equations. Some examples are given in
Table 1.4.
TABLE 1.4
Coefﬁcients and Source Item in the Generalized Differential Transmission Equation
Equation F G
F
S
F
Continuity equation 1 0 0
Momentum equation u
j
0 rg
i
À
@p
@x
i
þhr
2
u
i
Energy equation T
K
c
p
h 2
@u
i
@x
i
_ _
2
þ
@v
@x
þ
@u
@y
_ _
2
þ
@w
@y
þ
@v
@z
_ _
2
þ
@w
@x
þ
@u
@z
_ _
2
_ _
þl(rV)
2
þrq
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 56 3.11.2008 3:17pm Compositor Name: BMani
56 Handbook of Thermal Process Modeling of Steels
1.6.4.1 Continuity Equation for Incompressible SourceFree Flow
The incompressible sourcefree ﬂow, for example, the ﬁlling process during casting, has a null
divergence of ﬂow velocity at any point in the domain ﬁlled with ﬂow; i.e., there is no source and
leaking of ﬂow. It follows the mass conservation law. Therefore, the continuity equation can be
simpliﬁed as follows:
@u
@x
þ
@v
@y
þ
@w
@z
¼ 0 (1:164)
1.6.4.2 Euler Equations for Ideal Flow
For the ideal ﬂow without viscous force, the forces acting on the inﬁnitesimal cubic element include
the gravity and pressure upon six surfaces. According to Newton’s second law, we have
(dx dy dz)rg
x
Àdp(dy dz) ¼ (dx dy dz)ra
x
(dx dy dz)rg
y
Àdp(dz dx) ¼ (dx dy dz)ra
y
(dx dy dz)rg
z
Àdp(dx dy) ¼ (dx dy dz)ra
z
_
_
_
(1:165)
where a
x
, a
y
, and a
z
are the three components of acceleration, and they are the derivatives of three
velocity components, respectively. That is,
a
x
¼
du
dt
, a
y
¼
dv
dt
, a
z
¼
dw
dt
The acceleration here means the velocity change in the motion of an inﬁnitesimal ﬂuid element,
instead of that of the different ﬂow passing through a certain position. The former is the substantial
derivative of velocity to time and spatial coordinates while the latter is the partial derivative to time
only, and their relationship is
du
dt
¼
@u
@t
þ
@u
@x
@x
@t
þ
@u
@y
@y
@t
þ
@u
@z
@z
@t
dv
dt
¼
@v
@t
þ
@v
@x
@x
@t
þ
@v
@y
@y
@t
þ
@v
@z
@z
@t
dw
dt
¼
@w
@t
þ
@w
@x
@x
@t
þ
@w
@y
@y
@t
þ
@w
@z
@z
@t
_
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
_
(1:166)
Substituting Equation 1.166 into Equation 1.167, the momentum equation for the ideal ﬂow,
namely, Eular equation, can be obtained as
r
@u
@t
þu
@u
@x
þv
@u
@y
þw
@u
@z
_ _
¼ À
@p
@x
þrg
x
r
@v
@t
þu
@v
@x
þv
@v
@y
þw
@v
@z
_ _
¼ À
@p
@y
þrg
y
r
@w
@t
þu
@w
@x
þv
@w
@y
þw
@w
@z
_ _
¼ À
@p
@z
þrg
z
_
¸
¸
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
¸
¸
_
(1:167)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 57 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 57
1.6.4.3 Volume Function Equation
In the simulation of the ﬁlling process during casting, the solution algorithm–volume of ﬂuid
(SOLA–VOF) method has been developed to determine the position of the free surface [52–57].
It describes the whole ﬂow domain by volume function F, which is deﬁned as
F ¼ The volume of flow in an element=element volume
Hence, solving the volume function equation, which can be expressed as follows, provides the state
of each element.
@F
@t
þ
@(Fu)
@x
þ
@(Fv)
@y
þ
@(Fw)
@z
¼ 0 (1:168)
The value of volume function ranges from 0 to 1. When it equals zero, the element is empty without
ﬂow; when it equals unit, the element is full; In other cases, namely, 0 <F<1, it means that the
element is a surface element, which has ﬂow inside but not fully ﬁlled.
Hence, only when the value of volume function has been calculated for each element can the
ﬁlling and ﬂowing state be obtained for the cast parts at any time.
1.6.5 NUMERICAL SOLUTION OF GOVERNING PDES
The governing PDEs discussed above are applicable to ﬂow and heat transfer processes of all the
Newtonian ﬂuid, and the difference between different processes is prescribed by the initial and
boundary conditions (generally called monodromy conditions). The combination of governing
PDEs and corresponding initial and boundary conditions constitutes the complete mathematical
formulation of a physical process.
The initial conditions refer to the spatial distribution of the solving variables at the starting
moment of the analysis, and they need to be set initially. However, it is not necessary for the steady
problem. The boundary conditions are the evolution of the solving variables or their ﬁrst derivatives
around the domain boundary with the time and position.
Usually, the governing PDEs can be divided into three classes, hyperbolic, parabolic, and
elliptic. If PDEs in the whole solving domain belong to the same class, the physical problem can
be called after the class of the PDE. Some physical problems, in which there are different classes of
PDEs in one solving domain, are the socalled hybrid problems. The main difference in different
PDE classes lies on their domain of dependence and domain of inﬂuence, which affect their solving
methods and strategies in turn.
For the PDEs mentioned here, describing the ﬂuid ﬂow and heat transfer, many mathematical
methods have been developed to obtain the analytical solution, which are continuous on the whole
solving domain. Up to now, these analytical solutions are subjected only to a few simple cases. The
numerical methods have been applied more and more in mass ﬂow and heat transfer problems of
practical engineering signiﬁcance.
The fundamental idea of CFD is ﬁrst, to replace the ﬁeld of certain physical variables (velocity
ﬁeld, temperature ﬁeld, etc.) that are continuous in the spatial and time coordinates with
the collection of the values on a series of ﬁnite discretized points (nodes); then, to establish the
algebraic equation (discretization equation) reﬂecting the relationship of the values on these
discretized nodes based on certain principles; ﬁnally, to solve the established algebraic equation
to obtain the approximation value of the solving variables. The corresponding ﬂow chart is shown in
Figure 1.14.
In the past several decades, many numerical methods have been developed, and their main
differences lie in their discretization method of domains and equations, and the algorithm of solving
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 58 3.11.2008 3:17pm Compositor Name: BMani
58 Handbook of Thermal Process Modeling of Steels
the algebraic equations. The frequently used methods in CFD are FDM, FEM, ﬁnitevolume method
(FVM), and ﬁniteanalytic method (FAM). The FDM and FEM have been introduced in the
previous parts of this chapter; while the FVM and FAM will not be expanded due to space limit;
the related studies [58–63] can be referred. In general sense, FVM has obvious superiority over
other methods in CFD through the comprehensive evaluation from the view of the ease of
implementation, the maturity of development and application, etc.
REFERENCES
1. Segerlind, L.J., Applied Finite Element Analysis, 2nd ed., New York: Wiley, 1984.
2. Wait, R. and Mitchell, A.R., Finite Element Analysis and Applications, New York: Wiley, 1985.
3. Zienkiewicz, O.C. and Taylor, R.L., The Finite Element Method, Vol. 2., New York: McGrawHill, 1991.
4. Stasa, F.L., Applied Element Analysis for Engineers, New York: Holt, Rinehart, & Winston, 1985.
Establishing governing PDEs,
determining initial and boundary conditions
Divinding subdomains, determining nodes
(domain discretization)
Constructing the discretization equation
(PDEs discretization)
Discretizing intial and boundary
conditions
Solving the discretization equation
Nonlinear problems
Yes
No
Reconstructing
the discretization
equation with
current values
Linear problems
Converge or not?
Analyzing the solution
FIGURE 1.14 Flow chart of numerical solving for physical problems.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 59 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 59
5. Huebner, K.H. and Thornton, E.A., The Finite Element Method for Engineers, 2nd ed., New York: John
Wiley & Sons, 1982.
6. Incropera, F.P. and DeWitt, D.P., Introduction to Heat Transfer, 3rd ed., New York: John Wiley & Sons,
1996.
7. Reddy, J.N. and Gartling, D.K., The Finite Element Method in Heat Transfer and Fluid Dynamics,
New York: CRC Press, 2000.
8. Johnson, A.W. and Mehl, R.F., Reactions of kinetics in processes of nucleation and growth, Transactions
AIME, 1939, 135:416–458.
9. Fernandes, F.M.B., Denis, S., and Simon, A., Mathematical model coupling phased transformation and
temperature evolution during quenching of steels, Materials Science and Technology, 1985, 1(10):838–844.
10. Avrami, M., Journal of Chemical Physics, 1939, 7:1103–1112.
11. Hsu Tsuyao (Xu Zuyao), Principles of Phase Transformation, Beijing: Science Press, 1988, pp. 408–419
(in Chinese).
12. Hawbolt, E.B., Chau, B., and Brimacombe, J.K., Kinetics of austeniteferrite and austenitepearlite
transformation in a 1025 carbon steel, Metallurgical Transactions A, 1985, 16A:565–577.
13. Reti, T. and Felde, I., A nonlinear extension of the additivity rule, Computational Materials Science, 1999,
15:466–482.
14. Tian Dong. Simulation and technology design of quenching of steel workpieces with complex shapes,
PhD thesis, Shanghai Jiao Tong University, Shanghai, 1998.
15. Wang, K.F., Chandrasekar, S., and Yang, H.T.Y., Experimental and computational study of the quenching
of carbon steel, Steel Research, 1996, 67(7):257–265.
16. Pan Jiansheng, Zhang Weimin, Yuan Wenqing, Hu Mingjuan, and Gu Jianfeng, Discussions on the factors
affecting the accuracy of computer simulation on heat treatment. The Proceedings of First Chinese Heat
Treatment Week, 2002, Dalian, pp. 1–11 (in Chinese).
17. Song Dongli, Research of quenching process for largesized plastic die and mould steel blocks and its
application, PhD thesis, Shanghai Jiao Tong University, Shanghai, 2005.
18. Koistinen, D.F. and Marburger, R.E., General equation prescribing the extent of the austenite transform
ation in pure ironcarbon alloys and plain carbon steels, Acta Metallurgica, 1959, 7:50–60.
19. Magee, L.C., Phase transformations, ASME., 1970, 115.
20. Hsu Tsuyao (Xu Zuyao), Progress in martensitic transformations (I), Shanghai Metals, 2003, 25(3):1–8.
21. Sjöström, S.D. and Simon, A., Coupled temperature, stress, phase transformation calculation model
numerical illustration of the internal stresses evolution during cooling of a eutectoid steel cylinder,
Metallurgical Transactions A, 1987, 18A:1203–1212.
22. Couonna, F., Miassonl, E., Denis, S., et al., On thermoelasticviscoplastic analysis of cooling processes
including phase changes, Journal of Materials Processing Technology, 1992, 34:525–532.
23. Hsu Tsuyao (Xu Zuyao), Progress and perspective of materials heat treatment, Transaction of Materials
and Heat Treatment, 2003, 24(1):1–13 (in Chinese).
24. Hsu Tsuyao (Xu Zuyao), Fundamentals of the uniﬁed technology combining plastic forming and heat
treatment of materials, Engineering Science in China, 2004, 6(1):16–21 (in Chinese).
25. Denis, S., Archambault, P., Gautier, E., Simon, A., and Beck, G., Prediction of residual stress and
distortion of ferrous and nonferrous metals: Current status and future developments, Journal of Materials
Engineering and Performance, 2002, 11(1):92–102.
26. Inoue, T. and Wang, Z., Coupling between stress, temperature, metallic structures during processes
involving phase transformations, Materials Science and Technology, 1985, 1:845–849.
27. Denis, S., Gautier, E., Simon, A., and Beck, G., Stressphase transformation interactions basic principles,
modeling, and calculation of internal stresses, Materials Science and Technology, 1985, 1:806–814.
28. Denis, S.P., Archambault, C.A., et al., Modelling of phase transformation kinetics in steels and coupling
with heat treatment residual stress production, Journal of Physics (France), 1999(9):323–332.
29. Reese, S., A micromechanically motivated material model for the thermoviscoelastic material behaviour
of rubberlike polymers, International Journal of Plasticity, 2003, 19:909–940.
30. Mooney, M., A theory of large elastic deformation, Journal of Applied Physics, 1940, 11:582–592.
31. Rivlin, R.S., Large elastic deformations of isotropic materials, Philosophical Transactions of the Royal
Society of London Series A, 1948, 240:459–490.
32. Rivlin, R.S., Large elastic deformations of isotropic materials. IV. Further developments of the general
theory, Philosophical Transactions of the Royal Society of London Series A, 1948, 241:379–397.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 60 3.11.2008 3:17pm Compositor Name: BMani
60 Handbook of Thermal Process Modeling of Steels
33. Ogden, R.W., Nonlinear elasticity, anisotropy, material stability and residual stresses in soft tissue, in
Holzapfel, G.A. and Ogden, R.W. (Eds.), Biomechanics of Soft Tissue in Cardiovascular Systems, Vol.
441, CISM Courses and Lectures Series, Wien: Springer, 2003, pp. 65–108.
34. Ogden, R.W., Saccomandi, G., and Sgura, I., Fitting hyperelastic model to experimental data, Computa
tional Mechanics, 2004, 34:484–502.
35. Fields D.F. and Backofen, W.A., Determination of strainhardening characteristics by torsion testing, in
Proceedings of the 60th Annual Meeting of the American Society for Testing and Materials, Vol. 57, 1957,
pp. 1259–1272.
36. Khoddam, S. and Hodgson, P.D., Conversion of the hot torsion test results into ﬂow curve with multiple
regimes of hardening, Journal of Materials Processing Technology, 2004, 153(special issue):839–845.
37. Chan, K.C. and Gao, L., On the susceptibility to localized necking of defectfree metal sheets under biaxial
stretching, Journal of Materials Processing Technology, 1996, 58(2–3):251–255.
38. Perzyna, P., Fundamental problems in viscoplasticity, in Kuerti, G. (Ed.), Advances in Applied Mechanics,
Vol. 9, New York: Academic Press, 1966, pp. 243–377.
39. Lenard, J.G., Modeling Hot Deformation of Steels, Berlin: SpringerVerlag, 1989, pp. 101–115.
40. Xing, H.L., Wang, C.W., Zhang, K.F. et al., Recent development in the mechanics of superplasticity and its
applications, Journal of Materials Processing Technology, 2004, 151(1–3):196–202.
41. Colak, O.U. and Krempl, E., Modeling of uniaxial and biaxial ratcheting behavior of 1026 Carbon steel
using the simpliﬁed viscoplasticity theory based on overstress (VBO), Acta Mechanica, 2003, 160:27–44.
42. Colak, O.U. and Krempl, E., Modeling of the monotonic and cyclic swift effects using an isotropic, ﬁnite
viscoplasticity theory based on overstress (FVBO), International Journal of Plasticity, 2005, 21:573–588.
43. Chaboche, J., Constitutive equations for cyclic plasticity and cyclic viscoplasticity, International Journal
of Plasticity, 1989, 5:247–302.
44. Hart, E.W., Constitutive relations for the nonelastic deformation of metals, ASME. Journal of Engineering
Material and Technology, 1976, 98:193–202.
45. Weber, G. and Annand, L., Finite deformation constitutive equations and a time integration procedure for
isotropic, hyperelastic–viscoelastic solids, Computer Methods in Applied Mechanics and Engineering,
1990, 79:173–202.
46. Miller, A.K., Uniﬁed Constitutive Equations for Creep and Plasticity, London: Elsevier Applied Science,
1987.
47. Anand, L., Constitutive equations for hot working of metals, Journal of Plasticity, 1985, 1:213–231.
48. Frost, H.J. and Ashby, M.F., Deformation Mechanism Maps: The Plasticity and Creep of Metals and
Ceramics, Oxford: Pergamon Press, 1982.
49. Patankar, S.V., Numerical Heat Transfer and Fluid Flow, New York: McGrawHill, 1980.
50. Yu Qizheng, The Principles of Radiation Heat Transfer, Harbing: Press of Harbing Institute of Technol
ogy, 2000, pp. 116–132 (in Chinese).
51. Fan Weicheng and Wan Yuepeng, Models and calculation of ﬂow and combustion, Hefei: University Press
of Science and Technology, China, 1992, pp. 202–243 (in Chinese).
52. Hirt, C.W. and Nichols, B.D., Volume of ﬂuid (VOF) method for the dynamics of free boundaries, Journal
of Computational Physics, 1981, 39:201–225.
53. Stoehr, R.A., Wang, C., Hwang, W.S., and Ingerslev, X., Modeling the ﬁlling of complex foundry molds,
in Kou, S. and Mehrabian, R. (Eds.), Modeling and Control of Casting and Welding Process, 1986,
pp. 303–313.
54. Torrey, M.D., Mjolsness, R.C., and Stein, L.R., NASAVOF3D: A threedimensional computer program
for incompressible ﬂows with free surfaces, Los Alamos National Laboratory Report LA11009MS, 1987.
55. Kim, W.S. and Im, I.T., Analysis of a mold ﬁlling using an implicit SOLAVOF, Numerical Heat
Transfer. Part A, 1999, 35(3):1040–7782.
56. Kuo, J.H. and Hwang, W.S., Development of an interactive simulation system for die cavity ﬁlling and its
application to the operation of a lowpressure casting process, Modeling and Simulation in Materials
Science and Engineering, 2000, 8(4):583–602.
57. Babaei, R., Abdollahi, J., Homayonifar, P. et al., Improved advection algorithm of computational modeling
of free surface ﬂow using structured grids, Computer Methods in Applied Mechanics and Engineering,
2006, 195(7–8):775–795.
58. Manole, D.M. and Lage, J.L., Nonuniform grid accuracy test applied to the natural convection ﬂow within
a porous medium cavity, Numerical Heat Transfer, Part B, 1993, 23:351–368.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 61 3.11.2008 3:17pm Compositor Name: BMani
Mathematical Fundamentals 61
59. Tan Weiyan, Computational Shallow Water Dynamics. Beijing: Tsinghua University Press, 1998
(in Chinese).
60. Tao Wenquan, Computational Heat Transfer, Xi’an: Xi’an Jiao Tong University Press, 1988 (in Chinese).
61. Versteeg, H.K. and Malalsekera, W., An Introduction to Computational Fluid Dynamics, The Finite
Volume Method, Essex: Longman Scientiﬁc & Technical, 1995, p. 4.
62. Wei, L.I., Hybrid Finite Analytic Method for Viscous Flow, Beijing: Science Press, 2000 (in Chinese).
63. Chen, C.J., Bernatz, R., Carlson, K.D., and Lin, W.L., Finite Analytic Method in Flows and Heat Transfer.
New York: Taylor & Francis, 2000.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C001 Final Proof page 62 3.11.2008 3:17pm Compositor Name: BMani
62 Handbook of Thermal Process Modeling of Steels
2
Thermodynamics of Thermal
Processing
Sivaraman Guruswamy
CONTENTS
2.1 Introduction........................................................................................................................ 63
2.2 Concepts of State, Equilibrium, and Stability of a System............................................... 65
2.3 Zeroth Law of Thermodynamics ....................................................................................... 65
2.4 First Law of Thermodynamics........................................................................................... 66
2.4.1 Internal Energy........................................................................................................ 66
2.4.2 Enthalpy, H............................................................................................................. 67
2.5 Second Law of Thermodynamics ...................................................................................... 68
2.6 Third Law of Thermodynamics ......................................................................................... 69
2.7 Concept of Gibbs and Helmholtz Free Energies ............................................................... 70
2.8 Evaluation of G.................................................................................................................. 72
2.9 Estimation of Gibbs Free Energy of a Phase .................................................................... 73
2.10 Chemical Potential and Activity........................................................................................ 76
2.11 Gibbs Free Energy Curves and Phase Diagrams............................................................... 79
2.12 Gibbs Phase Rule............................................................................................................... 80
2.13 Kinetic Theory ................................................................................................................... 80
2.14 Statistical Thermodynamics ............................................................................................... 82
2.15 Applications of Thermodynamics in Thermal Processing................................................. 85
Bibliography.................................................................................................................................... 87
Some of the Sources for Thermodynamic Data ............................................................................. 88
2.1 INTRODUCTION
Each milestone in the human history is marked by the development of an ability to produce and
manipulate a new material to meet human needs. A wide range of metals and metallic alloys play a
key role in our industrialized society. Their importance to our life can be seen from their extensive
use in household appliances, ground transportation, power generation=conversion, aerospace, elec
tronics and communication, biological systems, and other applications. Examples of their use are
carbon and alloy steels in numerous engineering structures, aluminum alloys in aircraft structures,
titanium alloy in the compressor sections of aircraft turbine engines, titanium–aluminum–vanadium
alloy used in human orthopedic implants and heart valves, nickel alloys in the combustor sections of
aircraft engines, magnesium alloys in space vehicle and satellite components, magnesium alloys for
laptop casings, zirconium alloys in nuclear reactor fuel elements, neodymium–iron–boron in high
energy density magnets used in a wide range of hightechnology products, and so on. Production of
components made of these materials involves various thermalprocessing steps. The required
properties for each of the applications are obtained through a judicious choice of the principal
metallic element, alloying elements, control of the microstructure and thermal or heat treatments.
An indepth understanding of how desirable microstructures in these alloys are obtained through the
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 63 1.11.2008 2:41am Compositor Name: JGanesan
63
control of phase transformations that take place during the different heat treatment steps, and how
such an understanding can help engineer new materials, requires a good knowledge of thermo
dynamics. Thermodynamics is also essential for a better understanding and control of the various
thermalprocessing operations.
The science of thermodynamics had its origin in the late eighteenth century and the beginning of
industrial revolution when the relationship between heat and mechanical work was established. The
science that dealt with the relationship between thermal energy (thermal) and mechanical movement
(dynamics) was thus given the name ‘‘thermodynamics.’’ Its subsequent developments paralleled
the development of atomic theory of matter, and have shown that thermodynamics deals with all
forms of energy, work, and ﬂow of energy from one form to another, and thus touches almost
all aspects of life.
Classical thermodynamics (or simply thermodynamics as it is often referred to) is an experimental
science based on a small number of principles that are generalizations made from experience. It deals
with macroscopic properties or behavior of matter and the relationships between them. It makes no
hypotheses about the microscopic structure. The system may be an alloy or a process system under
consideration. While it is impossible to provide a detailed treatment here, all the essential elements of
thermodynamics as it applies to thermal processing are presented, which will allow the readers to
effectively apply it in their work. While the topic is often considered abstract, an attempt is made to
convey that this is based on simple principles, ideas, or observations and exciting in its universal
scope. The material covered applies not only to engineering materials, systems, and processes of
interest, but also to geological, biological, and meteorological systems and processes.
The usefulness of classical thermodynamics lies in its ability to predict whether a material or an
engineering process system is stable under a given set of conditions or whether a process or
change in material state is likely to occur spontaneously in a given desired direction. Kinetic theory
and statistical mechanics complement classical thermodynamics. Kinetic theory applies the laws of
mechanics to individual particles of the system and calculates the values of the macroscopic
properties from such a treatment. It also allows the determination of the rate at which a process
can progress. Statistical mechanics ignores detailed considerations of particles and applies statistical
considerations to a collection of particles to ﬁnd the most probable distribution of a large number of
particles among the different energy states of the system. Kinetic theory and statistical mechanics
together form the basis of statistical thermodynamics. This chapter ﬁrst presents classical thermo
dynamic concepts followed by an introduction to statistical thermodynamics and its link to classical
thermodynamics.
Deﬁnition of some of the frequently used thermodynamic terms: system, surroundings, and
phase would be appropriate here. The term system refers to a region of interest that is examined.
Surroundings refer to all environments outside of this system with which it interacts. Together, the
system and surroundings constitute the universe. An alloy or a material of interest is refered to as a
system in physical metallurgy. In thermal processing, it may refer to a reactor containing
the reactants and products, or some other region of interest. The systems are classiﬁed as open,
closed, and isolated. In an open system, exchange of energy and mass with the surroundings
can occur. A closed system can exchange energy in various forms but cannot exchange mass
with the surroundings. An isolated system can exchange neither mass nor energy with the sur
roundings.
In an alloy or in engineering systems, one would ﬁnd one or more regions, each region with
homogeneous (or uniform) composition and properties. Each such region is referred to as a phase.
The composition of the phase is deﬁned by the relative proportions of different components
(elements or compounds) present in the phase. The phases may be spatially distributed in different
ways. When the temperature, pressure, and other thermodynamic conditions are changed, the phases
and their relative amounts present in the system also change. For a given set of equilibrium
conditions, the relative amount of each phase present is determined by the thermodynamic proper
ties of the system.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 64 1.11.2008 2:41am Compositor Name: JGanesan
64 Handbook of Thermal Process Modeling of Steels
2.2 CONCEPTS OF STATE, EQUILIBRIUM, AND STABILITY OF A SYSTEM
A basic idea in thermodynamics is the concept of state. In classical thermodynamics, the state of a
macroscopic system is deﬁned by a set of variables or properties that describe the observable
characteristics of the system. The properties that deﬁne the thermodynamic state include both
extensive and intensive properties. Intensive properties are those that do not depend on the amount
of material, such as temperature, pressure, magnetic ﬁeld, and electric ﬁeld. The extensive properties
are those that are dependent on the amount of material, such as volume. While there are a number of
macroscopic properties, there exists an interrelationship between them described by the equation
of state for the material, and this reduces the number of variables that deﬁne the thermodynamic
state. Properties needed to specify the state of a system are independent variables and the rest of the
properties are dependent variables. These variables that deﬁne the state of the system are also
referred to as state variables or state functions.
Another basic concept of importance is the thermodynamic equilibrium state of the system.
A system is said to be in a stable state when it is in a state of thermodynamic equilibrium for a given
set of conditions or state variables (such as T, P, V, . . . ). Thermodynamic equilibrium occurs when
the system is under mechanical, chemical, and thermal equilibrium.
Mechanical Equilibrium is a condition achieved when the system is in a state of mechanical rest and the
total potential energy is a minimum.
Chemical equilibrium is obtained when no further net reaction occurs between the reacting substances in
the system. The forward and reverse reaction rates are equal.
Thermal equilibrium is a condition resulting from the absence of thermal gradients.
The laws of thermodynamics allow us to establish the stability criterion of a system under a given
set of conditions such as temperature, pressure, electric ﬁeld, and magnetic ﬁeld. A brief review of
thermodynamic laws is essential for a good grasp of the concept of stability.
2.3 ZEROTH LAW OF THERMODYNAMICS
Through our sensory perception, can be felt intuitive the hotness on contact with an object that is
hotter than our body or cold chills on touching a cold object while it is commonly accepted that heat
ﬂows from a hightemperature object to a lowertemperature object it is in contact with, the rigorous
deﬁnitions of temperature and heat did not occur till late nineteenth century when Maxwell ﬁrst
deﬁned temperature based on such an observation. This deﬁnition is referred to frequently as the
zeroth law of thermodynamics. The zeroth law states that if two bodies are in thermal equilibrium
with a third, then they are in thermal equilibrium with each other. States of all systems in thermal
equilibrium can be assigned a number, representing a temperature. The equation of state could be
used (an equation that deﬁnes relationship between the various properties) for a gaseous material to
deﬁne its temperature in terms of pressure and volume. Boyle’s law and Charles law, together, lead
to the equation of state for a gas. In 1660, Boyle determined experimentally that at constant
temperature, pressure is proportional to the inverse of volume. Charles experimentally showed in
1787 that volume varies linearly with temperature at constant pressure. This formed the basis for the
quantiﬁcation of temperature using a mercury bulb thermometer to deﬁne the degree of hotness or
coldness. Two ﬁxedpoint temperatures, freezing point and boiling point of water, were deﬁned as the
reference points, with the freezing point as 08 and the boiling point as 1008 in the centigrade scale.
Joseph Louis GayLussac observed in 1802 that the thermal expansion coefﬁcient of all gases
was ﬁnite. Subsequent measurements by GayLussac and later reﬁned measurements by others
using ideal gases gave this constant as 1=273.16. A plot of volume versus temperature thus would
extrapolate to zero volume at negative 273.16 degree centigrade (later called Celsius) and thus
deﬁnes the limit of temperature decrease. This deﬁnes the lowest temperature that can be achieved
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 65 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 65
and the absolute zero temperature, and thus deﬁnes a fundamental scale called the absolute
temperature or Kelvin scale. The absolute temperature scale has the zero value at a temperature
corresponding to À273.168C and the unit of temperature is the same as that of arbitrary Celsius
scale, that is, T (degrees absolute or Kelvin) ¼T (degree Celsius) þ273.16. A combination of
Boyles law and Charles law leads to the equation of state for an ideal gas given by PV¼nRT
where P is the pressure, V is the volume, n is the number of moles of gas, R is the gas constant, and T
is the temperature in the absolute temperature scale.
2.4 FIRST LAW OF THERMODYNAMICS
Count Rumford (1753–1814) ﬁrst reported the production of heat during the boring of canons, and
thus established the linkage between heat and mechanical work. Joule quantiﬁed the linkage by
measuring the temperature rise in a known amount of water inside of a container as a function of
work done in rotating the paddle. He determined that 1 cal. was equivalent to 4.19 J. Various forms
of energy input to water in a container led to the same relationship. The various forms of work
included mechanical work, surface work, magnetic work, and electrical work. Thus, energy merely
changes form and is not lost. This general observation led to the ﬁrst law of thermodynamics, which
states, ‘‘Energy can neither be created nor destroyed but merely changes form.’’
Work done by the system could be conﬁgurational or dissipative in nature. The products of an
intensive variable and the change in an extensive variable that determines the conﬁguration of
the system give the conﬁgurational work. Examples of conﬁgurational work are (1) pressure volume
work, P dV, where P is the pressure and dV is the change in volume, (2) electrical work, E dQ,
where E is the electromotive force and dQ is the change in electric charge, (3) magnetic work, B dM,
where B is the magnetic ﬁeld and dM is the change in the magnetization of the material, and (4)
surface work, g dA, where g is the surface energy and dA is the change in surface area. These can be
done on or by the system. Dissipative work is the irreversible component of work involved. The
dissipative work is zero in reversible processes and has a ﬁnite value in irreversible processes.
Examples of dissipative work are work done to overcome friction and electrical heating. Reversing
the direction of motion or current does not affect the sign of work.
2.4.1 INTERNAL ENERGY
The implication of the ﬁrst law can now be taken into account by considering an interaction of
the system that involves exchanges of heat and work, which takes it from one thermodynamic state
to another. Each state is characterized by a set of independent thermodynamic properties and
functions that depend only on the state. One such function is internal energy, U, which is the sum
of kinetic and potential energies of all parts of the system. In an alloy, these parts are the atoms that
constitute the alloy. The ﬁrst law of thermodynamics thus would require that the internal energy of a
system isolated (no exchange of heat or no work) from its surrounding, remains constant. In a
crystalline solid, the potential energy is a sum of the energy of individual atoms that arise from their
position in the crystalline lattice (with its value measured in relation to the energy of an isolated
atom that serves as the reference state). This potential energy is manifest as the energy of the bonds
that have formed between the atoms. Kinetic energy arises from the different modes of vibration of
the atoms about their lattice position.
When energy is supplied to the systemfromthe surroundings under a condition of constant volume,
no pressure volume (P dV) work is performed by the systemand all the supplied energy goes to increase
the internal energy, with a consequent increase in the temperature of the system. Under conditions of
constant pressure, the energy supplied to the system goes to change both U and PV terms.
Adiabatic process is a process in which no heat exchange between the system and surroundings
is allowed across the boundary. The work done in an adiabatic process between two equilibrium
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 66 1.11.2008 2:41am Compositor Name: JGanesan
66 Handbook of Thermal Process Modeling of Steels
states is a constant and independent of the path. The work done on the system fully goes to increase
the internal energy of the system.
When the process is not adiabatic, exchange of heat is allowed between the system and
surroundings. In these cases,
Heat supplied to the system, dq ¼ dU þw (2:1)
where
dU is the change in internal energy
w is the work done by the system
Work done by (or on) the system includes pressure–volume, magnetic, electrical, and other work
performed by (or on) the system. Often PV work is the only one involved, and therefore Equation
2.1 is often written as
Heat supplied to the system, dq ¼ dU þPdV (2:2)
2.4.2 ENTHALPY, H
The amount of energy required to increase the temperature by 18C is referred to as the heat capacity
of the system. Heat capacity under constant volume is given by
C
V
¼ (dU=dT) (2:3)
When a process is carried out at constant pressure,
Work done, w ¼
ð
2
1
P dV ¼ P V
2
ÀV
1
ð Þ ¼ PV
2
ÀPV
1
(2:4)
dq ¼ dU þP dV ¼ U
2
ÀU
1
ð Þ þ PV
2
ÀPV
1
ð Þ
¼ U
2
þPV
2
ð Þ À U
1
þPV
1
ð Þ ¼ H
2
ÀH
1
ð Þ (2:5)
Heat capacity under conditions of constant pressure is given by
C
P
¼ (dq=dT)
P
¼ [d(U þPV)=dT]
P
¼ (dH=dT)
P
(2:6)
The quantity (UþPV) is referred to as the enthalpy or heat content, H, of the system. As the name
suggests, it refers to the quantity of heat input to the system to bring it from its standard state to its
current state. The function H is the ﬁrst of the auxiliary functions introduced.
C
P
¼ (dq=dT)
P
¼ [d(U þPV)=dT]
P
¼ (dH=dT)
P
(2:7)
C
P
ÀC
V
¼
dH
dT
P
À
dU
dT
V
¼ P
dV
dT
P
(2:8)
For an ideal gas, C
P
ÀC
V
¼ P
dV
dT
P
¼ R (2:9)
Change in enthalpy is then given by the equation
DH ¼
ð
T
2
T
1
C
P
dT (2:10)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 67 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 67
The absolute value of enthalpy depends on the reference state. In general, one is interested in the
difference in enthalpy between states, and thus the choice of the reference energy or enthalpy level is
not very critical. The reference state is by convention deﬁned by the material in its most stable state
at 298 K and 1 atmospheric pressure. By deﬁnition H at 298 K and 1 atmospheric pressure for all
elements in their stable form is deﬁned as zero. This allows one to calculate the enthalpy values of
elements and compounds under any other condition once the C
P
values at different temperatures are
known. This will be illustrated in Section 2.9.
2.5 SECOND LAW OF THERMODYNAMICS
The second law of thermodynamics is based on the observations of many natural processes that one
recognizes in daily life. It is observed that heat always ﬂows from a hightemperature region to a
lowtemperature region, gas ﬂows from a highpressure region to a lowpressure region, and so on.
An irreversible process is one in which the change in properties occurs in ﬁnite steps accompanied
by an irrecoverable energy loss through dissipative work. In these natural processes, the changes in
the system take place in an irreversible manner, with the system not in equilibrium with its
surroundings at each instant along the path of change. The natural processes take the system from
a nonequilibrium state to a state of equilibrium in a spontaneous manner and therefore, the natural
processes are also referred to as spontaneous processes.
Centuries of human experience and the knowledge of this experience led to the following
postulates or statements by Kelvin–Plank and Clausius. Clausius postulate states that it is impossible
to create a machine that operates in cycles and has the sole effect of transferring heat from a cooler
body to a hotter body. Kelvin–Plank postulate states that it is impossible to create a machine that
operates in cycles and has the sole effect of extracting heat from a hot reservoir and converting it to a
cold reservoir and converting it all into work. The work of Sadi Carnot on the efﬁciencies of heat it
engines operating in cycles led to a treatise in 1824 that provided a proof of these statements and led
to the formulation of the second law of thermodynamics. Carnot considered an engine operating
cycle that consisted of isothermal expansion, adiabatic expansion, isothermal compression, and
adiabatic compression. Such a cycle is referred to as a Carnot cycle and an engine operating in this
manner is referred to as a Carnot engine. He showed that the efﬁciency of the engine converting heat to
work depends only on the temperatures of the heat source and heat sink. No engine operating between
two reservoirs can be more efﬁcient than the Carnot engine operating between those reservoirs.
A reversible process goes from one equilibrium state to another through a series of inﬁnitesimal
changes in the system under equilibrium with the surroundings at each instant along the path of
change. A system that goes through a reversible process that takes it from state 1 to state 2 through a
reversible path a can now be considered. The process 1
a
!
2 involves absorption of heat q
1a2
and the
system performs work w
1a2
, and the change in the internal energy is u
12
. If the process occurs
reversibly, the system pressure and temperature are close to that of the surrounding, and therefore
the work
Ð
p dV performed will be a maximum and the heat absorbed will be a maximum. If
the process of change from state 2 to 1 occurs reversibly through path b (2
b
!
1), the work done by the
system, and therefore the heat absorbed q
2b1
, will be –q
1a2
, the work performed will be Àw
1a2
, and
the change in internal energy is Àu
12
. For the whole cyclic process, Du ¼0. In addition, the integral
Þ
(dq=T) ¼ 0. A new state function for the system, which depends only on the state and not the path
can now be deﬁned. This new function called entropy S of a system is deﬁned by a difference
function DS ¼
Ð
(dq=T). Clausius showed using Carnot Cycles that
Þ
(dq=T) 0 where ¼sign
corresponds to a reversible process, and <sign corresponds to an irreversible process. This inequal
ity is known as Clausius inequality.
This can be illustrated using a simple system. Now let us consider the heat exchange that occurs
between two regions inside an isolated system (Figure 2.1). This isolated system consists of two
regions at temperatures T
1
and T
2
where T
2
>T
1
. The system is clearly not at equilibrium, and the
heat ﬂow must occur from region 2 to region 1.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 68 1.11.2008 2:41am Compositor Name: JGanesan
68 Handbook of Thermal Process Modeling of Steels
In this isolated system, the heat ﬂow of dQ from the hightemperature region to the low
temperature region (a natural process that one expects to occur) increases the entropy could be
clearly seen.
dS ¼ [dq=T
2
Àdq=T
1
] > 0 (2:11)
In the limit, that is, when T
2
approaches T
1
the process occurs with dS ¼0. This leads to the
statement of the second law of thermodynamics.
In an isolated system (no exchange of energy or mass with surroundings), entropy always
increases for any spontaneous process and it is zero for reversible processes.
[dS]
Isolated system
! 0 (2:12)
The entropy, S, is deﬁned using the relation
DS ¼ (DH=T) (2:13)
where DS and DH are the change in entropy and enthalpy, respectively, for a process that is
occurring at temperature, T. DH can be obtained from the heat capacity at constant pressure, C
P
,
using the relation DH¼C
P
dT. The absolute value of entropy is obtained using the third law of
thermodynamics.
2.6 THIRD LAW OF THERMODYNAMICS
Nernst postulated that the heat capacity of all substances goes to zero as absolute zero temperature is
approached. Experimental observations and quantum mechanics show that the heat capacity goes to
zero as T approaches zero, in such a manner that C
P
=T does not diverge. Nernst also postulated that
any reaction between substances at equilibrium must always occur with no change in entropy.
Planck postulated that Nernst’s second postulate would be valid if the entropy of the system
approaches zero as T approaches zero. This is referred to as the third law of thermodynamics,
which is often stated in the following form.
The entropy at absolute zero temperature (0 K) is zero in perfectly crystalline materials.
Thus,
S
T
¼
ð
T
0
C
P
T
dT (2:14)
The stable state for an isolated system is thus deﬁned by a maximum in S. In many metallurgical
systems and other material systems, conditions of constant temperature and pressure are often
Region 2 Region 1
Surrounding
dQ
T
2
T
1
FIGURE 2.1 Heat transfer and entropy change in an isolated system.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 69 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 69
encountered. Often, one would like to know which phase or combination of phases will be stable
under a given T and P for a given alloy or material system. The concept of Gibb’s free energy
provides the answer to such a question.
2.7 CONCEPT OF GIBBS AND HELMHOLTZ FREE ENERGIES
Let us consider a system under constant pressure and temperature conditions. This system can
exchange energy with its surroundings. If you consider this system and its surroundings together as
a new large system, the new system can be treated as an isolated system. Then,
dS þdS
Surroundings
! 0 (2:15)
If the heat transferred from the surroundings to the system is dQ and the enthalpy change for the
surroundings at temperature, T, is
dH
Surroundings
¼ Àdq ¼ ÀdH
System
(2:16)
(dH
System
is simply written as ÀdH from now on)
dS
Surroundings
¼ À
dH
T
(2:17)
dS þdS
Surroundings
¼ dS À
dH
T
! 0 (2:18)
dH ÀTdS 0 (2:19)
d(H ÀTS)
T,P
0 (2:20)
The term (HÀTS) is denoted by the symbol G, called the Gibbs free energy. G is a minimum for a
system under a thermodynamically stable condition at constant T and P. In the case of condensed
matter, constant T and P conditions are often dealt with, and therefore Gibbs free energy is used very
frequently in materials processing.
Another deﬁned or auxiliary function of importance is Helmholtz free energy, F. For a system
operating under constant volume and temperature conditions,
dS
Surroundings
¼
Àdq
T
¼ À
dU
T
(2:21)
dS þdS
Surroundings
¼ dS À
dU
T
! 0 (2:22)
dU ÀTdS 0 (2:23)
d(U ÀTS)
T,P
0 (2:24)
The term (UÀTS) is deﬁned as Helmholtz free energy, A and is useful in deﬁning the system
stability of systems operating under constant volume and temperature. E, H, A, and G are different
thermodynamic potentials. Depending on the conditions of the system, Gibbs free energy (G),
Helmholtz free energy (A), enthalpy (H), or entropy (S) is used to determine the stability of the
system. Another frequently used auxiliary function is chemical potential, which will be introduced
later in this chapter. The following is a summary of stability criteria under different situations.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 70 1.11.2008 2:41am Compositor Name: JGanesan
70 Handbook of Thermal Process Modeling of Steels
Condition Stability Criteria
Constant T, P, n
i
, E, H, . . . G minimum
Constant T, V, n
i
, E, H, . . . A minimum
Constant S, P, n
i
, E, H. . . H minimum
Constant S, V, n
i
, E, H, . . . U minimum
Constant V, U, n
i
, E, H, . . . S maximum
In the case of both open and closed systems, the following relations are often used.
dU ¼ T dS ÀP dV þ
X
i
m
i
dn
i
þB dM þE dQ þÁ Á Á (2:25)
dH ¼ T dS þV dP þ
X
i
m
i
dn
i
þB dM þE dQ þÁ Á Á (2:26)
dA ¼ ÀS dT ÀP dV þ
X
i
m
i
dn
i
þB dM þE dQ þÁ Á Á (2:27)
dG ¼ V dP ÀS dT þ
X
i
m
i
dn
i
þB dM þE dQ þÁ Á Á (2:28)
The term S
i
m
i
dn
i
includes the effect due to changes in the number of moles of the various species.
The term B dM is the contribution from the magnetic work done on the system and E dQ is the
contribution from the electrical work done on the system. In most problems, these terms may not be
relevant and thus often omitted.
From Equations 2.25 through 2.28, the following thermodynamic relationships deﬁning P, T, V,
and S can be obtained.
T ¼
@U
@S
V,composition
¼
@H
@S
P,composition
(2:29)
P ¼ À
@U
@V
S,composition
¼ À
@F
@V
T,composition
(2:30)
V ¼
@H
@P
S,composition
¼
@G
@P
T,composition
(2:31)
S ¼ À
@F
@T
V,composition
¼ À
@G
@T
P,composition
(2:32)
Maxwell’s equations: Maxwell’s relations are a set of large number of equations that provide
interrelationships between thermodynamic parameters. These equations are obtained using the
characteristic that thermodynamic variables are state functions and are exact variables. Let us
consider a thermodynamic property Z in a closed system where composition is kept constant. Z is
an exact state variable, and let Z be a function of two other state variables x and y.
Z ¼ F(x, y) (2:33)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 71 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 71
dZ ¼
@Z
@x
y
dx þ
@Z
@y
x
dy ¼ L dx þM dy (2:34)
For this form of the equation, it can be shown that
@L
@y
x
¼
@M
@x
y
(2:35)
Applying this result to Equations 2.25 through 2.28, for a system with ﬁxed composition, electric,
magnetic, and other ﬁelds, the following relationships commonly referred to as Maxwell’s relations
can be obtained. The usefulness of these relationships lies in the fact that they contain some easily
measurable quantities.
@T
@V
S
¼ À
@P
@S
V
(2:36)
@T
@P
S
¼
@V
@S
P
(2:37)
@S
@V
T
¼
@P
@T
V
(2:38)
@S
@P
T
¼ À
@V
@T
P
(2:39)
Another equation that provides a similar convenience is Gibbs–Helmholtz relationship applicable to
a closed system of ﬁxed composition.
d G=T ð Þ
@T
¼ À
H
T
2
(2:40)
d DG=T ð Þ
@T
¼ À
DH
T
2
(2:41)
2.8 EVALUATION OF G
In most thermal processing situations, constant pressure and temperature conditions are dealt with,
and thus G deﬁnes the stability. To evaluate G, one needs
1. Data of the heat capacity under constant pressure, C
P
, for the different phases of interest
2. Heat of formation data for the different compounds=phases in the system
This data are used to evaluate H and S, and therefore G using the following equation.
G ¼ H ÀTS (2:42)
For pure elements, when there is no phase change involved
H
T
¼ H
298
þ
ð
T
298
C
P
dT ¼ 0 þ
ð
T
298
C
P
dT ¼
ð
T
298
C
P
dT (2:43)
S
T
¼
ð
T
0
[C
P
=T] dT (2:44)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 72 1.11.2008 2:41am Compositor Name: JGanesan
72 Handbook of Thermal Process Modeling of Steels
If a phase change occurs between 298 K and T, the equation is modiﬁed as follows:
H
T
¼ H
298
þ
ð
T
trf
298
C
P
dT þDH
trf
þ
ð
T
T
trf
C
P
dT (2:45)
S
T
¼
ð
T
trf
0
[C
P
=T] dT þDS
trf
þ
ð
T
T
trf
(C
P
=T) dT (2:46)
where
DH
trf
is the enthalpy of transformation
DS
trf
is the entropy of transformation
T
trf
is the transformation temperature
In a similar way, one could evaluate the values of G for different multicomponent phases or
compounds using their heat of formation data and heat capacity data. An illustration of how the H,
S, and G values of a compound can be obtained is given below. Consider the reaction
A þB !AB (2:47)
At 298 K,
DH
formation
of AB ¼ H
AB
ÀH
A
ÀH
B
¼ H
AB
À0 À0 ¼ H
AB
(2:48)
That is, enthalpy of the compound AB at 298 K¼heat of formation of AB at 298 K
Now at a temperature of T,
H
T
AB
¼ DH
formation
þ
ð
T
trf
298
C
P
dT (2:49)
Similarly, one can also calculate
S
T
AB
¼ DS
formation
þS
A
þS
B
þ
ð
T
298
C
P
=T ð Þ dT (2:50)
Measurement of C
P
is made using calorimeters of various types. A common modern instrument is a
differential scanning calorimeter that requires only a small amount of the material, about a few tens
of milligrams. There are several thermodynamic data sources that provide C
P
, G, H, and S data for
various elements and compounds at various temperatures. Some of these sources are provided at the
end of the chapter.
2.9 ESTIMATION OF GIBBS FREE ENERGY OF A PHASE
In a singlecomponent system, composition does not vary, and the Gibb’s free energy of different
phases in the system is normally a function of P and T. Starting from standard temperature and
pressure (STP) conditions, G is evaluated using the relationship dG¼V dPÀS dT. If electrical or
magnetic potentials are involved, G is also a function of electric and magnetic ﬁelds. However, these
will not be considered here.
In systems containing two or higher number of components, G for a given phase in the system is
a function of P, T, and composition. Formation of a solid solution phase or a compound phase
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 73 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 73
involves atomic level mixing of various component elements. The free energy of the phase per mole
is given by the free energy associated with the elements involved before mixing and the change in
free energy associated with the mixing process.
G ¼ G
before mixing
þDG
mix
¼
X
i
X
i
G
i
þDG
mixing
(2:51)
DG
mix
¼ H
after mixing
ÀH
before mixing
À Á
ÀT S
after mixing
ÀS
before mixing
À Á
¼ DH
mix
ÀTDS
mix
(2:52)
The values of G
i
for each element i is obtained as described earlier at a given temperature and
pressure. The evaluation of DG
mix
requires the estimations of DH
mix
and DS
mix
, which are not
trivial. Simplifying assumptions are often made that make these easier.
One assumption that may be made is that mixing of the atoms occurs in an ideal manner, that is,
the behavior of atoms of each individual species in solution does not change from that in the original
pure material of that species. In such a case the kinetic and potential energies associated with each of
these atoms remains the same. It can also be assumed that the volume change and PV work is
negligible, resulting in no change in internal energy and enthalpy after mixing. In such ideal
systems, DH
mix
is therefore zero. However, there is a large entropy change during ideal mixing.
In the pure state, there is only one unique way of arranging the atoms in the pure lattices of
individual species, and the randomness in spatial arrangement is minimal or zero. The concept of
entropy from a statistical mechanics approach is now introduced and, described in more detail in the
next section. Thermodynamic properties are concerned with the average behavior of a large number
of particles. These particles have an associated distribution in their individual properties such as
kinetic energies, velocities, spatial positions, and so on. Let us consider the binary solid solution of
two elements A and B. In a perfect lattice containing X
A
mol of A and X
B
mol of B in solid solution,
6.023Â10
23
X
A
and 6.023Â10
23
X
B
atoms are distributed among 6.023Â10
23
(X
A
þX
B
) sites. There
are numerous possible ways of distributing the atoms in these sites. A given conﬁguration can be
reached in a number of ways, V, and the number of ways V by which a given conﬁguration is
reached is a measure of the statistical probability of obtaining the conﬁguration. A detailed analysis
would show that the most probable distribution has a probability that approaches unity, and the
property of the system corresponds to this distribution.
When the value of V is larger, the randomness of the system is higher. Statistical mechanics
provides a linkage between the increased randomness measured by the parameter V and the
entropy through the relation S ¼k ln V. This is discussed further in the section on statistical
thermodynamics.
In the above discussion, the randomness associated with only the spatial distribution of atoms have
been considered. However, the particles have associated energy or other characteristics with a
corresponding randomness in that characteristic. Thus, v is a measure of all possible randomness
and given by
v ¼ pv
i
¼ v
thermal
Âv
configurational
ÂÁ Á Á (2:53)
where
v
thermal
is the number of different ways the energy can be distributed among the different
vibrational modes
v
conﬁgurational
is the number of different ways of arranging the atoms in different lattice sites
The randomness associated with other characteristics that change on mixing can be included as
appropriate.
If only the spatial distribution changed after the atomic level mixing to form a particular phase,
then S before mixing is given by
S ¼ k ln v
configurational
¼ k ln 1 ¼ 0 (2:54)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 74 1.11.2008 2:41am Compositor Name: JGanesan
74 Handbook of Thermal Process Modeling of Steels
After mixing to form a random solid solution,
S ¼ k ln N
A
þN
B
ð Þ!=N
A
! N
B
! ½ (2:55)
where
N
A
is the number of atoms of A
N
B
is the number of atoms of B
N
A
þN
B
¼Na, Avogadro number
DS
mix
¼ k ln N
A
þN
B
ð Þ!N
A
! N
B
! ½ (2:56)
Using Stirling’s approximation (for large N, N! ¼N ln NÀN),
DS
mix
¼ k ln N
A
þN
B
ð Þ!=N
A
! N
B
! ½
¼ ÀN
a
k X
A
ln X
A
þX
B
ln X
B
½
¼ ÀR X
A
ln X
A
þX
B
ln X
B
½ (2:57)
In ideal solutions, DH
mix
¼0, and the A and B atoms are distributed randomly in the lattice sites.
DG
mix
¼ DH
mix
ÀTDS
mix
¼ ÀT DS
mix
ð Þ
random
(2:58)
DS
mix
ð Þ
random
or DS
mix
ð Þ
ideal
¼ ÀR X
A
ln X
A
þX
B
ln X
B
ð Þ (2:59)
DG
mix,ideal
¼ RT X
A
ln X
A
þX
B
ln X
B
ð Þ (2:60)
G ¼ X
A
G
A
þX
B
G
B
þRT X
A
ln X
A
þX
B
ln X
B
ð Þ (2:61)
Figures 2.2 show the plots of G as a function of composition. A very important feature of this plot is
that the slope of the curve at X
B
¼0 and X
B
¼1 is À1 and þ1, respectively. The implication of
this is that an addition of trace amounts of impurity decreases the Gibbs free energy at an inﬁnite
rate, and thus contamination is a natural or spontaneous process. It is thus impossible to have the
0 1
X
B
I
n
c
r
e
a
s
i
n
g
l
y
n
e
g
a
t
i
v
e
0
G
A
ΔG
mix
G
B
FIGURE 2.2 Free energy versus composition curve for a given solid solution phase in a binary system
containing elements A and B.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 75 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 75
purest form of any material. Trying to obtain materials of increasingly higher purity comes at an
exponentially increasing cost as is commonly realized. This is particularly relevant in thermal
processing.
2.10 CHEMICAL POTENTIAL AND ACTIVITY
An important and valuable concept is the chemical potential of an atomic species i, m
i
, in solution. It
is the rate at which the Gibbs free energy of a solution system, G
0
, is increased when the atoms of
that species are added while maintaining the P, T, and other compositional parameters constant.
Here one can assume the system to be so large that the addition of 1 mol of a species does not
essentially change the composition of the system. The chemical potential m
i
is given by the relation:
m
i
¼ dG
0
=dn
i
ð Þ
T,P,n
j6¼i
(2:62)
with T, P, and the number of moles of all other species remaining constant.
One can see that the chemical potential of a species is the Gibbs free energy associated with one
mole of the species in solution. If one considers two systems I and II, where the chemical potential
of the species is greater in system I than in II, then the species would prefer to be in system II, and
hence there will be a tendency to move from system I to II. The transfer will occur when the kinetics
becomes favorable for the transfer. If there is a chemical potential difference for a species between
one region and another within an alloy system, then transport of the species will occur from the
higher chemical potential region to the lower chemical potential region. The transport will occur
through vapor phase transport, solidstate diffusion, or liquidstate diffusion, depending on the
geometry, nature of phases present in the system, and the temperature.
As the chemical potential of a species is the Gibbs free energy associated with one of the moles
of the species in solution, the Gibbs free energy of a binary solution can therefore be written in terms
of the chemical potentials of A and B atoms in solution:
G ¼ X
A
m
A
þX
B
m
B
(2:63)
Comparing with the equation for an ideal solution,
G
ideal
¼ X
A
G
A
þX
B
G
B
þRT X
A
ln X
A
þX
B
ln X
B
ð Þ (2:64)
it can be seen that
m
A
¼ G
A
þRT ln X
A
(2:65)
m
B
¼ G
B
þRT ln X
B
(2:66)
Figure 2.3 illustrates the determination of the chemical potential of elements A and B in an ideal
solution containing A and B atoms from the Gibbs free energy versus composition curve for the
solution phase.
The condition of ideal solution can be approached in some systems, very often this is not the
case. On mixing, the behavior of atoms of each individual species in solution changes from that in
the original pure material of that species. In some cases, a given species may behave as if it is
present in a higher concentration than is the case. Thus, the atomic species has a relatively higher or
lower activity due to nonideal behavior in solution. This is analogous to a situation that can be seen
in daily life. Association with certain individuals enhances our productivity and it may not be the
case with others. Thus, in nonideal solutions, the distribution of atoms in the lattice will not be
random. This is because of the preference of atoms to have certain species as their neighbor. When
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 76 1.11.2008 2:41am Compositor Name: JGanesan
76 Handbook of Thermal Process Modeling of Steels
such preferences are relatively weak, that is, (DH
mix
) has a small positive or negative value, the
solution may be assumed to have a random solute distribution and DS
mix
%(DS
mix
)
random
. This
model of solid solution is referred to as a quasichemical model and such solutions are referred to as
regular solutions.
In a binary regular solution,
DG
mix,ideal
¼ VX
A
X
B
þRT X
A
ln X
A
þX
B
ln X
B
ð Þ (2:67)
where
V¼X
A
X
B
Z«, z is the number of nearest neighbors
« ¼1=2 («
AB
–(«
AA
þ«
BB
)=2)
«
AB
, «
AA,
and «
BB
are AB, AA, and BB bond energies, respectively
Thus for a regular solution,
G ¼ X
A
G
A
þX
B
G
B
þVX
A
X
B
þRT X
A
ln X
A
þX
B
ln X
B
ð Þ (2:68)
It may be shown that
G ¼ X
A
G
A
þVX
A
X
B
þRT ln X
A
ð Þ þX
B
G
A
þVX
A
X
B
þRT ln X
B
ð Þ
¼ X
A
m
A
þX
B
m
B
(2:69)
where
m
A
¼ G
A
þVX
A
X
B
þRT ln X
A
m
B
¼ G
A
þVX
A
X
B
þRT ln X
B
m
A
and m
B
could be written in a form similar to that in an ideal solution, by deﬁning a new
parameter referred to as the chemical activity of species i, a
i
.
m
A
¼ G
A
þVX
A
X
B
þRT ln X
A
¼ G
A
þRT ln a
A
(2:70)
m
B
¼ G
B
þVX
A
X
B
þRT ln X
B
¼ G
B
þRT ln a
B
(2:71)
I
n
c
r
e
s
i
n
g
l
y
n
e
g
a
t
i
v
e
−
R
T
I
n
X
A
−
R
T
I
n
X
B
0 1
X
B
0
G
A
G
B
ΔG
mix
m
A
m
B
FIGURE 2.3 Determining the chemical potential of the various species in a solution phase from the free
energy curve for that phase.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 77 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 77
Figure 2.4 illustrates the determination of chemical potential of elements A and B in a nonideal
solution containing A and B atoms from the Gibbs free energy versus composition curve for the
solution phase.
The variation of activity with concentration will follow the same trend as that of chemical
potential. Activity values usually will vary from zero to one if the pure species in its standard state is
taken to be the reference. However, in slagmetal reactions or metal reﬁning operations, it is often
convenient to use a dilute solution as a reference state when considering activities of a minor or
solute species. In such cases, the activity values can take values from zero to well over one.
The ratio a
i
=X
i
is referred to as the activity coefﬁcient of species i, g
i
. As can be seen in the
following equations, it is a measure of the nonideal behavior in solution.
ln
a
A
X
A
¼
V
RT
1 ÀX
A
ð Þ
2
(2:72)
ln
a
B
X
B
¼
V
RT
1 ÀX
B
ð Þ
2
(2:73)
For a dilute solution of B in A, with X
B
! 0,
g
A
¼
a
A
X
A
¼ 1 Rault’s law (2:74)
g
B
¼
a
B
X
B
¼ constant Henry’s law (2:75)
When preferences for certain species are strong, that is, (DH
mix
) has a large positive or negative
value, the solution will not be anywhere near random, and a statistical approach is needed to
determine the entropy and enthalpy of mixing, which is beyond the scope of this chapter.
Another important equation that is very valuable in experimental work is the Gibbs–Duhem
equation. Consider an extensive thermodynamic property G
0
of a solution, which is a function of
T, P, and the number of moles of components n
1
, n
2
, n
3
,. . . . At constant T and P,
I
n
c
r
e
a
s
i
n
g
l
y
n
e
g
a
t
i
v
e
0
G
A
ΔG
mix
X
B
G
B
m
A
m
B
0 1
−
R
T
I
n
a
A
−
R
T
I
n
a
B
FIGURE 2.4 Determining the chemical potential and activity of the various species in a nonideal solution
phase from the free energy curve for that phase.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 78 1.11.2008 2:41am Compositor Name: JGanesan
78 Handbook of Thermal Process Modeling of Steels
dG
0
¼
X
i
n
i
(VdP ÀSdT) þ
X
i
m
i
dn
i
¼
X
i
m
i
dn
i
(2:76)
G
0
¼
X
i
m
i
n
i
(2:77)
dG
0
¼
X
i
m
i
dn
i
þ
X
i
n
i
dm
i
(2:78)
X
i
n
i
dm
i
¼ 0 (2:79)
This equation is referred to as Gibbs–Duhem equation. The interdependence of chemical potentials
of various species in solution allows the determination of the chemical potential of one species if the
chemical potentials of other species are known. Using the relationship between chemical potential
and activity, this equation can be written in terms of chemical activity as given below.
X
i
n
i
d( ln a
i
) ¼ 0 (2:80)
2.11 GIBBS FREE ENERGY CURVES AND PHASE DIAGRAMS
The stability of a given phase depends on the Gibbs free energy value relative to other possible
phases in a system under a given set of conditions. In order to determine the phases that will be
stable for a given composition of the system at a given temperature and pressure condition, free
energy curves of all possible phases are drawn across the composition range. Figure 2.5 shows a plot
of free energy versus composition for two possible phases, solid and liquid, at a given temperature
and pressure. If the curve for one of the phases is below that of the other for all compositions, the
curves will not intersect and this phase would be stable at all compositions. If the curves intersect as
is the case shown in Figure 2.5, one can draw a common tangent line to the two curves. Outside of
0 1 X
B
G
Liquid
G
Solid
b
d
I
n
c
r
e
a
s
i
n
g
l
y
n
e
g
a
t
i
v
e
0
c
a
P
Q
FIGURE 2.5 Determining the phases present at a given temperature, pressure, and composition using the
plots of free energy versus composition of various possible phases in a material system.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 79 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 79
the common tangent points, the solid or the liquid phase has the lower G value and will be stable.
However, between the common tangent points, a mixture of the two phases has lower Gibbs free
energy than either of the phases. Therefore, the two phases coexist and the composition of each
phase would correspond to the two common tangent points. By drawing such diagrams over a range
of T and P, the regions of stability of various possible phases can be determined. The same approach
could be used for determining ternary and other multicomponent diagrams.
As the phase diagrams are obtained from free energy curves of phases, thermodynamic
information could be obtained from the phase diagrams.
2.12 GIBBS PHASE RULE
If all the phases, P present in a system are in equilibrium, then the chemical potential of each species
must be the same in all the phases.
m
a
A
¼ m
b
A
¼ m
g
A
¼ Á Á Á !(P À1) independent equations in this line
m
a
B
¼ m
b
B
¼ m
g
B
¼ Á Á Á
. . .
. . .
m
a
N
¼ m
b
N
¼ m
g
N
¼ Á Á Á
C lines: There are C components and therefore C lines. In each line, there are (PÀ1) equations as
there are P phases. Therefore, there are C (PÀ1) equations.
Number of variables ¼(CÀ1) components ÂP phases þNumber of intensive variables f.
If the intensive variables considered are only T and P, then f¼2.
If the intensive variable considered is only T and P is already selected and constant, then f¼1.
The difference between the number of variables and the available independent equation is the degree
of freedom. This refers to a number of variables that can be independently speciﬁed.
Unknowns No: of equations Degrees of freedom
# # #
(C À1)P þf ¼ (P À1)C þ F
CP ÀP þf ¼ PC ÀC þF
P þF ¼ C þf
(2:81)
2.13 KINETIC THEORY
Kinetic theory considers the movements of individual particles in a system and predicts the
properties by applying the laws of classical mechanics to their motion. It assumes that the system
is made up of a large number of molecules uniformly distributed within the walls of the system.
Average separation between molecules is large compared to the molecular dimension. Motions of
molecules are random and involve elastic collisions with other molecules and the walls of the
system or container. No other interaction exists between the molecules. Thus, the kinetic theory is
more appropriate for gases at low pressure, but is often used in the consideration of other systems. It
provides an insight into concepts such as pressure, temperature, internal energy, and speciﬁc heat.
The collision of molecules with the surface of the walls leads to a change in the momentum. The
change in momentum, dp, produced by all the molecules colliding on the surface of the container is
given by
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 80 1.11.2008 2:41am Compositor Name: JGanesan
80 Handbook of Thermal Process Modeling of Steels
dp ¼
1
3
nmv
2
dAdt (2:82)
where
n is the mean number of molecules per unit volume
m is the mass of the molecule
v is the mean velocity normal to the surface
The mean velocity v is estimated from Maxwell’s velocity distribution function. The average force
unit area acting on the surface, that is, pressure, is then given by
P ¼
dp
dAÁ dt
¼
1
3
nmv
2
(2:83)
P ¼
2
3
1
2
nmv
2
¼
2
3
1
2
N
V
mv
2
¼
2
3
1
2
Nmv
2
1
V
(2:84)
where
N is the number of molecules in the system
V is the volume
Thus,
PV ¼
2
3
1
2
Nmv
2
(2:85)
Comparing with the ideal gas law, PV¼NkT,
NkT ¼
1
3
Nmv
2
À Á
(2:86)
3
2
kT ¼
1
2
mv
2
À Á
(2:87)
The temperature is thus shown to be proportional to the average kinetic energy of molecules. This
average kinetic energy of molecules can be equally partitioned between motions associated with the
three components of velocity. The mean velocity can be expressed in terms of its components in
the x, y, and z directions as
v
2
¼ v
2
x
þv
2
y
þv
2
z
(2:88)
As the motions of molecules are random, the components in the three directions are equal.
Thus,
1
2
mv
2
x
¼
1
6
mv
2
À Á
¼
1
2
kT (2:89)
Each component of velocity represents a degree of freedom, and thus each degree of freedom is
associated with average energy of 1=2kT. While this result was derived for the case of ideal gas, the
result more generally applies to all solids, liquids, and gases. Kinetic theory can be used to predict
the various transport properties such as diffusivity, viscosity, and conductivity. However due to the
limited scope of this chapter, they are not considered here.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 81 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 81
2.14 STATISTICAL THERMODYNAMICS
A thermodynamic system is made up of an assembly of particles that may be atoms, molecules, or
microconstituents. Statistical thermodynamics deals with the prediction and interpretation of macro
scopic properties in terms of the properties of atoms, molecules, or microconstituents in the system.
Unlike kinetic theory, it however ignores detailed considerations of particles and applies statistical
considerations to a collection of particles to ﬁnd the most probable distribution of a large number of
particles among the very different energy states of the system.
A thermodynamic system consists of submicroscopic particles in enormous numbers. These
particles have an associated and constantly changing distribution of their individual properties such
as kinetic energies, velocities, and spatial positions. Central to statistical thermodynamics is the
prediction of the most probable distribution of these particles among the various energy states and
the prediction of properties based on these predictions.
Deﬁnitions of a few important concepts are ﬁrst needed. The ﬁrst term to be deﬁned is assembly.
Assembly is a term used to refer to a number N of identical entities. The macrostate of a system is
characterized by a deﬁned set of occupation numbers N
1
, N
2
, N
3
, . . . , N
i
associated with energy
levels «
1
, «
2
, «
3
, . . . , «
i
, that is, N
1
particles occupy energy level «
1
, N
2
particles occupy energy level
«
2
, and so on. Conﬁguration and distribution are other alternate terms used for the macrostate. Each
macrostate is associated with a speciﬁc total energy for the system.
In each energy level, there could be one or more energy states, that is, the energy level is
degenerate. Microstate speciﬁes the number of particles in each of the energy states of the system.
For each macrostate, there can be a large number of different microstates, and the number of
microstates that lead to the same macrostate is referred to as the thermodynamic probability of that
macrostate and can vary from zero to inﬁnity. True probability is a normalized value of thermo
dynamic probability, is normalized with the total number of all possible microstates for the system.
Its value can vary from zero to one.
When the particles of the system occupy the set of energy levels «
1
, «
2
, «
3
, . . . , «
I
, . . . , with
degeneracies g
1
, g
2
, g
3
, . . . , g
i
and occupation numbers N
1
, N
1
, N
3
, . . . , N
i
, . . . the number of ways
the particles can be distributed between the various energy states, that is, the number of microstates
for this macrostate, depends on the nature of statistics and the constraints on the system. Three
common statistics of interest in a thermodynamic system are
1. Maxwell–Boltzmann (MB) statistics: Particles are distinguishable, and there is no restric
tion on the number of particles in any given energy state.
The number of microstates in a distribution or macrostate D
V
MB
D
¼
N!
N
1
!N
2
! . . . N
i
! . . .
g
N
1
g
N
2
. . . g
N
i
. . . :
À Á
(2:90)
2. Fermi–Dirac (FD) statistics: Particles are indistinguishable and obey the Pauli exclusion
principle, that is, no more than one particle can occupy any given energy state.
The number of microstates in a distribution or macrostate D
V
FD
D
¼
g
1
!
g
1
ÀN
1
! ð ÞN
1
!
g
2
!
g
2
ÀN
2
! ð ÞN
2
!
Á Á Á ¼ P
i
g
i
!
g
i
ÀN
i
! ð ÞN
i
!
(2:91)
3. Bose–Einstein (BE) statistics: Particles are indistinguishable, and there are no restrictions.
The number of microstates in a distribution or macrostate D
V
BE
D
¼
g
1
þN
1
À1 ð Þ!
g
1
À1 ð Þ!N
1
!
g
2
þN
2
À1 ð Þ!
g
2
À1 ð Þ!N
2
!
Á Á Á ¼ P
i
g
i
þN
i
À1 ð Þ!
g
i
À1 ð Þ!N
i
!
(2:92)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 82 1.11.2008 2:41am Compositor Name: JGanesan
82 Handbook of Thermal Process Modeling of Steels
Particles obeying MB statistics subject to the constraints of constant N, U, and V could now be
considered. To ﬁnd the most probable distribution, the N
i
values corresponding to this most
probable distribution are obtained through the maximization of V
D
. If V
D
is a maximum, then
inﬁnitesimal changes in N
i
values will not change the value of V
D
. Thus,
X
i
@ ln V
D
@N
i
@N
i
¼ 0 (2:93)
with the constraints
X
n
j¼1
N
j
¼ N (2:94)
X
n
j¼1
N
j
«
j
¼ U (2:95)
where
N
j
is the number of particles at the energy level «
j
N is the total number of particles
U is the internal energy of the system
With these constraints, not all variables are independent. The Lagrangian multiplier method can be
used to solve this problem. Multiplying Equation 2.94 by a and Equation 2.95 by b, and subtracting
them from the equation above, gives
X
i
@ ln V
D
@N
i
Àa Àb«
i
@N
i
¼ 0 (2:96)
For each value of i, the bracketed term is zero.
@ ln V
D
@N
i
Àa Àb«
i
¼ 0 for all values of i (2:97)
For the MB distribution,
V
MB
D
¼
N!
N
1
!N
2
! . . . N
i
! . . .
g
N
1
g
N
2
. . . g
N
i
. . . :
À Á
ln V
MB
D
¼ N
1
þN
2
þÁ Á Á N
i
þÁ Á Á ð Þ ln N
1
þN
2
þÁ Á Á N
i
þÁ Á Á ð Þ
þ N
1
ln g
1
þN
2
ln g
2
þÁ Á Á N
i
ln g
i
þÁ Á Á ð Þ
À N
1
ln N
1
þN
2
ln N
2
þÁ Á Á N
i
ln N
i
þÁ Á Á ð Þ (2:98)
@ ln V
MB
D
@N
i
¼ ln N þ(N=N) þln g
i
Àln N
i
À N
i
=N
i
ð Þ
¼ ln Ng
i
=N
i
ð Þ ¼ a þb«
i
(2:99)
N
i
N
¼ g
i
e
Àa
e
Àb«
i
(2:100)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 83 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 83
X
i
N
i
=N ð Þ ¼ 1 ¼ e
Àa
X
g
i
e
Àb«
i
(2:101)
N
i
N
¼
g
i
e
Àb«
i
P
g
i
e
Àb«
i
¼
g
i
e
Àb«
i
Z
(2:102)
The denominator S
i
g
i
e
Àb«i
is called the partition function, Z. This important parameter depends on
temperature and on parameters that determine the energy levels and quantum states. From the
partition function Z, all the thermodynamic properties can be derived. The parameter b is equal to
(1=kT). Occupation numbers are fully determined by the temperature and volume. The set of
occupation numbers that maximize thermodynamic probability deﬁne the equilibrium state.
The internal energy is given by
U ¼
X
n
j¼1
N
j
«
j
¼ N
P
i
g
i
«
i
e
Àb«
i
P
g
i
e
Àb«
i
¼ N
P
i
g
i
«
i
e
Àb«
i
Z
(2:103)
U ¼ ÀN
@ ln Z
@b
V
¼
N
Z
kT
2
À Á
@ ln Z
@T
V
(2:104)
To have constant «
i
, it is required that the system has a constant volume. The system under
consideration has constant N, U, and V, and is an isolated system. For the constraints of constant
N, U, and V, the system approaches the macrostate, for which the number of microstates tends to be
maximum. Due to the large number of particles in the system, the number of microstates in the most
probable state is signiﬁcantly larger than the contribution from all other distributions. Thus, it is the
only term that makes the most dominant contribution to the total number of all possible microstates
for the system. As mentioned earlier in the discussion of classical thermodynamics, entropy tends
toward a maximum value as the system approaches equilibrium. This suggests the connection
between entropy and the number of microstates corresponding to the most probable state. It can be
shown from the expression for V
MB
D
that
ln V
MB
D
¼ n ln Z þ
U
kT
(2:105)
For constant V, the change in internal energy du ¼heat input dq
@ ln V
MB
D
¼
@U
@T
¼
@Q
kT
¼
TdS
kT
¼
dS
k
(2:106)
dS ¼ k@ ln V
MB
D
(2:107)
S ¼ k ln V (2:108)
This equation provides the link between entropy and the number of microstates for the system.
Expressions for other thermodynamic functions in terms of Z are summarized below.
Helmholtz free energy, A ¼ U ÀTS ¼ ÀNkT(ln Z Àln N þ1) (2:109)
Gibbs free energy, G ¼ H ÀTS ¼ ÀNkT(ln Z Àln N) (2:110)
Enthalpy, H ¼ NkT 1 þT
@ ln Z
@T
V
(2:111)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 84 1.11.2008 2:41am Compositor Name: JGanesan
84 Handbook of Thermal Process Modeling of Steels
Chemical potential, m ¼ ÀkT( ln Z Àln N) ¼ kT ln
N
Z
(2:112)
Pressure, P ¼ NkT
@ ln Z
@V
T
(2:113)
The expression N
i
=N ð Þ ¼
g
i
e
Àb«
i
Sg
i
e
Àb«
i
¼
g
i
e
Àb«
i
Z
can now be rewritten in terms of the chemical
potential,
N
i
=g
i
ð Þ ¼
Ne
Àb«
i
Z
¼ e
N
Z
e
À
«
i
kT
¼ e
mÀ«
i
kT
¼
1
e
«
i
Àm ð Þ=kT
(2:114)
For the case of FD statistics,
N
i
=g
i
ð Þ ¼
1
e
«
i
Àm=kT ð Þ
þ1
(2:115)
For the case of BE statistics,
N
i
=g
i
ð Þ ¼
1
e
«
i
Àm=kT ð Þ
À1
(2:116)
As the molecular energy can be separated into translation, rotational, vibrational, electronic, nuclear,
and other components, probability functions and partition functions can be evaluated for each of the
components. Their contribution to the total function is obtained by multiplying the individual
functions obtained. The discussions so far dealt with statistical thermodynamics applied to systems
at equilibrium. Statistical mechanics can also deal with systems not at equilibrium as they change
from a given equilibrium condition to another. This involves incorporation of both kinetic theory
and statistical mechanics. Due to the limitation in the scope of this chapter, statistical mechanics
could not be presented here, and the reader should refer to advanced texts in statistical thermo
dynamics, some of which are listed in the bibliography section.
2.15 APPLICATIONS OF THERMODYNAMICS IN THERMAL PROCESSING
Thermal processing involves processes controlled by thermal activation. These processes include
nucleation and growth of product phases, interfacial=grain boundary migration, homogenization,
diffusion in lattice and along dislocations, recovery, recrystallization, grain growth, and others.
These processes involve at the fundamental level random migration of atoms and barriers for such
movements. When you examine a material system, the atoms in the system being a solid, liquid, or
gas have an average energy of kT for each mode of vibration. The actual energy possessed by the
atom at a given instant can be any one of the possible energy states for the system. The probability
that the atom has an energy corresponding to a particular energy level is the probability of
occupancy (N
i
=N) corresponding to the most probable state of the system. Thus, when there is a
barrier the atom must overcome during a process, the atom always has a deﬁnite probability of
having energy greater than the barrier height, even though the average energy may be signiﬁcantly
less than the barrier height.
Another common example of the use of statistical thermodynamics is the determination of
solute distribution between lattice, defects sites at the grain interior, grain boundaries, and surfaces.
When solutes are added to a solid, there is a difference in energy in the solute atom that occupies a
site in the grain interior as opposed to a location on the surface or a grain boundary. The distribution
of solute atoms between the grain interior and grain boundaries can be evaluated by assigning
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 85 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 85
appropriate values for the energy levels corresponding to these sites. Theoretical estimation of heat
capacity is also made possible by statistical thermodynamics.
Application of thermodynamics to the diffusion process is important with regard to thermal
processing. A brief description of how thermodynamic concepts are used in obtaining a generalized
expression for diffusion ﬂux is given here.
When an electrical potential (voltage) gradient is present in a conductor, electrons in the
conductor are subjected to an electrical force qE, where q is the electrical charge on the electrons
and E is the electrical ﬁeld given by (ÀdV=dx), and accelerated up the voltage gradient.
These accelerated electrons are also subjected to scattering by the various scattering centers
(Figure 2.6), resulting in a steadystate drift velocity. This drift (or effective) velocity is proportional
to the ﬁeld. The proportionality constant is referred to as mobility, M
e
, and this is a measure of the
resistance to the electron motion in the material.
Drift velocity of the electron ¼ M
e
(ÀdE
P
=dx) (2:117)
where E
P
is the potential energy of the electron.
As in the case of electron motion in a conductor, the diffusional ﬂux of atoms in a solid can also
be expressed in terms of drift velocity n
B
superposed on the random jumping motion of each atom.
The drift velocity n
B
is proportional to the negative value of the chemical potential gradient driving
the diffusion process.
n
B
¼ ÀM
B
@m
B
@x
(2:118)
where M
B
is the proportionality constant and is deﬁned as the mobility of B atoms in the solid. If C
B
is the concentration of B atoms, the ﬂux of B atoms can be expressed as
J
B
¼ n
B
C
B
¼ ÀM
B
C
B
@ m
B
@ x
(2:119)
As shown earlier, the chemical potential of species B, m
B
, is given by
m
B
¼ G
B
þRT ln (a
B
) ¼ G
B
þRT ln (g
B
Á X
B
) (2:120)
x
−
v
1
e
+
v
2
FIGURE 2.6 Illustration of the motion of electrons along a potential gradient in a conductor.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 86 1.11.2008 2:41am Compositor Name: JGanesan
86 Handbook of Thermal Process Modeling of Steels
Substitution of the derivative of m
B
in the expression for diffusional ﬂux leads to
J
B
¼ ÀM
B
RT 1 þ
d ln g
B
d ln X
B
!
@C
B
@x
(2:121)
Comparing this with the expression for Fick’s ﬁrst law of diffusion for a condition where the
chemical potential gradient in the system depends only on the variation in concentration, it can be
seen that
D
B
¼ M
B
RT 1 þ
d ln g
B
d ln X
B
!
¼ M
B
RTF (2:122)
where F is thermodynamic factor.
For an ideal solution
d ln g
B
d ln X
B
!
¼ 0, and F ¼ 1: (2:123)
The diffusion coefﬁcient D
B
simpliﬁes to
D
B
¼ M
B
RT
In a more generalized form, the chemical potential can be written as
m
B
¼ G
B
þRT ln (g
B
Á X
B
) þS
i
E
i
(2:124)
where E
i
refers to various potential ﬁelds mainly relevant to the problem under consideration. A
generalized expression for diffusional ﬂux in an ideal solid solution phase, which is inﬂuenced by
the presence of various potential ﬁelds such as electric (E
e
), magnetic (E
m
), thermal (E
th
), or strain
(E
s
), can be written as
J
B
¼ ÀD
B
@C
B
@x
ÀM
B
C
B
@E
e
@x
ÀM
B
C
B
@E
s
@x
ÀM
B
C
B
@E
m
@x
Á Á Á
¼ ÀD
B
@C
B
@x
À
D
B
C
B
RTF
@E
e
@x
À
D
B
C
B
RTF
@E
s
@x
À
D
B
C
B
RTF
@E
m
@x
Á Á Á (2:125)
These are very relevant as the processing and use of materials often involve electrical, magnetic, and
elastic strain gradients besides the thermal gradient.
BIBLIOGRAPHY
C.H.P. Lupis, Chemical Thermodynamics of Materials, North Holland=Elsevier Science Publishing, New York,
1983.
D.R. Gaskell, Introduction to Thermodynamics of Materials, 4th ed., Taylor & Francis, New York, 2003.
O.F. Devereux, Topics in Thermodynamics of Materials, Wiley, Hoboken, NJ, 1983.
A. Prince, Alloy Phase Equilibria, Elsevier, New York, 1966.
B. Linder, Thermodynamics and Introductory Statistical Thermodynamics, WileyInterscience, Wiley,
Hoboken, NJ, 2004.
A. Carter, Classical and Statistical Thermodynamics, PrenticeHall, Upper Saddle River, NJ, 2001.
B.S. Bokstein, M.I. Mendelev, and D.J. Srolowitz, Thermodynamics and Kinetics in Materials Science, Oxford
University Press, Oxford, England, 2005.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 87 1.11.2008 2:41am Compositor Name: JGanesan
Thermodynamics of Thermal Processing 87
R.E. Sonntag and G.J. Van Wylen, Fundamentals of Statistical Thermodynamics, Robert E. Kreiger, Malabar,
FL, 1985 and Wiley, Hoboken, NJ, 1966.
E.S. Machlin, An Introduction to Aspects of Thermodynamics and Kinetics Relevant to Materials Science,
Eugene S. Machlin, New York, 1991.
D.A. Porter and K.E. Easterling, Phase Transformations in Metals and Alloys, 2nd ed., Chapman & Hall,
New York, 1992.
N.M. Laurendeau, Statistical Thermodynamics: Fundamentals and Applications, Cambridge University Press,
Cambridge, England, 2005.
C.L. Tien and J.H. Lienhard, Statistical Thermodynamics: Fundamentals and Applications, Hemisphere
Publishing, New York, 1979.
SOME OF THE SOURCES FOR THERMODYNAMIC DATA
L.B. Pankratz, Thermodynamic Properties of Elements and Oxides, U.S. Geological Survey Bulletin 672,
Bureau of Mines, U.S. Department of the Interior, 1982.
L.B. Pankratz, J.M. Stuve, and N.A. Gokcen, Thermodynamic Data for Mineral Technology, U.S. Geological
Survey Bulletin 677, Bureau of Mines, U.S. Department of the Interior, 1984.
L.B. Pankratz, Thermodynamic Properties of Halides, U.S. Geological Survey Bulletin 674, Bureau of Mines,
U.S. Department of the Interior, 1984.
R. Hultgren, P.D. Desai, D.T. Hawkins, M. Gleiser, K.E. Kelley, and D.D. Wagman, Selected Values of the
Thermodynamic Properties of Elements, ASM International, 1973.
R. Hultgren, P.D. Desai, D.T. Hawkins, M. Gleiser, K.E. Kelley, and D.D. Wagman, Selected Values of the
Thermodynamic Properties of Binary Alloys, ASM International, 1973.
H.G. Lee, Chemical Thermodynamics for Metals and Materials, Imperial College Press, London, 1999.
Outokumpu HSC Chemistry for Windows software Version 5.0, Outokumpu Research Oy, Pori, Finland, 2002.
GTT Chemsage Software, GTT Technologies, Herzogenrath, Germany, 1994.
JANAF Thermochemical Tables, 2nd ed., NSRDSNBS 37, U.S. Department of Commerce, Washington, DC,
1971.
JANAF Thermochemical Tables, 3rd ed., NSRDSNBS 37, U.S. Department of Commerce, Washington, DC,
1985.
I. Barin and O. Knacke, Thermodynamic Properties of Inorganic Substances, SpringerVerlag, Berlin, 1972.
CRC Handbook of Chemistry and Physics, CRC Press, Boca Raton, FL, all editions.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C002 Final Proof page 88 1.11.2008 2:41am Compositor Name: JGanesan
88 Handbook of Thermal Process Modeling of Steels
3
Physical Metallurgy
of Thermal Processing
Wei Shi
CONTENTS
3.1 Defects in Crystal Structure................................................................................................. 90
3.1.1 Point Defects ............................................................................................................. 90
3.1.2 Dislocations............................................................................................................... 91
3.1.3 Surface Defects ......................................................................................................... 93
3.2 Recovery, Recrystallization, and Grain Growth.................................................................. 94
3.2.1 Annealing of ColdWorked Steels ............................................................................ 95
3.2.2 Dynamic Recovery and Recrystallization during Hot Working............................... 95
3.2.3 Recovery ................................................................................................................... 96
3.2.4 Kinetics of Recrystallization..................................................................................... 97
3.2.5 Dynamic Recovery and Recrystallization................................................................. 97
3.2.6 Grain Growth ............................................................................................................ 98
3.3 Fundamentals of Solidiﬁcation ............................................................................................ 99
3.3.1 Nucleation ................................................................................................................. 99
3.3.1.1 Homogenous Nucleation ............................................................................. 99
3.3.1.2 Heterogeneous Nucleation......................................................................... 101
3.3.2 Crystal Growth........................................................................................................ 102
3.4 Diffusion ............................................................................................................................ 103
3.4.1 Mechanism for Diffusion........................................................................................ 103
3.4.2 Activation Energy for Diffusion ............................................................................. 103
3.4.3 Rate of Diffusion..................................................................................................... 104
3.4.4 Factors Affecting Diffusion .................................................................................... 105
3.5 Fundamentals of SolidState Phase Transformation.......................................................... 105
3.5.1 Phase Transformation.............................................................................................. 105
3.5.2 Phase Rule............................................................................................................... 106
3.5.3 Phase Diagram ........................................................................................................ 106
3.5.4 Nucleation in SolidState Phase Transformation.................................................... 106
3.5.5 Growth of New Phase............................................................................................. 107
3.5.6 Kinetics of SolidState Phase Transformation........................................................ 108
3.6 SolidState Phase Transformation in Undercooled Austenite ........................................... 108
3.6.1 Pearlite Transformation........................................................................................... 108
3.6.1.1 Mechanism of Phase Nucleation and Growth........................................... 109
3.6.1.2 Kinetics of Pearlite Transformation .......................................................... 109
3.6.2 Bainitic Transformation .......................................................................................... 110
3.6.2.1 Mechanism of Bainitic Transformation .................................................... 110
3.6.2.2 Kinetics of Bainitic Transformation.......................................................... 112
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 89 3.11.2008 1:49pm Compositor Name: JGanesan
89
3.6.3 Martensitic Transformation..................................................................................... 113
3.6.3.1 Mechanism of Martensitic Transformation............................................... 113
3.6.3.2 Kinetics of Martensitic Transformation .................................................... 113
3.6.3.3 Tempering.................................................................................................. 114
3.7 Formation of Austenite ...................................................................................................... 114
3.8 Precipitation Strengthening................................................................................................ 116
3.9 Melting............................................................................................................................... 118
References ..................................................................................................................................... 119
Further Reading ............................................................................................................................ 119
By controlling the atomic arrangement and the microstructure, some control over all of the
physical properties of a metal, particularly mechanical properties, may be exercised. In order to
understand how the thermal processes control the atomic arrangement and microstructure, the
theoretical base of physical metallurgy is of high importance. Within the frame of thermal
processing of steels, this chapter covers kinetics of melting, solidiﬁcation, phase transformation
on heating=cooling, nucleation, diffusion, grain growth, precipitation, vacancy, dislocation, recov
ery, and recrystallization.
3.1 DEFECTS IN CRYSTAL STRUCTURE
It is hard to obtain perfect metals in industries, and there are always some defects in the crystal
arrangement of atoms in engineering steels. Defects can be introduced sometimes by solidiﬁcation
or thermal processing. The defects often affect the properties of steels and are useful in industrial
applications. However, there are also some applications where minimization of a particular type of
defect is necessary. Three basic types of defects are discussed in this section: point defects, line
defects, and surface defects.
3.1.1 POINT DEFECTS
Point defects are localized disruptions in perfect atomic arrangements in a crystal structure. A point
defect typically involves one atom, or a pair of atoms. Point defects may be introduced by the
movement of atoms when they gain energy during thermal processes. Although a point defect
occurs at one or two sites, it affects surrounding atoms over quite a large distance in the crystalline
material.
As illustrated in Figure 3.1, there are different types of point defects. Atoms or ions at normal
lattice sites are not still; they vibrate around the normal lattice sites. If the movement caused by
vibration is large enough, an atom can leave its normal location. A vacancy is produced when an
atom misses from its normal site in a crystal structure. Crystalline materials have vacancy defects;
when vacancies are present, the entropy of the material increases.
An interstitial defect is formed if an extra atom situates at a normally unoccupied position in the
crystal structure. Although interstitial atoms are much smaller than the atoms located at the lattice
points, they are still larger than the spaces of interstitial sites that they occupy; the surrounding
crystal region is compressed and distorted.
A substitutional defect is formed when one atom located at a normal lattice site is replaced by a
different type of atom. Substitutional atoms have a different size from the original atoms,
so substitutional defects affect the surrounding interatomic spacing and create distortion in the
crystal structure.
When an atom leaves its original lattice site and moves into an interstitial site, then a vacancy
interstitial pair is formed. When an atom moves to surfaces of crystalline materials or the normal
lattice sites, it leaves a vacancy inside crystalline materials.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 90 3.11.2008 1:49pm Compositor Name: JGanesan
90 Handbook of Thermal Process Modeling of Steels
The concentration of vacancies increases as the temperature increases, and under thermal
equilibrium it follows the Arrhenius equation [1]:
n
v
¼ n exp À
Q
v
RT
_ _
(3:1)
where
n
v
is the number of vacancies per cubic centimeter
R is the gas constant
T is the temperature in Kelvin
n is the number of atoms per cubic centimeter
Q
v
is the energy required to produce 1 mol of vacancies
It should be noted that the concentration of vacancies is more than that of the equilibrium state
under some conditions. When crystal materials are rapidly cooled to low temperatures, the concen
tration of point defects is much higher than that formed in the equilibrium state.
3.1.2 DISLOCATIONS
Dislocations are line defects in crystalline materials that are formed during either solidiﬁcation or
plastic deformation. There are three types of dislocations discussed in this section: the edge
dislocation, screw dislocation, and the mixed dislocation.
It is assumed that there is a simple crystalline cubic, and an extra plane of atoms is inserted into
the top half of the cubic, as illustrated. There is a mismatch of the atom arrangement at the bottom
edge of the inserted plane, so the mismatch is called the edge dislocation. The intersection line
between the extra plane and the sliding plane is called the edge dislocation line. The schematic
diagram of an edge dislocation is shown in Figure 3.2.
(a) (b)
(c) (d)
FIGURE 3.1 Point defects: (a) vacancy, (b) interstitial atom, (c) and (d) substitutional atom.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 91 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 91
If a simple cubic is partly sliced, then the cubic is twisted by one atom spacing. There is
a boundary line between the slipped area and the nonslipped area; the atom arrangement around
the boundary line is screwing in the threedimensional space. This kind of mismatch of atom
arrangement is called the screw dislocation. The schematic diagram is shown in Figure 3.3.
A mixed dislocation has an edge dislocation component and a screw dislocation component.
There is a transition region between them. The schematic diagram of a mixed dislocation is shown
in Figure 3.4.
Dislocation line
b
−
FIGURE 3.2 Schematic view of edge dislocation. (Adapted from Askelan, D.R. and Phule, P.P., Essentials of
Materials Science and Engineering, Thomson Learning, 2004.)
b
−
FIGURE 3.3 Schematic view of screw dislocation. (Adapted from Askelan, D.R. and Phule, P.P., Essentials
of Materials Science and Engineering, Thomson Learning, 2004.)
FIGURE 3.4 Schematic view of mixed dislocation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 92 3.11.2008 1:49pm Compositor Name: JGanesan
92 Handbook of Thermal Process Modeling of Steels
There are many dislocations in steels. Dislocations developed during fatigue in the austenitic
phase of DSS SAF 2507 steel is shown in Figure 3.5. Plastic deformation in metals and alloys is
explained as the result of the glide and climb of dislocations. The dislocation slip process explains
why the strength of metals is much lower than the value calculated from the metallic bonds.
Insitu transmission electron microscope (TEM) deformation experiments of Mo showed the
propagation of edge dislocations at low stress (Figure 3.6).
3.1.3 SURFACE DEFECTS
Surface defects are the boundaries or planes that separate a block of material into different regions
that have the same crystal structure but different orientations. Surface defects are usually regions
with the thickness of several atoms, and the atoms are not properly spaced. There are two classes of
surface defects: material surfaces and internal boundary surfaces (grain boundaries).
Materials end at surfaces, so atoms on the surfaces are partly surrounded by other atoms; the
composition of surfaces is often different from that of the internal part. The microstructure of
engineering steels comprises many grains. A grain is a portion of material within which the
orientation of the atom arrangement is nearly identical. The crystal structure is the same for each
adjoining grain, but the orientation of the atom arrangement is different for each adjoining grain
(Figure 3.7). A grain boundary is the surface that separates different grains.
0.5 μm
FIGURE 3.5 Ladderlike dislocations with austenitic grains. (From Marinelli, M.C., Degallaix, S., and
AlvarezArmas, I., Mater. Sci. Eng. A, 435–436, 305, 2006.)
Edge
Edge
Screw
t = 0 s t = 3 s
FIGURE 3.6 Propagation of edge dislocations in Mo away froma tangle at lowstress. As the edge dislocations
move, they trail long screw segments behind them. (From Lagow, B.W., et al., Mater. Sci. Eng. A, 309–310,
445, 2001.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 93 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 93
3.2 RECOVERY, RECRYSTALLIZATION, AND GRAIN GROWTH
Grain size is an important property that greatly affects the physical properties of steels. For example, the
yield strength is related to the grain size, as described by the Hall–Petch equation:
s
y
¼ s
0
þKd
À1=2
where
s
y
is the yield strength
d is the average diameter of the grains in millimeters
s
0
and K are the constants for the metal
The proper value of the exponent of d is veriﬁed to be À1=2 by Morrison and Miller, which covers a
range of grain sizes from 0.35 to 400 mm [4].
Making the grain size ﬁne is a method to increase the yield strength of steels. The micrographs
of coarse and ﬁne austenite grains in 33NiCrMoV steel are illustrated in Figure 3.8. Among factors
inﬂuencing the ﬁnal grain size of steels after thermal processing, recovery, recrystallization, and
grain growth are important.
Grain
boundary
FIGURE 3.7 Schematic view of grain and grain boundary.
FIGURE 3.8 Coarse and ﬁne austenite grains in 33NiCrMoV steel.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 94 3.11.2008 1:49pm Compositor Name: JGanesan
94 Handbook of Thermal Process Modeling of Steels
3.2.1 ANNEALING OF COLDWORKED STEELS
Recovery and recrystallization are fundamental issues at all levels of steel metallurgy, from grain
size control to understanding the kinetics of grain boundary motion. While steels are deformed
under coldworking conditions, part of the deformation energy is stored within formed materials as
the increment of the dislocation density. Further evolution of the microstructures occurs during
subsequent annealing process of the coldformed materials, including recovery, recrystallization,
and growth of grains as illustrated in Figure 3.9.
Dislocations are locally rearranged into lower energy patterns during the annealing process.
There is no translation of large granular boundary, so the shape of grains keeps the same as that of
deformed ones, e.g., the ﬁber or ﬂatter form is kept.
When heated above a certain temperature, new small grains ﬁrst nucleate at the cell boundaries
in a heavily deformed area and then grow up until the deformation structures are composed of the
newly generated uniaxial grains without distortion. The ﬁnal stage is grain growth. New grains
swallow each other under the driving energy provided by boundary energy until stable sizes of
grains are obtained under that condition.
The strength and hardness of materials vary along with the progress of recovery and recrystal
lization; the internal stresses also change. Reduction in hardness is small during recovery, which is
near 1=5 of the total variance, and the hardness reduces remarkably during recrystallization. The
changing of material strength is almost the same as that of hardness. Almost all macroscopic stresses
are released during recovery, but microscopic stresses are only released by recrystallizing.
3.2.2 DYNAMIC RECOVERY AND RECRYSTALLIZATION DURING HOT WORKING
Hot working is deﬁned as plastic deformation of metallic materials at temperatures above the
recrystallization temperature. The recrystallization temperature is about 0.4T
m
; T
m
represents the
melting temperature. Deformation of steels at high temperatures is much more complicated than
deformation at room temperature. Hot deformation of steels involves work hardening, recovery, and
recrystallization. The ﬂow stress of metals under hot forming strongly depends on the work
hardening, recovery, and recrystallization process as shown in Figure 3.9. Figure 3.10 shows
schematically the change in ﬂow stress during hot working. The ﬂow stress increases as the
deformation increases at the beginning of hot forming, then the curve deviates from the work
hardening curve because of dynamic recovery, and the ﬂow stress ﬁnally decreases because of the
dynamic recrystallization. The change in dislocation density during hot forming explains why the
ﬂow stress changes that way.
Recovery and recrystallization processes occurring during annealing are different than those
occuring during hot working while materials are deformed. Recovery and recrystallization during hot
working is called dynamic recovery, and dynamic recrystallization. The reduction of ﬂow stress
during hot working is the evidence that dynamic recrystallization occurs.
(a) (b) (c) (d)
FIGURE 3.9 (a) Work hardening, (b) recovery, (c) recrystallization, and (d) grain growth during annealing.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 95 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 95
3.2.3 RECOVERY
At the early stage of annealing of cold formed metals, external thermal energy permits the
dislocations to move and form the boundaries of a polygonized subgrain structure while
the dislocation density stays almost unchanged. This process also removes the residual stresses
formed due to cold working signiﬁcant [1,5]. The recovering of physical and mechanical properties
varies with the temperature and time. Recovery is a relaxation process with the following charac
teristics:
1. There is no incubation period
2. Recovery rate is large at the beginning, then it slows down till it is near zero
3. Recovery has a limit value varying with temperature; the higher the temperature, the
greater is the limit value and the shorter is the time needed to reach the limit value
4. The greater the deformation, the greater is the initial recovery rate, and decrease in grain
size helps to accelerate the recovery process
The characteristic of recovery can be expressed as the following equation [5]:
dx
dt
¼ Àcx (3:2)
where
t is the time of heating under constant temperature
x is the fraction of property increase caused by cold work after heating
c is a constant related with material and temperature
The value of constant parameter c can be described with the Arrhenius equation:
c ¼ c
0
exp À
Q
RT
_ _
(3:3)
Flow stress
s
s
e
c
e
max
Work hardening
Work hardening + dynamic recovery
Plastic strain
Work hardening + dynamic recovery +
dynamic recrystallization
FIGURE 3.10 Variation in ﬂow stress due to hot forming. (From Yanagida, A. and Yanagimoto, J., J. Mater.
Process. Technol., 151, 33, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 96 3.11.2008 1:49pm Compositor Name: JGanesan
96 Handbook of Thermal Process Modeling of Steels
Mechanisms of recovery are different in different temperature ranges. At low temperatures recovery
is mainly caused by dislocation translation whereas at medium temperatures, it is caused by
dislocation movement and redistribution. Dislocations on the same slip surface attract and then
are eliminated. Recovery at higher temperatures is caused by climbing of edge dislocations which is
activated at about 0.3T
m
.
3.2.4 KINETICS OF RECRYSTALLIZATION
The process of formation of new grains by heat treating coldworked steels is known as recrystal
lization. The temperature at which materials are recrystallized is known as the recrystallization
temperature.
Recrystallization is a heterogeneous process and dependent on the deformation state of steels.
The kinetics of recrystallization depends on nucleation rate,
_
N, and growth rate, G. For uniform
nucleation, spherical kernel, constant nucleation rate, and growth, Johnson and Mehl expressed
recrystallized fraction volume for a time period t under constant temperature:
V ¼ 1 Àexp
Àp
_
NG
3
t
4
3
_ _
(3:4)
The nucleation rate decreases when time increases under constant temperature. The kinetics of
recrystallization can be described by Johnson–Mehl–Avrami equation as follows:
V ¼ 1 Àexp (ÀBt
n
) (3:5)
where the constants B and n contain information about the nucleation rate and grain growth.
The kinetics of recrystallization can be measured by the fraction of the microstructure that has
been transformed, and experimental data are converted to a Kolmogorov–Johnson–Mehl–Avrami
(KJMA) plot. For sitesaturated nucleation conditions and twodimensional growth, n ¼2; for
continuous nucleation and threedimensional growth, n ¼4.
3.2.5 DYNAMIC RECOVERY AND RECRYSTALLIZATION
While strain increases during hot work, the dislocation density also increases. Dislocations begin to
tangle and form a subgrain structure. For metals with high stack fault energy, the glide and climb of
dislocations easily occur, and dislocation density decreases when dislocations encounter and diminish
at new slip planes. When the dislocation density is low in subgrain structures, only dynamic recovery
occurs. When the generation of dislocations is balanced by the diminishing of dislocation, there is
no further strain hardening.
For metals with low stacking fault energy, the glide and climb of dislocations are difﬁcult, so the
decrease in ﬂow stress of metal with recrystallization mainly results from the decrease in dislocation
density due to dynamic recrystallization. The decrease in dislocation density with dynamic recrys
tallization can be expressed as follows [6]:
r ¼ (1 ÀX
D
)r*j
«
þX
D
r
0
(3:6)
where
r is the average dislocation density in the metal
r*j
«
is the dislocation density of the metal under work hardening and dynamic recovery
r
0
is the dislocation density of dynamic recrystallized grains
X
D
is the volume fraction of dynamically recrystallized grains
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 97 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 97
X
D
is expressed by Avrami equation [6]:
X
D
¼ 1 Àexp ÀG
D
( « À«
c
)
p
_ ¸
(3:7)
where
G
D
is the rate of dynamic recrystallization
«
c
is the critical strain for the onset of dynamic recrystallization
p is a material constant
Values of «
c
are measured from ﬂow stress curves obtained from hot compression tests.
The grain size distribution at the steady state of hot upsetting of AISI304L steel specimens is
observed by the technique of electron back scattering diffraction (EBSD). The crystallography maps
show that equiaxed grains are developed in the steady state under uniaxial compression at high
temperatures (Figure 3.11). The results suggest that the deformed grains are replaced by new
equiaxed grains, and dynamic recrystallization is fully carried out.
3.2.6 GRAIN GROWTH
If the temperature of metals is high enough after recrystallization, grains begin to grow via grain
boundary immigration; this phenomenon is called grain growth. Grain growth is driven by the
=100 μm; Map2; Step=2 μm; Grid 50ϫ100
Initial state
T = 1050ЊC and e = 0.001 s
−1
0 45
T = 1100ЊC and e = 0.001 s
−1
T = 950ЊC and e = 0.001s
−1
=100 μm; Grise; Step=2 μm; Grid 50ϫ100
=100 μm; Grise; Step=2 μm; Grid 50ϫ100
=100 μm; Grise; Step=2 μm; Grid 50ϫ100
FIGURE 3.11 Grain boundary and orientation distribution in AISI304L steel. (From Wahabi, M.E. et al.,
Mater. Sci. Eng. A, 393, 83, 2005.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 98 3.11.2008 1:49pm Compositor Name: JGanesan
98 Handbook of Thermal Process Modeling of Steels
tendency to decrease the total boundary surface energy by decreasing the grain boundary area. The
kinetics of grain growth is represented by the wellestablished Becktype equation [8]:
d
n
Àd
n
0
¼ k(T)t (3:8)
k ¼ k
0
exp À
Q
m
RT
_ _
(3:9)
where
d is the mean grain size achieved at the end of grain growth
T is the temperature at which isothermal annealing is carried out
t is the time duration
n is the grain growth exponent
k
0
is the preexponential coefﬁcient
Q
m
is the activation energy for grain growth
R is the gas constant
Analytical models of grain growth kinetics cannot describe the complex spatial and time
dependent topology of the grain that evolves during grain growth. Numerical methods have
provided a powerful tool to analyze grain growth in recent years. The numerical models developed
to simulate grain growth include vertex, phase ﬁeld, Monte Carlo, and cellular automata models [9].
3.3 FUNDAMENTALS OF SOLIDIFICATION
Solidiﬁcation is considered as one of the most important thermal processing techniques. Besides
being used as a primary processing step to produce steel slabs or ingots, solidiﬁcation is also applied
to joint techniques such as welding, brazing, and soldering. Most metallic alloys with desired
components are produced by solidiﬁcation from the liquid phase. The original microstructure
and properties in steel alloys are also determined largely by solidiﬁcation. The solidiﬁcation process
of crystalline materials is divided into two stages. First, ultraﬁne crystallites of a solid phase, known
as the nuclei, are formed in the liquid. In the second stage, the nuclei begin to grow as atoms from
the liquid attached to it until no liquid remains.
3.3.1 NUCLEATION
Nucleation refers to the formation of the ﬁrst ultraﬁne crystallites from molten materials. There are
two types of nucleation: homogeneous and heterogeneous nucleation. Because it is very hard to
obtain pure molten metals without any impurity, homogeneous nucleation never occurs in liquid
except in controlled laboratory experiments.
3.3.1.1 Homogenous Nucleation
The free energy difference between the liquid and solid is the driving force for solidiﬁcation. When
the temperature of metals is below the melting temperature, the free energy of the solid is smaller
than the free energy of the liquid, so there is a thermodynamic tendency for the liquid to transform to
the solid. During the solidiﬁcation, a liquid–solid interface is created. The surface free energy
associated with this interface is s
sl
. The total change in energy during solidiﬁcation is DG,
DG ¼
4
3
pr
3
DG
v
þ4pr
2
s
sl
(3:10)
The free energy change per unit volume, DG
v
, is negative because the phase transformation is
thermodynamically feasible. The total free energy change during nucleation is shown in Figure 3.12.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 99 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 99
An embryo is a tiny particle of the solid phase that forms from the liquid as atoms cluster. The
embryo is also unstable; i.e., it may either grow into a stable nucleus or redissolve. At the temperature
where the solid and liquid phase are in thermodynamic equilibrium, which is the melting temperature,
the free energy of the solid phase and liquid phase are equal; the total free energy change will be
positive. When the solid mass is with a radius less than the critical radius for nucleation (r*), the
further growth of the solid mass causes the total free energy to increase. Instead of growing, the solid
mass has a tendency to remelt. Embryos are unstable at the freezing temperature.
The formation of embryos is a statistical process. If a solid particle forms with a radius that is
larger than r* by chance, the newsolid is then stable and sustainable since nucleation has occurred and
the solid particle begins to grow. This solid particle is now called nucleus. At the thermodynamic
melting or freezing temperature, the probability of forming stable, sustainable nuclei is extremely
small. Solidiﬁcation does not begin when the temperature of the liquid decreases to the melting
temperature. If the temperature continues to decrease, the tendency of liquid to transform to solid
increases. When the temperature of the liquid is below the melting temperature, the liquid is
called undercooled. The difference between the melting temperature and the actual undercooled
liquid temperature is undercooling DT.
When melted steels are cooled to temperatures below the equilibrium temperature, there are two
factors that favor nucleation. First, the atoms in the liquid lose their thermal energy, and the
probability of forming clusters of large size embryos increases. Second, the larger different free
energy between solid and liquid phase makes the critical radius of nuclei small. The size of the
critical radius r* is given by the following equation:
r* ¼
2s
sl
T
m
DH
f
DT
(3:11)
F
r
e
e
e
n
e
r
g
y
c
h
a
n
g
e
0
ΔG
s
= (4πr
2
) ⋅
s
sl
r
∗
ΔG= Total freeenergy change
ΔG
v
⋅ πr
3
4
3
FIGURE 3.12 The total change in energy of homogeneous nucleation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 100 3.11.2008 1:49pm Compositor Name: JGanesan
100 Handbook of Thermal Process Modeling of Steels
where
DH
f
is the latent heat of fusion
T
m
is the melting temperature in Kelvin
DT ¼T
m
ÀT is the undercooling when the liquid temperature is T
It is clear that the critical radius of nuclei decreases as the degree of undercooling increases.
When the undercooling is great enough, homogeneous nucleation occurs. The latent heat of fusion
represents the heat given up during the liquid–solid transformation or the heat absorbed during
solid–liquid transformation.
The rate of nucleation, which is the number of nuclei formed per unit time, is a function of
temperature. There is no nucleation at temperatures above the solidiﬁcation point, so the rate of
nucleation is zero. In homogenous nucleation, the nucleation rate is controlled by the energy needed
for nucleation and the probability of atom diffusion:
N ¼ Aexp À
DG*
kT
_ _
exp À
Q
RT
_ _
(3:12)
DG* ¼
16ps
3
sl
T
2
m
3(DH
f
DT)
2
(3:13)
where
A is a constant
DG* is the energy needed for nucleation
Q is the activation energy for atoms to pass through the liquid–solid boundaries
R is the gas constant
k is the Boltzmann constant
T is the temperature in Kelvin
As the temperature of molten metal decreases below the solidiﬁcation point, the driving force
for nucleation increases. However, atomic diffusion decreases at the same time. The rate of
nucleation reaches the maximum at some temperature below the transformation temperature.
3.3.1.2 Heterogeneous Nucleation
Impurities suspended in the liquid or contacted on the liquid surface provide a surface on which the
solid can form, as shown in Figure 3.13. The curvature radius of preexisting small solid surfaces is
greater than the critical radius; so much less undercooling is needed to achieve the critical radius.
This kind of nucleation on preexisting surfaces is known as heterogeneous nucleation. The energy
needed in heterogeneous nucleation is smaller than that needed in homogeneous nucleation.
Impurity
Liquid
Solid
q
FIGURE 3.13 Nucleation at surface of impurity.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 101 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 101
In heterogeneous nucleation, the rate of nucleation is decided by the concentration of the nucleating
agent introduced.
3.3.2 CRYSTAL GROWTH
Once the solid nuclei form in a liquid, more atoms attach to the solid surface; thus, the crystal
growth begins. The speciﬁc heat in the liquid and the latent heat of fusion must be removed during
the solidiﬁcation process. The growth of the solid nuclei depends on how heat is removed from the
molten metal.
The speciﬁc heat in the liquid is removed either by radiation into the surrounding atmosphere or
by conduction into the surrounding solid phase or mold. When the temperature is reduced down to
the solidiﬁcation temperature, the fusion heat is generated at the liquidsolid interface. The fusion
heat or latent heat must be removed before the solidiﬁcation is ﬁnished. How the fusion heat is
removed at the solidiﬁcation front decides the crystal growth mechanism and the ﬁnal structure of
the solidiﬁed crystalline materials.
When a wellinoculated molten metal is cooled to the solidiﬁcation temperature, the solidiﬁca
tion process begins because of heterogeneous nucleation, so no undercooling is needed. The fusion
heat is conducted from the liquid–solid surface into the solid, the temperature of the liquid ahead of
the solidiﬁcation front is higher than the melting temperature, and the temperature of the solid is at
or below the melting temperature. Although there are some protuberances at the solidiﬁcation front,
the solid–liquid surface steadily moves into the liquid as shown in Figure 3.14. This kind of growth
mechanism is called planar growth.
For a liquid that is not inoculated, the nucleation is poor because there is a little heterogeneous
nucleation. When a small solid protuberance forms at the solid–liquid interface under this condition,
the small protuberance continues to grow because the liquid ahead of the solidiﬁcation front is
undercooled. And the grown solid protuberance is called dendrite. Minor dendritic branches on the
primary dendrite also grow into the liquid. As illustrated in Figure 3.15, the latent heat of fusion is
conducted into the undercooled liquid during the dendrite growth, and the growth of the dendrite
stops when the temperature of the molten metal increases to the solidiﬁcation temperature. This
crystal growth mechanism is known as dendritic growth.
When dendritic growth stops, any remaining liquid is solidiﬁed by planar growth, and hetero
geneous nucleation is then carried out easily. The method of removing the latent heat of fusion
makes planar and dendritic growth different. The fusion heat is conducted into the solid or container
for planar growth, but the fusion heat is conducted into the undercooled liquid for dendritic growth.
Dendritic growth is only a small part of the total growth in pure metals; the fraction of dendritic
growth is given by the following equation:
f ¼
cDT
DH
f
(3:14)
Solid
Heat
Growth direction
Liquid
FIGURE 3.14 Schematic view of planar growth.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 102 3.11.2008 1:49pm Compositor Name: JGanesan
102 Handbook of Thermal Process Modeling of Steels
where c is the speciﬁc heat of the liquid. The equation represents howmuch the latent heat of fusion can
be absorbed by the undercooled liquid, so the fraction of dendritic growth is obtained. When a liquid
is well inoculated, the undercooling is almost zero; the solidiﬁcation is dominated by planar growth.
The crystal growth behavior of steel alloys is much more complicated than that of pure metals.
The temperature, concentration of compositions, and liquid ﬂow mode inﬂuence the crystal growth
and cause microsegregation and macrosegregation in the solid.
3.4 DIFFUSION
Controlling the diffusion of atoms is important for some thermal processing techniques, for
example, carburizing, nitriding, and surface hardening of steels.
3.4.1 MECHANISM FOR DIFFUSION
An atom may move from a normal crystal structure location to occupy a nearby vacancy in materials
containing vacancies. An atom may move from one interstitial site to another. Thus, there are two
importance mechanisms by which atoms or ions can undergo diffusion: vacancy diffusion and
interstitial diffusion.
An atom leaves its lattice site to ﬁll a nearby vacancy in diffusion involving substitutional
atoms, so a new vacancy is created at the original lattice site. As diffusion continues, countercurrent
ﬂows of atoms and vacancies are generated, called vacancy diffusion.
When a small interstitial atom is present in the crystal structure, the atom moves from one
interstitial site to another. No vacancies are required for this mechanism, so it is called interstitial
diffusion. Because there are many more interstitial sites than vacancies, interstitial diffusion occurs
more easily than vacancy diffusion.
3.4.2 ACTIVATION ENERGY FOR DIFFUSION
Atoms pass through the surrounding atoms to reach their new sites during diffusion, so energy
must be supplied to force an atom to its new position (Figure 3.16). Atoms originally locate in
Solid (dendrite)
Solid (dendrite)
Growth direction
Liquid
Fusion heat
FIGURE 3.15 Schematic view of dendritic growth.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 103 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 103
lowenergy, relatively stable locations; an atom must overcome an energy barrier, known as the
activation energy Q, to move to a new location. Diffusion occurs when the supplied thermal energy
exceeds the activation energy. Less energy is required for an interstitial atom to pass the surrounding
atoms; activation energies are lower for interstitial diffusion than for vacancy diffusion.
3.4.3 RATE OF DIFFUSION
The rate at which atoms or ions diffuse in a material can be measured by the ﬂux. The ﬂux is deﬁned
as the number of atoms passing through a plane of unit area per unit time, as shown in the ﬁgure.
Fick’s ﬁrst law explains the net ﬂux of atoms:
J ¼ ÀD
dc
dx
(3:15)
where
J is the ﬂux
D is the diffusivity or diffusion coefﬁcient
dc=dx is the concentration gradient
The negative sign in the equation indicates that the species diffuse from higher to lower concentra
tion. The concentration gradient shows how the composition of the material varies with distance and
serves as the thermodynamic driving force for diffusion at a microscopic scale.
The ability of atoms to diffuse increases as the temperature or thermal energy possessed by the
atoms increases. The rate of atom movement is related to temperature by the Arrhenius equation:
Rate ¼ c
0
exp À
Q
RT
_ _
(3:16)
E
n
e
r
g
y
Substitutional
(vacancy)
Interstitial
Q
ν
Q
i
FIGURE 3.16 Energy barrier in diffusion. (From Askelan, D.R. and Phule, P.P., Essentials of Materials
Science and Engineering, Thomson Learning, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 104 3.11.2008 1:49pm Compositor Name: JGanesan
104 Handbook of Thermal Process Modeling of Steels
where
c
0
is a constant
R is the gas constant
T is the absolute temperature in Kelvin
Q is the activation energy
3.4.4 FACTORS AFFECTING DIFFUSION
The diffusion coefﬁcient D is related to temperature by an Arrheniustype equation:
D ¼ D
0
exp À
Q
RT
_ _
(3:17)
where
Q is the activation energy for the diffusion of species under consideration
R is the gas constant
T is the absolute temperature
D
0
is a constant for a given diffusion system and is equal to the value of the diffusion
coefﬁcient at 1=T¼0
When the temperature of a material increases, the diffusion coefﬁcient D increases, and the ﬂux
of atoms increases as well. At higher temperatures, the thermal energy possessed by atoms permits
them to overcome the energy barrier generated by the surrounding atoms and easily move to new
sites. At low temperature, which is often below 0.4T
m
, diffusion is very slow and not signiﬁcant,
because it is hard to overcome the energy barrier.
Besides the temperature, other factors inﬂuence the activation energy and the rate of diffusion.
Activation energies are usually lower for atoms diffusing through open crystal structure than for
closepacked crystal structure. The activation energy depends on the strength of atomic bonding, so
it is higher for atoms diffusing in materials with a high melting temperature. The diffusion
coefﬁcient also depends on the concentration of diffusing species and compositions of the matrix.
However, the dependence of D on the concentration of diffusing species can be ignored in many
situations.
3.5 FUNDAMENTALS OF SOLIDSTATE PHASE TRANSFORMATION
3.5.1 PHASE TRANSFORMATION
Alloys rather than pure materials are usually used in many applications. Plain carbon steel is an alloy
of iron (Fe) and carbon (C); stainless steels usually consist of iron, carbon, chromium(Cr), nickel (Ni),
and some other elements. Alloys are divided into two groups: singlephase alloys and multiphase
alloys. Steels are the integration of atoms or molecules; the integration may be homogeneous or
nonhomogeneous. The homogenous part of such an assembly is called phase. A phase is deﬁned as
any portion of a systemthat is physically homogeneous within itself and mechanically separable from
any other portions. A phase has the following characteristics:
1. The same crystal structure or atomic arrangement throughout
2. Roughly the same composition and properties throughout
3. A deﬁnite interface between the phase and any surrounding or adjoining phases
Solid, liquid, and gaseous forms of a material are different phases obviously. However, phases do
not always have to be these three forms. Iron can have face centered cubic (FCC) and body centered
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 105 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 105
cubic (BCC) crystal structure. These two solid forms of iron correspond to two different phases
(austenite and ferrite).
When temperature or pressure changes, microstructure or crystal structure also changes in metals,
so phase transformation occurs; this kind of phase transformation is known as solidstate phase
transformation. Although most solidstate phase transformation processes consist of nucleation and
growth, solidstate phase transformation is quite different from liquid–solid transformation.
3.5.2 PHASE RULE
J.W. Gibbs developed the phase rule to describe the relationship between the number of com
ponents and the number of phases for a given system. The phase rule is presented in the following
general form:
2 þC ¼ F þP (3:18)
where
C is the number of chemically independent components, usually elements or compounds in a
system
F is the number of degrees of freedom, or the number of variables that are allowed to change
independently without changing the number of phases in equilibrium
P is the number of phases present
The Gibbs phase rule is developed under the condition of thermodynamic equilibrium. Equilibrium
condition is usually not maintained in material processing, so the number and compositions of
phases in practice are different from those predicted by the Gibbs phase rule.
3.5.3 PHASE DIAGRAM
An equilibirium phase diagram shows the phases and their compositions at any combination of the
temperature and alloy composition. A phase diagram is important for both material design and the
choice of parameters of thermal processing. If only two elements are present in a material, a binary
phase diagram can be constructed.
The wellknown iron–iron carbide phase diagram is important to understand solidstate phase
transformation in steels. Plotting phase diagrams using computer databases has recently achieved
considerable development. However the equilibiriumdiagramis of little value under nonequilibrium
conditions. Therefore, there is a need for the timetemperaturetransformation (TTT) diagram that is
determined for a speciﬁc steel type which has been austenitized under deﬁned conditions and with a
known average austenite grain size. There are two types of TTT diagrams: isothermaltransformation
(IT) and continuouscoolingtransformation (CCT) diagrams.
For the establishment of IT diagrams, the austenitized samples are quenched to temperatures at
which austenite is metastable. At these temperatures, beginning and end of the transformation
reactions is measured by dilatometric and metallographic investigations. The diagram is read
along the time axes.
In the case of CCT diagrams, the samples are cooled continuously at different cooling rates. The
temperatures at which transformations take place are measured and similar diagrams are obtained
that have to be read along the cooling curves. CCT diagrams also give information about the critical
cooling rate which indicates how fast a steel must be cooled to form only martensite.
3.5.4 NUCLEATION IN SOLIDSTATE PHASE TRANSFORMATION
Most solidstate phase transformations are accomplished by the nucleation and growth
process. Nucleation begins when nuclei form from the parent phase. If the nuclei are generated
without any preference in the parent phase, it is known as homogeneous nucleation. The driving
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 106 3.11.2008 1:49pm Compositor Name: JGanesan
106 Handbook of Thermal Process Modeling of Steels
force for homogenous nucleation is the free energy difference between the parent phase and new
phase. For homogeneous nucleation, the total free energy change after a spherical solid precipitate
nucleates from the matrix is [1]
DG ¼
4
3
pr
3
DG
v
þ4pr
2
s
ab
þ
4
3
pr
3
« (3:19)
where
DG
v
is the free energy change per unit volume
s
ab
is the energy change needed to create a unit area of phase interface
« is the strain energy per unit volume
Nucleation occurs only when DG is negative.
Assuming that d(DG)=dr ¼ 0, the critical radius of nucleation is obtained:
r
c
¼ À
2s
ab
DG
v
þ«
(3:20)
Then, the work for nucleation is described by the following equation:
W ¼
16ps
3
ab
3(DG
v
þ«)
2
(3:21)
If the difference of free energy between the parent and new phase is great and the critical radius is
small, it is easy to nucleate. If the surface energy or the strain energy is great and the critical radius
is large, the energy needed for nucleation is also great.
Similar to the case of solidiﬁcation, the rate of nucleation for homogeneous nucleation is
represented by the equation:
I ¼ nn exp À
Q þW
KT
_ _
(3:22)
where
n is the number of atoms per unit volume in the parent phase
n is the atomic frequency
Q is the activation energy for diffusion
Nucleation begins easily at the crystal defects in the parent phase. Heterogeneous nucleation
dominates the nucleation process during solidstate phase transformations. When nuclei of the new
phase form at sites of the defects, the total change of free energy is
DG ¼
4
3
pr
3
DG
v
þ4pr
2
s
ab
þ
4
3
pr
3
« þDG
d
(3:23)
where DG
d
is the reduction of free energy caused by the elimination of the defects. The defects
existing in the material promote the process of nucleation.
3.5.5 GROWTH OF NEW PHASE
Growth of the nuclei of the new phase is the moving of the phase boundary surface into the parent
phase. There is no longrange diffusion during the growth of the new phase. The composition of the
new phase is the same as that of the parent phase for solidstate phase transformation.
The growth rate of the new phase depends on the immigration speed of the phase boundaries. If
the immigration of the phase boundaries is accomplished through surface diffusion, the phase
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 107 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 107
transformation is called diffusive transformation. If the immigration of the phase boundary surface
is accomplished by the lattice array displacement, the phase transformation is called diffusionless
transformation or sheartype transformation.
3.5.6 KINETICS OF SOLIDSTATE PHASE TRANSFORMATION
Kinetics of solidstate phase transformations depends on both nucleation and growth of the new
phase. It is supposed that there is a phase transformation system at a certain temperature, the parent
phase is a, and the new phase is b. If the nucleation rate of the new phase is I; the number of nuclie
of the new phase formed during a period of dt is
n ¼ I(V
0
ÀV) dt (3:24)
where
V
0
is the total volume of the phase transformation system
V is the volume of the new phase
The relationship of the nuclei radius R and time is represented as
R ¼ G(t Àt) (3:25)
where
G is the linear growth rate of the new phase
t is the incubation period
The volume of the new phase generated until time t is
V ¼
_
t
0
4
3
pG
3
(t Àt)
3
I(V
0
ÀV)dt (3:26)
Assuming X as the volume fraction of the new phase, the wellestablished John–Mehl equation is
obtained:
X ¼ 1 Àexp À
p
3
IG
3
t
4
_ _
(3:27)
In the John–Mehl equation, the nucleation rate is assumed to be constant. However, since the
nucleation rate is actually not constant in phase transformations, the kinetics of diffusive transform
ations is represented by the Avrami equation:
X ¼ 1 Àexp (Àct
n
) (3:28)
where
c is a coefﬁcient dependant on temperature, composition of the parent phase, and grain size
n is a coefﬁcient dependant on the type of phase transformation
The experimental data obtained from most diffusive transformations meet the Avrami equation well.
3.6 SOLIDSTATE PHASE TRANSFORMATION IN UNDERCOOLED AUSTENITE
3.6.1 PEARLITE TRANSFORMATION
Pearlite is a twophase lamellar structure of ferrite and cementite, which develops in steels cooled in
a normal fashion or isothermally transformed at a relatively high temperature. Ferrite is the BCC
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 108 3.11.2008 1:49pm Compositor Name: JGanesan
108 Handbook of Thermal Process Modeling of Steels
crystal structure of iron. Cementite is the hard, brittle ceramiclike compound Fe
3
C. The structure of
pearlite and formation of pearlite in cooled austenite are shown in Figure 3.17; ferrite and cementite
are ordered as alternative plates. Pearlite greatly contributes to the strength of steels. Pearlite that
consists of ﬁne plates is harder and stronger than that consisting of coarse plates. The morphology of
pearlite is largely determined by the austenite–pearlite phase transformation.
3.6.1.1 Mechanism of Phase Nucleation and Growth
The nucleation mechanism of pearlite involves the formation of two phases, ferrite and cementite.
Although there is debate on the pearlite nucleation mechanism, the mechanism can be explained
as follows. The proeutectoid ferrite nucleates ﬁrst and continues to grow with the same crystallo
graphic orientation during the pearlite formation in hypoeutectoid steels. The nucleation of cemen
tite is a ratelimiting step in this case. For hypereutectoid steels, the roles of nucleation of ferrite and
cementite are reversed. In perfectly eutectoid steel, the pearlite nucleation is assumed to occur at the
austenite grain corners, edges, and boundaries.
Two different theories are proposed for the growth of pearlite. The Zener–Hillert theory
assumes that the volume diffusion of carbon in the austenite is the ratecontrolling mechanism. In
addition, Hillert theory assumes that grain boundary diffusion of the carbon atoms is the rate
controlling mechanism. The Kolmogorov–Johnson–Mehl–Avrami (KJMA) theory predicts the
overall transformation rate on the basis of nucleation and growth rates. It is the most widely used
model to describe the austenite–pearlite transformation kinetics [11].
3.6.1.2 Kinetics of Pearlite Transformation
The nucleation rate follows the general nucleation theory [12]:
_
N ¼ N
n
b*Z exp À
DG*
KT
_ _
exp À
t
t
_ _
(3:29)
M
3
C
g
a
500 nm
100 nm
γ
/
M
3
O
E
D
S
p
r
o
f
i
l
e
γ
/
α
E
D
S
p
r
o
f
i
l
e
FIGURE 3.17 TEM micrograph of pearlite–austenite growth front. (From Hutchinson, C.R., Hackenberg,
R.E., and Shiﬂet. G.J., Acta Mater., 52, 3565, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 109 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 109
According to the Zener–Hillert model, the volume diffusion of carbon is the ratelimiting step and
the growth rate is given by [11,12]
G
v
¼
D
g
C,V
k
v
1
f
a
f
b
C
ga
eq
ÀC
gu
eq
C
u
eq
ÀC
a
eq
1
l
1 À
l
c
l
_ _
(3:30)
where
G
v
is the pearlite growth velocity
D
g
C,V
is the diffusion coefﬁcient for volume diffusion of carbon
k
v
is the geometric constant
C
ga
eq
and C
gb
eq
are the equilibriumcompositions of austenite in contact with ferrite and cementite,
respectively
C
a
eq
and C
b
eq
are the equilibrium compositions of ferrite and cementite, respectively
f
a
and f
b
are the volume fractions of the two pearlitic phases
l is the adopted interlamellar spacing
l
c
is the theoretical minimum spacing
The overall pearlite transformation rate can be described by the KJMA theory, and the fraction
f of the formed pearlite phase is deﬁned as a function of the isothermal transformation time t [11]:
f (t) ¼ 1 Àexp Àk
g
G
d
_
t
0
_
N
u
(t
0
)(t Àt
0
)
d
dt
0
_
_
_
_
(3:31)
where
G is a constant growth rate
d is the dimensionality of the growth
k
g
is a constant
_
N
u
is the nucleation rate
Coefﬁcients d and k
g
depend on the geometry of the particle, for example, k
g
is 4p=3 and d is 3 for
spherical particles.
Some parameters need to be determined before the analytical model is used to predict the kinetics of
pearlitic transformation. It is convenient to ﬁt the experimental data to a generalized KJMA equation:
f (t) ¼ 1 Àexp (Àct
n
) (3:32)
3.6.2 BAINITIC TRANSFORMATION
The decomposition product of austenite that forms at a temperature below that of the austenite–
pearlite transformation and above the temperature of martensitic transformation is called bainite. It
is commonly classiﬁed into upper bainite and lower bainite which consist of aggregates of plates of
ferrite, separated by untransformed austenite, cementite, or martensite. Upper bainite forms at higher
temperatures and comprises a series of parallel ferrite laths, which are separated by carbide layers.
Lower bainite (lath bainite) forms at lower temperatures, ferrite takes the form of laths or plates in
lower bainite, and carbide particles distribute inside ferrite.
3.6.2.1 Mechanism of Bainitic Transformation
Austenite g phase transforms into ferrite a and cementite during bainitic transformation. Carbon
diffusion takes place and the transformation rate is also controlled by the diffusion rate of carbon
atoms. Bainite can form in an isothermal process in a certain temperature range; the continuous
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 110 3.11.2008 1:49pm Compositor Name: JGanesan
110 Handbook of Thermal Process Modeling of Steels
cooling transformation (CCT) curve of bainitic transformation is a ‘‘C’’ shaped curve, which is
similar to the CCT curve of pearlite transformation. Both the temperature and the concentration of
carbon have inﬂuence on the process of bainitic transformation.
The carbon diffusion rate is great at relatively high temperatures, and the oversaturated carbon
atoms in plate ferrite immigrate into remaining austenite when bainitic transformation takes place.
The products of transformation are plate bainitic ferrite and austenite of high carbon concentration,
known as carbonfree bainite (see Figure 3.18).
When the temperature of austenite decreases to about 3508C–5508C, the carbon diffusion rate
also decreases. The oversaturated carbon atoms in plate ferrite move to interfaces via diffusion, and
the carbon concentration in austenite between the plates of ferrite increases. Cementite forms in
these regions when the carbon concentration in austenite increases to a critical value; and typical
upper bainite forms. An optical micrograph and a TEM micrograph of upper bainite in silicon steel
is shown in Figure 3.19.
200 mm
FIGURE 3.18 Morphology of carbide free bainite. (From Yang, Z.G. and Fang, H.S., Curr. Opin. Solid State
Mater. Sci., 9, 277, 2005.)
(a) (b)
BF
M−RA
M+RA
0.15 μm
BF
FIGURE 3.19 (a) Optical micrograph, and (b) TEM micrograph of upper bainite. M: martensite, RA: retained
austenite, BF: bainitic ferrite. (From Liu, C., Zhao, Z., and Bhole, S.D., Mater. Sci. Eng. A, 434, 289, 2006.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 111 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 111
If the temperature of austenite continues to decrease, carbon atoms cannot diffuse in austenite.
The carbon diffusion rate in ferrite is small, and carbon atoms move only via the shortrange
diffusion. When ferrite grows, small pieces of carbide form inside ferrite and lower bainite forms.
Although bainitic transformation is a widely used heat treatment technique for steels, the argument
on whether bainitic transformation is a diffusive or diffusionless (sheartype) transformation is still
continuing. Experimental evidence supporting the diffusionless mechanism is the fundamental work
of Ko and Cottrell. It has been found that bainite and Widmanstatten ferrite in hypereutectoid steel
grow slowly but simultaneously and produce a martensitelike upheaval on the free surface of the
specimens [13]. Based on the experimental observations, it is deduced that bainite forms the way
martensite does.
3.6.2.2 Kinetics of Bainitic Transformation
Similar to the formation of pearlite, bainite can form in an isothermal process, and there is also a
period of incubation. The experimental data of the volume fraction of bainite are well ﬁtted to the
Avrami equation. Because bainitic transformation is usually not complete, kinetics of bainite
transformation is described by a modiﬁed Avrami equation:
V ¼ V
max
_
1 Àexp½Àb(t Àt)
n
¸_
(3:33)
where
V is the volume fraction of bainite
V
max
is the maximum volume fraction of bainite
t is the incubation period in seconds
t is the time of transformation in seconds
b and n are constants
The following factors affect the process of bainitic transformation: chemical composition,
austenite grain size, austenitization temperature, stress state, and plastic deformation history of
austenite. Kinetics of the diffusional transformation mechanism is successfully applied to predict the
volume fraction of bainite formed in some steels during heat treatments, but the debate on the
mechanism makes the kinetics of bainitic transformation an open question (Figure 3.20).
FIGURE 3.20 Morphology of plate lower bainite.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 112 3.11.2008 1:49pm Compositor Name: JGanesan
112 Handbook of Thermal Process Modeling of Steels
3.6.3 MARTENSITIC TRANSFORMATION
Martensite is a metastable phase that forms in steels and other materials by a diffusionless,
nonisothermal transformation. In order to obtain martensite, the steel must be rapidly cooled from
the stable austenite region. The optical micrograph of the plate martensite obtained in a highcarbon
iron–chromium steel is shown in Figure 3.21.
3.6.3.1 Mechanism of Martensitic Transformation
Martensitic transformation begins when the temperature of undercooled austenite decreases to a
certain value, which is known as the martensite start temperature. The growth rate in martensitic
transformation is so high that nucleation becomes the controlling process. The amount of martensite
increases as the temperature decreases; the steel should contain 100% martensite when the tem
perature is below the martensite ﬁnish temperature. Martensitic transformation takes place at
relatively low temperatures, and there is almost no diffusion occurring. Martensitic transformation
is mainly accomplished by the rearrangement of the lattice array.
Martensitic transformation is distinguished from diffusional transformations in steels with the
following characteristics:
1. There are upheavals on the free surface of specimens after transformation
2. Atoms transfer from the parent phase to martensite pass via the movement of ‘‘military’’
mode; the moving distance is no more than an atomic spacing
3. There are habit planes and strict orientation relationship between the parent phase and
product
4. Martensitic transformation is reversible for some alloys such as shape memory alloys
3.6.3.2 Kinetics of Martensitic Transformation
The overall kinetics of martensitic transformation depends on both the nucleation and growth
stages and is largely dominated by the slower one of the two stages. The volume fraction of
martensite is a function of the temperature. The widely used model to describe the kinetics
of martensitic transformation assumes that nucleation and growth of martensite are accomplished
in a very short time. The kinetics of martensitic transformation is described by the Koinstinen–
Marburger equation:
V ¼ 1 Àexp [Àc(M
s
ÀT)] (3:34)
50 μm
(a) (b)
50 μm
FIGURE 3.21 (a) Homogenized austenite and (b) martensite after subzero treatment. (Yang, J.R., Yu, T.H.,
and Wang, C.H., Mater. Sci. Eng. A, 438, 276, 2006.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 113 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 113
where
V is the volume fraction of martensite
M
s
is the martensite start temperature
T is the temperature
c is a material constant
Predeformed austenite, and applied=residual stresses inﬂuence martensitic transformation.
3.6.3.3 Tempering
Tempering always follows the hardening process of steels. The purpose is to reduce the brittleness
due to martensite, to relieve internal stresses, and to produce deﬁnite physical properties. Tempering
consists of holding the steel at a speciﬁc temperature (between 1508C to 6508C) for the required
period of time, and then cooling it, usually in still air. For some steels, tempering in the range of
2608C–3708C results in temperembrittlement.
The amount of decrease in strength and hardness, and improvement of ductility depend on the
tempering temperature. As an exception, tempering of highspeed steels results in secondary
hardening due to precipitation of alloying elementcarbides.
Martensite is a supersaturated solid solution with a high density of dislocations. If it is heated
from room temperature depending on the temperature various processes can be observed:
.
Room temperature to 2508C: Catoms segregate and «carbides precipitate at the disloca
tions. Tetragonality of the crystal structure of martensite decreases and disappears.
.
2008C–3008C: Retained austenite decomposes to bainite.
.
2508C–4008C: Cementite starts to form at previous austenite grain boundaries; dislocations
anneal out by dissolution of ecarbide precipitates.
.
4008C–7208C: Cementite particles coarsen and tend to have spherical shape.
3.7 FORMATION OF AUSTENITE
Heat treatments for steels generally consist of heating above the austenitization temperature, holding
at austenitization temperature for a certain period of time, and cooling processes. The recrystalliza
tion of austenite has great inﬂuence on microstructures in steel. Austenite is a solid solution of
carbon in an FCC structure. It is stable at high temperatures, its strength and hardness are low, and
its speciﬁc volume is small. The formation of austenite follows the general rule of solidstate phase
transformations; it is accomplished through nucleation and growth.
Nucleation of austenite grains generally takes place at the interfaces between ferrite and
cementite or the interface between pearlite colonies at the eutectoid temperature (Figure 3.22).
The TEM micrograph shows the nucleation of austenite at the interface between ferrite and
cementite. Pearlite is quickly dissolved because the diffusion distances for carbon are relatively
short. The diffusion distance is of the order of the interlamellar spacing of pearlite. Ferrite then
transforms into austenite. The homogenization of carbon distribution follows, and ﬁnally grain
growth of austenite is predominant at high temperature or long austenization times.
Formation of austenite depends on the nucleation rate and growth speed of austenite. Under the
condition of homogeneous nucleation, the nucleation rate of austenite grains is deﬁned as a function
of temperature:
I ¼ Cexp À
Q þW
kT
_ _
(3:35)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 114 3.11.2008 1:49pm Compositor Name: JGanesan
114 Handbook of Thermal Process Modeling of Steels
where
C is a constant
Q is the activation energy of diffusion
W is the critical nucleation work
T is the temperature in Kelvin
k is the Boltzmann constant
When the strain energy is omitted, the critical nucleation work is represented by [17]
W ¼ A
s
3
DG
2
v
(3:36)
where
A is a constant
s is the energy of austenite–pearlite interface per unit area
DG
v
is the difference of free energy between pearlite and austenite
After nucleation, the growth speed of grains is determined by the immigration speed of phase
interfaces. It is assumed that the inﬂuence of carbon diffusion on interface immigration speed
is neglected; the linear speed of the interface immigration during the formation of austenite is
described by the following equation [17]:
G ¼ ÀKD
g
c
dC
dx
1
DC
B
(3:37)
where
K is a constant
D
g
c
is the diffusion coefﬁcient of carbon atoms in austenite
dC=dx is the gradient of carbon concentration in austenite at the interface
DC
B
is the difference of carbon concentration in two phases at the interface of austenite and
ferrite or at the interface of austenite and cementite
D
g
C
follows the Arrhenius equation
D
g
C
¼ D
0
exp À
Q
RT
_ _
(3:38)
0.1 μm
FIGURE 3.22 Crystallography of nucleation of austenite at the ferrite–cementite interface. (From Shtansky,
D.V., Nakai, K., and Ohmori, Y., Acta Mater., 47, 2619, 1999.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 115 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 115
where
D
0
is a constant
Q is the activation energy
R is the gas constant
T is the temperature in Kelvin
The linear speed of the austenite boundary surfaces moving into ferrite is described as the
following equation [17]:
G
g!a
¼ ÀKD
g
C
dC
dx
1
C
g
gÀa
ÀC
a
gÀa
(3:39)
where
C
g
gÀa
is the concentration of carbon near the phase boundary in austenite
C
a
gÀa
is the concentration of carbon near the phase boundary in ferrite
The linear speed of the austenite boundary surfaces moving into cementite is described as
follows [17]:
G
g!a
¼ ÀKD
g
C
dC
dx
1
6:67 ÀC
a
gÀc
(3:40)
Temperature and heating rate play important roles in the formation of austenite. A continuous
heating transformation diagram of pearlite–austenite transformation for a 0.4% C to 5% Cr steel
obtained by dilatometry experiments, is shown in Figure 3.23. The diagram shows that the pearlite–
austenite transformation period decreases while overheating increases.
3.8 PRECIPITATION STRENGTHENING
Percipitation strengthening is one of the commonly used methods to improve the strength of alloys.
For the application of this process, the equilibrium phase diagram must show partial solid solubility;
and the slope of the solvus line must be such that there is a lower solubility at a lower temperature
1000 100
800
850
900
T
e
m
p
/
Њ
C
T (s)
Ac1
Ac3
A
B
C
D
E
FIGURE 3.23 Continuous heating transformation diagram of a 0.4%C–5%Cr steel, curves A, B, C, D, and E
refer to heating rate of 508C min, 2008C min, 3008C min, 4008C min, and 6008C=min, respectively.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 116 3.11.2008 1:49pm Compositor Name: JGanesan
116 Handbook of Thermal Process Modeling of Steels
than at a higher temperature. Two stages are generally required to produce age hardening: solution
treatment and aging. Material is ﬁrst heated to the singlephase region and held there until a
homogeneous solid solution is obtained. The alloy is then cooled rapidly to room temperature at
which two phases exist in equilibrium. A supersaturated solution results, with the excess second
phase trapped in solution. The excess solute tends to come out of the solution. The speed at which
precipitation occurs varies with temperature.
According to the coherent lattice theory, after solution treatment and rapid cooling, the alloy is
in a supersaturated condition, with the solute atoms randomly distributed in the lattice structure.
During an incubation period, the excess solute atoms tend to migrate to certain crystallographic
planes, forming embryos of the precipitate. During aging, these embryos form an intermediate
crystal structure maintaining coherency with the existing lattice structure. The excess phase will
have different lattice parameters from those of the matrix phase, and as a result of atom matching
the matrix distorts considerably. This distortion interferes with the movement of dislocations and
causes a remarkable increase in hardness and strength during aging. As aging continues the
formation of the equilibrium excess phase with its own lattice structure results in coherency loss
with the matrix and less distortion. Then, hardness and strength starts to decrease (overaging).
The degree of improvement in hardness and strength by aging process varies depending on the
alloy type, mainly due to the effect of the precipitate on lattice distortion. In the case of steels, there
are some applications of precipitation hardening process such as precipitationhardening of stainless
steels, maraging, and ausforming.
.
Precipitation hardening in stainless steels occurs if coherent or partially coherent precipi
tates form during slow cooling or aging of a supersaturated solid solution. An effective
strengthening can be obtained if particles that contain Cu or Al, with interstitials bonded
strongly to the alloying elements, like Nb, Ti, or V, are precipitated. These steels are
usually solutionannealed at the mill and supplied in that condition; after forming process
they are aged to improve hardness and strength.
.
Maraging steels are lowC iron, containing 10%–25% Ni, and the substitutional elements
which produce precipitates. They are considered to be martensitic as annealed and attain
ultrahigh strength after aging in martensitic condition. The martensite formed is soft and
tough compared to that of conventional lowalloy steels.
.
Ausforming consists of deforming unstable austenite of moderately alloyed steels during
cooling by avoiding transformations to pearlite or bainite, and immediate quenching below
M
s
(martensite start) temperature. Ausform steels should have a CCT diagram having a
wide gap between pearlite and bainite formation temperatures. The resultant microstructure
consists of ﬁne martensitic plates, the size and dispersion of which are determined by prior
austenite grain size and the amount of plastic deformation. The improvement in the
strength can be explained by combined effects of increased dislocation density and
precipitation hardening.
When a precipitate of phase b forms from a solid matrix of phase a, both nucleation and growth
occur. Nucleation of precipitates of a new phase easily occurs on surfaces that exist in the matrix;
heterogeneous nucleation dominates the nucleation process in precipitation.
After nucleation, precipitates continue to grow by longrange diffusion and redistribution of
atoms. Diffusing atoms are ﬁrst detached from their original location, then move to the nucleus, and
are ﬁnally incorporated into the nucleus of precipitates. If the diffusing atoms are tightly bonded
within their original locations, the detachment process is difﬁcult and limits the growth rate. If the
attaching process is difﬁcult, the growth rate is also limited.
The precipitation process is completed in a shorter time at a higher temperatures, because the
diffusion coefﬁcient is greater provided the number of nuclei is the same. The rate of transformation
is given by the Avrami type equation (see equation 3.4).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 117 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 117
Both nucleation and growth are heat activation processes; the temperature has a strong effect on
the precipitation process.
The increase in the yield strength results from the interaction of precipitates with dislocations.
When a dislocation moves on its slip plane containing a distribution of precipitates, it may cut
through the particles or avoid them by moving out of its slip plane or by bending between the
particles. Energy is consumed in these processes. The way that a dislocation encounters precipitates
depends on the nature of precipitation and the applied stress. The presence of precipitates makes the
dislocation movement difﬁcult, and the resistance to plastic deformation increases.
Figure 3.24 gives a series of video images showing how screw dislocations overcome an
obstacle. Dislocation 1 continues to slip past the obstacle and eventually the dipole is pinched
off. Dislocation 2 requires the crossslip of a segment of the dislocation at the obstacle.
3.9 MELTING
Phenomenon of melting can be found in many thermal processes, such as welding, laser processing,
and remelting hardening. Generally, the thermal vibration aptitude of atoms or ions increases with
the temperature. When the thermal vibration distance of atoms exceeds the critical value, the
arrangement of atoms changes from the longrange order to the shortrange order, and the solid
metal transforms to the molten liquid.
It is easy to undercool the hightemperature molten metal, but it is difﬁcult and often impossible
to overheat a free metal. Metals always have inherent defects, such as vacancies, dislocations, grain
boundaries, and material surfaces. Atoms are in some local disorder in the sites of these defects,
where the atomic structure is similar to that of amorphous structures to some extent. The bulk metal
does not begin to melt simultaneously; defects act as preferential nucleation sites for the liquid
phase. When the melting temperature is approached, the liquidlike molecular mobility ﬁrst appears
in these defects. There are enough nuclei in defects to initiate melting close to or at the equilibrium
melting temperature. No activation energy is needed during melting; the metals can seldom be
overheated.
When a metal is heated to the melting temperature, an atomically thin liquid ﬁlm is generated at
the surface. The surface melting behavior is known as surface melting. Thin ﬁlms exhibit different
thermodynamic properties and lower thermal stability against phase transformation than their bulk
t = 0 s t = 4 s
t = 7 s t = 7.1 s
0.5 μm
2
2 2
1
1
2
1
1
FIGURE 3.24 Interaction of dislocations with obstacles. (From Lagow, B.W., et al., Mater. Sci. Eng. A,
309–310, 445, 2001.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 118 3.11.2008 1:49pm Compositor Name: JGanesan
118 Handbook of Thermal Process Modeling of Steels
solid counterparts. The melting point of thin ﬁlms is lower than the equilibrium melting temperature
of bulk solids.
For noneutectoid steels, the liquidphase fraction is calculated from the local temperature and
the solidus temperature (T
S
), and the liquidus temperature (T
L
) of the melting mushy region by the
following equation:
f ¼ 0, for T T
S
(solidphase region)
f ¼
T ÀT
S
T
L
ÀT
S
, for T
S
< T < T
L
(mushy region)
f ¼ 1, for T ! T
L
(liquidphase region)
REFERENCES
1. Askelan, D.R. and Phule, P.P., Essentials of Materials Science and Engineering, Thomson Learning, 2004.
2. Marinelli, M.C., Degallaix, S., and AlvarezArmas, I., Dislocation structure developed in the austenitic
phase of SAF 2507 duplex stainless steel, Materials Science and Engineering A, 435–436, 305, 2006.
3. Lagow, B.W., Robert, I.M., Jouiad, N., Lassila, D.H., Lee, T.C., and Birnbaum, H.K., Observation of
dislocation dynamics in the electron microscope, Materials Science and Engineering A, 309–310, 445, 2001.
4. Leslie, W.C., The Physical Metallurgy of Steels, McGrawHill, Kogakusha, 1982.
5. Hu, G.X., Cai, X., and Rong, Y.H., Fundamentals of Materials Science, 2nd ed., Shanghai Jiao Tong
University Press, 2006 (in Chinese).
6. Yanagida, A. and Yanagimoto, J., A novel approach to determine the kinetics for dynamic recrystallization
by using the ﬂow curve, Journal of Materials Processing Technology, 151, 33, 2004.
7. Wahabi, M.E. et al., EBSD study of purity effects during hot working in austenitic stainless steels,
Materials Science and Engineering A, 393, 83, 2005.
8. Sahay, S.S., Malhotra, C.P., and Kolkhede, A.M., Accelerated grain growth behavior during cyclic
annealing, Acta Materialia, 51, 339, 2003.
9. Raghavan, S. and Sahay, S.S., Modeling the grain growth kinetics by cellular automaton, Material Science
and Engineering A, 445–446, 203, 2007.
10. Hutchinson, C.R., Hackenberg, R.E., and Shiﬂet, G.J., The growth of partitioned pearlite in FeCMn
Steels, Acta Materialia, 52, 3565, 2004.
11. Offerman, S.E. et al., Insitu study of pearlite nucleation and growth during isothermal austenite decom
position in nearly eutectoid steel, Acta Materialia, 51, 3927, 2003.
12. Whiting, M.J., A reappraisal of kinetic data for the growth of pearlite in high purity FeC eutectoid alloys,
Scripta Materialia, 43, 969, 2000.
13. Yang, Z.G. and Fang, H.S., An overview on bainite formation in steels, Current Opinion in Solid State and
Materials Science, 9, 277, 2005.
14. Liu, C., Zhao, Z., and Bhole, S.D., Lathlike upper bainite in a silicon steel, Materials Science and
Engineering A, 434, 289, 2006.
15. Yang, J.R., Yu, T.H., and Wang, C.H., Martensitic transformations in AISI 440C stainless steel, Materials
Science and Engineering A, 438, 276, 2006.
16. Shtansky, D.V., Nakai, K., and Ohmori, Y., Pearlite to austenite transformation in an Fe2.6Cr1C alloy,
Acta Materialia, 47, 2619, 1999.
17. Xu, Z. and Zhao, L.C., Principles of SolidState Phase Transformation in Metals, Science Press,
Beijing, 2004.
FURTHER READING
1. Cahn, R.W. and Haasen, P. (eds.), Physical Metallurgy (4 volumes), NorthHolland Physics Publishing,
Amsterdam, 1983.
2. Verhoeven, J.D., Fundamentals of Physical Metallurgy, John Wiley & Sons, New York, 1975.
3. Avner, S.H., Introduction to Physical Metallurgy, McGrawHill, New York, 1974.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 119 3.11.2008 1:49pm Compositor Name: JGanesan
Physical Metallurgy of Thermal Processing 119
Gur/Handbook of Thermal Process Modeling of Steels 190X_C003 Final Proof page 120 3.11.2008 1:49pm Compositor Name: JGanesan
4
Mechanical Metallurgy
of Thermal Processing
Božo Smoljan
CONTENTS
4.1 Introduction........................................................................................................................ 122
4.2 Elements of Theory of Elasticity and Plasticity ................................................................ 122
4.2.1 Deﬁnition of Stress ................................................................................................. 123
4.2.2 Deﬁnition of Strain ................................................................................................. 127
4.2.3 Stress–Strain Relations in the Elastic Range .......................................................... 131
4.2.4 Yielding Criteria...................................................................................................... 133
4.2.5 Stress–Strain Relations in the Plastic Range .......................................................... 135
4.3 Fracture .............................................................................................................................. 141
4.3.1 Stress at a Crack Tip............................................................................................... 141
4.3.2 Grifﬁth Theory of Brittle Fracture .......................................................................... 144
4.3.3 Crack Opening Displacement ................................................................................. 145
4.3.4 JIntegral ................................................................................................................. 146
4.3.5 Strain Energy Density Failure Criterion ................................................................. 146
4.3.6 Statistical Nature of Fracture .................................................................................. 148
4.3.7 Fatigue Crack Propagation...................................................................................... 149
4.3.8 Creep ....................................................................................................................... 151
4.4 Micromechanisms of Plastic Deformation and Fracture ................................................... 154
4.4.1 Elements of Theory of Dislocations ....................................................................... 154
4.4.2 Micromechanisms of Plastic Deformation.............................................................. 158
4.4.3 Microstructural Aspects of Deformation Properties ............................................... 160
4.4.4 Micromechanism of Fracture .................................................................................. 162
4.4.5 Micromechanisms of Creep .................................................................................... 168
4.4.6 Micromechanism of Fatigue Cracking.................................................................... 169
4.4.7 Micromechanisms of Strengthening ....................................................................... 170
4.4.8 Toughening Mechanisms ........................................................................................ 172
4.5 Metal Behavior at Elevated Temperature .......................................................................... 174
4.5.1 Effect of Temperature on Flow Properties ............................................................. 174
4.5.2 Effect of Temperature on Fatigue........................................................................... 176
4.5.3 Transformation Plasticity ........................................................................................ 176
4.5.4 Effect of Stress on Microstructure Transformations............................................... 177
4.6 Residual Stresses, Distortions, and Fracture...................................................................... 178
References ..................................................................................................................................... 181
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 121 1.11.2008 2:46am Compositor Name: JGanesan
121
4.1 INTRODUCTION
The aim of this chapter is to give nonspecialists in mechanical metallurgy a general review of the
mechanical metallurgy ﬁelds related to thermal processing. It consists of a summary of the classical
knowledge in specialized works of mechanical metallurgy, needed for successful development of
mathematical modeling of thermal processes. The main references among these are given in the
bibliography. Mechanical metallurgy involves many disciplines, but it is the area of metallurgy that
is primarily concerned with the response of metals to loads.
Each type of engineering component failure is related with characteristic mechanical property,
and in designing of engineering components it is necessary to understand the types of failure. To
understand mechanical metallurgy problems it is necessary to understand theories of elasticity and
plasticity and strength of materials [1].
Generally, engineering components can fail due to fracture or due to excessive deformation [2].
The deformation could be elastic or plastic and fracture could be ductile or brittle. Yield and crack
resistance as well as modulus of elasticity are fundamental properties, which are used in design of
safety of engineering component applications. For speciﬁc conditions of member application,
speciﬁc material properties must be used in the design of failure risk due to excessive deformation.
For example, creep, or deformation at longtime constant load, cannot be successfully predicted
based on shorttime yield resistance.
Important assumptions in the theory of elasticity and plasticity are that the analyzed body is
continuous, homogeneous, and isotropic [3]. A continuous body does not contain empty spaces of
any kind. A body is homogeneous if it has identical properties at all points. A body is isotropic when
the analyzed property does not vary with orientation. There is no doubt that engineering materials
are heterogeneous, anisotropic, and not continuous on microscale, but on macroscale they are
usually statistically continuous, homogeneous, and isotropic. However, when metals are severely
deformed in a particular direction, as in rolling or forging, the mechanical properties may be
anisotropic on a macroscale. Discontinuity on macroscale may be present in porous castings or
powder metallurgy parts.
The determination of the relationship between mechanical behavior and structure is an import
ant subject in mechanical metallurgy. Without the dislocation concept, it is not possible to
understand the mechanical behavior of crystalline solids. Micromechanisms of the ﬂow and fracture
of metals are broadly considered in mechanical metallurgy [4]. When mechanical behavior is
understood in terms of metallurgical structure, it is generally possible to improve the mechanical
properties or at least to control them. In order to be able to predict the limitations of the engineering
applications of theories of elasticity and plasticity and fracture mechanics, designers need know
ledge of the physical principles of deformation and fracture.
Additionally, thermal metallurgy processes have to be considered with respect to temperature.
Especially the inﬂuence temperature on mechanical behavior of metals has to be taken into account in
designing these processes. The mechanical properties and metallurgical structure may continuously
change with time during thermal processing. Workpiece mechanical behavior at elevated temperature
could be important for tool life prediction and manufacture effectiveness in thermal processing.
During thermal processes, the tools are affected by many degradation processes, such as creep,
hightemperature fatigue, and thermal fatigue. Analysis of distortion and fracture appearances due to
residual stresses is an important part of mechanical metallurgy of thermal processing.
4.2 ELEMENTS OF THEORY OF ELASTICITY AND PLASTICITY
Mechanical designs of engineering components consist of prediction of stresses and strains that exist
in application. These stresses and strains have to be compared with characteristic material proper
ties, that is, characteristic strength and critical deformation of material that limit material application.
Characteristic material properties must be used with an appropriate failure criterion. The value of
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 122 1.11.2008 2:46am Compositor Name: JGanesan
122 Handbook of Thermal Process Modeling of Steels
stress for a particular material used in a particular way that is considered to be a safe stress is called
the working stress [5].
The theory of elasticity and plasticity is concerned with the relation between the mechanical
loads acting on the member, internal forces, and deformation. This section attempts to provide a
basic understanding of elasticity and plasticity, which are discussed in detail in Refs. [6–10].
Deﬁnition of stresses and strains at member points as result of internal forces existence is
usually the ﬁrst step in the analysis of the theory of elasticity or plasticity.
4.2.1 DEFINITION OF STRESS
Mathematical deﬁnition of stress at a point can be done by considering a body subjected to a system
of external forces. Figure 4.1 represents a body in equilibrium under the action of external forces
F
1
, F
2
, . . . , F
7
. To determine the stress at point O in a plane such as xy, part I of the body is removed
and replaced by the system of continuously distributed forces on plane xy, which will retain each
point in part II of the body in the same position as before the removal of part II. The stress acting at
point O in plane xy is deﬁned as
s = lim
DS÷0
DF
DS
(4:1)
The stress swill be in the direction of the force Fand must be referred to a particular plane. Stress will
be different for any other plane passing through the same point O. The total stress s can be resolved
into two components, a normal stress s
z
perpendicular to xy plane and a shearing stress t lying in the
plane xy of the area. Shearing stress can be decomposed in two components t
zx
and t
zy
. The ﬁrst
subscript indicates the normal to the plane under consideration, and the second subscript indicates the
direction of the stress. To specify the stress at a point completely it is necessary to specify the stresses
at that point on three mutually perpendicular planes passing through the point (Figure 4.2).
The stress at a point is given by nine quantities:
s
x
t
yx
t
zx
t
xy
s
y
t
zy
t
xz
t
yz
s
z
which are components of stress tensor s
ij
. Static equilibrium of forces in the corresponding
coordinate axis requires that stress at the point satisfy the following equations of equilibrium:
A A
I
II
F
6
F
1
F
3 F
21
F
12 F
4
F
7
F
5
F
2
(a) (b)
O
x
y
DS
z
S
s
t
s
z
t
zx
t
zy
FIGURE 4.1 Stresses at points: (a) body in equilibrium and (b) stress components.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 123 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 123
@s
x
@x
÷
@t
yx
@y
÷
@t
zx
@z
= ÷F
x
@t
xy
@x
÷
@s
y
@y
÷
@t
zy
@z
= ÷F
y
@t
xz
@x
÷
@t
yz
@y
÷
@s
z
@z
= ÷F
z
(4:2)
where F
i
are the components of the body forces per unit volume. In addition, from static equilibrium
it can be obtained that
t
yx
= t
xy
t
yz
= t
zy
t
xz
= t
zx
(4:3)
In tensor notation s
ij
=s
ji
. This expresses the fact that the stress tensor is symmetric and there are in
general only six independent components of stress:
s
ij
=
s
x
t
yx
t
zx
t
xy
s
y
t
zy
t
xz
t
yz
s
z
_
_
_
_
(4:4)
If the six components of stress at a point with respect to some coordinate system (x, y, z) are known,
it is possible to determine the stresses acting on any plane through this point. In Figure 4.3 an
inﬁnitesimal tetrahedron is shown with the known stresses acting on the three coordinate planes and
with unknown total stress t acting on the inclined plane ABC. The normal of the inclined plane has
direction cosines l, m, and n. Components of stress t in x, y, and z direction are t
x
, t
y
, and t
z
. Let
x
z
s
y
t
yx
O y
t
yz
t
xy
t
xz
s
x
s
z
t
zx
t
zy
dy
dx
dz
dy
∂y
∂t
yx
+ t
yx
dx
∂x
∂t
xy
+ t
xy
dx
∂x
∂s
x
+ s
x
dy
∂y
∂s
y
+ s
y
s
y
t
yx
x
y
s
x
t
xy
dy
∂y
∂s
y
+ s
y
dz
∂z
∂t
zy
+ t
zy
dy
∂y
∂t
yz
+ t
yz
dz
∂z
∂s
z
+ s
z
dx
∂x
∂s
x
+ s
x
s
z
t
zy
y
z
s
y
t
yz
dx
∂x
∂s
x
+ s
z
dz
∂z
∂t
zx
+ t
zx
dx
∂x
∂t
xz
+ t
xz
sx
t
xz
z
x
s
z
t
zx
FIGURE 4.2 Stresses acting on an inﬁnitesimal cube.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 124 1.11.2008 2:46am Compositor Name: JGanesan
124 Handbook of Thermal Process Modeling of Steels
magnitudes of stress vector t and its x, y, and z components be designated by t, t
x
, t
y
, and t
z
. Static
equilibrium of forces requires that these stresses satisfy Equation 4.5:
t
x
DS = s
x
lDS ÷t
yx
mDS ÷t
zx
nDS
t
x
= ls
x
÷mt
yx
÷nt
zx
(4:5)
Similarly
t
y
= lt
xy
÷ms
y
÷nt
zy
t
z
= lt
xz
÷mt
yz
÷ns
z
(4:6)
or in tensor notation:
t
j
= l
i
s
ij
(4:7)
where l, m, and n are replaced by l
1
, l
2
, and l
3
.
The magnitude of normal stress t
n
on the inclined plane can be determined by projecting the
stresses t
x
, t
y
, and t
z
onto the normal to inclined plane:
t
n
= lt
x
÷mt
y
÷nt
z
(4:8)
t
n
= l
2
s
x
÷m
2
s
y
÷n
2
s
z
÷2 lmt
xy
÷mnt
yz
÷nlt
zx
_ _
(4:9)
The magnitude of resultant shear stress t
s
on inclined plane can be expressed by
t
2
s
= t
2
÷t
2
n
= t
2
x
÷t
2
y
÷t
2
z
÷t
2
n
(4:10)
A plane with only normal stresses is deﬁned as principal stress plane, and its normal direction is
called a principal direction. Since total stress acting on principal plane has the same direction
cosines as the normal, its components in the x, y, and z directions are t
x
=lt; t
y
=mt; t
z
=nt; and then
l s
x
÷t ( ) ÷mt
yx
÷nt
zx
= 0
lt
xy
÷m s
y
÷t
_ _
÷nt
zy
= 0
lt
xz
÷mt
yz
÷n s
z
÷t ( ) = 0
(4:11)
x
z
s
y
t
yx
N
B
A
C
O
D
y
A
B
C
x
z
t
z
t
y
t
x
t
s
t
n
t
D
y
t
yz
t
xy
s
x
s
z
t
xz
t
zx
t
zy
FIGURE 4.3 Stresses acting on an inﬁnitesimal tetrahedron.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 125 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 125
or in tensor notation:
l
i
s
ij
÷d
ij
t
_ _
= 0 (4:12)
where l, m, and n are replaced by l
1
, l
2
, and l
3
. Since the nontrivial solution of Equation
4.11 gives
s
ij
÷d
ij
t
¸
¸
¸
¸
=
s
x
÷t t
yx
t
zx
t
xy
s
y
÷t t
zy
t
xz
t
yz
s
z
÷t
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
= 0 (4:13)
a characteristic cubic equation for t is
t
3
÷I
1
t
2
÷I
2
t ÷I
3
= 0 (4:14)
where
I
1
= s
x
÷s
y
÷s
z
I
2
= t
2
xy
÷t
2
yz
÷t
2
zx
÷ s
x
s
y
÷s
y
s
z
÷s
z
s
x
_ _
I
3
= s
x
s
y
s
z
÷2t
xy
t
yz
t
zx
÷ s
x
t
2
yz
÷s
y
t
2
zx
÷s
z
t
2
xy
_ _
(4:15)
Values I
1
, I
2
, and I
3
are called the ﬁrst, second, and third invariants of the stress tensor. The
characteristic cubic Equation 4.14 has three real roots, which are referred to as principal stresses
designated by s
1
, s
2
, and s
3
. The orientation of principal stress can be computed from Equation
4.11 substituting any of solution t in Equation 4.14. The stress tensor can be written in terms of the
principal stresses as
s
ij
=
s
1
0 0
0 s
2
0
0 0 s
3
_
_
_
_
(4:16)
The magnitude of shear stress acting on some inclined plane can be expressed by principal
stresses:
t
2
s
= s
1
÷s
2
( )
2
l
2
m
2
÷ s
1
÷s
3
( )
2
l
2
n
2
÷ s
2
÷s
3
( )
2
m
2
n
2
(4:17)
where l, m, and n are the direction cosines between the normal to the inclined plane and three
principal axes. The maximum or principal shear stresses act on the planes with the following
combinations of direction cosines that bisect the angles between two of the three principal axes:
l m n
0 ±2
÷1=2
±2
÷1=2
±2
÷1=2
0 ±2
÷1=2
±2
÷1=2
±2
÷1=2
0
(4:18)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 126 1.11.2008 2:46am Compositor Name: JGanesan
126 Handbook of Thermal Process Modeling of Steels
The magnitude of maximum shear stresses can be calculated by
t
1
= ±
1
2
s
2
÷s
3
( )
t
2
= ±
1
2
s
1
÷s
3
( )
t
3
= ±
1
2
s
1
÷s
2
( )
(4:19)
Octahedral shear stresses are acting on the octahedral planes. The direction cosines between the
normal to these planes and the three principal axes are then l =m=n =±3
÷1=2
and
t
oct
=
1
3
s
1
÷s
2
( )
2
÷ s
2
÷s
3
( )
2
÷ s
3
÷s
1
( )
2
_ _
1=2
(4:20)
or
t
oct
=
1
3
s
x
÷s
y
_ _
2
÷ s
y
÷s
z
_ _
2
÷ s
z
÷s
x
( )
2
÷6 t
2
xy
÷t
2
yz
÷t
2
zx
_ _ _ _
1=2
(4:21)
The stress tensor can be decomposed into the spherical part, which involves only pure tension or
compression, and the deviatoric part, which represents shear stresses:
s
ij
= s
H
d
ij
÷s
/
ij
(4:22)
where s
H
is the hydrostatic stress:
s
H
=
1
3
s
1
÷s
2
÷s
3
( ) =
1
3
s
x
÷s
y
÷s
z
_ _
(4:23)
The deviatoric stress tensor is important in causing plastic deformation and it is deﬁned by
s
/
ij
= s
ij
÷s
H
d
ij
=
s
x
÷s
H
t
yx
t
zx
t
xy
s
y
÷s
H
t
zy
t
xz
t
yz
s
z
÷s
H
_
_
_
_
(4:24)
The principal directions are the same for the deviatoric stress tensor as for the original stress tensor.
The principal values of stress deviators s
/
1
, s
/
2
, and s
/
3
are deﬁned as s
/
i
= s
i
÷s
H
. The charac
teristic cubic equation for deviatoric stress is
s
/3
÷J
1
s
/2
÷J
2
s
/
÷J
3
= 0 (4:25)
where s
/
=t ÷1=3I
1
(Equation 4.14). The ﬁrst invariant J
1
is equal to zero; invariant J
2
is
J
2
=
1
6
s
x
÷s
y
_ _
2
÷ s
y
÷s
z
_ _
2
÷ s
z
÷s
x
( )
2
÷6 t
2
xy
÷t
2
yz
÷t
2
zx
_ _ _ _
(4:26)
and the third invariant J
3
is the determinant of Equation 4.24.
4.2.2 DEFINITION OF STRAIN
The displacements of a body’s points can be the result of translation, rotation, and deformation.
If the length of a line joining any two points in a body is changed, the body is deformed or strained.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 127 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 127
Deformation of materials can be recognized as dilatation, that is, change in volume, or as angular
distortion, that is., change in shape. The speciﬁcation of strain at a point is the same for elastic and
for plastically deforming bodies and it is a geometric problem. After straining a vector r is deformed
into the vector r
/
in Figure 4.4. A vector dr is a result of different displacements of points P
0
and P
and it represents rigid body motions and deformation. The components of vector dr are
dr
x
= r
/
x
÷r
x
= u ÷u
0
dr
y
= r
/
y
÷r
y
= v ÷v
0
dr
z
= r
/
z
÷r
z
= w ÷w
0
(4:27)
where
u
0
, v
0
, and w
0
are the components of displacement of the point P
0
u, v, and w are the components of displacement of the point P
Displacements of P expanded in a Taylor series about P
0
are (higher order terms are neglected)
u = u
0
÷
@u
@x
r
x
÷
@u
@y
r
y
÷
@u
@z
r
z
v = v
0
÷
@v
@x
r
x
÷
@v
@y
r
y
÷
@v
@z
r
z
w = w
0
÷
@w
@x
r
x
÷
@w
@y
r
y
÷
@w
@z
r
z
(4:28)
and
dr
x
=
@u
@x
r
x
÷
@u
@y
r
y
÷
@u
@z
r
z
dr
y
=
@v
@x
r
x
÷
@v
@y
r
y
÷
@v
@z
r
z
dr
z
=
@w
@x
r
x
÷
@w
@y
r
y
÷
@w
@z
r
z
(4:29)
z
x
y
P(x, y, z)
P
0
(x
0
, y
0
, z
0
)
PЈ
0
(x
0
Ј
, y
00
Ј
, z
0
Ј )
PЈ(xЈ, yЈ, zЈ)
r
rЈ
dr
r
FIGURE 4.4 Displacement vectors.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 128 1.11.2008 2:46am Compositor Name: JGanesan
128 Handbook of Thermal Process Modeling of Steels
or in tensor notation dr
i
=u
i,j
r
j
. The tensor
u
i,j
=
@u
@x
@u
@y
@u
@z
@v
@x
@v
@y
@v
@z
@w
@x
@w
@y
@w
@z
_
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
_
(4:30)
is referred to as the relative displacement tensor. It is not difﬁcult to prove that for rigid body motion
the relative displacement tensor u
i,j
is antisymmetric: u
i,j
=÷u
j,i
. If the relative displacement tensor
is decomposed into symmetric and antisymmetric parts
u
i,j
= «
ij
÷v
ij
=
1
2
u
i,j
÷u
j,i
_ _
÷
1
2
u
i,j
÷u
j,i
_ _
(4:31)
the strain tensor «
ij
represents pure deformation and rotation tensor v
ij
represents rigid body motion:
«
ij
=
@u
@x
1
2
@u
@y
÷
@v
@x
_ _
1
2
@u
@z
÷
@w
@x
_ _
1
2
@u
@y
÷
@v
@x
_ _
@v
@y
1
2
@v
@z
÷
@w
@y
_ _
1
2
@u
@z
÷
@w
@x
_ _
1
2
@v
@z
÷
@w
@y
_ _
@w
@z
_
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
_
(4:32)
v
ij
=
0
1
2
@u
@y
÷
@v
@x
_ _
1
2
@u
@z
÷
@w
@x
_ _
1
2
@v
@x
÷
@u
@y
_ _
0
1
2
@v
@z
÷
@w
@y
_ _
1
2
@w
@x
÷
@u
@z
_ _
1
2
@w
@y
÷
@v
@z
_ _
0
_
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
_
(4:33)
Strain components «
11
=«
x
; «
22
=«
y
; and «
33
=«
z
are normal strains, while «
12
=«
xy
; «
13
=«
xz
; and
«
23
=«
yz
are shear strains (Figure 4.5).
As for stress, it is possible to deﬁne principal strains, which correspond with principal planes
and principal direction. The principal planes are three mutually orthogonal planes where only
normal strains exist. The roots of cubic equation are the solution for three principal strains:
«
3
÷I
/
1
«
2
÷I
/
2
« ÷I
/
3
= 0 (4:34)
x
y
A
B
D
C
CЈ
DЈ
BЈ
u
e
xy
e
yx
FIGURE 4.5 Angular distortion.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 129 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 129
Invariants appearing in a characteristic cubic equation are
I
/
1
= «
x
÷«
y
÷«
z
I
/
2
= «
2
xy
÷«
2
yz
÷«
2
zx
÷ «
x
«
y
÷«
y
«
z
÷«
z
«
x
_ _
I
/
3
= «
x
«
y
«
z
÷2«
xy
«
yz
«
zx
÷ «
x
«
2
yz
÷«
y
«
2
zx
÷«
z
«
2
xy
_ _
(4:35)
The orientation of principal strains can be determined from
«
x
÷« ( )l ÷«
xy
m ÷«
xz
n = 0
«
xy
l ÷ «
y
÷«
_ _
m ÷«
yz
n = 0
«
xz
l ÷«
yz
m ÷ «
z
÷« ( )n = 0
(4:36)
where l, m, and n are direction cosines of the principal strain plane. In analogy to the deﬁnition of
stresses at the point, maximum shearing strains can be determined by
g
1
= ±
1
2
«
2
÷«
3
( )
g
2
= ±
1
2
«
1
÷«
3
( )
g
3
= ±
1
2
«
1
÷«
2
( )
(4:37)
Octahedral shear strain can be calculated in terms of principal strains
g
2
oct
=
1
9
«
1
÷«
2
( )
2
÷ «
2
÷«
3
( )
2
÷ «
3
÷«
1
( )
2
_ _
(4:38)
or in terms of actual strains:
g
2
oct
=
1
9
«
x
÷«
y
_ _
2
÷ «
y
÷«
z
_ _
2
÷ «
z
÷«
x
( )
2
÷6 «
2
xy
÷«
2
yz
÷«
2
zx
_ _ _ _
(4:39)
Since the total strain involves both the volume or dilatational strain and shape or deviatoric strain,
the strain tensor can be decomposed into a spherical strain tensor, which represents volumetric
strain, and deviatoric strain tensor «
/
ij
, which represents the distortional strain.
«
ij
= «
/
ij
÷«
H
d
ij
(4:40)
where «
H
=1=3 («
1
÷«
2
÷«
3
). The change in volume per unit volume is
DV
V
= 1 ÷«
1
( ) 1 ÷«
2
( ) 1 ÷«
3
( ) ÷1 (4:41)
and after neglecting products of the strains for small strains:
DV
V
= «
1
÷«
2
÷«
3
= «
x
÷«
y
÷«
z
= 3«
H
(4:42)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 130 1.11.2008 2:46am Compositor Name: JGanesan
130 Handbook of Thermal Process Modeling of Steels
The deviatoric strain tensor is
«
/
ij
=
«
x
÷«
H
«
xy
«
xz
«
xy
«
y
÷«
H
«
yz
«
xz
«
yz
«
z
÷«
H
_
_
_
_
(4:43)
The invariants of the deviatoric strain tensor are
J
/
1
= 0
J
/
2
=
1
3
I
/2
1
÷3I
/
2
_ _
J
/
3
=
1
27
2I
/
3
1
÷9I
/
1
I
/
2
÷27I
/
3
_ _
(4:44)
Since the three displacements u, v, and w are deﬁned by six equations in terms of strain tensor
components, strains must satisﬁed equations of compatibility in order that the displacements be
continuous functions of the coordinates:
@
2
«
x
@y
2
÷
@
2
«
y
@x
2
= 2
@
2
«
xy
@x@y
@
2
«
y
@z
2
÷
@
2
«
z
@y
2
= 2
@
2
«
yz
@y@z
@
2
«
z
@x
2
÷
@
2
«
x
@z
2
= 2
@
2
«
xz
@x@z
@
@x
÷
@«
yz
@x
÷
@«
zx
@y
÷
@«
xy
@z
_ _
=
@
2
«
x
@y@z
@
@y
÷
@«
zx
@y
÷
@«
xy
@z
÷
@«
yz
@x
_ _
=
@
2
«
y
@z@x
@
@z
÷
@«
xy
@z
÷
@«
yz
@x
÷
@«
zx
@y
_ _
=
@
2
«
z
@x@y
(4:45)
4.2.3 STRESS–STRAIN RELATIONS IN THE ELASTIC RANGE
The relation between stress and strain for an isotropic material, including thermal strains, and using
the tensor notation, can be expressed as
«
ij
=
1
2m
s
ij
÷d
ij
n
E
Q÷aT
_ _
(4:46)
or in engineering notation:
«
x
=
1
E
s
x
÷n s
y
÷s
z
_ _ _ ¸
÷aT
«
y
=
1
E
s
y
÷n s
x
÷s
z
( )
_ ¸
÷aT
«
z
=
1
E
s
z
÷n s
x
÷s
y
_ _ _ ¸
÷aT
«
xy
=
1
2m
t
xy
=
1 ÷n
E
t
xy
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 131 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 131
«
yz
=
1
2m
t
yz
=
1 ÷n
E
t
yz
«
zx
=
1
2m
t
zx
=
1 ÷n
E
t
zx
(4:47)
Solved for the stresses the relation between stress and strain is
s
ij
= 2m «
ij
÷d
ij
aT
_ _
÷d
ij
l u ÷3aT ( ) (4:48)
where
E is the elastic modulus
n is the Poisson’s ratio
a is the coefﬁcient of linear thermal expansion
T is the temperature above some arbitrary reference temperature
l and m are the Lame’s constants related to E and n by the relations
l =
nE
1 ÷n ( ) 1 ÷2n ( )
(4:49)
m = G =
E
2 1 ÷n ( )
(4:50)
where G is the shear modulus, and
Q = s
x
÷s
y
÷s
z
= I
1
(4:51)
u = «
x
÷«
y
÷«
z
= I
/
1
(4:52)
The relation between deviatoric strain and deviatoric stress tensors is
«
/
ij
=
1
2m
s
/
ij
(4:53)
If external forces deform some body, the accumulated elastic strain energy per unit volume is
U =
1
2
s
ij
«
ij
(4:54)
Substituting Equation 4.48 into Equation 4.54, the elastic strain energy can be expressed as
U = m«
ij
«
ij
÷
l
2
u
2
÷
2m ÷3l
2
aTu (4:55)
Equation 4.55 can be written as
U =
2m ÷3l
6
I
/2
1
÷3aTI
/
1
_ _
÷2mJ
/
2
(4:56)
The ﬁrst term in Equation 4.56 is the energy of the volume changing or dilatation, and the
second term is the energy of shape changing or deviation. The energy of deviation can be
expressed by
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 132 1.11.2008 2:46am Compositor Name: JGanesan
132 Handbook of Thermal Process Modeling of Steels
U
d
= 2mJ
/
2
= 3mg
2
oct
=
1
2m
J
2
=
3
4m
t
2
oct
(4:57)
To solve an elastic problem, the stresses and the strains that will satisfy the equilibrium equations,
the compatibility equations, and the boundary conditions have to be found. The stresses have to be
related to the strains through the stress–strain relations. These equations can be reduced to three
equations, involving only the displacement u
i
:
m\
2
u
i
÷ l ÷m ( )u
,i
÷F
i
= 0 (4:58)
where
\
2
is the Laplacian operator
F
i
are the body forces
Invariant u can be written as
u =
@u
@x
÷
@v
@y
÷
@w
@z
(4:59)
4.2.4 YIELDING CRITERIA
From the viewpoint of design, plasticity is concerned with predicting the maximum load that can be
applied to a body without causing yielding, or predicting the dilatation and angular distortions of the
purposely stressed member into the plastic region. Resistance to the failure due to excessive plastic
deformation is usually deﬁned by the yield resistance, which must be used with an adequate yield
criterion.
The yield criterion depends on the state of stress at the point. The criterion for initial yield at
which a material will start yielding can be written as
f s
ij
_ _
= s
s
(4:60)
Equation 4.60 represents a hypersurface in the sixdimensional stress space. Points on this hypersur
face represent a stress states at which yielding can begin. The surface described by Equation 4.60 in
the stress space is called the yield surface and the function is called the yield function. For an
isotropic material, the yield criterion must be independent of the choice of axes, that is, f(s
ij
) must
be an invariant function. Since experimental observations give that pure hydrostatic pressure does
not cause yielding, Equation 4.60 can be expressed by principal stress deviators:
g s
/
1
, s
/
2
, s
/
3
_ _
= s
s
(4:61)
or by the stress deviators invariants:
h J
2
, J
3
( ) = s
s
(4:62)
For a perfectly plastic material, the yield surface remains constant. The yield stress is rising in the
workhardened material. If the workhardened material is unloaded and then loaded again, add
itional yielding will not occur until the new value of s
s
in Equation 4.60 is achieved. The function
s
s
is then the strainhardening function and will depend on the complete previous stress and strain
history of the material and its strainhardening properties [11].
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 133 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 133
At present, there are two generally accepted criteria for predicting the onset of yielding in
ductile metals: distortion energy theory and maximum shear stress theory.
The distortion energy theory or the von Mises yield criterion presumes that yielding begins
when the distortion energy equals the distortion energy at yield in uniaxial tension:
U
d
=
1
2m
J
2
=
3
4m
t
2
oct
(4:63)
At the yield point in uniaxial tension, the second invariant of the deviatoric stress is
J
2
=
1
3
s
2
Y
(4:64)
where s
Y
is the yield stress in uniaxial tension test. The yield condition can be written as
1
2
s
1
÷s
2
( )
2
÷ s
2
÷s
3
( )
2
÷ s
3
÷s
1
( )
2
_ _
= s
2
Y
(4:65)
A distortion energy theory plot for biaxial case is shown in Figure 4.6. This criterion shows good
agreement with experimental observations and is often used by designers.
Maximum shear stress theory, or Tresca criterion, assumes that the yielding will occur when the
maximum shear stress reaches the value of the maximum shear stress at yield in uniaxial tension test
1=2s
Y
, that is, when any one of the following conditions is reached.
s
1
÷s
2
= ±s
Y
s
2
÷s
3
= ±s
Y
s
3
÷s
1
= ±s
Y
(4:66)
A plot illustrating Tresca criterion for the case of biaxial stress is shown in Figure 4.6. The Tresca
criterion shows good agreement with experimental observations.
s
2
Distortion energy
theory
Maximum shear
stress theory
s
1
s
Y
s
2
−s
Y
−s
Y
FIGURE 4.6 Plots of yield criteria in s
1
s
2
plane.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 134 1.11.2008 2:46am Compositor Name: JGanesan
134 Handbook of Thermal Process Modeling of Steels
4.2.5 STRESS–STRAIN RELATIONS IN THE PLASTIC RANGE
In the plastic range the strains are not generally uniquely determined by the stresses, they depend on
the history of loading. Figure 4.7 shows a typical stress–strain curve for a uniaxial stressed
specimen.
The true stress and strain values have to be calculated in respect of the actual crosssectional
area S and the actual specimen length L:
s
true
=
F
S
; «
true
=
_
dL
L
= ln
L
L
0
= ln
S
0
S
(4:67)
where
S
0
is the original crosssectional area
L
0
is the original length
The true stress–true strain diagram is also known as ﬂow diagram. For loading in the plastic range,
which is beyond the initial yield point s
Y
, the total strain consists of elastic and plastic strain. On
unloading to a point of zero stress, a permanent deformation exists. For small strains, it can be
accepted that total strain is [12]
« = «
e
÷«
p
(4:68)
After unloading to a state of zero stress, elastic strain «
e
decreases to zero and only plastic strain «
p
remains after the load is removed. On subsequent reloading a new yield point will be established
whose value is greater than the original yield stress. The plastic strain does not occur until the
original stress–strain curve is reached. This effect is known as stress hardening. Further loading
continues along the original stress–strain curve. For many metal materials and many other engin
eering materials it is common for the behaviors in the elastic range to be linear. In the following
developments of stress–strains relations linear elastic behavior is presumed to be in the elastic range
and it is accepted that elastic and plastic strains are additive Equation 4.68.
The total strains should be obtained by integration or summation of differentials or increments
of elastic and plastic strain, which are computed throughout the whole loading history.
Strain
S
t
r
e
s
s
True stresstrue strain curve
e
p
FIGURE 4.7 Stress–strain curves.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 135 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 135
The equations relating the increments of total strain to the stress deviations are given by the
Levy–Mises equations:
d«
x
s
/
x
=
d«
y
s
/
y
=
d«
z
s
/
z
=
d«
xy
t
xy
=
d«
yz
t
yz
=
d«
zx
t
zx
= dl (4:69)
In these equations, the elastic strains are ignored and these equations can be applied to problems of
large plastic ﬂow. In the elastic–plastic regions, it is a necessary to consider both elastic and plastic
strain. The generalized equations that include both the elastic and plastic components are known as
the Prandtl Reuss equations. In Prandtll–Reuss equations, it is assumed that the plastic strain
increment is proportional to the instantaneous deviatoric stress:
d«
p
x
s
/
x
=
d«
p
y
s
/
y
=
d«
p
z
s
/
z
=
d«
p
xy
t
xy
=
d«
p
yz
t
yz
=
d«
p
zx
t
zx
= dl (4:70)
In Equation 4.70 the increments of plastic strain depend only on the current values of the deviatory
stress state:
d«
p
ij
= dls
/
ij
(4:71)
Equation 4.71 can be written in terms of the actual stresses as
d«
p
x
=
2
3
dl s
x
÷
1
2
s
y
÷s
z
_ _
_ _
d«
p
y
=
2
3
dl s
y
÷
1
2
s
z
÷s
x
( )
_ _
d«
p
z
=
2
3
dl s
z
÷
1
2
s
x
÷s
y
_ _
_ _
d«
p
xy
= dlt
xy
d«
p
yz
= dlt
yz
d«
p
zx
= dlt
zx
(4:72)
In Equation 4.72 the relationship between the ratios of plastic strain increments in the different
directions is given as functions of actual stresses and the constant dl. From Equation 4.70 it
follows that
d«
p
x
÷d«
p
y
_ _
2
÷ d«
p
y
÷d«
p
z
_ _
2
÷ d«
p
z
÷d«
p
x
_ _
2
÷6 d«
p
xy
_ _
2
÷6 d«
p
yz
_ _
2
÷6 d«
p
zx
_ _
2
= (dl)
2
s
x
÷s
y
_ _
2
÷ s
y
÷s
z
_ _
2
÷ s
z
÷s
x
( )
2
÷6t
2
xy
÷6t
2
yz
÷6t
2
zx
_ _
(4:73)
or from Equations 4.21 and 4.39:
dg
p
oct
= dlt
oct
(4:74)
The constant dl is equal:
dl =
dg
p
oct
t
oct
(4:75)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 136 1.11.2008 2:46am Compositor Name: JGanesan
136 Handbook of Thermal Process Modeling of Steels
In theory of plasticity it is common to express the constant dl by equivalent or effective stress s
e
and equivalent or effective plastic strain increment d«
p
:
dl =
3
2
d«
p
s
e
(4:76)
where
s
e
=
3
ﬃﬃﬃ
2
_ t
oct
=
1
ﬃﬃﬃ
2
_ s
x
÷s
y
_ _
2
÷ s
y
÷s
z
_ _
2
÷ s
z
÷s
x
( )
2
÷6 t
2
xy
÷t
2
yz
÷t
2
zx
_ _ _ _
1=2
(4:77)
and
d«
p
=
ﬃﬃﬃ
2
_
dg
oct
=
ﬃﬃﬃ
2
_
3
d«
p
x
÷d«
p
y
_ _
2
÷ d«
p
y
÷d«
p
z
_ _
2
÷ d«
p
z
÷d«
p
x
_ _
2
_
÷6 d«
p
xy
_ _
2
÷6 d«
p
yz
_ _
2
÷6 d«
p
zx
_ _
2
_
1=2
(4:78)
Equation 4.71 becomes
d«
p
ij
=
3
2
d«
p
s
e
s
/
ij
(4:79)
For practical use of Equation 4.79 the relations between equivalent stress and equivalent strain have
to be determined experimentally, preferably using the uniaxial tension test. The equivalent stress s
e
and equivalent plastic strain increment d«
p
can be deﬁned based on a true stress–strain curve for
uniaxial tension test, substituting components of stresses and strains for uniaxial tension test in
Equations 4.77 and 4.78:
s
e
= s
x
d«
p
= d«
p
x
(4:80)
Since for uniaxial stress state yielding begins at the point where s
x
=s
Y
, the equivalent stress is
equal to the von Mises yield function, and it is seen that Prandtl–Reuss assumptions imply the von
Mises yield criterion.
The general stress–strain relations for any yield criterion are based on a uniﬁed approach will be
expressed as follows [8,13–15]. To obtain general stress–strain relations in the plastic range the
precise deﬁnition of work hardening is required. Work hardening implies that for such added
stresses a positive work will be done, and moreover, that there is no additional useful net energy
over the elastic energy that can be extracted from the material after removal of acting stresses.
A mathematical deﬁnition of work hardening can be written due to Drucker [13]:
ds
ij
d«
e
ij
÷d«
p
ij
_ _
> 0
ds
ij
d«
p
ij
_ 0
(4:81)
If h(s
ij
) is the loading function, a further plastic deformation takes place for workhardening material if
@h
@s
ij
ds
ij
> 0 (4:82)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 137 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 137
Loading function h can depend on the stress and strain history. Furthermore, linear relation between
inﬁnitesimals of stress and plastic strain can be assumed:
d«
p
ij
= C
ijkl
ds
kl
(4:83)
In this case the superposition principle is valid for stress and strain increments, and the increment of
stress d s
kl
in plastic range can be decomposed into two parts, ds
I
kl
, which produces no plastic ﬂow
and ds
II
kl
, which produces plastic ﬂow:
ds
kl
= ds
I
kl
÷ds
II
kl
(4:84)
Since increment ds
I
kl
produces no plastic work,
@h
@s
kl
ds
I
kl
= 0 (4:85)
and increment ds
II
kl
is proportional to the gradient of h(s
ij
):
ds
II
kl
= D
@h
@s
kl
(4:86)
where D>0 is scalar, and since
@h
@s
kl
ds
kl
=
@h
@s
kl
ds
II
kl
(4:87)
scalar D can be written as
D =
@h=@s
kl
@h=@s
mn
( ) @h=@s
mn
( )
ds
kl
(4:88)
Since every component of d«
p
ij
is proportional to D, Equation 4.87 gives
d«
p
ij
= g
ij
@h
@s
kl
ds
kl
(4:89)
where g
ij
depends on the stress and strain history. From Equation 4.81 it follows that
ds
I
ij
d«
p
ij
= ds
I
ij
g
ij
@h
@s
kl
ds
kl
= 0 (4:90)
and
ds
I
ij
g
ij
= 0 (4:91)
Equation 4.85 gives
g
ij
= G
@h
@s
ij
(4:92)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 138 1.11.2008 2:46am Compositor Name: JGanesan
138 Handbook of Thermal Process Modeling of Steels
where G is a scalar that depends on the stress and strain history. The general stress–strain relation is
derived by substituting Equation 4.92 into Equation 4.89:
d«
p
ij
= G
@h
@s
ij
@h
@s
kl
ds
kl
(4:93)
or
d«
p
ij
= G
@h
@s
ij
dh (4:94)
For the von Mises yield criterion:
h = J
2
=
1
6
s
1
÷s
2
( )
2
÷ s
2
÷s
3
( )
2
÷ s
3
÷s
1
( )
2
_ _
(4:95)
@h
@s
1
=
2
3
s
1
÷
1
2
s
2
÷s
3
( )
_ _
(4:96)
Then,
d«
p
1
=
2
3
dl s
1
÷
1
2
s
2
÷s
3
( )
_ _
(4:97)
where dl =G dh, which is in accordance with Prandtl–Reus equations.
For the Tresca yield condition, if the maximum principal stress s
1
and minimum principal stress
s
3
are known:
h =
1
2
s
1
÷s
3
( ) (4:98)
@h
@s
1
=
1
2
;
@h
@s
2
= 0;
@h
@s
3
= ÷
1
2
(4:99)
Then,
d«
p
1
=
1
2
dl; d«
p
2
= 0; d«
p
3
= ÷
1
2
dl (4:100)
For practical use of Equation 4.94 the relation between equivalent stress (function of stresses) and
equivalent strain (function of strains) has to be determined experimentally, preferably using the
uniaxial tension test. Since in uniaxial tension the yielding follows the stress–strain curve in the
uniaxial tension test, it is reasonable to accept the true ﬂow stress as equivalent stress and the true
strain as equivalent strain. Values of stress and strain for some other stress states can be found by
assuming that strain hardening effect depend only on the total plastic work and it is independent of
strain path. In this case the stress and increment of strain can be deﬁned by the equivalence of
increment of plastic work. An increment of plastic work can be written as
dW
p
= s
/
ij
d«
p
ij
= s
e
d«
p
(4:101)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 139 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 139
and
d«
p
=
s
/
ij
d«
p
ij
s
e
(4:102)
For von Mises yield criterion it follows that
d«
p
=
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
3
d«
p
ij
d«
p
ij
_
(4:103)
which agrees with Equation 4.78, and for Tresca criterion, if h =s
1
÷s
3
and s
1
>s
2
>s
3
:
d«
p
= d«
p
1
(4:104)
Although Equation 4.103 is derived for h =J
2
, it is reasonable to use it for nearly any h(J
2
J
3
) [8].
Substituting Equation 4.94 into Equation 4.103 gives
G dh =
ﬃﬃﬃ
3
2
_
d«
p
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
@h=@s
ij
_ _
@h=@s
ij
_ _
_ (4:105)
and the general stress–strain equation is [8,15]
d«
p
ij
=
ﬃﬃ
3
2
_
@h=@s
ij
_ _
d«
p
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
@h=@s
mn
( ) @h=@s
mn
( )
_ (4:106)
or for von Mises yield criterion where h =J
2
:
d«
p
ij
=
ﬃﬃﬃ
3
2
_
s
/
ij
s
e
d«
p
(4:107)
Since the plastic deformation depends on stress and strain history, to deﬁne the total plastic strain
components, it is necessary to integrate the inﬁnitesimals of plastic strain over the actual loading
path. If the total loading path is divided into N increments of load, the total strains at the end of the
mth increment can be written with thermal strains as [8]
«
ij
~
1
2m
s
ij
÷d
ij
n
E
Q÷aT
_ _
÷
m÷1
k=1
D«
p
ij,k
÷D«
p
ij,m
(4:108)
The fourth term on the righthand side of Equation 4.108 is the plastic strain accumulated in the ﬁrst
mth ÷1 increments of the load, and the ﬁfth term is the plastic strain due to the mth increment of the
load. The plastic strain increments for the current or mth increment of the load for the von Mises
criterion is
D«
p
ij
=
3
2
D«
p
s
e
s
/
ij
(4:109)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 140 1.11.2008 2:46am Compositor Name: JGanesan
140 Handbook of Thermal Process Modeling of Steels
where
D«
p
=
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
3
D«
p
ij
D«
p
ij
_
(4:110)
Values of D«
p
and s
e
are related to each other through the uniaxial tensile true stress–true strain
curve (Figure 4.7).
The displacements, the stresses, and the strains have to be determined to solve an elastic–plastic
problem. There are 15 unknowns, but there are 15 equations: the three equilibrium equations, the six
equations relating displacements with strains, and the six constitutive equations relating strains with
stresses. These equations are independent of the material properties and will hold irrespective of
whether the body behaves elastically or whether plastic ﬂow occurs. The difference between the
elastic problem and the plastic problem occurs in the nature of the stress–strain relations. As was
shown for elastic problems these equations can be reduced to three equations, involving only the
displacements.
In order to carry out such calculations in thermal processes it is necessary to know the material
properties, that is, yield stress, workhardening coefﬁcient, coefﬁcient of expansion, Young’s
modulus, and Poisson’s ratio in terms of temperature.
4.3 FRACTURE
Fracture destroys the cohesion of a material by creating a surface within the material. The risk of
fracture can be predicted using adequate fracture criteria based on relevant mechanical properties
determined by short or longtime tests. Since excessive plastic deformation occurs before ductile
fracture, prevention of excessive plastic deformation usually means prevention of ductile fracture,
but in brittle conditions the maximum stress has to be less than the critical stress for brittle fracture.
The critical stress for brittle fracture is a function of both material crack resistance and the nature of
existing cracks. Since highstrength materials have a low crack resistance, that is, low fracture
toughness, the conventional design criteria are nonadequate when these cracks exist. The occurrence
of lowstress fracture in highstrength materials persuaded the development of fracture mechanics.
Fracture mechanics considers the calculations of strength as a function of crack size or answers
questions such as the following: What is the maximum crack size that a material can sustain safely?
What is the strength of a structure as a function of crack size? How does the crack size relate to the
applied loads? What is the critical load required to extend a crack of known size, and is the crack
extension stable or unstable? How does the crack size increase as a function of time? How long does
it take for a crack to grow to the critical size [2].
Progressive fracture can occur in parts that are subjected to static or alternating stresses.
Fractures that occur under static loads are referred to as delayed and creep fractures [16]. Fractures
under alternating loads are referred to as fatigue fractures.
This section attempts to provide a basic understanding of the fracture phenomenon, which is
discussed in greater detail in Refs. [2,4,5,17].
4.3.1 STRESS AT A CRACK TIP
A crack in a material can be stressed in three different modes (Figure 4.8). Normal stresses result in
an increase in the openingmode denoted as mode I. The displacements in the crack surfaces
are perpendicular to the plane of the crack. In mode II or in slidingmode the displacement of the
crack surface is in the plane of the crack and perpendicular to the leading edge of the crack. In mode
III or in tearingmode, crack surface displacements are in the plane of the crack and parallel to
the leading edge of the crack. The superposition of the three modes describes the general case of
cracking [2].
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 141 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 141
Fracture stress s
f
in brittle condition decreases with decreasing radius r in the vicinity of a hole
until a limiting radius r
0
[18]. Below radius r
0
, fracture stress is approximately constant. The effect
of the notch radius r on the fracture stress s
f
is shown in Figure 4.9.
Stresses in the vicinity of the transverse crack (r =0) of length 2a in an inﬁnite plate plane
subjected to a tensile stress at the inﬁnity are equal (Figure 4.10) [19,20]:
s
x
= s
ﬃﬃﬃﬃﬃ
a
2r
_
cos
u
2
1 ÷sin
u
2
sin
3u
2
_ _
s
y
= s
ﬃﬃﬃﬃﬃ
a
2r
_
cos
u
2
1 ÷sin
u
2
sin
3u
2
_ _
t
xy
= s
ﬃﬃﬃﬃﬃ
a
2r
_
sin
u
2
cos
u
2
cos
3u
2
s
z
= 0 (plane stress)
s
z
= n s
x
÷s
y
_ _
(plane strain)
(4:111)
Mode I
openingmode
Mode II
openingmode
Mode III
openingmode
III II
I
FIGURE 4.8 Modes of cracking.
F
r
a
c
t
u
r
e
s
t
r
e
s
s
,
s
f
r
0
Notch radius, r
FIGURE 4.9 Effect of the notch radius (schematically).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 142 1.11.2008 2:46am Compositor Name: JGanesan
142 Handbook of Thermal Process Modeling of Steels
Equation 4.111 is valid only for small r compared to the crack size. Each term of Equation 4.111
represents the ﬁrst term of the general solution f
ij
(u). The distribution of the stress s
y
as a function of
r at u =0 is illustrated in Figure 4.11.
Generally, the stress and strain ﬁelds at the crack tip can be deﬁned by the factor K
I
known as
stress intensity factor:
s
ij
=
K
I
ﬃﬃﬃﬃﬃﬃﬃﬃ
2pr
_ f
ij
(u) with K
I
= s
ﬃﬃﬃﬃﬃﬃ
pa
_
(4:112)
where the subscript I stands for mode I.
Equations 4.111 and 4.112 are an elastic solution, which allows for the stresses to become
inﬁnite at the crack tip. In reality plastic deformation taking place at the crack tip keeps the stresses
ﬁnite. The size of the crack tip plastic zone can be determined using some of the yield criteria.
Crack extension occurs when the stresses and strains at the crack tip reach a critical value. This
means that fracture must be expected to occur when K reaches a critical value K
Ic
. The critical stress
intensity factor K
Ic
can be accepted as a material parameter and it is equal:
K
Ic
= s
c
ﬃﬃﬃﬃﬃﬃ
pa
_
(4:113)
If the K
Ic
value is known, the fracture strength of cracks of any size in the same material can be
predicted. For a specimen of ﬁnite size the stress intensity factor can be expressed by
K
I
= Ys
ﬃﬃﬃﬃﬃﬃ
pa
_
(4:114)
2a
y
s
x
q
r
t
xy
s
y
s
x
t
yx
dx
d
y
FIGURE 4.10 Crack in an inﬁnite plate.
σ
y
x
2a
FIGURE 4.11 Stress distribution at the crack tip.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 143 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 143
where Y is the parameter that depends on the specimen and crack geometry. A consistent K
Ic
can be
determined for a plane strain condition where it does not depend on plate thickness. For that reason
K
Ic
is referred to as plane strain fracture toughness. Highstrength materials usually have low
fracture toughness. Typical values of the fracture toughness of some metal alloys are given in
Table 4.1.
4.3.2 GRIFFITH THEORY OF BRITTLE FRACTURE
The basic equations of fracture mechanics have already been developed in 1921 by Grifﬁth [21].
Grifﬁth recognized that crack propagation would occur if the energy released upon crack growth is
sufﬁcient to provide all the energy that is required for crack growth:
dU
da
=
dW
da
(4:115)
where
U is the elastic energy
W is the energy required for crack growth
The term dU=da per inﬁnite cracked plate of unit thickness is (Figure 4.10)
dU
da
=
2ps
2
a
E
(4:116)
where E is the Young’s modulus. Term dU=da can be replaced by 2G:
G =
ps
2
a
E
(4:117)
G is the socalled elastic energy release rate per crack tip or the crack driving force. The energy
consumed in crack propagation is speciﬁed by R=dW=da, which is referred to as the crack resistance.
If R is a constant, the catastrophic crack growth occurs when G exceeds a certain critical value G
Ic
:
ps
2
c
a
E
= G
Ic
or s
c
=
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
EG
Ic
pa
_
(4:118)
The energy criterion Equation 4.118 is equivalent to the stress criterion Equation 4.113. Grifﬁth
derived his equation for glass, which is a very brittle material, and he assumed that R consists just of
surface energy. In ductile materials, such as metals, much work is dissipated in producing a plastic
TABLE 4.1
Fracture Toughness of HighStrength Materials
Material
Yield Strength,
s
Y
MPa
Fracture Toughness,
K
Ic
MPa m
1=2
Carbon steel 240 220
Maraging steel 2200 50
Bainitic steel 1450 65
Hightemperature steel 350 200
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 144 1.11.2008 2:46am Compositor Name: JGanesan
144 Handbook of Thermal Process Modeling of Steels
zone at the crack tip. For metals the surface energy is so small that it can be neglected and R is
mainly equal to plastic energy [22].
4.3.3 CRACK OPENING DISPLACEMENT
The linear elastic fracture mechanics concept can be used if the size of the crack tip plastic zone is
small compared to the size of the crack. Since plastic zone size is proportional to K
I
=s
Y
, the linear
elastic fracture mechanics is not usually available for low yield strength materials. Cottrell [23] and
Wells [24] independently introduced the concept of a critical crack opening displacement (COD) as
a fracture criterion for the study of crack initiation in situations where signiﬁcant plastic deformation
precedes fracture (Figure 4.12).
In the COD concept, crack extension can take place when the material at the crack tip has
reached a critical strain. Since the crack tip strain is related to the COD, it is assumed that crack
extension will begin when the COD reaches some critical value [2]. A measure of the amount of
crack tip plastic strain is the separation of the crack faces or crack tip opening displacement
(CTOD), usually denoted by d (Figure 4.12). It is accepted that CTOD is a material constant,
independent of specimen geometry and crack length. Criterion COD is equivalent to the K
Ic
and G
Ic
criterions in the case of linear elastic fracture mechanics. General relations between the CTOD
denoted by d, and K
I
and G
I
are:
d =
G
I
ls
Y
=
K
2
I
1 ÷n
2
( )
Els
Y
(4:119)
where factor l depends on the stress state and increases with increasing plane strain conditions
1 _ l _ 2. Critical stress can be expressed in terms of critical CTOD, denoted by d
c
:
s
c
=
1
Y
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
lEs
Y
pa
d
c
_
(4:120)
CTOD can be determined indirectly by measuring the COD at x =0 Figure 4.12, denoted by D:
d =
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
D ÷
16s
2
E
2
a
2
_
(4:121)
Disadvantage of the COD concept arises from the difﬁculty to involve a critical value of COD in
fracture stress calculation. Critical COD is primarily a comparative toughness value for high
toughness materials.
0
2a
y
D
Plastic
zone
d
x
FIGURE 4.12 Crack opening displacement.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 145 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 145
4.3.4 JINTEGRAL
Exact calculation of the effect of plasticity on energy release rate can be done by Jintegral [25,26]:
J =
_
G
W dy ÷T
@u
@x
ds
_ _
(4:122)
where
G is the closed contour of an area in a stressed material (Figure 4.13)
T is the tension vector perpendicular to G in an outside direction
u is the displacement in xdirection
ds is an element of G
Moreover, W is the strain energy per unit volume. For the elastic case, in polar coordinates (r, u)
J = r
_
p
÷p
W(r, u) cos u ÷T(r, u)
@u
@x
(r, u)
_ _
du (4:123)
Solution of Equation 4.122 leads to
J = G (4:124)
The criterion for crack initiation under openingmode loading is
J = J
Ic
(4:125)
where J
Ic
is a material property for a given thickness under speciﬁed environmental conditions. In
the elastic case it follows that
J
Ic
= G
Ic
=
1 ÷n
2
E
K
2
Ic
(4:126)
Jintegral is a generalized deﬁnition for the energy release rate, which can be applied if there is
substantial crack tip plasticity.
4.3.5 STRAIN ENERGY DENSITY FAILURE CRITERION
Yielding and fracture are unique features of material damage and in energy density criterion they are
treated simultaneously by a unique failure criterion. The strain energy density (SED) criterion was
G
A
C
T
B
y
x
ds
FIGURE 4.13 Deﬁnition of Jintegral.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 146 1.11.2008 2:46am Compositor Name: JGanesan
146 Handbook of Thermal Process Modeling of Steels
the generalized failure criterion developed by Sih [27,28]. The basic measure in the SED
criterion is the density dW=dV. The energy per unit volume referred to as the SED function
dW=dV is deﬁned by
dW
dV
=
_
«
ij
0
s
ij
d«
ij
÷f (DT, DC) (4:127)
where
s
ij
and «
ij
are the stress and strain components
f(DT, DC) takes the inﬂuence of temperature and moisture concentration on SED
According to the total energy theory referred to as Beltrami–Haigh theory, and distortional energy
theory referred to as von Mises or Hubert–von Mises–Hencky theory, failure in a material by
yielding occurs when the total, or the distortional, strain energy per unit volume absorbed by the
material equals the energy per unit volume stored in the material loaded in uniaxial tension at yield.
Usually, both the dilatational and the distortional part of SED play a role in the material damage
process and they both have to be taken into account for a complete description of material damage.
It is expected that SED varies from one location to another (Figure 4.14).
The SED function dW=dV decays with distance r from the crack tip. The SED function dW=dV
can be expressed by [27]
dW
dV
=
S
r
(4:128)
where
S is the SED factor
r is the distance from the crack tip to the site of possible failure initiation, which can be
inclusion, void, etc.
d
W
/
d
V
(dW/dV)
c
= constant
r
r
FIGURE 4.14 Change of SED with a distance from crack tip.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 147 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 147
The SED criterion is stated in terms of three basic hypotheses:
1. The location of the fracture corresponds with the location of relative minimum SED,
(dW=dV)
min
, and yielding corresponds with relative maximum SED (dW=dV)
max
.
2. Fracture occurs when (dW=dV)
min
reaches its respective critical value, and yielding
occurs when (dW=dV)
max
reaches its respective critical value.
3. The crack growth increments r
1
, r
2
, . . . , r
j
, . . . , r
c
during stable crack growth satisfy the
equation
dW
dV
=
S
1
r
1
=
S
2
r
2
= =
S
j
r
j
= =
S
c
r
c
(4:129)
Unstable fracture or yielding occurs when the critical ligament size r
c
is reached. Thus, crack
initiation occurs when S =S
c
. Critical value of SED factor S
c
is a material property and characterizes
the fracture resistance of the material:
S
c
= r
c
dW
dV
_ _
c
(4:130)
In some cases, it is suitable to predict the critical value of SED factor S
c
through the uniaxial true
stress–true strain diagram. For the general case of an openingmode, SED factor S is in relation with
stress intensity factor K
I
:
S =
(1 ÷n)(1 ÷2n)
2pE
K
I
(4:131)
The SED criterion is a general failure criterion, which can be applied in the solution of many
engineering problems: the static and dynamic crack problems; elastic and plastic problems; ductile
and brittle fracture; prediction of crack growth under mixedmode loading; prediction of crack
initiation; slow stable crack growth and ﬁnal separation; fracture at normal and elevated tempera
ture; failures at different loading types, such as monotonic, cyclic, or residual stresses, etc. This
criterion is suitable for the design of thermal processes in which it is common that plastic
deformation and fracture occur simultaneously, and in which both nucleation and calescence of
damages in global and local range occur.
4.3.6 STATISTICAL NATURE OF FRACTURE
Mechanical properties of brittle materials show large variability, which requires the statistical
approach. For statistical analysis of mechanical behaviors of brittle materials, Weibull distribution
is most appropriate. The Weibull distribution for stresses s can be written as
p(s) = m
s ÷s
u
s
m
0
_ _
m÷1
e
÷ s÷s
u
( )=s
0
[ [
m
(4:132)
where
s
u
, s
0
, and m are Weibull parameters that are constant for a given population that has a
Weibull distribution:
s
u
is stress below which there is a zero probability of failure
s
0
is the characteristic strength that corresponds with 63.2% point
m is the Weibull modulus or slope parameter
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 148 1.11.2008 2:46am Compositor Name: JGanesan
148 Handbook of Thermal Process Modeling of Steels
Parameter m characterizes the variability in strength of the material and is analogous to the standard
deviation. The three Weibull parameters can be obtained experimentally on a specimen with volume
V
0
equal to unity. In accordance with the weakestlink concept, the probability that fracture stress is
less or equal to s is
P(s) =
_
s
÷·
p(s)ds (4:133)
The probability of survival is
S(s) = 1 ÷P(s) (4:134)
The strength of brittle materials shows a size effect in which the strength decreases with increasing
volume of the specimen. The probability of survival of a specimen of volume V=nV
0
is
S(V) = S(s)
n
(4:135)
or
S(V) = exp ÷V
s ÷s
u
s
0
_ _
m
_ _
(4:136)
Equation 4.136 can be written in the form
ln ln
1
S(V)
_ _
= ln V ÷mln s ÷s
u
( ) ÷mln s
0
(4:137)
The risk of failure is
R(V) = 1 ÷exp ÷V
s ÷s
u
s
0
_ _
m
_ _
(4:138)
4.3.7 FATIGUE CRACK PROPAGATION
Under cyclic loads engineering components can fail at stresses that are less than shorttime strength.
The stress intensity factor is a good measure of the stress and strain situation at the crack tip. Fatigue
crack propagation per cycle is usually in the relation with the stress intensity factor. The rate of
fatigue crack propagation per cycle da=dN can be expressed as:
da
dN
= f (DK) (4:139)
where DK=K
max
÷K
min
is the amplitude of the stress intensity factor, and K
min
=0. The load is
usually sinusoidal with constant amplitude and frequency. The crack propagation rate progres
sively increases with increasing crack size. A typical plot of the characteristic shape of a
log(DK) – log(da=dN) fatigue crack growth rate curve is shown in Figure 4.15.
Three regions can be distinguished in Figure 4.15. In region I, da=dN decreases rapidly, and for
some materials there is a minimum value of the stress intensity factor amplitude under which no
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 149 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 149
crack propagation takes place. In region II there is a linear log(DK) ÷log(da=dN) relation. Finally,
in region III the crack growth rate curve rises and the maximum stress intensity factor K
max
becomes
equal to the critical stress intensity factor K
c
leading to catastrophic failure.
In accordance with experimental results, the relation of Equation 4.139 can be written in
the form
da
dN
= C DK ( )
n
(4:140)
where C and n are constants. Values of n usually vary between 2 and 4. Equation 4.140 is one of the
most widely used fatigue crack propagation laws proposed by Paris and Erdogan [29] and is usually
referred to in the literature as the ‘‘Paris law.’’ If K
min
,= 0 the crack growth rate can be predicted by
da
dN
= f
1
K
max
, DK ( ) = f
2
R, DK ( ) (4:141)
where R is known as the cycle ratio:
R =
K
min
K
max
(4:142)
At the conventional fatigue test a specimen is held under ﬁxed stress amplitude and each time the
maximum stress is held constant until the specimen fails. Tests may be performed at a number of
different values of stress to determine the S–N curve, but each time the maximum stress is held
10
0
10
−1
10
−2
10
250 500 1000
d
N
d
a
(
μ
m
/
c
y
c
l
e
)
ΔK (N/mm
3/2
)
I II III
FIGURE 4.15 Fatigue crack propagation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 150 1.11.2008 2:46am Compositor Name: JGanesan
150 Handbook of Thermal Process Modeling of Steels
constant until the test is completed. However, there are many practical applications where cyclic
stress does not remain constant, but instead there are periods when the average cycle stress is either
above or below certain average design level. In this case, the linear cumulative damage rule, also
called Miner’s rule, can be used. This rule assumes that the total life of a part can be estimated by
adding up the percentage of life consumed by each overstress cycle. If n
1
, n
2
, . . . , n
k
represent the
number of cycles of operation at speciﬁc overstress levels and N
1
, N
2
, . . . , N
k
represent the life (in
cycles) at these same overstress levels, then
i=k
i=1
n
i
N
i
= 1 (4:143)
Many deviations from Miner’s rule have been observed, and numerous modiﬁcations to this
relationship have been proposed.
4.3.8 CREEP
The time dependent, progressive plastic deformation of a material at constant stress (or load) is
called creep. At elevated temperature the strength is very dependent on both strain rate and time of
exposure. A number of metals under these conditions behave as viscoelastic materials. Generally,
creep in engineering is signiﬁcant at a temperature greater than half of melting temperature T
m
. An
idealized creep curve is shown in Figure 4.16.
A creep curve could be represented by the following empirical equation [30]:
« = «
0
÷«
t
1 ÷e
÷rt
( ) ÷ _ «
s
t (4:144)
where
t is the time
«
0
is the instantaneous strain on loading
«
t
is the limit for transient creep
r is the ratio of transient creep rate to the transient creep strain
_ «
s
is the steadystate creep rate
Primary
creep
Secondary creep Tertiary
creep
Fracture
S
t
r
a
i
n
Time
FIGURE 4.16 Schematic representation of creep curve.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 151 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 151
The ﬁrst stage of creep, which is known as primary creep, represents a region of decreasing
creep rate. Primary creep is the predominant creep process at low temperatures and stresses. Primary
creep is a period of predominantly transient creep. In primary creep, creep resistance of the material
increases with its own deformation. The second stage of creep, known also as secondary creep, is
a period of nearly constant creep rate. For this reason, secondary creep is usually referred to as
steadystate creep. The third stage of creep, or tertiary creep, occurs at high stresses at high
temperatures. Tertiary creep occurs when there is an effective reduction in crosssectional area
either because of necking or internal void formation. Thirdstage creep is often associated with
metallurgical changes such as coarsening of precipitate particles, secondary recrystallization, or
diffusional changes in the phases [5].
Below onehalf the melting point, transient creep predominates. A creep curve could be
represented by the following empirical equation:
_ « = At
÷n
/
(4:145)
where A and n
/
are the empirical constants. If n
/
=0, the creep rate is constant and Equation 4.145
represents steadystate creep. When n
/
=1, Equation 4.145 becomes
« = aln t (4:146)
where a is a constant. This is the logarithmic creep law found at low temperatures. Simplify by
assuming that creep is an activated process strain rate and can be expressed by an Arrheniustype
equation
_ « = Ae
÷DH=RT
(4:147)
where
DH is the activation energy for the ratecontrolling process
A is the coefﬁcient that depends on the structure of the material, containing the frequency of
vibration of the ﬂow unit and entropy change
T is the absolute temperature
R is the universal gas constant
Activation energy for creep is equal to the activation energy of the ratecontrolling process,
which is a thermally activated process. Dislocation climb is the ratecontrolling process in high
temperature creep, and since dislocation climb undergoes diffusion of vacancies, activation energy
of the hightemperature creep is that for selfdiffusion.
The creep equations can be developed from a model of the effect of stress on an activated creep
process. In order for the creep process to activate, an energy barrier shown in Figure 4.17 must be
overcome. The most applicable empirical equation for relating the steadystate creep rate to stress
and temperature is [5]:
_ «
s
= A(sinh aW)
n
/
e
÷Q=RT
(4:148)
The value of W depends on the applied stress, and can be expressed by
W = Vs (4:149)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 152 1.11.2008 2:46am Compositor Name: JGanesan
152 Handbook of Thermal Process Modeling of Steels
where V is the activation volume. The rate of creep strain can be expressed as
_ «
s
= Ae
÷(Q÷W)=RT
(4:150)
Creep strength at a given temperature can be deﬁned as the stress that produces a steadystate creep
rate of a ﬁxed amount. Moreover, it may be deﬁned as the stress to cause a creep strain of a ﬁxed
amount, usually of 1%. Rupture strength is the stress which cause rupture after some amount of time
deﬁned as the rupture life. It is important to be able to extrapolate creep or stress–rupture data into
regions where data are not available [5]. More than 30 different parameters have been proposed for
the extrapolation of rupture data [30]. Mostly they are derived from the following:
P =
s
( log t=Q)÷log t
A
T ÷T
A
( )R
(4:151)
where
P is the parameter
T is the temperature
s is the stress
T
A,
t
A
, Q, and R are empirical constants
In hightemperature materials design, the Larson–Miller method is used very often:
P = f (s) = T( log t ÷25) (4:152)
Subcritical ﬂow growth can occur by other mechanisms than fatigue or creep, for example, there are
stress corrosion cracking, delayed failure and so on. The classical approaches of the theory of
plasticity and fracture mechanics sometimes cannot be successfully applied in the prediction of the
progressive deterioration of material preceding the macroscopic fracture.
Recently, the theory of damage mechanics was applied for solving this kind of task [12]. The
theory of damage is concerned with all materials at low or high temperatures under any kind of load.
The damage represents discontinuities in the form of microcracks or cavities. Microcracks and
cavities that constitute the damage leave traces of different forms in a damaged material. In volume
element of a damaged material, which is sufﬁciently large size with respect to the inhomogeneities, the
total area of the defect traces is usually corrected for stress concentrations and interactions between the
defects. The state of mechanical damage is deﬁned by the density damages at the microstructure scale.
Damage evolution, which is not always easily distinguishable from the deformation phenomena,
which usually accompany it, is due to several mechanisms, such a ductile plastic damage, brittle
viscoplastic or creep damage, fatigue damage, macrobrittle damage, and so on.
Q−W
Q+W
B
A
C
FIGURE 4.17 Creep energy barrier.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 153 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 153
Damage mechanics describes the evolution of the damage phenomena between the virgin state
and macroscopic crack initiation. Knowing the stress and strain history for the given volume
element of a structure, the damage laws can provide, by integration with respect to time, prediction
of the damage evolution in the element up to the point of macroscopic crack initiation. The theory
provides the time or the number of cycles corresponding to the initiation of such a crack at the most
stressed point of the structure. Existence of a macroscopic crack of 0.1–1 mm length is usually
referred to as the ﬁnal damage state for metals.
4.4 MICROMECHANISMS OF PLASTIC DEFORMATION AND FRACTURE
The phenomenological description of the inﬂuence of stress and temperature on elastic and plastic
behavior of materials has been shown in the previous section. The mechanical behavior of materials
is described based on the simple assumption that materials are homogeneous and isotropic.
Knowledge of basic mechanisms of deformations and fracture of the metal materials is necessary
for the formulation of hypotheses upon which the phenomenological theories of deformation and
fracture can be based. In metals, the introduction of the concept of dislocations was the crucial step
in explaining nearly all mechanical phenomena such as slip of crystals, strain hardening, the yield
point, creep, fatigue, and brittle fracture. The dislocation concept of mechanical behaviors was
proposed independently by Taylor, Orowan, and Polany in 1934 [31]. This section attempts to
provide a basic understanding of the micromechanisms of plastic deformation and fracture, which
are discussed in greater detail in Refs. [4,5,31].
4.4.1 ELEMENTS OF THEORY OF DISLOCATIONS
A dislocation is a line defect and it can be deﬁned as the region of localized lattice disturbance
separating the slipped and unslipped regions of a crystal. Basic types of dislocations are edge and
screw dislocation [32]. Figure 4.18 shows an edge (Figure 4.18a) and screw (Figure 4.18b)
dislocation schematically.
A B
C D
Slip
vector
Slip vector
A B
C
D
Slip
vector
A
B
D
Slip
vector
C
(a)
(b)
FIGURE 4.18 Dislocation types: (a) edge dislocation and (b) screw dislocation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 154 1.11.2008 2:46am Compositor Name: JGanesan
154 Handbook of Thermal Process Modeling of Steels
In both cases slip occurs over area ABCD. Slip occurs in the area over which dislocation moves.
Slip displacement is deﬁned by slip or the Burgers vector b. The Burgers vector deﬁnes the
magnitude and direction of slip; its magnitude deﬁnes the translation which would be necessary
to create a dislocation; it is constant along the whole dislocation line; the closing vector of a contour
surrounding several dislocations is the sum of the Burgers vector of these dislocations. For a pure
edge dislocation, the Burgers vector is perpendicular to the dislocation line, while for a pure screw
dislocation the Burgers vector is parallel to the dislocation line. If the dislocation line and Burgers
vector are at certain angle to each other, the dislocation is a mixed one. Dislocation lines are not
necessarily straight [31]. In its general form the dislocation line takes the form of a curve or loop as
shown in Figure 4.19.
A dislocation loop is made up of elements of edge and of screw, and the magnitude and
direction of the slip are deﬁned by the Burgers vector, which is speciﬁed in terms of the components
along the principal crystal axes.
Dislocations in metals can be detected by transmission electron microscopy of thin foils, xray
microscopy, and diffraction, or by ion microscope. The density of dislocations is very high in
most metals; it varies from 10
3
cm
÷2
for annealed crystals to 10
11
cm
÷2
for quenched or cold
worked state.
The free energy of a crystal is increased by the introduction of a dislocation [33,34]. Since
dislocations produce pure elastic distortions of the lattice, it is possible to calculate the elastic energy
U stored in the material with dislocation. The tangential elastic stress in a ﬁeld surrounding a
dislocation (Figure 4.20) can be expressed by:
Slipped
Burgers
vector
Unslipped
FIGURE 4.19 Dislocation loop.
r
0
b
y
x
r
0
q
FIGURE 4.20 Elastic deformation surrounding edge dislocation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 155 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 155
t
El
= a
1
m
b
r
(4:153)
where factor a
1
is of the order of 0.5.
For dislocation of unit length, with Burgers vector of magnitude b, the accumulated elastic
energy can be expressed by [4,12]
U = amb
2
(4:154)
where factor a is between 0.5 and 1.
During the dislocation movement, the slip displacement requires the breaking of bonds only in
the vicinity of the dislocation line, and the slip continues from one atom to the next. The dislocation
under an external load moves across a crystal by irreversible displacement [12] (Figure 4.21).
Because the Burgers vector of a screw dislocation is parallel to the dislocation line, it is free to
slip on any plane that contains the dislocation line and the Burgers vector.
An edge dislocation can move perpendicularly to its slip plane, directly above or below the slip
plane by climb. Dislocation climb occurs by the diffusion of vacancies or interstitial atoms to the site
of the dislocation [12,33,35]. If vacancy is close to a dislocation line and an applied external load
causes a distortion of the lattice, an atom of the lattice can jump by half a plane on the empty side of
the crystal and result in the rearrangement of the whole row of atoms (Figure 4.22).
If the movement of a dislocation on its slip plane is prevented, it may climb to another plane. In
the case of screw dislocations, climb is not possible. A dislocation that moves in its slip plane
usually intersects other dislocations that cross the slip plane (Figure 4.23).
The dislocation intersection produces a jog and it can glide with the rest of the dislocation. Jogs
can be made by the intersections of different kinds of dislocations. Jogs that lie in the slip plane
instead of normal to it are referred to as kinks. Since a jog is roughly equivalent to a dislocation of
length b, the energy of a jog is about mb
3
[4,33].
FIGURE 4.21 Slip displacement of dislocation.
FIGURE 4.22 Displacement by climb.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 156 1.11.2008 2:46am Compositor Name: JGanesan
156 Handbook of Thermal Process Modeling of Steels
Plastic deformations increase the density of dislocations, which, in turn, increases the number of
blockings and contributes to the process of hardening. Dislocations can be introduced in crystals
during crystal formation by formation of dislocation loops due to the aggregation and collapse of
vacancies, furthermore, by nucleation of dislocations as a result of phase transformations, or due to
high local stresses at grain or secondphase particle boundaries.
The important source of dislocations is deﬁned as Frank–Read mechanism of dislocation
generation: Under the applied stress, the dislocation is forced to move around the points A and B
(Figure 4.24). When the two branches join at C, they have the same Burgers vector but with
opposite unit vectors. They combine to create a large loop and a new segment AB, which in turn
generates another loop, and so on.
Dislocations can pileup at barriers such as grain boundaries and second phases. The disloca
tions in the pileup will be tightly packed together (Figure 4.25).
The number of dislocations that can occupy a distance L along the slip plane between the source
and the obstacle is [4,5,33]
n =
ktL
mb
(4:155)
where
t is the average resolved shear stress in the slip plane
k is a factor
A
B
x
y
A
x
y
B CЈ
C
b
2
b
2
b
1
b
1
FIGURE 4.23 Intersection of dislocations.
s
12
A
b
B
A B
A
B
A B
C
A B
A B
FIGURE 4.24 Generation of dislocations by the Frank–Read mechanism.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 157 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 157
The number of dislocations that can be kept by an obstacle will depend on the type of barrier,
material, and the temperature.
4.4.2 MICROMECHANISMS OF PLASTIC DEFORMATION
In a purely elastic deformation, the initial conﬁguration of atoms is restored after the removal of the
load. Permanent or plastic deformation is in addition to elastic deformation and it corresponds to a
relative displacement of atoms that remains when the load is removed.
The usual method of plastic deformation in metals is by the sliding of blocks of the crystal over
one another along slip planes. The resistance to slip is generally less for the planes of greatest atomic
density than for any other set of planes. Figure 4.26 schematically illustrates a deformation by slip.
When a shear stress is applied to a perfect crystal, the same force opposing the displacements
acts on all the atoms. The theoretical shear strength of a perfect crystal is approximately equal to the
shear modulus divided by 2p, (m=2p). The discrepancy between the real and theoretical shear
strengths of metals has been explained by the concept of dislocation [36]. The motion of a
dislocation through a crystal lattice requires a stress smaller than the theoretical shear stress. The
movement of the dislocation produces a step at the free surface, that is, produces a permanent
deformation (Figure 4.27a). In Figure 4.27b the energy change from an unslipped to slipped
state is shown. The force required to move a dislocation through the crystal lattice is known as
Peierls–Nabarro force [34,37,38]. The stress associated with Peierls–Nabarro force, that is, the shear
stress required to move a dislocation through a crystal lattice in a particular direction is
Obstacle
Source
L
FIGURE 4.25 Dislocation pileup at an obstacle.
Slip plane
Slip plane
FIGURE 4.26 Schematic representation of slip.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 158 1.11.2008 2:46am Compositor Name: JGanesan
158 Handbook of Thermal Process Modeling of Steels
t
p
~
2G
1 ÷n
e
÷2pw=b
~
2G
1 ÷n
e
÷[2pa=(1÷n)b]
(4:156)
where
a is the distance between slip planes
b is the distance between atoms in the slip direction
A slip begins when the shear stress on the slip plane in the slip direction is equal to the critical
resolved shear stress. The critical resolved shear stress should increase as the density of defects
increases [39]. The magnitude of the critical resolved shear stress is determined by the interaction of
dislocations with each other and with defects such as vacancies, interstitials, and impurity atoms. An
increased concentration of solute atoms around a dislocation is called an impurity atmosphere.
Isolated solute atoms and vacancies are centers of elastic distortion. Point defects and dislocations
interact elastically and exert forces on each other. Interaction of dislocations with point defects is
important for the yieldpoint behavior, strain aging, and solidsolution strengthening [5,39].
The second important mechanism by which metals deform is the process known as twinning
[40]. Twinning generally occurs when the slip systems are restricted. Twins can be formed by
annealing in metals if they are mechanically deformed before annealing. During twinning a part of
the crystal takes up an orientation that is related to the orientation of the rest of the untwinned lattice
in a symmetrical way (Figure 4.28).
(a)
Interfacial
region
Unslipped
region
Slipped
region
(b)
S
l
i
p
p
e
d
s
t
a
t
e
U
n
s
l
i
p
p
e
d
s
t
a
t
e
DE
E
n
e
r
g
y
Displacement
FIGURE 4.27 Permanent deformations by dislocation movements: (a) movement of an edge dislocation and
(b) energy change from unslipped to slipped state.
Twin plane
A
A
FIGURE 4.28 Classical picture of twinning.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 159 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 159
In a slip, the orientation of the crystal above and below the slip plane is the same after
deformation as before, while in twinning an orientation is different across the twin plane.
In twinning, the atom movements are much less than an atomic distance.
4.4.3 MICROSTRUCTURAL ASPECTS OF DEFORMATION PROPERTIES
The mechanical behaviors of metals can be easily approximated by a stress–strain diagram [12,40].
In Figure 4.29 schematically represents the stress–strain diagram of a metal deformed by a uniaxial
external loading, which ﬁrst increases and then decreases.
Elastic deformations (OE in Figure 4.29) are the result of reversible relative movements of
atoms. They are almost without permanent deformations.
The elastic limit is characterized by the state of stress that causes the ﬁrst irreversible slip. As it
is difﬁcult to detect, for engineering purposes the usable limit for elastic behavior is described by the
yield stress, s
Y
, (Point Y in Figure 4.29) by convention to be the stress for a ﬁxed amount of
permanent strain, usually «
p
=0.002. Hall and Petch have established that tensile yield stress is
related to grain size by [4,5]
s
Y
= s
i
÷k
Y
d
÷1=2
(4:157)
where
s
i
is friction stress opposing motion of dislocation
k
Y
is the constant
d is the grain diameter
Strengthening due to grain boundaries results from mutual interference to a slip within the grains.
The Hall–Petch equation has been found to express the grainsize dependence of ﬂow stress at any
plastic strain.
s
Y
E
P
V
R RЈ
B
0
Y
VЈ
Strain
S
t
r
e
s
s
FIGURE 4.29 Stress–strain diagram for metal.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 160 1.11.2008 2:46am Compositor Name: JGanesan
160 Handbook of Thermal Process Modeling of Steels
During tension test, many metals show a localized type of transition from elastic to plastic
deformation, which produces a yield point in the stress–strain curve (Figure 4.30) [40].
Since deformation throughout the yieldpoint elongation is heterogeneous, discrete bands
(Lüders bands) of deformed metal will form at several points of stress concentration. These bands
are generally at approximately 458 to the tensile axis. The yielding occurs at a stress where the
average dislocation sources can create slip bands through a critical part of the material. The stress
required to deform the specimen decreases when dislocations begin to move and mobile dislocations
begin to multiply. After the Lüders bands have covered the entire length of the specimen, the ﬂow
will increase with strain in the usual manner.
The plastic deformation (EP in Figure 4.29) by slip occurs in crystals with crystallographic slip
planes oriented at the same direction to the maximum shear stress t. The reorientation of crystals,
necessary to ensure the compatibility of deformations, activates other slip systems and the deform
ation appears macroscopically homogeneous. It is a stable deformation and each state is that of
elastic–plastic equilibrium.
The characteristic of the common plastic deformation of metals is the fact that the shear stress
required to produce a slip continuously increases with increasing shear strain. If the stress continues
to rise, the dislocation density is increased but the number of barriers is increased even more, so that
the deformation cannot progress unless the load is increased. The increase in the stress required
to cause a slip because of previous plastic deformation is wellknown work or strain hardening.
In the basic equation relating strain hardening to structure, one component is the yield stress and
the second one is the contribution arising from the dislocation multiplication and interaction.
In many investigations, it was shown that stress ﬂow increases as the total dislocation density
increases [41]. Flow stress in terms of dislocation density r can be expressed by the Taylor
dislocation model [39,42]
s
flow
= ambr
1=2
(4:158)
where
a is a constant
m is the shear modulus
b is the magnitude of the Burgers vector
Yield strain
Lower yield
stress
Strain
S
t
r
e
s
s
Upper yield
stress
FIGURE 4.30 Schematic diagram of yieldpoint behavior.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 161 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 161
Besides the model originally proposed by Taylor in 1934, a linear relation between shear ﬂow stress
and square root of the dislocation density is also suggested by other hardening mechanisms
and conﬁrmed by experimental observations proposed.
Strain hardening is caused by dislocation multiplication and by dislocations interacting with
other dislocations or with other barriers. Dislocation multiplication can arise by the Frank–Read
mechanism or by emission of dislocations from a highangle grain boundary.
Strain hardening could be connected with dislocations pileup at barriers on slip planes. If
strain hardening arises from dislocation pileup at barriers it is relatively independent of temperature
and strain rate. In addition, strain hardening can be due to dislocation intersection. This hardening
can be overcome with the help of thermal ﬂuctuations, and it is temperature and strainrate
dependent.
If the stress continues to rise, the slips can cross and follow the grain boundaries producing the
viscoplastic deformation (PV in Figure 4.29). This phenomenon of intergranular slip is favored by
thermal activation and is especially signiﬁcant at high temperatures. A large part of the deformation
remains intragranular and consists of slip and climb of dislocations.
When the slip direction is reversed, the structure is not unstable enough and dislocation lines
will not move back synchronically by removing the load. Recovery (RR
/
in Figure 4.29) is a
function of time and is favored by thermal activation. On a macroscopic level, it manifests itself in
a partial recovery of the deformation or in a decrease in the hardening [12].
Since slip deformations do not change a crystal structure, the total volume of aggregate remains
unchanged during plastic deformation. The volume change due to the increase in the density of
dislocations always remains very small. Only elastic deformations can produce a volume change.
As permanent deformations differ from one crystal to the next, the compatibility of strains at
grain boundaries is assured only by elastic microdeformations. When the load is removed, elastic
microdeformations remain to a certain extent locked resulting in selfequilibrated microscopic
residual stresses. These microscopic stresses can increase or decrease the external load necessary
to produce new slips in a way that varies according to the direction orientation. Anisotropy results
from permanent deformations.
Since the barriers to the rear of the accumulated dislocations are not closely spaced to them, the
yielding can occurs at a lower stress level when the loading direction is reversed. This explains the
fact that the ﬂow curve in the reverse direction lies below the curve for continued ﬂow in the original
direction. The lowering of the ﬂow stress when deformation in one direction is followed by
deformation in the opposite direction is called the Bauschinger effect (point B in Figure 4.29).
4.4.4 MICROMECHANISM OF FRACTURE
Crack extension occurs when the stresses or strains at the crack tip exceed a critical value. Critical
stress or strains criteria, that is, fracture criteria should be based on the fracture mechanism. Applied
mechanics can, with some limitation, describe the stresses and strains in the vicinity of a notch or
crack tip. Two main basic mechanisms of local fracture are brittle fracture by cleavage and ductile
fracture resulting from localized plastic deformation. The tendency for brittle fracture is increased
with decreasing temperature, increasing strain rate, and triaxial state of stress. Brittle failures of
materials are characterized by low toughness or crack resistance. Ductile failures of materials are
characterized by high toughness.
Cleavage fracture consists of a direct separation of particular crystallographic planes. At grain
boundaries, the cleavage surfaces change direction to follow the crystallographic cleavage planes of
the next crystal (Figure 4.31a).
In a crystal, several parallel surfaces of cleavage can develop, which join each other perpen
dicularly in the form of steps arranged in rivers (Figure 4.31b). Intergranular fracture is a cleavage
that follows the grain boundaries. At low temperature brittle fracture, impurities segregated at
the grain boundaries lower the energy of cohesion at these boundaries. At hightemperature
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 162 1.11.2008 2:46am Compositor Name: JGanesan
162 Handbook of Thermal Process Modeling of Steels
intergranular fracture, a local viscoplastic deformation at grain boundaries is a process that supports
intergranular fracture. Therefore, the propagation of a microcrack produced by deformation through
a strong barrier such as a grain boundary is the most difﬁcult step in the fracture process of metals.
The grain size will have a strong inﬂuence on stress for brittle fracture:
s
f
= s
of
÷k
f
d
÷1=2
(4:159)
This is the Petch equation for fracture stress analogous to Equation 4.157 for the grainsize
dependence of the yield point and the ﬂow stress.
The process of brittle fracture starts by microplastic deformation, which involves the pileup of
dislocations at an obstacle until a microcrack is nucleated. A crack forms when the applied work is
equal to the work for moving the dislocations against the friction stress and the work for producing
the new fracture surfaces. The crack grows by microplastic deformation so long as the dislocation
source continues to force dislocations into the pileup. A microcrack will start to propagate when the
applied stress reaches a critical value. Tensile stresses are not involved in the microcrack nucleation
process; only shearing stresses are involved in forcing the dislocations together. Tensile stress is
needed to make the microcracks propagate. Zeener [43] proposed that stresses at the head of a
dislocation pileup can produce fracture; the model is shown in Figure 4.32.
Grain boundary
River pattern
(a) (b)
FIGURE 4.31 Schematic illustration of cleavage fracture: (a) propagation of cleavage fracture and
(b) formation of river pattern.
t
n
b
Obstacle
Source
L
2
c
FIGURE 4.32 Model of microcrack formation at a pileup of edge dislocations.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 163 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 163
The shear stress acting on the slip plane tightens the dislocations together. At some critical value
of stress, the dislocations coalesce into a microcrack of height nb and length 2c. If the dislocation
source is at the center of a grain of diameter d, the tensile stress at the pileup can be equated to the
theoretical cohesive stress [5,20,44]
t ÷t
i
( )
d
2r
_ _
1=2
=
Eg
S
a
_ _
1=2
(4:160)
and microcrack nucleation occurs at
t = t
i
÷
2Erg
S
da
_ _
1=2
(4:161)
where
r is the distance from the tip of the pileup to the point where the crack is forming
a is the distance between slip planes
If it is accepted that r ~a
0
, and since E is the modulus of elasticity, E=2m then becomes
t = t
i
÷
4mg
S
d
_ _
1=2
(4:162)
However, from Equation 4.155, the number of dislocations in the slip band can be expressed as [44]
nb ~ d
t ÷t
i
m
(4:163)
and
t ÷t
i
( )nb ~ 4g
S
(4:164)
and substituting in Equation 1.164 that s=2(t ÷t
i
), it follows that
snb ~ 8g
S
(4:165)
Substituting Equation 4.163 into Equation 4.165,
s t ÷t
i
( )d = 8mg
S
(4:166)
Since yield stress is equal to
s
Y
= s
i
÷k
Y
d
÷1=2
(4:167)
taking in account that microcracks from by micro yielding, the fracture stress is [44]
s
f
=
4mg
S
k
Y
d
1=2
(4:168)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 164 1.11.2008 2:46am Compositor Name: JGanesan
164 Handbook of Thermal Process Modeling of Steels
Parameter k
Y
depends on dislocation pileup properties. Effective surface energy g
S
includes the
energy of plastic deformation. The effective surface energy can be generally expressed as
2g
S
~ s
cm
d (4:169)
where
s
cm
is the strength of an element at the crack tip
d is the displacement at the crack tip
High toughness materials have large values of s
cm
and d.
Equation 4.164 can be written in the form [44]
t
i
d
1=2
÷k
Y
_ _
k
Y
= bmg
S
(4:170)
Factor b depends on the stress–strain condition. Equation 4.170 is a general expression of the
microstructural aspect for nucleation and propagation of cleavage crack. The lefthand side of
Equation 4.170 deﬁnes the formation of crack and the righthand side deﬁnes the propagating of
crack. If the lefthand side of the equation is smaller than the righthand side, a microcrack can be
formed but it cannot grow. If the left side of the equation is greater than the right, a brittle crack will
propagate at the stress equal to yield stress.
All strengthening mechanisms that are based on dislocation locking result in brittleness. The
contribution of plastic deformation will depend on the effective surface energy, number of available
slip systems, and the number of mobile dislocations at the crack tip. A high value of friction stress t
i
lowers ductility. In an alloy containing ﬁne secondphase particles, particle distance determines the
slip distance.
Since values in Equation 4.170 depend on temperature, Equation 4.170 describes a ductile to
brittle transition. There is transition temperature at which a ductile fracture changes to brittle. Fine
carbide particles in quenched and tempered steel lower the transition temperature. In Figure 4.33,
the effect of grain size on the yield and fracture stresses is shown.
Corrosion or hydrogen penetration may lower the surface energy. The presence of a notch
greatly increases the tendency for brittle fracture. The effect of a notch in decreasing the ratio of
shear stress to tensile stress is covered in Equation 4.170 by the constant b.
(Grain diameter)
−1/2
S
t
r
e
s
s
Strain to
fracture
S
t
r
a
i
n
Fracture
stress
Fracture
stress
Yield
stress
Yield
stress
FIGURE 4.33 Effect of grain size on the yield and fracture stresses for carbon steel (schematically).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 165 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 165
Based on the stress concentration phenomenon due to dislocation pileup at the carbide grain
boundary, the fracture criterion for steel can take the form [45,46]
c
0
d
_ _
s
2
f
÷t
2
eff
1 ÷
4
p
c
0
d
_ _1
2 t
i
t
eff
_ _
2
_
8mg
c
p 1 ÷n
2
( )d
(4:171)
where
c
0
is the carbide thickness
t
eff
is the maximum stress that can be attained before yielding occurs
g
c
is the surface energy of the carbide
It can be seen that larger carbides lower the fracture stress. The second term on the lefthand side of
Equation 4.171 represents the contribution of dislocation to cleavage initiation. When this term is
removed, this equation reduces to Grifﬁth’s equation for a grain boundary microcrack.
The mechanical behavior can be generally predicted as a function of stress–strain state, strain
rate, and temperature. Ductiletobrittle transition behavior in body centered cubic (BCC) metal can
be expressed by cleavage resistance s
f
and resistance to plastic ﬂow s
Y
(Figure 4.34a).
Below the transition temperature, the material fractures without plastic ﬂow and needs little
energy. The cleavage resistance s
f
is similar in uniaxial and triaxial stress states, that is, it is similar
in notched and unnotched specimen. The ﬂow stress is higher in the triaxial stress state and
transition temperature increases with an increase in triaxial stress state produced by a notch in the
notched specimen. The effect of the loading rate is represented in Figure 4.34b.
Ductile fracture arises from the instability that results when large local deformations occur. The
amount of plastic deformation required to produce a fracture may be limited when a state of plane
strain reduces the possibilities for plastic deformation. At ductile fracture, defects, such as grain
boundaries, secondphase particles, inclusions, solution precipitates, and pileup of dislocations
often cannot accommodate the plastic deformation of the matrix. They fail early on, and voids or
cavities could be formed. Fractures of bridges between voids cause coalescence of cavities, which
leads to the ﬁnal fracture. The fracture surface consists of small dimples that represent the
coalesced voids. Dimples always have an irregular shape, due to the random occurrence of voids.
Triaxial
stress state
S
t
r
e
s
s
Transition temperature
simple tension
Temperature
s
f
s
Y
qs
Y
C
h
a
r
p
y
i
m
p
a
c
t
e
n
e
r
g
y
Loading
rate
Temperature (a) (b)
FIGURE 4.34 Schematic representation of ductiletobrittle transition: (a) effect of loading rate and (b) effect
of triaxial stress state.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 166 1.11.2008 2:46am Compositor Name: JGanesan
166 Handbook of Thermal Process Modeling of Steels
Equiaxed dimples may be formed if the stresses are predominantly tensile (Figure 4.35a), and
elongated dimples occur in both the shear and tear modes (Figure 4.35b and c).
Models of ductile rupture by void nucleation and growth usually refer to a single void in an
inﬁnite plate and do not take into account void interaction [47–49]. A model that studies plastic ﬂow
and void interaction is proposed by Ref. [50]. The yield condition in this case is
3
2
s
/
ij
s
/
ij
s
2
Y
÷2 f cosh
3
2
s
H
s
Y
_ _
÷ 1 ÷f
2
_ _
= 0 (4:172)
where
s
/
ij
is the deviatoric stress deﬁned by Equation 4.24
s
H
is the hydrostatic stress deﬁned by Equation 4.23
f is the void volume fraction
For f =0 reduces to the von Mises yield criterion. Improvement of Equation 4.172 was done by
experimental work [51]:
3
2
s
/
ij
s
/
ij
s
2
Y
÷2fq
1
cosh
3
2
q
2
s
H
s
Y
_ _
÷ 1 ÷ q
1
f
2
_ _ _ ¸
= 0 (4:173)
Using experimental results it was found that good predictions of yielding are obtained when q
1
=2
and q
2
=1.
(a)
(b)
(c)
Equiaxed dimples
Shear dimples
Opposite direction
Same direction
Tear dimples
FIGURE 4.35 Occurrence of different types of dimples.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 167 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 167
4.4.5 MICROMECHANISMS OF CREEP
The mechanisms of creep in metals are primarily based on both atomic diffusion and dislocation
motion [40]. A different creep process will be dominant at different temperatures and stresses.
Generally, the creep rate can be expressed by [5,52,53]
_ « = A
s
m
_ _
n
exp
÷Q
kT
_ _
(4:174)
where
m is the shear modulus
Q is the activation energy of creep
n is the exponent
k is the Boltzmann constant
The Boltzmann constant is closely related to the molal gas constant R by k =R=N, where N is the
Avogadro number. Speciﬁc values of n and Q are associated with speciﬁc mechanisms of creep.
At high temperatures and low stresses, diffusional creep is favored. The diffusional creep rate
can be expressed by Ref. [30]
_ « = A
b
d
_ _
2
Dmb
kT
s
m
_ _
(4:175)
where
D is the diffusion coefﬁcient
b is the magnitude of Burgers vector
d is the grain size
Grains in polycrystalline metals are able to move easily relative to each other at elevated
temperature. The principal deformation processes at elevated temperature are slip, subgrain forma
tion, and grainboundary sliding. Many secondary deformation processes, such as the formation of
extremely coarse slip bands, kink bands, fold formation at grain boundaries, and grainboundary
migration have been observed in metals at elevated temperature. In a ﬁnegrain size material
vacancy diffusion is rapid and the creep rate in this case can be higher.
At intermediate temperature and stresses, the dominant mechanism of deformation could be
based on dislocation motion. In the primary creep stage, the dislocations can arrange themselves into
a lowangle grain boundary. During the steadystate creep dislocation generation and annihilation
are in equilibrium. The dislocation mechanism of creep is sensitive to stress, and the creep rate can
be expressed by [54]
_ « = A
Dmb
kT
s
m
_ _
n
(4:176)
where exponent n varies between 3 and 10. Measurement of the diffusivity coefﬁcient D reveals that
climbcontrolled dislocation motion is involved in creep mechanism. This mechanism is usually
referred to as dislocation glide plus climb. Since creep depends on dislocation climb, the rate at
which vacancies can diffuse to edge dislocations is usually the controlling factor.
The third stage of creep is involved with the development of cracks at the grain boundaries.
These cracks could be wedgetype cracks or spherical cavities [55,56].
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 168 1.11.2008 2:46am Compositor Name: JGanesan
168 Handbook of Thermal Process Modeling of Steels
Wedgeshaped cracks are associated with creep conditions involving moderate creep tempera
tures and higher stress levels. Wedgeshaped cracks (wtype) initiate mostly at grainboundary triple
points and propagate along the grain boundaries, which are predominantly normal to the applied
stress. These cracks initiate by grainboundary sliding, at a point of stress concentration. The
manner of wedgeshaped cracks is illustrated in Figure 4.36.
The initiation of spherical cavities usually occurs at grainboundary ledges and at ﬁne second
phase particles. Grainboundary sliding is necessary to initiate rtype cavities. The initiation of
cavities in the grain boundary usually occurs at grainboundary ledges and at ﬁne secondphase
particles at the grain boundary [57].
4.4.6 MICROMECHANISM OF FATIGUE CRACKING
Under cyclic loads, cracks can be initiated as a result of accumulation of local plastic deformation on
a microscale. Even if the nominal stresses are low, locally the stresses may be above yield due to
stress concentrations. A model of the initiation of fatigue cracks by local plastic deformation is
represented in Figure 4.37 [44].
The cyclic slip can give rise to an extrusion or an intrusion in the metal surface. An intrusion can
grow into a crack by continuing plastic ﬂow during subsequent cycles.
s
t
s
t
FIGURE 4.36 Intergranular crack form.
Extrusion
S
u
r
f
a
c
e
Intrusion
s
t
s
t
FIGURE 4.37 Fatigue crack initiation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 169 1.11.2008 2:46am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 169
The cyclic opening and closing of the crack will develop a typical pattern of ripples (fatigue
striations). Fatigue fracture can be intergranular if the deformation possibilities in the grains are poor.
A corrosive environment that would not normally attack the metal may cause cracking under
mechanical stresses. Stress corrosion cracks are usually intergranular due to a potential difference
between the grain boundaries and the interior of the grains, or due to the presence of secondphase
particles at grain boundaries.
The presence of hydrogen can be very important in metal cracking. The presence of hydrogen in
steels can cause cracking even during processing. Hydrogeninduced cracking may occur when no
loss of shorttime tension properties can be detected. Hydrogen can also cause cracking of high
strength steel after a considerable period of continued loading (delayed failure). Hydrogen cracking
is usually intergranular [58].
4.4.7 MICROMECHANISMS OF STRENGTHENING
The strength of a metal is determined by its crystal structure. The crystal structure determines the
number and type of slip systems, ﬁxes the Burgers vector, and determines the lattice friction stress
(Peierls stress), which sets the base strength level. Cubic metals and other highsymmetry metals,
opposite to lowsymmetry metals, generally have high ductility at room temperature in polycrys
talline form. Dislocations motion can be blocked at grain boundaries and additionally by imperfec
tions such as foreign atoms, precipitated particles, and other dislocations. In Figure 4.38 the possible
mechanisms of metal strengthening by microstructural changes are presented.
Finegrain size, large additions of solute atoms, and ﬁne particles usually result in higher
strength of metals. The crystallographic orientation changes abruptly in passing from one grain to
the next. The ordinary highangle grain boundary represents a region of random misﬁt between the
adjoining crystal lattices. As the grain size decreases and strain increases, the deformation becomes
more homogeneous. The strain hardening of a ﬁnegrain size metal is more intensive than in that of
a coarse grain.
The formation of subgrains usually results in an increase in the yield stress with good ductility.
Subboundaries or lowangle boundaries can be produced during crystal growth, during high
temperature deformation, or as the result of a phase transformation, as well as by combination of
deformation with an annealing treatment to rearrange the dislocations into subgrain boundaries. This
process has been called polygonization.
Precipitation
hardening
Precipitation
hardening of
martensite
Martensite
strengthening
Solidsolution
strengthening
Grain boundary
strengthening
Secondphase
strengthening
FIGURE 4.38 Micromechanism of metal strengthening (schematically).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 170 1.11.2008 2:46am Compositor Name: JGanesan
170 Handbook of Thermal Process Modeling of Steels
The introduction of solute atoms into the solid solution in the solvent–atom lattice invariably
produces an alloy that is stronger than the pure metal, but only a relatively small hardening effect
can be produced in most alloy systems by solidsolution additions. The usual result of solute
additions is an increase in ﬂow stress. Solute atoms fall into two categories with respect to their
relative strengthening effect. Atoms that obstruct nonspherical distortions, such as most interstitial
atoms, have a relative strengthening effect per unit concentration of about three times their
shear modulus (3m). Carbon, nitrogen, boron, and oxygen are the elements that commonly form
interstitial solid solutions. Solute atoms such as substitutional atoms, which produce spherical
distortion and have no effect on nonspherical distortions, have a relative strengthening of about
m=10 [4,5].
The transformation of austenite to martensite in quenching of steel is one of the most common
strengthening processes. Although martensitic transformations occur in a number of metallurgical
systems, only alloys based on iron and carbon show such a visible strengthening effect.
The martensite strength increases if carbon content increases. The ﬁrst important contribution to
the high strength of martensite arises from the effective barriers to slip provided by the ﬁne twin
structure or the high dislocation density. The contribution to strength from the barriers in the
martensite structure is essentially independent of carbon content. The second contribution to the
strength of martensite arises from the carbon atoms that strain the martensite lattice. This strain can
be relieved by redistribution of carbon atoms by diffusion. One result is that a strong binding is set
up between dislocations and the carbon atoms. Another result is the formation of carbon atom
clusters. Many common alloys usually contain a heterogeneous microstructure consisting of two or
more phases. The secondphase particles strengthening is usually additive to the strengthening
produced by solid solution in the matrix. Strengthening from secondphase particles is dependent on
the shape, number, and distribution of the secondphase particles. The strength, ductility, and strain
hardening behavior of both the matrix and second phase are important for the intensity of second
phase strengthening. The alloy is brittle if the brittle phase surrounds a grain boundary. An optimum
strength and ductility is obtained when the brittle phase is present as a ﬁne dispersion uniformly
distributed throughout the softer matrix, as is in case of heattreated steel with a tempered
martensitic structure [4,5].
Generally, the particle size is much ﬁner for the dispersed structure and may extend down to
submicroscopic dimensions in the early stages of precipitation. Dispersion hardening is strengthen
ing produced by a ﬁnely dispersed insoluble second phase in a metallic matrix. The second phase in
dispersionhardening systems has very little solubility in the matrix, and there is no coherency
between the secondphase particles and the matrix. It is possible to produce various dispersion
hardening systems by mixing metallic powders and secondphase particles, as well as carbides,
nitrides, oxides, borides, and so on, and consolidating them by powder metallurgy techniques.
Generally, dispersionhardening systems could be thermally stable at very high temperatures. Fine
particles can act as strong impenetrable noncoherent particles, through which the dislocations can
move only by sharp changes in the curvature of the dislocation line. On the other hand, they can act
as coherent or incoherent particles through which dislocations can pass, but only at stress levels
much above those required to move dislocations through the matrix phase. The stress required to
force the dislocation between the obstacles is usually expressed by Hall–Petch type of equations [4]:
Ds = kL
÷1=2
(4:177)
where L is the distance between the particles.
Precipitation hardening consists of solution treating and quenching followed by aging. A second
phase is in solid solution at the elevated temperature, but precipitates upon quenching and aging at a
lower temperature. After quenching from solid solution, the alloy contains regions of solute
segregation. With additional aging coherent with the matrix, clusters will be formed. These clusters
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 171 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 171
are referred to as GP zones or precipitates. These coherent precipitates produce an increased strain
ﬁeld in the matrix and a further increase in hardness. In further aging the equilibrium, the second
phase will be formed. These particles are not coherent with the matrix, and the strength is lower than
at the stage when coherent particles were present. Particles lose coherency when they grow to some
critical size.
Generally, the strength of dispersionstrengthened alloys is dependent on yielding or fracture of
the particles, or the tearing of the matrix away from the particles. However, since the particles are
usually very ﬁne intermetallic compounds, they possess high strength. Therefore, the degree of
strengthening resulting from secondphase particles depends on the distribution of particles in the
ductile matrix, the volume fraction, average particle diameter, and interparticle spacing [4,5].
Strain hardening or coldworking is an important industrial process that is used to harden
metals. Increasing the temperature lowers the rate of strain hardening. Strain hardening is a result
of the interaction of dislocations with other dislocations and with other barriers to their motion
through the lattice. Because of the mutual interference of adjacent grains in a polycrystalline
specimen, multiple slips occur readily, and there is appreciable strain hardening. Coldwork is
plastic deformation with no the strain hardening relief. The dislocation density of coldworked metal
is 10
10
to 10
12
cm
÷2
. Coldworked dislocation cell structure is not thermodynamically stable and the
coldworked state is a condition of higher free energy. With increasing temperature, the metal
softens and reverts to a strainfree condition. The process of annealing can be divided into recovery,
recrystallization, and grain growth. Recovery is usually deﬁned as the restoration of the mechanical
properties of the coldworked metal without observable change in microstructure. Recrystallization
is the replacement of the coldworked structure by a new set of strainfree grains. Recrystallization is
evidenced by a decrease in strength and an increase in ductility. The density of dislocations
decreases considerably on recrystallization, and all effects of strain hardening are eliminated.
Some of the grains of a ﬁnegrained recrystallized metal can begin to grow rapidly at a higher
temperature. It is called the secondary recrystallization [5].
Very good results in metal strengthening can be achieved by the combination of metal
strengthening by microstructural changes in conjunction with strain hardening. Very high strengths
can be achieved by ausforming, that is, thermalmechanical processes in which martensite is formed
from an austenitic matrix, which had been previously strengthened by plastic deformation. The
dislocation density of ausformed martensite is very high (10
13
cm
÷2
), and the dislocations are
usually uniformly distributed and the inﬂuence of precipitation contribution to the strength of
ausformed martensite is more intensive than in ordinary quenched martensite. Strong alloys are
those in which the particles are formed in dense dislocation cell structures of the deformed matrix.
Very strong alloys are produced by combining the effects of dispersion and strain hardening [4,5].
4.4.8 TOUGHENING MECHANISMS
Generally, an inverse relation exists between strength and toughness, but some micromechanisms
exist for simultaneously increasing strength and toughness. It could be achieved by processes such
as proper alloy chemistry, melting practice, suitable microstructure and phase distribution, and
microstructure reﬁnement.
The combination of alloy strength and toughness is dependent on alloying elements [4]. For
example, in steel alloys carbon is the strengthener but carbide locations are places for crack
nucleation. Nickel lowers transition temperature and provides better toughness. Chromium is a
solidsolution strengthener but strong carbide former. Molybdenum is a solidsolution strengthener,
strong carbide former, and temper embrittlement suppresser. Silicon increases yield strength and
transition temperature. Manganese prevents hot cracking and lowers transition temperature. Cobalt
increases martensite formation and precipitation hardening in maraging steels. Titanium is a carbide
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 172 1.11.2008 2:47am Compositor Name: JGanesan
172 Handbook of Thermal Process Modeling of Steels
and nitride former and in maraging steels it is used for precipitation hardening. Aluminum lowers
lattice resistance to dislocation motion and lowers transition temperature.
Austenitic steels are tough because of the tough austenite face centered cube (FCC) crystal
structure. Lower bainite and lowcarbon martensite provide greater fracture resistance than ferrite,
pearlite, or upper bainite. Intermetallic precipitates in maraging steel are much ﬁner and more
resistant to fracture than analogous carbides in quenched and tempered steel. The fracture toughness
increases with an increase in the amount of austenite retained in the microstructure. The retained
austenite phase is a crack arrester and crack blunter, since it is softer and tougher than the martensite
phase. However, retained austenite in highcarbon steels can transform by stressinduced trans
formation to untempered martensite, which is a brittle microconstituent.
Microstructural reﬁnement represents a unique opportunity by which the material may be
simultaneously strengthened and toughened (Figure 4.39). The toughness and strength superiority
of ﬁnegrained materials has been recognized for many years. Microstructures achieved by cyclic
heat cyclic treatment are superior to those associated with quenching and tempering as well as with
normalizing processes cyclic heat processes give ﬁner transformation products, such as lower
bainite and martensite, than quenching and tempering processes [59]. Grain reﬁnement serves to
increase yield strength and fracture strength, while lowering the ductilebrittle transition tempera
ture. By using dislocation models, fracture stress and yield stress could be given by Equations 4.168
and 4.157, respectively.
If more dislocations are generated near the crack tip, more blunting can take place and the
tougher the material would be. When dislocation sources are pinned by solute interstitials, such as
nitrogen and carbon in the case of steel alloys, effective surface energy and fracture stress are
reduced. The beneﬁcial effect of increased test temperature may be traced to a reduction in the
Peierls stress and an increase in dislocation velocity.
The fatigue properties of metals are quite structuresensitive. Most highstrength materials are
fatiguelimited. Fatigue strength is directly proportional to the difﬁculty of dislocation cross slip.
Grain size has its greatest effect on fatigue life in the lowstress, highcycle regime.
In general, quenched and tempered microstructures result in the optimum fatigue properties in
heattreated lowalloy steels. However, a bainitic structure produced by austempering results in
better fatigue properties than a quenched and tempered structure with the same hardness. Only a
small amount of nonmartensitic transformation products can cause an appreciable reduction in the
fatigue limit. The presence of only a trace of decarburization on the surface may drastically reduce
the fatigue properties [5].
F
r
a
c
t
u
r
e
s
t
r
e
s
s
Yield stress
d
1
>
d
2
d
2
d
1
FIGURE 4.39 Inﬂuence of grain size on fracture and yield stress.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 173 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 173
The fatigue properties at high hardness levels are extremely sensitive to surface preparation,
residual stresses, and inclusions. Improvements in fatigue performance result from design changes
by which the stress concentration is reduced and by using beneﬁcial compressive residual stress.
Fatigue strength could be increased by the control of microstructure through thermomechanical
processing.
4.5 METAL BEHAVIOR AT ELEVATED TEMPERATURE
The inﬂuence of temperature on mechanical behavior of metals has to be taken into account in
thermal processing. In thermal processes, the properties and metallurgical structure may continu
ously change with time. Mechanical behaviors data tool and work piece at elevated temperature are
important for tool life prediction and quality of manufacturing. During thermal processes, tools
are affected by many degradation processes, such as creep, hightemperature fatigue, and thermal
fatigue. Analysis of distortion and fracture appearances due to residual stresses is an important part
of mechanical metallurgy of thermal processing.
Since the mobility of atoms increases rapidly with temperature, diffusioncontrolled processes
can have a signiﬁcant effect on mechanical properties at elevated temperature. An unpredictable
combination of deformation mechanisms can exist at elevated temperatures. However, microstruc
tural changes such as precipitation, strain aging, or recrystallization may occur in certain tempera
ture ranges to alter metal behavior. The concentration of vacancies increases with temperature and
high temperature can result in greater mobility of dislocations. Therefore, coldworked metals and
agehardening alloys may lose strength at elevated temperature. New interaction of the metal with
its environment can be a problem at high temperature.
This section attempts to provide a basic understanding of steel behaviors at elevated tempera
ture, which is discussed in greater detail in Refs. [5,15,60,61].
4.5.1 EFFECT OF TEMPERATURE ON FLOW PROPERTIES
High external loadings of metallic materials usually exhibit under elastic–plastic deformation at
elevated temperature. In general, ductility increases and strength decreases with temperature increase.
The change of the stress–strain curve in mild steel with temperature is shown schematically in
Figure 4.40.
The yield strength, which limits the elastic strain range, is strongly temperature dependent and
decreases with increasing temperature. According to these facts, steels are relatively soft at high
temperatures. The temperature dependency of yield strength s
Y
for low carbon steel are shown in
Figure 4.41.
T
1
>
T
2
>
T
3
T
3
T
2
T
1
Strain
S
t
r
e
s
s
FIGURE 4.40 Engineering stress–strain curves of mild steel (schematically).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 174 1.11.2008 2:47am Compositor Name: JGanesan
174 Handbook of Thermal Process Modeling of Steels
Thermally activated processes assist deformation and reduce strength at elevated temperatures.
The temperature dependence of ﬂow stress at constant strain and strain rate generally can be
expressed by
s
flow
= C
2
e
Q=RT
(4:178)
where
Q is an activation energy for plastic ﬂow
R is the universal gas constant
T is the absolute testing temperature
Activation energy Q can be predicted by two tests at different temperatures:
Q = Rln
s
1
s
2
_ _
T
1
T
2
T
2
÷T
1
(4:179)
The rate at which strain is applied to a specimen can have an important inﬂuence on the ﬂow stress.
Strain rate is deﬁned as _ « = d«=dt. Ranges of strain rates are from 10
÷8
to 10
÷5
s
÷l
for creep
processes at constant load to 10
2
to 10
4
s
÷l
for high strain rate at impact loads. The ﬂow stress
increases with increasing strain rate. A relationship between ﬂow stress and strain rate at constant
temperature and strain is [40]
s = C_ «
m
(4:180)
where m is known as the strainrate sensitivity. Strainrate sensitivity of metals is low at room
temperature, but increases with temperature, especially at temperatures above half of the absolute
melting point T
m
. By combining Equations 4.178 and 4.180 we can obtain
s = f _ «e
DH=RT
_ _
(4:181)
where DH is an activation energy, related to the activation energy Q from Equation 4.178. The
thermal process at temperatures above 0.6 of T
m
and at high strain rates in the range of 1 to 500 s
÷1
Temperature, u (
º
C)
200
600
400
200 0 400 600 800 1000
Y
i
e
l
d
s
t
r
e
n
g
t
h
,
s
Y
(
N
/
m
m
2
)
Carbon content 0.2 %
Carbon content 0.15 %
FIGURE 4.41 Temperature dependency of yield strength for low carbon steel.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 175 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 175
with a strain of 1–5 is referred to as hotworking. Recovery processes occur substantially during the
deformation process of hotworking. At high temperature, large strains can be achieved with
essentially no strain hardening. The softening mechanism in metals such as alphabased iron alloys
is recovery. This occurs by the formation of a subgrain structure. In this case the activation energy
for hotworking is equal to that for creep. In metals such as austenitic stainless steel, the softening
process is recrystallization and the activation energy for these metals for softening in hotwork is
higher than that for creep.
4.5.2 EFFECT OF TEMPERATURE ON FATIGUE
In general, the fatigue strength of metals decreases with increasing temperature. Ferrous materials,
which ordinarily exhibit a sharp fatigue limit in roomtemperature tests, will no longer have a
fatigue limit when tested at temperatures above approximately 4508C. At high temperatures, creep
will be the principal cause of failure. The transition from fatigue failure to creep failure with
increasing temperature will result in a change in the type of failure from the usual transcrystalline
fatigue failure to the intercrystalline creep failure. At any given temperature the amount of creep will
increase with increasing mean stress. Local grainboundary oxidation can contribute signiﬁcantly to
crack initiation at elevated temperature. In addition, fatigue tests at high temperature will depend on
the frequency of stress application. It is customary to report the total time to failure as well as the
number of cycles to failure. In general, a higher creep strength gives a higher hightemperature
fatigue strength [5].
Finegrain size results in better fatigue properties at lower temperatures. As the test temperature
is increased, the difference in fatigue properties between coarsegrain and ﬁnegrain material
decreases. At high temperatures, where creep predominates, a coarsegrain material has a higher
strength.
Procedures that are successful in reducing fatigue failures at room temperature may not be
effective in hightemperature fatigue. For example, compressive residual stresses may be annealed
out at operating temperature [5].
Fatigue tests on metals at temperatures below room temperature show that fatigue strength
increases with decreasing temperature. There is no evidence to indicate any sudden change in
fatigue properties at temperatures below the ductiletobrittle transition temperature.
4.5.3 TRANSFORMATION PLASTICITY
An additional dimensional change to that produced by the applied stress can be recognized during
phase transformation. This effect can occur even at stresses below the yield stress of the material.
This phenomenon is referred to as transformation plasticity [62–64]. Transformation plasticity
accompanying martensitic transformations under stress has provided extraordinary mechanical
behavior ranging from the reversible plasticity phenomena in thermoelastic shape memory alloys.
There is a good opportunity that the theory of plasticity may be used in the calculation of
stresses during thermal processes that are accompanied by transformation plasticity.
Several assumptions have been used for the explanation of transformation plasticity. First,
transformation plasticity can be explained with the temporarily lost yield stress during microstruc
ture transformation, as a result of the movement of atoms from their original positions as they
participate in the transformation. This assumption can be used in the prediction of transformation
plasticity during the formation of martensite.
In the Greenwood–Johnson complex model of transformation plasticity, it is assumed that the
combination of an applied stress and internal stresses created locally by the transformation produces
stress concentrations that locally exceed the yield stress of the material [65]. Moreover, in the
Greenwood–Johnson model it is assumed that the slip at the interface between the phases is
responsible for the observed additional strain. Finally, it is assumed that the preferred orientation
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 176 1.11.2008 2:47am Compositor Name: JGanesan
176 Handbook of Thermal Process Modeling of Steels
of the product phase creates an abnormally high strain in certain directions relative to the direction
of the applied stress. There is experimental evidence that the diffusionless transformations, such as
the formation of martensite in steel, are orientated under the inﬂuence of an applied stress. An
abnormally high strain in convinced directions relative to the direction of the applied stress can be
created by preferred orientation of the product phase.
In the Greenwood–Johnson model a rigid plastic model of the deformation process is used
together with the assumption that all plastic ﬂow occurs in only one of these phases that was weaker
than the others [15]. In accordance with Greenwood–Johnson research, the tensile strain expected
when transformations occur under tensile stress can be described by [65]
« =
5s
6s
Y
DV
V
_ _
(4:182)
where
DV is the volume dilation
s
Y
is the yield strength of the weaker phase
s is the uniaxial applied stress
Taking into account the fraction of transformation product in the structure, the transformation
plasticity can be expressed by Ref. [15]
« =
3sDV
4s
Y
V
m ÷
2
3
m
3
2
_ _
(4:183)
where m is the fraction of the structure transformed.
In the case where the transformation plasticity is related to the orientation of the product with
respect to the direction of the applied stress, the contribution of the constituent formed at each
orientation should be summated to compute the value of the total transformation plasticity.
As an addition to the above given models, the transformation plasticity could be explained by
the assumption that transformation strains cause the creation of point defects that accelerate the
creep effects [15].
Although most of the work has been based on the martensitic transformation in steels, the
phenomenon of transformation plasticity is not restricted to diffusionless transformations and
the same equations may be applied to other reactions, such as the formation of pearlite and bainite.
4.5.4 EFFECT OF STRESS ON MICROSTRUCTURE TRANSFORMATIONS
It is known that parameters of martensite transformation in steels can be slightly inﬂuenced by
applied stresses. Both the shear stress and the uniaxial tensile stress raise the M
s
temperature, but a
hydrostatic compression has the opposite effect. High M
s
temperature is associated with an
enhanced amount of transformation product. Hydrostatic compression reduces equilibrium tem
perature and displaces the eutectoid composition. Uniaxial tensile stress displaces the characteristic
curves for the start and the ﬁnish of the transformations in TTT diagrams to higher temperatures and
shorter times. On the other hand, a hydrostatic compression displaces the same curves in the
opposite sight. It is known that an increase in the dislocation density of the austenite lowers
the M
s
temperature. The increment of straininduced martensitic nucleation sites can be expressed
as the product of the density of the new nucleation sites introduced by the plastic strain and of the
probability that these sites will be activated [65,66]:
dN
d«
= i
0
exp ÷
W
RT
_ _
(4:184)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 177 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 177
where
W is the activation energy
i
0
is the potent number of nucleation sites per unit volume, which increases when the plastic
strain increases
The applied elastic stress can alter the driving force for the transformation and can change the
appearance of the microstructure transformation.
Activation energy W is a function of the driving force G, which is a sum of chemical and
mechanical contribution:
G = G
s=0
÷
@G
@s
s
Y
(4:185)
4.6 RESIDUAL STRESSES, DISTORTIONS, AND FRACTURE
Hightemperature distortion and fracture are important engineering problems, which may consid
erably increase the costs of operations that involve thermal processing [61]. In the heated=cooled
components, stresses that exist in a body in the absence of mechanical loads are referred to as
thermal stresses. Thermal stresses result when the thermal change in the dimensions of a member is
prevented by some kind of constraint. In this case the thermal stress developed by a temperature
change DT is proportional to aEDT, where a is the linear thermal coefﬁcient of expansion, and E is
the elastic modulus [61,67]. The resultant force produced by thermal stresses is equal to zero.
Thermal stresses are often introduced in a material during thermal processes, such as casting,
welding, or heat treatment [61,68].
Thermal stresses could locally cause plastic deformations if the corresponding stresses exceed
the local yield strength of the material state. If sudden failure occurs by thermal stress, it is referred
to as thermal shock. Failure that occurs after repeated applications of thermal stresses is referred to
as thermal fatigue. Conditions for thermalshock and thermalfatigue failure are frequently present
in hightemperature equipment.
The resistance to both thermalshock and thermalfatigue failure is related to the parameter
l=Ea, where l is the thermal conductivity. A high value of this parameter gives good resistance to
thermal failures. Austenitic stainless steel is particularly sensitive to this phenomenon because of its
low thermal conductivity and high thermal expansion.
Thermal or residual stress generation is dependent on temperature distributions, elastic–plastic
stress and strain behaviors, physical and mechanical properties, and microstructural changes in
material.
Residual stresses are of three different types, that is, residual stresses of the ﬁrst, second, and
third kind [60]. Residual stresses of the ﬁrst kind are equilibrated with respect to the whole body.
They are homogeneous across large areas, for example, several grains of a material.
The ﬁrst kind of macroresidual stress state develops as a consequence of nonuniform heating or
cooling at different points of volume body.
The second kind of residual stresses is induced by microstructure heterogeneity. Residual
stresses of the second kind are in equilibrium across a sufﬁcient number of grains and are
homogeneous across microscopically small areas, for example, one grain, or subgranular regions.
Macroscopic changes in the dimensions of a volume element possessing these stresses are usually
not possible.
The third kind of residual stresses is usually the result of lattice defects, dislocations, or
interactions of dissolved atoms. Residual stresses of the third kind are in equilibrium across
subgranular sections and are inhomogeneous across submicroscopically small areas of a body,
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 178 1.11.2008 2:47am Compositor Name: JGanesan
178 Handbook of Thermal Process Modeling of Steels
that is, regions with dimensions of several atomic distances. No macroscopic changes in the
dimensions of a volume element possessing these stresses are possible.
Residual stresses of the ﬁrst kind are also called macro or volumeresidual stresses, but the
second and third kind residual stresses are called micro or texturalresidual stresses. In Figure 4.42
residual stresses of the three different types are schematically illustrated.
Superposition of residual stresses of the ﬁrst, second, and third kind determines the total residual
stress state acting locally at a given material state:
s
RS
ij
= s
RS
ij
_ _
first
÷ s
RS
ij
_ _
second
÷ s
RS
ij
_ _
third
(4:186)
Since the phase transformations are accompanied by volume changes, residual stresses can be
induced by microstructure transformations. For example, the volume changes at room temperature
due to total martensitic transformation of plain carbon steels with various carbon contents can be
predicted by [60]
DV=V
20
·
C
=
3:216 ÷0:859 wt% C ÷(0:34 wt% C)
2
100
(4:187)
up to carbon contents of approximately 1.0 wt%.
Residual stresses, distortion, and fracture in complex components are not easy to predict
precisely. In Figure 4.43 residual longitudinal stresses are shown in specimens made of lowalloy
steel, quenched in water [60].
Residual stresses, distortion, and fracture result either from coupled thermomechanical inter
actions or from the complex superposition of thermomechanical with transformation effects [68].
Potential interactions between the time, temperature, deformation, and microstructure are shown in
Figure 4.44.
The calculations of thermal stresses and strains have to be based on theoretical considerations
given in Section 4.2. Therefore, it is necessary to possess a set of relationships between temperature,
microstructure, and material properties that affect the generation of residual stresses, such as yield
stress, workhardening constants, coefﬁcient of expansion, Young’s modulus, Poisson’s ratio, and
R
e
s
i
d
u
a
l
s
t
r
e
s
s
e
s
+

+

+

s
RS
first
s
RS
second
s
RS
third
FIGURE 4.42 Residual stresses (schematically).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 179 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 179
so on. The inﬂuence of temperature on mechanical and physical properties of steel microstructure
constituents is shown in Figure 4.45.
In order to calculate the stress and strain distribution during thermal processes, usually the
volume residual stresses and transformations stresses must be taken into account if microstructure
transformations exist during thermal processing. Additionally, the kinetics of these phase trans
formations and their interactions with the local stress states have to be considered. As a conse
quence, more or less pronounced transformation plasticity effects as an important phenomenon
during the microstructure transition have to be added to the list of relevant property data used in
residual stresses and distortions predictions. Proper failure criteria have to be accepted based on the
consideration given in Sections 4.3, and 4.4. Analyses speciﬁc to the designed thermal process have
to involve the introduction of failure criteria as functions of time and temperature.
200
600
400
200
0
−400
800
1000
T
e
m
p
e
r
a
t
u
r
e
,
Њ
C
−200
0
L
o
n
g
i
t
u
d
i
n
a
l
r
e
s
i
d
u
a
l
s
t
r
e
s
s
(
N
/
m
m
2
)
d=100 mm
d=30 mm
d=10 mm
Core
Surface
1 10 10
2
10
3
Time (s) Ratio of cross section
Surface Core
0 0.5 1.0
d=100 mm
d=30 mm
d=10 mm
FIGURE 4.43 Residual stresses in cylindrical specimens of lowalloy steel.
Time,
temperature
Stress, strain
Thermal stress
Heat of
deformation
Time,
temperature
Microstructure
Phase transformation
Latent heat
Microstructure Stress, strain
Transformation
strain
Stressinduced
transformation
FIGURE 4.44 Time, temperature, deformation, and microstructure interactions.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 180 1.11.2008 2:47am Compositor Name: JGanesan
180 Handbook of Thermal Process Modeling of Steels
REFERENCES
1. Tegart, W.J. McG., Elements of Mechanical Metallurgy, Macmillan, New York, 1968, p. 12.
2. Gdoutos, E.E., Fracture Mechanics: An Introduction, (Solid Mechanics and Its Applications) 2nd Ed., Vol.
123, Springer, Dordrecht, 2005, Chapter 1–7.
3. Hill, R., The Mathematical Theory of Plasticity, Oxford University Press, New York, 1998, Chapters 2, 5,
9, 10.
4. Hertzberg, R., Deformation and Fracture Mechanics of Engineering Materials, John Wiley & Sons,
New York, 1995, Chapters 2–4, 7, 10.
5. Dieter, G., Mechanical Metallurgy, McGrawHill, New York, 1977, Chapters 1, 4–6, 9.
6. Solecki, R. and Conant, R., Advanced Mechanics of Materials, Oxford University Press, Oxford, 2003,
Chapters 2, 3, and 4.
7. Khan, A. and Huang, S., Continuum Theory of Plasticity, WileyInterscience, New York, 1995, Chapters 2,
3, 5, 9, 10.
8. Mendlson, A., Plasicity, Theory and Application, Macmillan, New York, 1968, Chapters 3–7.
Y
i
e
l
d
s
t
r
e
n
g
t
h
,
M
P
a
Temperature (ºC)
M
B
F + P
A
E
l
a
s
t
i
c
m
o
d
u
l
u
s
,
G
P
a
M
B; F + P
A
Temperature (ºC)
S
p
e
c
i
f
i
c
h
e
a
t
c
a
p
a
c
i
t
y
,
1
0

6
J
/
m
3
K
M
A
B; F + P
Temperature (ºC)
T
h
e
r
m
a
l
c
o
n
d
u
c
t
i
v
i
t
y
,
W
/
m
K
M
B; F + P
A
Temperature (ºC)
P
o
i
s
s
o
n
r
a
t
i
o
M (until 550 ºC)
A
200
400
800
1600
0
1200
400 600 800
200 0 400 600 800
Temperature (ºC)
250
200
150
100
50
200 0 400 600 800
200 0 400 600 800
200
0
400 600 800
6
4
2
50
40
30
20
10
B; F + P
0.35
0.30
0.25
0.20
0.15
0.10
0.05
M
B
F + P
A = Austenite
= Ferrite + pearlite
= Bainite
= Martensite
FIGURE 4.45 Mechanical and physical properties of steel microstructure constituents.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 181 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 181
9. Yu, M. et al., Generalized Plasticity, SpringerVerlag, 2005, Chapters 2–4, 6, 13.
10. Boresi, A.P. and Chong, K.P., Elasticity in Engineering Mechanics, Elsevier Science, New York, 1987,
Chapter 2.
11. Prager, W., Strain hardening under combined stress, Journal of Applied Physics, 16, 837, 1945.
12. Lemaitre, J. and Chaboche, J., Mechanics of Solid Materials, Cambridge University Press, Cambridge,
2002, Chapters 1, 2, 3.
13. Drucker, D.C., Some implications of work hardening and ideal plasticity, Quarterly Applied Mathematics,
7, 411, 1950.
14. Drucker, D.C., A More Fundamental Approach to Plastic StressStrain Relations, 1st U.S. Congress on
Applied Mechanics, ASME, New York, 1952, 487.
15. Fletcher, A.J., Thermal Stress and Strain Generation in Heat Treatment, Elsevier Applied Science, London
and New York, 1989, Chapters 2, 4, 5.
16. Smoljan, B., Delayed failure of quenched steel 42CrMo4 (in Croatian), Ljevarstvo, 1–2, 19, 1986.
17. Broek, D., Elementary Engineering Fracture Mechanics, Sijthoff & Noordhoff, Alphen aan den Rijn, the
Netherlands, 1978, Chapters 1, 2, 3, 5, 9.
18. Irwin, G.R., Structural aspects of brittle fracture, Applied Materials Research, 3, 65, 1964.
19. Paris, P.C. and Sih, G.C., Stress Analysis of Cracks, ASTM STP 381, 1965, 30.
20. Tetelman, A.S. and Mc Evily, A.J., Fracture of Structural Materials, John Wiley & Sons, New York,
1971, Chapters 2, 3, 4.
21. Grifﬁth, A.A., The phenomena of rupture and ﬂow in solids, Philosophical Transactions of the Royal
Society of London, A221, 1921, 163.
22. Orowan, E., Energy criteria of fracture, Welding Journal, 34, 1575, 1955.
23. Cottrell, A.H., Theoretical aspects of radiation damage and brittle fracture in steel pressure vessels, Iron
Steel Institute Special Report, 69, 1961, 281.
24. Wells, A.A., Unstable crack propagation in metals: cleavage and fracture, Proceedings of the Crack
Propagation Symposium, Vol.1, College of Aeronautics, Cranﬁeld, 1961, p. 210.
25. Rice, J.R., A path independent integral and the approximate analysis of strain concentration by notches and
cracks, Journal of Applied Mechanics, Transactions ASME 35, 379, 1968.
26. Begley, J.A. and Landes, J.D., The Jintegral as a fracture criterion, In: Fracture Toughness, ASTM STP
514, American Society for Testing and Materials, Philadelphia, 1972, Chapter 1.
27. Sih, G.C., Strainenergydensity factor applied to mixedmode crack problems, International Journal of
Fracture, 10, 305, 1974.
28. Sih, G.C., Energydensity concept in fracture mechanics, Engineering Fracture Mechanics, 5, 1037,
1973.
29. Paris, P. and Erdogan, F., A critical analysis of crack propagation laws, Journal of Basic Engineering,
Trans. ASME 85, 528, 1963.
30. Jeffery, C., Creep deformation of metals, polymers, ceramics, and composites, in ASM Handbook—
Mechanical Testing and Evaluation, Vol. 8, ASM International, Materials Park, OH, 2000, p. 361.
31. Quarrell, A.G., Dislocations in metals, In: The Structure of Metals, a Modern Concept, The institution of
metallurgists, London, 1959, Chapter 3.
32. Read, W.T. Jr., Dislocations in Crystals, McGrawHill, New York, 1953, Chapter 2.
33. Weertman, J. and Weertman, J.H., Elementary Dislocation Theory, Oxford University Press, New York,
1992, Chapters 4–6.
34. Hull, D. and Bacon, D., Introduction to Dislocations, Pergamon Press, New York, 1965, Chapters 3, 4.
35. Hirth, J.P. and Lothe, J., Theory of Dislocations, McGrawHill, New York, 1968, p. 506.
36. Cottrell, H., An Introduction to Metallurgy, Edward Arnold Publishers, London, 1967, p. 266.
37. Guyot, P. and Dorn, E., A critical review of the peierls mechanism, Canadian Journal of Physics, 45, 983,
1967.
38. Rosenﬁeld, et al., Dislocation Dynamics, McGrawHill, New York, 1968, Chapter 2.
39. Taylor, G.I., The mechanism of plastic deformation of crystals, Part I—Theoretical, Proceedings of the
Royal Society of London, Series A 145, 1934, p. 362.
40. Bernstein, M.L. and Zaimovsky, V.A., Mechanical Properties of Metals, Mir publisher, Moskov, 1983,
Chapters 2, 3, 5.
41. Bonade, R. et al., Plastic ﬂow of martensitic model alloys, Materials Science and Engineering, Series A
387–389, 2004, 16.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 182 1.11.2008 2:47am Compositor Name: JGanesan
182 Handbook of Thermal Process Modeling of Steels
42. Taylor, G.I., Plastic strain in metals, Journal of Institute of Metals, 62, 307, 1938.
43. Zeener, C., The micromechanisms of fracture, In: Fracturing of Metals, American Society for Metals,
Metals Park, OH, 1948, p. 3.
44. Cottrel, A.H., Theoretical aspects of fracture, In Fracture, Averbach, L. et al., Eds., John Wiley & Sons,
New York, 1959, p. 20.
45. Knott, J.F., Fundamentals of Fracture Mechanics, Butterworths, London, 1981, p. 188.
46. Smith, E., The nucleation and growth of cleavage microcracks in mild steel, Proceedings of the Conference
on Physical Basis of Yield and Fracture, Institute of Physics and Physical Society, Oxford, 1966, p. 36.
47. Argon, A.S., Im, J., and Safogly, R., Cavity formation from inclusions in ductile fracture, Metallurgical
Transactions, 6A, 1975, 825.
48. Beremin, F.M., Cavity formation from inclusions in ductile fracture of A 508 Steel, Metallurgical
Transactions, 12A, 723, 1981.
49. Goods, S.H. and Brown, L.M., The nucleation of cavities by plastic deformation, Acta Mechanica, 27,
l, 1979.
50. Gurson, A.L., Continuum theory of ductile rupture by void nucleation and growth: Part I—Yield criteria
and ﬂow rules for porous ductile media, Journal of Engineering Materials and Technology, 99, 2, 1977.
51. Tvergaard, V., On localization in ductile materials containing spherical voids, International Journal of
Fracture, 18, 237, 1982.
52. Sherby, O.D. and Burke, P.M., Mechanical behavior of crystalline solids at elevated temperature, Progress
in Material Science, 325, 1967.
53. Bird, J.E., Mukherjee, A.K., and Dorn, J.F., Correlations Between HighTemperature Creep Behavior and
Structure, Haifa University Press, Israel, 1969, p. 255.
54. Mukherjee, A.K., Bird, J.E., and Dorn, J.E., Experimental correlation of high temperature creep, ASM
Transactions Quarterly, 62, 155, 1969.
55. Gifkins, R.C., Mechanisms of intergranular fracture at elevated temperature, In: Fracture, Averbach, L.
et al., Eds., John Wiley & Sons, New York, 1959, p. 613.
56. Garofalo, F., Ductility, American Society for Metals, Metals Park, OH, 1968, Chapter 4.
57. Hull, D. and Rimmer, D.E., The growth of grainboundary voids under stress, Philosophical. Magazine,
Eighth Series, 4, 673, 1950.
58. Le May, I., Failure and mechanisms and metallography, In Metallography in Failure Analysis, McCall, J.
and French, P., Eds., Plenum Press, New York, 1977, p. 1.
59. Smoljan, B., An analysis of combined cyclic heat treatment performance, Journal of Materials Processing
Technology, 155, 1704, 2004.
60. Macherrauch, E. and Wohringer, O., Residual stresses after quenching, Theory and Technology of
Quenching, SpringerVerlag, Berlin, 1992, Chapter 6.
61. Totten, G. et al., Failures related to heat treating operations, in ASM Handbook—Failure Analysis and
Prevention, Vol. 11, ASM International, Berlin, 2002, p. 192.
62. Leblond, J.B. et al., A theoretical and numerical approach to the plastic behavior of steels during phase
transformations—derivation of general relations, Journal of Mechanical Physics and Solids, 34, 395, 1986.
63. Leblond, J.B. et al., Mathematical modelling of transformation plasticity in steels—Case of idealplastic
phases, International Journal of Plasticity, 5, 551, 1989.
64. Stringfellow, R.G. et al., A constitutive model for transformation plasticity accompanying straininduced
martensitic transformations in metastable austenitic steels, Acta Metalurgica et Materialia., 40, 1703,
1992.
65. Greenwood, G.W. and Johnson, R.H., The deformation of metals under small stresses during phase
transformations, Proceedings of the Royal Society London, A283, 403, 1965.
66. Perlade, A., Bouaziz, O., and Furne’mont, Q., A physically based model for TRIPaided carbon steels
behavior, Materials Science and Engineering, A356, 145, 2003.
67. Berns, H., Distortion and crack formation by heat treatment of tools, Radex Rundsch, 1, 40, 1989.
68. Bhadeshia, H., Material factors, In: Handbook of Residual Stress and Deformation of Steel, Totten, G. and
Inoe, T., Eds., ASM International, 2002.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 183 1.11.2008 2:47am Compositor Name: JGanesan
Mechanical Metallurgy of Thermal Processing 183
Gur/Handbook of Thermal Process Modeling of Steels 190X_C004 Final Proof page 184 1.11.2008 2:47am Compositor Name: JGanesan
5
Modeling Approaches and
Fundamental Considerations
Bernardo HernandezMorales
CONTENTS
5.1 Mathematical Modeling ..................................................................................................... 185
5.2 Governing Equation........................................................................................................... 187
5.3 Boundary and Initial Conditions........................................................................................ 189
5.4 Numerical Solution of the Governing Equation................................................................ 190
5.5 FiniteElement Method ...................................................................................................... 192
5.6 FiniteVolume Method....................................................................................................... 202
5.7 Monte Carlo Method.......................................................................................................... 214
5.8 PhaseField Method ........................................................................................................... 218
5.9 Concluding Remarks.......................................................................................................... 222
References ..................................................................................................................................... 222
Mathematical modeling and computer simulation of thermal processing of metals have become
essential tools to simulate, understand, optimize, and control existing processes as well as to test
new ones before actually constructing a physical prototype.
In this chapter, the general methodology to develop a mathematical model is discussed,
followed by a detailed presentation of the numerical methods commonly used to solve mathematical
models of thermal processes.
5.1 MATHEMATICAL MODELING
Materials process engineering is a discipline that has evolved over the last few decades to aid in the
design, optimization, and troubleshooting of manufacturing industrial processes. Building upon
previous papers [1,2], Brimacombe [3] identiﬁed ﬁve tools that the metallurgical=materials process
engineer has at his=her disposal to understand, simulate, and predict the behavior of materials in
extractive and manufacturing processes. These tools are: (1) plant measurements, (2) pilot plant
measurements, (3) physical models, (4) mathematical models, and (5) laboratory measurements.
Plant measurements are carried out as the ﬁrst step to understand the process behavior. They can
also provide valuable data for model validation but are usually very difﬁcult to conduct due to
conditions that cannot be easily controlled. On the other hand, pilot plant measurements are carried
out in the design stage of a new process. A pilot plant is a scaleddown version of the fullscale
plant. The measurements are directed toward demonstrating the technical and economical feasibility
of a new process and its ability to deliver the required product. Therefore, they must accurately
represent the physical and chemical aspects of the process. Physical models are also scaleddown
versions of the fullscale prototype but they are more approximate than pilot plants and are usually
applied to simulate only certain aspects of a process. For example, isothermal models (such as that
in Ref. [4]) have been extensively used to study the nature of ﬂuid ﬂow in tundishes for continuous
casting of steel even though there are always thermal gradients in the real process. Physical models
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 185 7.11.2008 5:44pm Compositor Name: BMani
185
are built with alternative materials that allow the measurements to be carried out in a more
convenient fashion. In contrast, mathematical models are mathematical representations of the
physical behavior of the system. They consist of equations based on conservation principles and
constitutive equations, together with boundary and=or initial conditions. Finally, controlled meas
urements are carried out in the laboratory to determine: (1) the parameters (thermophysical,
thermodynamic, and thermomechanical properties; boundary conditions; and kinetics of chemical
reactions and phase transformations) needed in mathematical models and (2) the response of the
system. The latter is used to validate the mathematical model.
As pointed out by Brimacombe [3], the strength of these tools lies in their combined use.
However, in this chapter the focus is on mathematical model development, with particular emphasis
on the numerical methods that are commonly applied to solve them.
The roadmap to develop a mathematical model [5] is shown schematically in Figure 5.1. During
the preparation stage, the objectives of the overall work are clearly established. An important point
to consider is the nature of the research itself as different approaches are required for fundamental as
opposed to applied research. In addition, previous work in the ﬁeld needs to be considered. In order
to successfully apply this methodology, the objective of the mathematical modeling exercise must
be accompanied by a statement specifying its scope, which will necessarily consider the conditions
(human, material, and technological resources as well as deadlines) under which the project will be
carried out. Therefore, a number of assumptions are laid out in order to make the problem tractable.
These assumptions are related to topics such as whether to consider steady or unsteady state; a one,
two, or threedimensional (3D) domain; constant or variable properties; coupled or uncoupled
phenomena; and so on. The assumptions made determine how closely the mathematical model will
reproduce the actual behavior of the process; they will also deﬁne the complexity of the model and,
therefore, the effort (technical and computational) required to develop it. With all this information,
the mathematical model can be formulated, i.e., the differential equations* that mathematically
Preparation
Mathematical formulation
Numerical solution
Verification
Validation
Sensitivity analysis
Application
FIGURE 5.1 Roadmap to develop a mathematical model.
* The mathematical formulation may also be cast in other forms: integral, integrodifferential, stochastic, etc.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 186 7.11.2008 5:44pm Compositor Name: BMani
186 Handbook of Thermal Process Modeling of Steels
describe the response of the system can be written down, together with initial and=or boundary
conditions.
To model industrialscale thermal processes accurately, the assumptions are kept to a minimum,
which usually produces highly nonlinear differential equations. Thus, it is virtually impossible to
solve the mathematical model analytically and, therefore, one has to resort to numerical solutions.
To compute the numerical solution efﬁciently, the corresponding algorithm needs to be implemen
ted in a computer. Once the numerical solution has been coded, two steps must be taken to ensure
the accuracy and reliability of the code: veriﬁcation and validation. In the veriﬁcation step,
numerical predictions are compared to analytical solutions of simpliﬁed mathematical formulations
of the system of interest. If the code passes this test, a direct comparison between numerical
solutions against available experimental measurements is carried out; if this stage is successfully
passed the code may be considered as validated. Once validated, the code can be used to conduct
sensitivity (parametric) studies and, ﬁnally, applied to reach the objectives of the project.
5.2 GOVERNING EQUATION
The fundamental block of a mathematical model is the governing equation. This equation is a
mathematical statement of the conservation of a transport property (momentum, mass, energy, etc.)
within a sampling volume in the system:
Rate of change ¼ Net flow þInterphase flow þRate of creation=destruction (5:1)
In Equation 5.1 the net ﬂow (input minus output) may occur by two mechanisms acting within the
system: diffusion and convection; for heat transfer a third mechanism (radiation) may occur.
Interphase ﬂow refers to transport property exchange with the surrounding media and, therefore,
takes place at the interphases between the system and the surroundings; it applies to both heat and
mass transfer. Both the rate of change and the rate of creation=destruction are volumerelated
quantities. Evidently, more than one governing equation may apply for a given system. In fact,
for thermal problems where convective transport is signiﬁcant at least three governing equations
(energy, continuity, and momentum) are needed.
Turning to each term in the conservation equation, the diffusive ﬂux is expressed mathematic
ally using the socalled constitutive equation:
Diffusive flux ¼ ÀrÁ G(~x, t)rf(~x, t) (5:2)
where
G(~x, t) is a physical property of the system related to its ability to transfer a transport property
by a diffusive mechanism
f(~x, t) is a speciﬁc transport property, i.e., transport property=mass
The convective ﬂux arises due to the macroscopic motion of the system, which, therefore,
carries with it the transport property.
On the other hand, the interphase ﬂux is characterized by a property transfer coefﬁcient and the
value of a primitive property* in the bulk of the surrounding media:
Interphase flux ¼ l(x
i
, t) Â Æ c
B
(t) Àc
1
(t) ð Þ ½ (5:3)
* Examples of primitive properties are temperature and concentration.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 187 7.11.2008 5:44pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 187
where
l(x
i
, t) is an interphase transfer coefﬁcient along a particular boundary of the system
c is a primitive property
B and 1 are the subindices that stand for boundary and bulk, respectively
The sign in Equation 5.3 is selected depending on whether the interphase ﬂux is directed into the
element (the interphase ﬂux must be positive) or leaving the element (the interphase ﬂux must
be negative). The interphase ﬂux should not be confused with the boundary condition given below;
the former is included in the conservation equation only for directions other than those where
transport occurs by diffusive and convective mechanisms.
When considering an inﬁnitesimally small sampling volume ﬁxed in space* the governing
equation, after grouping the convection and rate of change terms in the right hand side (RHS), is
given by
rÁ G(~x, t)rf(~x, t) þS
f
(~x, t) þl(x
i
, t) Æ c
B
(t) Àc
1
(t) ð Þ ½
¼
@ r(~x, t)f(~x, t) ð Þ
@t
þrÁ r(~x, t)~v(~x, t)f(~x, t) ð Þ (5:4)
where
S
f
(~x, t) is the volumetric rate of creation=destruction (also known as the source term) of the
transport property
r(~x, t) is the density
~v(~x, t) is the velocity ﬁeld
The form of the governing equation given by Equation 5.4 is known as the conservative form,
which is the most useful one for computational transport phenomena [7]. The nonconservative form
is based on the balance of an element moving with the ﬂow. Although mathematically different,
both forms of the governing equation represent the same physical principle and either one of them
may be mathematically manipulated to obtain the other.
The dependent variable, f, may be either a scalar (speciﬁc thermal energy, mass=mole fraction)
or a vector (momentum per unit mass, i.e., velocity). In the latter case, the governing equation may
have up to three components, corresponding to each one of the directions of an orthogonal
coordinate system. Physically, the variables G, S
f
, l, and r are functions of one or more dependent
variables; however, since the dependent variables are themselves functions of spatial and time
coordinates, the variables listed above are represented as functions of the spatial and, possibly,
time coordinates.
The governing equation may be written for particular dependent variables of interest in
thermal processes as follows. The conservation of mass for a pure ﬂuid is given by the continuity
equation:
@(r(~x, t))
@t
þrÁ (r(~x, t)~v(~x, t)) ¼ 0 (5:5)
The conservation of momentum (in the x direction) for a Newtonian ﬂuid is described by
rÁ m(~x, t)ru(~x, t) þS
u
(~x, t) ¼
@(r(~x, t)u(~x, t))
@t
þrÁ (r(~x, t)~v(~x, t)u(~x, t)) (5:6)
* This is the socalled Eulerian approach, as opposed to the Lagrangian approach [6].
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 188 7.11.2008 5:44pm Compositor Name: BMani
188 Handbook of Thermal Process Modeling of Steels
where
m is the viscosity
u is the xcomponent of the velocity ﬁeld (i.e., the speciﬁc momentum in the xdirection)
S
u
includes all possible sources of momentum in the xdirection
Similar equations may be written for the other directions.
The conservation of internal energy,* without considering interphase ﬂow, may be expressed as
rÁ
k(~x, t)
^
C
p
(~x, t)
" #
re(~x, t) þS
e
(~x, t) ¼
@(r(~x, t)e(~x, t))
@t
þrÁ (r(~x, t)~v(~x, t)e(~x, t)) (5:7)
where
k is the thermal conductivity
^
C
p
is the speciﬁc heat capacity at constant pressure
e is the speciﬁc internal energy and all possible sources of internal energy are included in S
e
Since temperature rather than internal energy is of interest in thermal processes, Equation 5.7
may be written in terms of temperature:
y
rÁ k(~x, t)rT(~x, t) þS
e
(~x, t) ¼ r(~x, t)
^
C
p
(~x, t)
@T(~x, t)
@t
þr(~x, t)
^
C
p
(~x, t)rÁ (~v(~x, t)T(~x, t)) (5:8)
Note that the source term is still related to internal energy and that, although written in a form
suitable for computing the evolution of the thermal ﬁeld, each term in Equation 5.8 has units of
energy=(time Âvolume).
In thermochemical processes, such as carburizing, a governing equation for the conservation of
the individual chemical species is required in addition to the energy equation. The governing
equation, without considering interphase ﬂow, is
rÁ D
k
(~x, t)rg
k
(~x, t) þS
g
k
(~x, t) ¼
@(r(~x, t)g
k
(~x, t))
@t
þrÁ (r(~x, t)~v(~x, t)g
k
(~x, t)) (5:9)
where
D
k
is the diffusion coefﬁcient of species k
g
k
is the mass fraction of species k
S
g
k
includes the sources of species k within the system
5.3 BOUNDARY AND INITIAL CONDITIONS
To complete the mathematical formulation, boundary and initial conditions are required. Systems in
steady state require only boundary conditions; the corresponding mathematical formulation is
known as a boundaryvalue problem. On the other hand, if there are no gradients of the property
of interest in an unsteadystate system, one only needs to deﬁne an initial condition, which
originates an initialvalue problem. The more common situation is that of a timevarying ﬁeld
* Other forms of the energy equation may be written for: (a) total energy and (b) kinetic energy.
y
Recall that the speciﬁc internal energy is e ¼
^
C
p
T.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 189 7.11.2008 5:44pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 189
with signiﬁcant gradients, which requires the deﬁnition of boundary as well as initial conditions; this
formulation is known as a mixed boundary and initialvalue problem.
The initial condition represents the state of the system at the beginning of the simulation; in
most cases a uniform property ﬁeld is assumed.
On the other hand, the boundary conditions reﬂect known interactions between the system and
the surroundings in the direction(s) of transport by diffusion or convection. Along the domain
boundary surface several boundary conditions may be applicable. For thermal processing systems
typical boundary conditions may be classiﬁed as [8]
1. Prescribed thermal ﬁeld (Dirichlet)
T ¼ f
i
x
i
, t ð Þ on a particular section of the boundary surface (5:10a)
2. Prescribed normal derivative of temperature or heat ﬂux (Neumann)
@T
@^n
¼ f
i
x
i
, t ð Þ on a particular section of the boundary surface (5:10b)
3. Prescribed energy exchange with the surroundings (third kind)*
k
i
@T
@^n
þ
h
i
T ¼ f
i
(x
i
, t) on a particular section of the boundary surface (5:10c)
where
^n is the outwarddrawn normal to the boundary surface i
h
i
is a combined heat transfer and radiation coefﬁcient
If the RHS of any of Equation 5.10 is equal to zero, then the boundary condition is said to be
homogeneous.
5.4 NUMERICAL SOLUTION OF THE GOVERNING EQUATION
As mentioned before, the governing equation and the boundary conditions are highly nonlinear for
all problems of interest, which precludes the use of analytical solutions. Therefore, numerical
techniques (ﬁnite differences, ﬁnite element, boundary element, control volume, Monte Carlo,
and phaseﬁeld, among others) need to be implemented to obtain approximate solutions to the
governing equation(s). Through the numerical algorithm, the nonlinear problem is converted into a
set of linear problems—which may still involve iterative procedures and further approximations.
The resulting system of linear algebraic equations may be either explicit or implicit. The latter will
require the implementation of efﬁcient numerical techniques to solve banded matrices [9].
The particular form of the governing equation will lead to differences in the solution method. In
this regard, the partial differential equations of interest may be classiﬁed as (a) elliptic (associated
with pure boundaryvalue problems), (b) parabolic (associated with mixed initialboundaryvalue
problems), and (c) hyperbolic [10].
The transformation from partial differential to algebraic equations is achieved through a
discretization process that translates the continuous equations into their discrete equivalents.
The discretization is applied, in a different manner, to (1) the problem domain and (2) the governing
partial differential equation.
* Note that the boundary condition of the third kind is a linear combination of the other two.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 190 7.11.2008 5:44pm Compositor Name: BMani
190 Handbook of Thermal Process Modeling of Steels
The domain discretization involves partitioning a continuous system into a set of discrete
volumes. This operation is carried out by laying out a grid, known as a mesh, which consists of a
large number of points, known as nodes. For 2D problems, the discrete volumes become areas
whereas for a 1D domain they simplify to lines. Through domain discretization, the dependent
function becomes discrete, i.e., it only exists at the nodes in the computational domain.
Although the same general principle is applied to generate a mesh for any of the commonly used
numerical methods, each mesh has its own distinct characteristics, as it can be seen later. In addition,
regular grids, i.e., those based on Cartesian, cylindrical, or spherical coordinate systems, are not
well suited for irregular geometries. In that case, the use of nonorthogonal bodyﬁtted grids offers
many advantages although further manipulation during the discretization of the governing equation
is required [11].*
The discretization of the governing equation takes into account the fact that values of the
dependent variables can only be computed (or are known, in the case of boundary and initial
conditions) at the nodal positions of the mesh and, for unsteadystate problems, at discrete time
intervals. For mixed boundary and initialvalue problems the governing equation contains deriva
tives with respect to both space and time (see Equation 5.4) and is discretized, regardless of the
particular numerical method implemented, in the following order: space discretization is ﬁrst
applied, yielding a system of ﬁrstorder differential equations, which are then discretized in time.
Details of the procedures concerning each of the numerical methods reviewed are given in later
sections. A numerical method is robust only if it shows convergence and stability; however, due to
space limitations, these issues will not be discussed in this chapter.
The complex geometries modeled and the massive amount of data generated by a mathematical
model have prompted the development of powerful pre and postprocessing capabilities built into
commercial and inhouse packages to facilitate their use. Preprocessing allows the user to input the
desired geometry, build a mesh, and assign thermophysical properties and boundary conditions as
well as parameters related to the numerical method itself (relaxation factors, convergence criteria,
and the like). Postprocessing is related to the presentation of the results in forms (such as contour
plots) that permit the user to better identify the ﬂow patterns and the dynamic response of the
variables of interest (temperature, concentration, velocity, etc.).
Given the importance of numerical methods in process engineering, the objective of this chapter
is to lay down the fundamentals of several numerical methods applied in modeling thermal
processes: ﬁnite element, ﬁnite volume, Monte Carlo, and phaseﬁeld. It is expected that it will
provide a good starting point for those interested in writing their own computer codes as well as
those using commercial packages. The literature on these subjects is vast; in this chapter only the
salient features of each method will be presented.
Because it is extensively used in the various numerical methods presented in this chapter, the
approximation of ﬁrstorder derivates using the Taylor’s series expansion is now presented [12]. If a
continuous function f(x) has continuous derivatives in the interval [x
0
, x
0
þDx], then the function
may be represented by an inﬁnite power series known as a Taylor’s series:
f (x
0
þDx) ¼ f (x
0
) þDx
f
0
(x
0
)
1!
þ(Dx)
2
f
00
(x
0
)
2!
þÁ Á Á (5:11)
where f
0
(x
0
) denotes the ﬁrst derivative evaluated at the base point (x
0
), and so on. If the series is
ﬁnite, i.e., if the higher order derivatives are zero then the series approximates the function exactly.
On the other hand, if, for example, the secondorder derivative is nonzero but one chooses to
truncate the series up to the ﬁrst derivative term, then an error of the order of (Dx)
2
is generated
* The methodology is based on dealing with two distinct domains: physical and computational. In particular, mathematical
transformations between the curvilinear grid used to discretize the physical domain and an orthogonal grid for the
computational domain are required.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 191 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 191
into the modeling of the function and, as a consequence, in all derived quantities. Evidently, as
Dx ! 0 the modeling error diminishes.
Using the Taylor’s series expansion, several expressions may be derived to approximate the
ﬁrst derivative of a continuous function (such as temperature, pressure, mass fraction, and compon
ents of the velocity vector). The forward, backward, and central difference approximations are,
respectively,
@f x
i
, . . . ð Þ
@x
i
¼
f x
i
þDx
i
, . . . ð Þ Àf x
i
, . . . ð Þ
Dx
i
þO Dx
i
½ (5:12)
@f x
i
, . . . ð Þ
@x
i
¼
f x
i
, . . . ð Þ Àf x
i
ÀDx
i
, . . . ð Þ
Dx
i
þO Dx
i
½ (5:13)
@f x
i
, . . . ð Þ
@x
i
¼
f x
i
þDx
i
, . . . ð Þ Àf x
i
ÀDx
i
, . . . ð Þ
2Dx
i
þO Dx
i
ð Þ
2
h i
(5:14)
where
x
i
is a spatial or a time coordinate
O[Dx
i
] denotes ‘‘of the order of x
i
,’’ and so on
Therefore, the error associated with a central difference approximation is smaller than that of either
a forward or a backward difference approximation.
Another tool applied to develop numerical methods for solving governing equations for thermal
processing is the Gauss divergence theorem [13]:
ð
V
rk dV ¼
þ
S
^nk dS (5:15a)
ð
V
rÁ
~
A dV ¼
þ
S
^n Á
~
AdS (5:15b)
where
k is a scalar function
~
A is an arbitrary vector function
^n is an outwarddrawn normal unit vector
This theorem permits to reduce the volume integral on the left hand side (LHS) of Equations 5.15 to
the surface integral given on the RHS.
5.5 FINITEELEMENT METHOD
The ﬁniteelement method (FEM) has been extensively used in structural problems to calculate the
mechanical response of a structure for a given mechanical load [14–16]. In thermal processes such
as casting and heat treating, the loads are usually nonmechanical in nature but are related to changes
in the thermal and microstructural ﬁelds as well as changes in the thermomechanical properties [17].
A typical mesh for a 2D problem is shown in Figure 5.2. The domain is discretized in N ﬁnite
elements, each one containing a deﬁnite number of nodes. Note that the nodes are placed along the
boundaries of each element. In addition, there is no gap between elements and no elements overlap.
The basis of the FEM is the approximation, within any given element, of the continuous
function, f(~x, t), by a piecewise trial solution
~
f(~x, t,{a(t)}), the latter being deﬁned by a ﬁnite
sum of simple functions. Thus, the approximate solution at the element level is
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 192 7.11.2008 5:45pm Compositor Name: BMani
192 Handbook of Thermal Process Modeling of Steels
f
e
(~x, t) %
~
f
e
(~x, t,{a(t)}) (5:16)
with
~
f
e
(~x, t,{a(t)}) ¼ a
1
(t)N
(e)
1
(~x) þÁ Á Á þa
N
(t)N
(e)
N
(~x) ¼
X
N
i¼1
a
j
(t)N
(e)
j
(~x) (5:17)
where
N
(e)
j
(~x) are known functions called element shape functions (or trial functions)
a
j
(t) are the parameters (known as degrees of freedom) to be computed
Note that the shape functions are functions of position only while the degrees of freedom at each
node are functions of time only. The number of shape functions required for each element depends
on the degree of approximation desired. For example, in a 1D domain, two shape functions are
needed for a linear element whereas a quadratic element requires three shape functions. The number
of degrees of freedom at each node is problem dependent. If a thermal ﬁeld is simulated, then
only one degree of freedom (the temperature) is associated with each node. On the other hand, a
displacement ﬁeld may consist of up to three degrees of freedom at each node (e.g., x, y, and
zdisplacement). The total number of degrees of freedom for the entire system is the number of
degrees of freedom per node times the total number of nodes in the mesh.
In the FEM, the discretization of the governing equation that describes the system may be
accomplished through one of three approaches: (1) the principle of virtual work, (2) variational
methods, and (3) weightedresidual methods. In the following latter is focussed upon.
Among the weightedresidual methods, the Galerkin method is the preferred choice in setting up
the FEM. The basic steps are as follows [18]:
1. Use the governing equation to deﬁne a residual by transferring all terms to the LHS and
substituting the trial solution. In general, the residual is a nonzero quantity.
2. Minimize the residual by using a weighted average. For the Galerkin method the
weighting functions are the shape functions themselves.
3. Integrate by parts to reduce the order of the higher order integrand. This operation also
generates diffusive ﬂuxlike terms.
FIGURE 5.2 Portion of a 2D Cartesian ﬁniteelement mesh for quadratic shape functions.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 193 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 193
4. Substitute the general form of the trial solution into interior integrals in the residual
equations. This operation will originate the element equations.
5. Substitute speciﬁc expressions of the shape functions in the element equations and
integrate.* To simplify the integration, evaluate the thermophysical properties at speciﬁc
points within the element instead of allowing them to vary within the element.
6. Prepare expressions for the ﬂux terms using the shape functions.
7. Assemble the element equations into system equations.
8. Apply the boundary conditions to the system equations.
9. Numerically evaluate all the terms in the system equations.
10. Solve the system equations.
11. Evaluate the ﬂux, if needed.
The resulting system equations (after step 9 above) have the general form:
[C(t)]
da(t)
dt
& '
þ[K(t)]{a(t)} ¼ {F(t)} (5:18)
where
[C(t)] is the timevarying capacitance matrix
[K(t)] is the timevarying stiffness matrix
{F(t)} is the timevarying load vector
{a(t)} is the timevarying degrees of freedom vector, i.e., the solution sought
The methodology described above will result in a system of equations where both derivatives
and integrals of the shape functions need to be evaluated. Thus, the shape functions should be as
simple as possible. The most common form of the shape functions is that of a polynomial function.
For a 1D Cartesian domain, the shape functions may be deﬁned as
N
(e)
i
(x; {A}) ¼ A
1
þA
2
x þA
3
x
2
þÁ Á Á (5:19)
However, in order to take full advantage of the FEM a special case, that of an interpolation
polynomial, must be used. To rewrite Equation 5.19 as an interpolation polynomial, the shape
functions must take the value of 1 at a given node within the element and 0 at any other node within
that same element. This interpolation property is mathematically stated as
N
(e)
j
(x
i
) ¼ d
ji
(5:20)
where d
ij
is the Kronecker delta. As a consequence, the trial solution at the element level,
~
f
e
(~x, t),
takes the value
~
f
(e)
x
i
; {a(t)} ð Þ ¼ a
i
(t) (5:21)
at each node, i, within the element.
Using either Equations 5.20 or 5.21, the shape functions may be rewritten in the desired form.
For example, the linear shape functions for a 1D Cartesian domain would be
N
(e)
1
(x) ¼
x
2
Àx
x
2
Àx
1
(5:22a)
* The integration may be carried out analytically or numerically.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 194 7.11.2008 5:45pm Compositor Name: BMani
194 Handbook of Thermal Process Modeling of Steels
N
(e)
2
(x) ¼
x Àx
1
x
2
Àx
1
(5:22b)
where x
1
and x
2
are the coordinates of the local nodes 1 and 2. Note that each shape function is still a
linear polynomial itself. Another characteristic is that the element shape functions are local in the
sense that they are nonzero only within a particular element.
Following the same procedure, quadratic shape functions are deﬁned by quadratic polynomials:
N
(e)
1
(x) ¼
x Àx
2
ð Þ x Àx
3
ð Þ
x
1
Àx
2
ð Þ x
1
Àx
3
ð Þ
(5:23a)
N
(e)
2
(x) ¼
x Àx
1
ð Þ x Àx
3
ð Þ
x
2
Àx
1
ð Þ x
2
Àx
3
ð Þ
(5:23b)
N
(e)
3
(x) ¼
x Àx
1
ð Þ x Àx
2
ð Þ
x
3
Àx
1
ð Þ x
3
Àx
2
ð Þ
(5:23c)
Although ﬁnite elements using the types of shape functions described above are useful for analyzing
regular geometries, the need to simulate systems of irregular geometry has prompted the develop
ment of a very powerful type of shape functions: the isoparametric shape functions. They are based
on mapping a parent element onto a real element, the former with a regular geometric form (say a
triangle or a quadrilateral) while the latter has a distorted shape, well suited to represent irregular
boundaries.* A discussion of isoparametric elements is beyond the scope of this chapter.
The interested reader should consult Ref. [19].
Once all the element equations are built, the next step consists in assembling them to form the
system equations. To accomplish this task, the local trial solution is ﬁrst generalized so that it can be
applied to any element within a computer program by using the local notation:
~
f
e
(~x, t,{a(t)}) ¼ a
1
(t)N
(e)
1
(~x) þÁ Á Á þa
N
(t)N
(e)
N
(~x) (5:24)
where the subscripts refer to local node numbers. The advantage of this notation is that the shape
functions need to be developed only once and then applied in a repetitive manner for all elements.
However, in order to relate the local node numbers to the global node numbers, a connectivity table
must be built. The procedure is illustrated in Figure 5.3 and Table 5.1.
At the interelement boundaries the actual property is continuous. This fact can be used to
assemble the local trial solutions by enforcing their continuity at the interelement boundaries.
For two contiguous elements
~
f
(k)
boundary
¼
~
f
(l)
boundary
(5:25)
Element
number
Node
number
i j k l m
F E D C
FIGURE 5.3 Finiteelement mesh for a 1D problem using linear shape functions.
* It is evident that this type of ﬁnite element is directly related to the concept of bodyﬁtted grids presented before.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 195 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 195
which, taking into account the interpolation property of the shape functions, reduces to
a
i
¼ a
j
(5:26)
where global nodes i and j are located at the interelement boundary between elements k and l.
To assemble the element equations into the system equations within the framework of a
computer program. The following general rule is applied [20]: Add the element stiffness, K
(e)
ij
,
term to the ith row and jth column of the system stiffness matrix, and add the element load term,
F
(e)
i
, to the ith row of the system load vector, where the subscripts refer to global node numbers.
The assembly step will produce banded matrices, i.e., with most of the nonzero terms placed
around the main diagonal only if a proper node numbering scheme is adopted. This is a desirable
characteristic because it will dramatically lower computational costs. For 1D problems, the optimum
node numbering scheme is a sequential one starting from one end of the system and ﬁnishing at the
other; for 2D problems, a basic guideline is to number the nodes sequentially along paths that
contain the least number of nodes from one boundary to another [21]. These criteria are based on
matrix bandwidth minimization.
Once the system equations have been assembled, the next step is to impose the boundary
conditions. Because its application within the context of building the system equations depends on
its mathematical form, it is convenient to classify the boundary conditions applicable to second
order partial differential equations (which is the most common form of the governing equation in
thermal problems) as follows:
.
Essential boundary condition: the value of f at a boundary is prescribed
.
Natural boundary condition: the value of the ﬁrst spatial derivative of f is prescribed
Given that the system has been discretized into a number of ﬁnite elements, additional interior
(interelement) boundaries are automatically created. Thus, interior boundary conditions need to be
deﬁned and applied to the system equations. These interior boundary conditions are of the same
form of the system boundary conditions described here. It is immediately realized (see Equation
5.26) that the assembly step has already introduced the interior essential boundary conditions into
the system equations. The interior natural boundary conditions are applied by enforcing the
continuity of ﬂux across interelement boundaries:
ÀG(~x, t)
@
~
f
(k)
(~x, t;{a})
@x
1
boundary
¼ ÀG(~x)
@
~
f
(l)
(~x, t;{a})
@x
1
boundary
(5:27)
where x
1
is a direction normal to the interelement boundary between elements k and l. During the
assembly step this enforcement of continuity of the ﬂux at the interelement boundaries will
TABLE 5.1
Connectivity Table for the FiniteElement Mesh
Shown in Figure 5.3
Element
Global Node
Number Equivalent
to Local Node Number 1
Global Node
Number Equivalent
to Local Node Number 2
C i j
D j k
E k l
F l m
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 196 7.11.2008 5:45pm Compositor Name: BMani
196 Handbook of Thermal Process Modeling of Steels
combine two identical numbers but with opposite signs, resulting in a value of zero for all the
interior ﬂux components of the load vector that result from the integration by parts of the residual.*
There are two methods to apply essential boundary conditions to the system equations [22].
Both introduce the value of the known degrees of freedom at each time step while preserving the
characteristics, such as symmetry, of the system equations. The ﬁrst method consists in substituting
the value of each known degree of freedom in all the equations where it appears. If a
i
is prescribed,
then use the following algorithm:
1. Substitute the value of a
i
for F
i
in the ith row
2. Set K
ij
¼0 for j 6¼ i in the ith row
3. In all other rows where a
i
occurs transfer the term K
ki
a
1
to the RHS of that equation
Considering again that a
i
is prescribed, the second method is based on
1. Add a very large number, say b, to the K
ii
term
2. Substitute the F
i
in the ith row by ba
i
As mentioned before, the boundary terms of the load vector resulting from the integration by parts
have the form of a diffusive ﬂux. Consequently, natural boundary conditions are applied by directly
substituting their value into the corresponding boundary terms of the load vector.
The system equations are then solved to compute the evolution of the ﬁeld of interest. Note that
Equation 5.18 represents a system of coupled ﬁrstorder ordinary differential equations, which itself
requires a numerical method. The solution of the system equations can be accomplished through
timestepping methods by using either a ﬁnitedifference scheme or a ﬁniteelement formulation to
integrate Equation 5.18 over time. These methods will transform the system of ordinary differential
equations into a series of systems of linear algebraic equations that are solved in a timemarching
fashion. The algebraic equations are also known as recurrence relations and are linear in nature.
Although they can be constructed using information from several time steps (multistep methods), it
is more common to use information from a single time step (onestep methods).
The most popular onestep method is the Crank–Nicolson method (a ﬁnitedifference scheme).
This method discretizes Equation 5.18 by approximating the time derivative using a ﬁnite central
difference (i.e., assuming a linear behavior for {a(t)} during any given time step
y
) [23]:
da(t)
dt
& '
nÀ1=2
ﬃ
{a}
n
À{a}
nÀ1
Dt
n
(5:28)
where the subindices indicate a particular time.
Evaluating Equation 5.18 at the center of the time step gives
[C]
nÀ1=2
da(t)
dt
& '
nÀ1=2
þ[K]
nÀ1=2
{a}
nÀ1=2
¼ {F}
nÀ1=2
(5:29)
From the assumption of linearity of {a(t)} it follows that
{a}
nÀ1=2
%
{a}
nÀ1
þ{a}
n
2
(5:30)
* Each element of the load vector may also have contributions from loads applied inside the system.
y
The time step may be either constant or variable with time.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 197 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 197
Thus, Equation 5.29 can be rewritten as
1
Dt
n
[C]
nÀ1=2
þ
1
2
[K]
nÀ1=2
{a}
n
¼ {F}
nÀ1=2
þ
1
Dt
n
[C]
nÀ1=2
À
1
2
[K]
nÀ1=2
{a}
nÀ1
(5:31)
where
[C]
nÀ1=2
ﬃ
[C]
nÀ1
þ[C]
n
2
(5:32)
[K]
nÀ1=2
ﬃ
[K]
nÀ1
þ[K]
n
2
(5:33)
{F}
nÀ1=2
ﬃ
{F}
nÀ1
þ{F}
n
2
(5:34)
Equation 5.31 is the recurrence relation to be applied at each time step to compute {a}
n
; it is a
system of linear algebraic equations. The system is implicit, given that the matrix on the LHS is
nondiagonal, but can be solved by any standard method such as Gaussian elimination [9].
As an example of the application of the FEM, let us consider an experiment such as the Jominy
endquench test for an aluminum alloy. The physical situation is depicted in Figure 5.4. One
dimensional transient heat ﬂow, in the zdirection, is assumed but interphase ﬂow is included in the
analysis. No heat sources are included.
Applying the temperature form of the energy equation, the mathematical formulation is
@
@z
k(z, t)
@T(z, t)
@z
þh
2
(z, t)[T(z, t) ÀT
f
]
P
A
?,1
¼ r
^
C
p
(z, t)
@(T(z, t))
@t
0 z Z, t > 0 (5:35)
where
A
?,1
is the area normal to heat ﬂow in the rdirection (interface ﬂow)
^
C
p
is the speciﬁc heat capacity at constant pressure
h
2
is a combined heat transfer coefﬁcient for the lateral surface in contact with still air
k is the thermal conductivity
2R
Z
z
r
Heat loss to the
water column
Heat loss to the
sorrounding air
FIGURE 5.4 Schematic representation of the endquench test.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 198 7.11.2008 5:45pm Compositor Name: BMani
198 Handbook of Thermal Process Modeling of Steels
P is the perimeter of the lateral surface
t is the time
T is the thermal ﬁeld
T
f
is the ﬂuid bulk temperature
z is the position along the longitudinal axis
r is the density
Note that r and
^
C
p
are grouped as the product r
^
C
p
.
The initial and boundary conditions are
IC T(z, 0) ¼ T
0
0 z Z (5:36)
BC1 Àk(z, t)
@T(z, t)
@z
¼ q
1
(t) z ¼ 0, t > 0 (5:37)
BC2 Àk(z, t)
@T(z, t)
@z
¼ 0 z ¼ Z, t > 0 (5:38)
where
T
0
is the initial temperature
q
1
(t) is the heat ﬂux extracted by the water column
The ﬁniteelement equations at the element level are derived as follows. First, the evolution of
the thermal ﬁeld at the element level is approximated by a trial solution:
T
(e)
(z, t) ﬃ
~
T
(e)
(z, t;{a}) ¼
X
N
j¼1
a
j
(t)N
(e)
j
(z) (5:39)
The residual of the governing equation is constructed by transferring all terms in Equation 5.35 to
the RHS and substituting the approximate solution:
R(z, t; a) ¼
@ rC
p
(z, t)
~
T
(e)
(z, t; {a})
À Á
@t
À
@
@z
k(z, t)
@
~
T
(e)
(z, t; {a})
@z
þh
2
(z, t)
~
T
(e)
(z, t; {a}) ÀT
f
Â Ã
P
A
?,1
(5:40)
Applying the Galerkin method one obtains
ð
V
e
rC
p
(z, t)
@
~
T
(e)
(z, t; {a})
@t
N
(e)
i
(z)dV
À
ð
V
e
@
@z
k(z, t)
@
~
T
(e)
(z, t; {a})
@z
!
N
(e)
i
(z)dV
þ
ð
V
e
h
2
(z, t)
~
T
(e)
(z, t; {a})*
P
A
?
N
(e)
i
(z)dV
À
ð
V
e
h
2
(z, t)T
f
*
P
A
?
N
(e)
i
(z)dV ¼ 0 i ¼ 1, 2, . . . , n (5:41)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 199 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 199
Since dV¼r du dr dz
pR
2
ð
z
2
z
1
rC
p
(z, t)
@
~
T
(e)
(z, t; {a})
@t
N
i
(z)dz
ÀpR
2
ð
z
2
z
@
@z
k(z, t)
@
~
T
(e)
(z, t; {a})
@z
!
N
i
(z)dz
þpR
2
ð
z
2
z
1
h
2
(z, t)
~
T
(e)
(z, t; {a})
P
A
?
N
i
(z)dz
ÀpR
2
ð
z
2
z
1
h
2
(z, t)T
f
P
A
?
N
i
(z)dz ¼ 0 i ¼ 1, 2, . . . , n (5:42)
Integrating the higher order term by parts and rearranging,
pR
2
ð
z
2
z
1
rC
p
(z, t)
@
~
T
(e)
(z, t; {a})
@t
N
(e)
i
(z)dz þpR
2
ð
z
2
z
1
@N
(e)
i
(z)
@z
k(z, t)
@
~
T
(e)
(z, t; {a})
@z
dz
þpR
2
ð
z
2
z
1
h
2
(z, t)
~
T
(e)
(z, t; {a})
P
A
?
N
(e)
i
(z)dz ¼ pR
2
ð
z
2
z
1
h
2
(z, t)T
f
P
A
?
N
(e)
i
(z)dz
ÀpR
2
~q
(e)
k,z
(z, t; {a})N
(e)
i
(z)
h i
z
2
z
1
i ¼ 1, 2, . . . , n (5:43)
where the heat ﬂux* is deﬁned as
~q
(e)
k,z
(z, t; {a}) ¼ Àk(z, t)
@
~
T
(e)
(z, t; {a})
@z
(5:44)
Substituting the general form of the trial solution into the interior terms of the residual equations,
pR
2
X
n
j¼1
ð
z
2
z
1
N
(e)
i
(z)rC
p
(z, t)N
(e)
j
(z)dz
0
@
1
A
da
j
(t)
dt
þpR
2
X
n
j¼1
ð
z
2
z
1
dN
(e)
i
(z)
dz
k(z, t)
dN
(e)
j
(z)
dz
dz
0
@
1
A
a
j
(t)
þpR
2
X
n
j¼1
ð
z
2
z
1
N
(e)
i
(z)
h
2
(z, t)
~
T
(e)
(z, t)
P
A
?
N
(e)
j
(z)dz
0
@
1
A
¼ pR
2
ð
z
2
z
1
h
2
(z, t)T
f
P
A
?
N
(e)
i
(z)dz
ÀpR
2
~q
(e)
k,z
(z, t; {a})N
(e)
i
(z)
h i
z
2
z
1
i ¼ 1, 2, . . . , N (5:45)
* This term results from the integration by parts.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 200 7.11.2008 5:45pm Compositor Name: BMani
200 Handbook of Thermal Process Modeling of Steels
which may be represented in matrix form as
[C]
(e)
da(t)
dt
& '
þ[K]
(e)
{a(t)} ¼ {F(t)}
(e)
(5:46)
where
C
(e)
ij
¼ pR
2
ð
z
2
z
1
N
(e)
i
(z)rC
p
(z, t)N
(e)
j
(z)dz (5:47a)
K
(e)
ij
¼ pR
2
ð
z
2
z
1
dN
(e)
i
(z)
dz
k(z, t)
dN
(e)
j
(z)
dz
dz þ
ð
z
2
z
1
N
(e)
i
(z)
h
2
(z, t)
P
A
?
N
(e)
j
(z)dz
2
4
3
5
(5:47b)
F
(e)
i
(t) ¼ pR
2
ð
z
2
z
1
N
(e)
i
(z)
h
2
(z, t)T
f
P
A
?
dz
2
4
3
5
ÀpR
2
À N
(e)
i
(z)~q
(e)
k,z
(z, t; {a})
h i
z
2
z
1
& '
(5:47c)
are the capacitance matrix, the stiffness (or conductance) matrix, and the load vector, respectively. Note
that, after integration by parts, the conductance matrix is symmetrical and that heat ﬂuxlike terms have
been generated. The shape functions correspond to a 1D problem and may be taken from any standard
reference and inserted in the integrals, as indicated by the equations above. After performing the
integrals, either analytically or numerically, the assembly step is carried out, which results in all the
interior terms of the load vector vanishing. For this problem, the domain boundary conditions are all
natural and, therefore, applied by direct substitution in the corresponding terms of the load vector. The
resulting systemequations can then be solved by applying the Crank–Nicolson method described above.
If a phase transformation occurs (as in the Jominy endquench test for steels), the governing
equation would include a heat generation term. In the ﬁniteelement formulation, this term results in
an additional integral. The only term that would be modiﬁed in the system equations is that of the
load vector, which would then have the additional term
F
(e)
i,gen
(t) ¼ pR
2
ð
z
2
z
1
N
(e)
i
(z)r(z, t)DH
@f (z, t)
@t
dz (5:48)
where
f is the fraction transformed of the new phase
DH is the speciﬁc enthalpy of transformation
From Equation 5.48 it is evident that the thermal ﬁeld depends on the evolution of the fraction
transformed which, in turn, is a function of the thermal ﬁeld evolution. Therefore, the following
iterative algorithm is required during the time the system is within the transformation region:
1. Compute a guessed thermal ﬁeld with the source term set to zero, T*.
2. Compute a guessed fraction transformed, f*, and the corresponding term for the load vector.
3. Recompute the guessed thermal ﬁeld, T**.
4. Recompute the guessed fraction transformed, f**, and the corresponding term of the load
vector.
5. Recompute the guessed thermal ﬁeld, T***.
6. If the absolute value of the difference between T** and T*** is less than a preset
convergence criterion, then the solution has converged. Move to the next time step.
7. Otherwise set T** T*** and go back to step 4.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 201 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 201
To reduce the number of iterations, an underrelaxation technique needs to be applied to the guessed
thermal ﬁeld before computing the evolution of the fraction transformed in all the corresponding
steps in the previous algorithm. Also, for small time steps the temperature at a given node would not
change much during the iteration procedure and, therefore, the thermophysical properties and, more
importantly, the conductance matrix need to be evaluated only once for each time interval.
5.6 FINITEVOLUME METHOD
The ﬁnitevolume method (FVM) is currently the numerical method of choice to solve heat or mass
transfer problems where ﬂuid dynamics plays a central role. If the velocity ﬁeld is unknown, the
domain discretization consists of a combination of two types of control volumes: regular control
volumes* and control volumes for the velocity vector components [24,25]. Figure 5.5 shows a
portion of such a mesh for a 2D Cartesian domain. It is built by placing nodes throughout the
domain; note that the nodes for the velocity components are staggered with respect to the scalar
nodes. In particular, they are backward staggered, i.e., form two grids shifted to the left
(ucomponent of the velocity ﬁeld) or the bottom (vcomponent of the velocity ﬁeld) of the scalar
grid. The control volumes, for each variable, are then generated by drawing boundaries midway
between adjacent nodes. The domain boundaries are also controlvolume boundaries.
The basic steps required to discretize the governing equations for the case where the velocity
ﬁeld is known are
1. Integrate, symbolically, the governing equation over a typical interior control volume
2. Approximate the derivatives associated with the diffusive and convective ﬂuxes at the
scalar controlvolume faces
3. Approximate the value of the physical properties at the scalar controlvolume faces
4. Linearize the source terms
5. Rearrange the resulting equation after steps 1–4 to obtain the general discretized equation
6. Modify the general discretized equation for control volumes adjacent to the domain
boundaries to incorporate boundary conditions
7. Solve the resulting implicit system of linear algebraic equations to compute the desired
ﬁelds
8. Evaluate the ﬂuxes, if necessary
(a) (b) (c)
FIGURE 5.5 Three components of a typical FVM mesh for a 2D Cartesian domain where the velocity ﬁeld is
not known a priori (a) regular (scalar); (b) ucomponent of the velocity ﬁeld; and (c) vcomponent of the
velocity ﬁeld. Solid circles: scalar nodes; open circles: ucomponent of the velocity ﬁeld nodes; and open
squares: vcomponent of the velocity ﬁeld nodes.
* Also known as scalar or pressure control volumes.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 202 7.11.2008 5:45pm Compositor Name: BMani
202 Handbook of Thermal Process Modeling of Steels
The general goal of the procedure is to transform the governing equations from their differential
form into a system of linear algebraic equations. For steadystate problems, they take the form
a
P
f
P
¼
X
a
i
f
i
þS
0
(5:49)
where the summation is performed over all the neighboring nodes.
In the following, this algorithm is developed in detail to obtain the discretized equations
required to compute the coupled velocity and pressure ﬁelds for an incompressible Newtonian
ﬂuid. Because the ﬁelds are coupled and there is no conservation equation for the pressure ﬁeld, a
few steps will be added to the previous algorithm. The system will be assumed to be under steady
or quasisteadystate and a 2D Cartesian domain where momentum is transferred by convective and
diffusive mechanisms but no source terms other than pressure gradients will be considered.
Moreover, laminar ﬂow will be assumed.* The derivation largely follows Ref. [24].
For a 2D ﬂow, two components of the momentum equation need to be considered, along with
the continuity equation. Thus, for the system described above
xComponent of the momentum equation:
@
@x
G(~x)
@u
@x
þ
@
@y
G(~x)
@u
@y
À
@p
@x
¼
@
@x
(r(~x)uu) þ
@
@y
(r(~x)vu) (5:50)
yComponent of the momentum equation:
@
@x
G(~x)
@v
@x
þ
@
@y
G(~x)
@v
@y
À
@p
@y
¼
@
@x
(r(~x)uv) þ
@
@y
(r(~x)vv) (5:51)
Continuity equation:
@
@x
(r(~x)u) þ
@
@y
(r(~x)v) ¼ 0 (5:52)
where u ¼ u(~x), v ¼ v(~x), and a Newtonian ﬂuid has been assumed.
y
These are the three governing equations needed to solve for the velocity and pressure ﬁelds. As
shall be seen later, the ﬁrst two will be used directly while the latter will be reformulated in terms of
the pressure ﬁeld.
The discretization of the governing partial differential equations in the FVM is based on the
integration of the governing equations over each control volume. However, the momentum equa
tions will be integrated over backward staggered control volumes while a scalar control volume will
be used to integrate the continuity equation. The former operation will result in two systems (one for
each component of the velocity ﬁeld) of implicit algebraic equations while the latter will produce a
system of implicit algebraic equations for the pressure ﬁeld.
The staggered control volume used to integrate the xcomponent of the momentum conservation
equation for any interior node is shown in Figure 5.6. In Figure 5.6a, uppercase letters denote
locations where the ucomponent of the velocity ﬁeld will be computed while lowercase letters mark
the boundaries of the staggered control volume; this notation will be useful for setting up the integral
of the corresponding governing equation. For computational purposes, the matrix notation shown in
Figure 5.6b is adopted: two uppercase letters are associated with a scalar location while combin
ations of upper and lowercase letters will denote a velocity component location. Note that the
indices for the velocity component location are backstaggered.
* For turbulent ﬂow, the modiﬁed momentum conservation equations and the auxiliary equations to model turbulence may be
treated in a similar fashion.
y
The ﬂuid viscosity is represented by G (instead of m) in order to keep the formulation as general as possible.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 203 7.11.2008 5:45pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 203
In Equation 5.50, the following terms may be recognized: the ﬁrst two terms on the LHS
represent diffusive transport of the xcomponent of the momentum ﬁeld while the third term, the
pressure gradient, is the only source of momentum considered for this case. The RHS of the
equation contains the convective contribution to momentum ﬂow.
The diffusive terms are integrated as follows:
Int
diff
¼
ð
CV
@
@x
G(~x)
@u
@x
þ
@
@y
G(~x)
@u
@y
!
dV (5:53)
or splitting the integral and applying the divergence theorem,
Int
diff
¼
ð
A
x
^n Á
@
@x
G(~x)
@u
@x
dA
x
þ
ð
A
y
^n
@
@y
G(~x)
@u
@y
dA
y
(5:54)
where
^n is the outwarddrawn normal unit vector to the controlvolume faces
A
x
and A
y
are controlvolume areas normal to the u and vcomponents of the ﬂow ﬁeld,
respectively, i.e., they correspond to the w and e, and the n and s boundaries of the control
volume of Figure 5.6a
Performing the integrals indicated in Equation 5.54, the following result is obtained:*
Int
diff
¼ G(~x)
@u
@x
A
x
e
À G(~x)
@u
@x
A
x
w
!
þ G(~x)
@u
@y
A
y
n
À G(~x)
@u
@y
A
y
s
!
(5:55)
This result is now discretized using a central differencing scheme. Referring to Figure 5.6a, the
general form of the central difference formula at, for example, node P in an equally spaced grid
(Dx ¼dx
WP
¼dx
PE
) is
y
(a)
(b)
u
v
W
w
S
s
P
n
e
E
N
v
u
i–1,J–1 I–1,J–1 i,J–1 I,J–1
I–1,j
I–1,J
i,J
i–1,J l–2,J
i–1,J+1 I–1,J+1
l–1,j+1
I,J+1
l,j+1
i+1,J+1
i+1,J
i+1,J–1
I+1,J I,J
I,j
i,J+1
FIGURE 5.6 An interior ucomponent backward staggered control volume for a 2D Cartesian domain:
(a) standard notation; (b) matrix notation. Solid circles: scalar nodes; open circles: ucomponent of the velocity
ﬁeld nodes; and open squares: vcomponent of the velocity ﬁeld nodes.
* Note that this result preserves the form of net diffusive ﬂuxes (now multiplied by an area) present in the governing equation.
y
In the remainder of this section, the notation dx
KL
or dx
kl
denotes the distance between two points, either grid nodes or
control volume boundaries, respectively.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 204 7.11.2008 5:45pm Compositor Name: BMani
204 Handbook of Thermal Process Modeling of Steels
@f(~x, t)
@x
P
¼
f(~x, t)j
E
Àf(~x, t)j
W
2Dx
þO Dx
2
À Á
(5:56)
that is, it has secondorder accuracy. Applying the central difference formula to the spatial deri
vatives in Equation 5.55 (noting that dx
Pe
¼dx
PE
=2, etc.) gives
Int
diff
¼ G
e
A
e
u
E
Àu
P
dx
PE
ÀG
w
A
w
u
P
Àu
W
dx
WP
!
þ G
n
A
n
u
N
Àu
P
dy
PN
ÀG
s
A
s
u
P
Àu
S
dy
SP
!
(5:57)
or
Int
diff
¼ D
e
A
e
u
E
Àu
P
ð Þ ÀD
w
A
w
u
P
Àu
W
ð Þ ½ þ D
n
A
n
u
N
Àu
P
ð Þ ÀD
s
A
s
u
P
Àu
S
ð Þ ½ (5:58)
where the diffusive conductance per unit area at the cell face i is D
i
¼G
i
=dx
i
. Comparing Equation
5.58 with the general discretized equation (Equation 5.49) it is apparent that
a
W,diff
¼ D
w
A
w
(5:59a)
a
E,diff
¼ D
e
A
e
(5:59b)
a
S,diff
¼ D
s
A
s
(5:59c)
a
N,diff
¼ D
n
A
n
(5:59d)
a
P,diff
¼ a
W,diff
þa
E,diff
þa
S,diff
þa
N,diff
(5:59e)
Note that values of G
i
are required at the controlvolume boundaries but they are known only at the
scalar nodal points. For this calculation, only G
w
and G
e
are known; the other G
i
’s are evaluated
applying a linear interpolation. Taking A
u
¼Dy Â1 and A
v
¼Dx Â1 for areas normal to the u and
vcomponents of the velocity ﬁeld, respectively, and using the notation of Figure 5.6b, the diffusive
contribution to the general discretized equation becomes*
a
iÀ1,J,diff
¼ D
IÀ1,J
Dy (5:60a)
a
iþ1,J,diff
¼ D
I,J
Dy (5:60b)
a
i,JÀ1,diff
¼
G
IÀ1,J
þG
I,J
þG
IÀ1,JÀ1
þG
I,JÀ1
ð Þ=4
y
J
Ày
JÀ1
ð Þ
Dx (5:60c)
a
i,Jþ1,diff
¼
G
IÀ1,Jþ1
þG
I,Jþ1
þG
IÀ1,J
þG
I,J
ð Þ=4
y
Jþ1
Ày
J
ð Þ
Dx (5:60d)
a
i,J,diff
¼ a
iÀ1,J,diff
þa
iþ1,J,diff
þa
i,JÀ1,diff
þa
i,Jþ1,diff
(5:60e)
The convective contribution presents a signiﬁcant difference with respect to the diffusive contribu
tion: the former has a preferential transport direction (the direction of ﬂow) whereas the latter
transports the property of interest in all directions (for isotropic materials). Therefore, the central
difference scheme cannot be used to discretize the convective terms in the integrated form of the
governing equation.
To solve this problem various schemes have been developed: upwind, hybrid, powerlaw, and
quick. Here the hybrid scheme is present, which is a combination of the central difference and the
upwind schemes [26]. First, the convective terms in the governing equation are integrated over
the control volume:
* For uniform (although not necessarily identical) grids in the x and ydirection.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 205 7.11.2008 5:46pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 205
Int
conv
¼
ð
CV
@
@x
r(~x)uu ð Þ þ
@
@y
r(~x)vu ð Þ
!
dV (5:61)
or, splitting the integral and applying the divergence theorem,
Int
conv
¼
ð
A
x
^n Á
@
@y
r(~x)uu ð ÞdA
x
þ
ð
A
y
^n
@
@y
r(~x)vu ð ÞdA
y
(5:62)
which yields (refer to Figure 5.6a):
Int
conv
¼ r(~x)uuA
x
ð Þ
e
À r(~x)uuA
x
ð Þ
w
Â Ã
þ r(~x)vuA
y
À Á
n
À r(ð~x)vuA
y
À Á
s
Â Ã
(5:63)
To linearize this equation the product of velocities is split into two terms: one will be assumed as
known (taken from the previous iteration) while the other will be computed. The result is
Int
conv
¼ r
e
u
e
A
e
u
e
ð Þ Àr
w
u
w
A
w
u
e
ð Þ ½ þ r
n
v
n
A
n
u
n
ð Þ Àr
s
v
s
A
s
u
s
ð Þ ½ (5:64)
or
Int
conv
¼ F
e
A
e
u
e
ð Þ ÀF
w
A
w
u
w
ð Þ ½ þ F
n
A
n
u
n
ð Þ ÀF
s
A
s
u
s
ð Þ ½ (5:65)
where the convective mass ﬂux per unit area at the cell face i is F
i
¼r
i
u
i
(or F
i
¼r
i
v
i
).
In the central difference scheme, a linear interpolation is applied within each control volume to
represent the ﬁeld of interest at the cell faces. Thus,*
u
w
¼
u
W
þu
P
2
(5:66a)
u
e
¼
u
P
þu
E
2
(5:66b)
u
s
¼
u
S
þu
P
2
(5:66c)
u
n
¼
u
P
þu
N
2
(5:66d)
Comparing Equation 5.65 (after introducing Equation 5.66) with the general discretized equation
(Equation 5.49),
a
W,conv
¼
F
w
2
A
w
(5:67a)
a
E,conv
¼ À
F
e
2
A
e
(5:67b)
a
S,conv
¼
F
s
2
A
s
(5:67c)
a
N,conv
¼ À
F
n
2
A
n
(5:67d)
* For uniform (although not necessarily identical) grids in the x and ydirection.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 206 7.11.2008 5:46pm Compositor Name: BMani
206 Handbook of Thermal Process Modeling of Steels
a
P,conv
¼ a
W,conv
þa
E,conv
þa
S,conv
þa
N,conv
þ F
e
ÀF
w
ð Þ þ F
n
ÀF
s
ð Þ (5:67e)
The upwind scheme takes into account the ﬂow direction. Thus, the value of the property of interest
is no longer computed assuming a linear variation within the control volume; instead, it takes the
value of the property at the immediate upstream node. For a ﬂow from east to west and from south to
north (Figure 5.6a),
u
w
¼ u
W
(5:68a)
u
e
¼ u
P
(5:68b)
u
s
¼ u
S
(5:68c)
u
n
¼ u
P
(5:68d)
Similar expressions may be derived when the directions of ﬂow are reversed. A general form is then
written as
a
W,conv
¼ max F
w
A
w
, 0 ð Þ (5:69a)
a
E,conv
¼ max 0, ÀF
e
A
e
ð Þ (5:69b)
a
S,conv
¼ max F
s
A
s
, 0 ð Þ (5:69c)
a
N,conv
¼ max 0, ÀF
n
A
n
ð Þ (5:69d)
a
P,conv
¼ a
W,conv
þa
E,conv
þa
S,conv
þa
N,conv
þ F
e
A
e
ÀF
w
A
w
ð Þ þ F
n
A
n
ÀF
s
A
s
ð Þ (5:69e)
The hybrid scheme combines both the central differencing and the upwind schemes. The local
Peclet number (which is a measure of the relative importance of the convective and diffusive
mechanisms) at a given cell face is deﬁned by
Pe
i
¼
F
i
D
i
(5:70)
If the local Peclet number is small then both mechanisms are important, whereas the convective
mechanism dominates when the local Peclet number is large. The hybrid scheme uses central
differencing for both the diffusive and convective components when jPe
i
j <2 and the upwind
scheme (canceling also the diffusive terms) when jPe
i
j ! 2. Mathematically
a
W
¼ max F
w
A
w
, a
iÀ1,J,diff
þ
F
w
2
A
w
, 0
!
(5:71a)
a
E
¼ max ÀF
e
A
e
, a
iþ1,J,diff
À
F
e
2
A
e
, 0
!
(5:71b)
a
S
¼ max F
s
A
s
, a
i,JÀ1,diff
þ
F
s
2
A
s
, 0
!
(5:71c)
a
N
¼ max ÀF
n
A
n
, a
i,Jþ1,diff
À
F
n
2
A
n
, 0
!
(5:71d)
a
P
¼ a
W
þa
E
þa
S
þa
N
þ F
e
ÀF
w
ð Þ þ F
n
ÀF
s
ð Þ (5:71e)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 207 7.11.2008 5:46pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 207
Now let us examine how to compute the F
i
terms. Referring to Figure 5.6a, the ucomponent of
the velocity ﬁeld is not available at the cell faces nor is the ﬂow per unit area. Also, values of r
i
are
required at locations different from the scalar nodal points. Therefore, averages using the neighbor
ing nodes are adopted for all of these quantities. Using the notation of Figure 5.6b
F
w
¼
F
i,J
þF
iÀ1,J
ð Þ
2
¼
r
i,J
u
i,J
þr
iÀ1,J
u
iÀ1,J
À Á
2
(5:72a)
F
e
¼
F
iþ1,J
þF
i,J
ð Þ
2
¼
r
iþ1,J
u
iþ1,J
þr
i,J
u
i,J
À Á
2
(5:72b)
F
s
¼
F
I,j
þF
IÀ1,j
À Á
2
¼
r
I,j
v
I,j
þr
IÀ1,j
v
IÀ1,j
À Á
2
(5:72c)
F
n
¼
F
I,jþ1
þF
IÀ1,jþ1
À Á
2
¼
r
I,jþ1
v
I,jþ1
þr
IÀ1,jþ1
v
IÀ1,jþ1
À Á
2
(5:72d)
with
r
i,J
¼
r
I,J
þr
IÀ1,J
À Á
2
(5:73a)
r
iÀ1,J
¼
r
IÀ1,J
þr
IÀ2,J
À Á
2
(5:73b)
r
iþ1,J
¼
r
Iþ1,J
þr
I,J
À Á
2
(5:73c)
r
IÀ1,j
¼
r
IÀ1,J
þr
IÀ1,JÀ1
À Á
2
(5:73d)
r
I,jþ1
¼
r
I,Jþ1
þr
I,J
À Á
2
(5:73e)
r
IÀ1,jþ1
¼
r
IÀ1,Jþ1
þr
IÀ1,J
À Á
2
(5:73f)
The momentum generated due to pressure gradients results in another integral:*
Int
p
¼
ð
CV
À
@p
@x
!
dV (5:74)
The pressure gradient is approximated assuming a linear behavior within the control volume:
@p
@x
%
p
e
Àp
w
ð Þ
Dx
(5:75)
Therefore,
Int
p
% À
p
e
Àp
w
ð Þ
Dx
DV
u
¼ À p
e
Àp
w
ð ÞA
u
(5:76)
* Other sources of momentum could be considered and would be associated with the source term.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 208 7.11.2008 5:46pm Compositor Name: BMani
208 Handbook of Thermal Process Modeling of Steels
where DV
u
( ¼A
u
Dx) is the volume of the staggered control volume for the ucomponent of the
velocity ﬁeld. In terms of the matrix notation,
Int
p
% À p
I,J
Àp
IÀ1,J
ð ÞA
i,J
¼ À p
I,J
Àp
IÀ1,J
ð ÞDy (5:77)
Collecting terms from all integrals, the discretized equation for the xcomponent of the momentum
equation may be written as
a
i,J
u
i,J
¼ a
iÀ1,J
u
iÀ1,J
þa
iþ1,J
u
iþ1,J
þa
i,JÀ1
u
i,JÀ1
þa
i,Jþ1
u
i,Jþ1
þ p
IÀ1,J
Àp
I,J
ð ÞDy (5:78)
The corresponding equation for the ycomponent of the momentum equation is developed based on
the backward staggered control volume shown in Figure 5.7 and is given by
a
I, j
v
I, j
¼ a
IÀ1, j
v
IÀ1, j
þa
Iþ1, j
v
Iþ1, j
þa
I, jÀ1
v
I, jÀ1
þa
I, jþ1
v
I,Jþ1
þ p
I,JÀ1
Àp
I,J
ð ÞDx (5:79)
Since the pressure ﬁeld is usually unknown, Equations 5.78 and 5.79 cannot be solved in their
present form. The strategy to approximate them is to use the continuity equation to generate an
equation for a corrected pressure ﬁeld and use it to solve the velocity ﬁelds iteratively.
In the semiimplicit method for pressurelinked equations (SIMPLE) algorithm [27], a guessed
pressure ﬁeld, p*, is applied to compute an initial estimate of the u and vcomponents of the
velocity ﬁeld: u* and v*, respectively. Then, Equations 5.78 and 5.79 are rewritten as
a
i, J
u
i, J
*
¼ a
iÀ1, J
u
iÀ1, J
*
þa
iþ1, J
u
iþ1, J
*
þa
i, Jþ1
u
i, Jþ1
*
þa
i, JÀ1
u
i,JÀ1
*
þ p
IÀ1, J
*
Àp
I, J
*
ð ÞDy (5:80)
and
a
I, j
v
I, j
*
¼ a
IÀ1, j
v
IÀ1, j
*
þa
Iþ1, j
v
Iþ1, j
*
þa
I, jÀ1
v
I, jÀ1
*
þa
I, jþ1
v
I, Jþ1
*
þ p
I, JÀ1
*
Àp
I, J
*
ð ÞDx (5:81)
(a) (b)
u u
I, J −2
I, j −1
I, J −1
i, J i +1, J
i +1, J −1 i, J −1
I −1, J −1
I −1, j
I −1, J
I +1, J −1
I +1, j
I +1, J
I, j +1
I, J
I, j
I, J +1
v
v
W w P
e
E
n
N
s
S
FIGURE 5.7 An interior vcomponent backward staggered control volume for a 2D Cartesian domain:
(a) standard notation; (b) matrix notation. Solid circles: scalar nodes; open circles: ucomponent of the velocity
ﬁeld nodes; and open squares: vcomponent of the velocity ﬁeld nodes.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 209 7.11.2008 5:46pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 209
Correction equations are then deﬁned by
p ¼ p* þp
0
(5:82)
u ¼ u* þu
0
(5:83)
v ¼ v* þv
0
(5:84)
Using Equations 5.78 through 5.84 the following formulae for the corrected velocities are
obtained:*
u
0
i,J
¼ u
i,J
*
þ
p
0
IÀ1,J
Àp
0
I,J
À Á
Dy
a
i,J
¼ u
i,J
*
þ p
0
IÀ1,J
Àp
0
I,J
À Á
d
i,J
(5:85)
v
0
I, j
¼ v
I, j
*
þ p
0
I,JÀ1
Àp
0
I,J
À Á
Dx
a
I, j
¼ v
I, j
*
þ p
0
I,JÀ1
Àp
0
I,J
À Á
d
I, j
(5:86)
where d
i,J
¼A
i,J
=a
i,J
and so on. Similar expressions apply to all neighboring nodes of each control
volume. Note that the pressure correction ﬁeld is yet unknown.
There is no transport equation for the pressure but the continuity equation may be discretized
and recast to provide an equation for the pressure correction ﬁeld. A scalar control volume such as
the one shown in Figure 5.8 is adopted.
Applying the controlvolume integration to the continuity equation,
ð
CV
@
@x
(r(~x)u) þ
@
@y
(r(~x)v)
!
dV ¼ 0 (5:87)
or splitting the integral and applying the divergence theorem,
(a) (b)
u
v
W
w
S
s
P
e
n
N
E
v
u
I, J −1
I, j +1
I, J +1
I, j
I, J
i, J i +1, J I +1, J
I −1, J
FIGURE 5.8 An interior regular (scalar) control volume: (a) standard notation; (b) matrix notation. Solid
circles: scalar nodes; open circles: ucomponent of the velocity ﬁeld nodes; and open squares: vcomponent of
the velocity ﬁeld nodes.
* Where (a
iÀ1,J
u
0
iÀ1,J
þa
iþ1,J
u
0
iþ1,J
þa
i,JÀ1
u
0
i,JÀ1
þa
i,Jþ1
u
0
i,Jþ1
) and (a
IÀ1,j
v
0
IÀ1,J
þa
Iþ1,j
v
0
Iþ1,j
þa
I,jÀ1
v
0
I,jÀ1
þa
I,jþ1
v
0
I,jþ1
)
have been set, temporarily, to zero. Note that these terms will approach zero as the solution approaches convergence.
A similar consideration applies to all neighboring nodes.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 210 7.11.2008 5:46pm Compositor Name: BMani
210 Handbook of Thermal Process Modeling of Steels
ð
A
x
^n Á
@
@x
(r(~x)u)dA
x
þ
ð
A
y
^n
@
@y
(r(~x)v)dA
y
¼ 0 (5:88)
which results in
ruA
x
ð Þ
e
À ruA
x
ð Þ
w
Â Ã
þ rvA
y
À Á
n
À rvA
y
À Á
s
Â Ã
¼ 0 (5:89)
In matrix notation,
ruA
x
ð Þ
iþ1,J
À ruA
x
ð Þ
i,J
Â Ã
þ rvA
y
À Á
I,jþ1
À rvA
y
À Á
I,j
h i
¼ 0 (5:90)
Substituting the velocities in terms of their guessed and corrected values and rearranging, the
pressure correction factor is computed as follows:
a
I,J
p
0
I,J
¼ a
Iþ1,J
p
0
Iþ1,J
þa
IÀ1,J
p
0
IÀ1,J
þa
I,Jþ1
p
0
I,Jþ1
þa
I,JÀ1
p
0
I,JÀ1
þb
0
I,J
(5:91)
with
a
Iþ1,J
¼ (rd)
iþ1,J
Dy (5:92a)
a
IÀ1,J
¼ (rd)
i,J
Dy (5:92b)
a
I,Jþ1
¼ (rd)
I,jþ1
Dx (5:92c)
a
I,JÀ1
¼ (rd)
I,j
Dx (5:92d)
a
I,J
¼ a
Iþ1,J
þa
IÀ1,J
þa
I,Jþ1
þa
I,JÀ1
(5:92e)
b
0
I,J
¼ ru* ð Þ
i,J
Dy À ru* ð Þ
iþ1,J
Dy þ rv* ð Þ
I,j
Dx À rv* ð Þ
I,jþ1
Dx (5:92f)
where d
i þ1,J
¼A
i þ1,J
=a
i þ1,J
and so on.
A general source term (other than a pressure gradient) may be incorporated through another
integral over the control volume:*
Int
source
¼
ð
CV
S
u
(~x)dV ¼
S
u
DV (5:93)
If the source is a function of the dependent variable, then a linear relationship is assumed to model
this function, giving
Int
source
¼ S
(u,0),P
þS
u,P
u
P
(5:94)
which results in two terms to be incorporated in the general discretized equation:
a
P,source
¼ S
u,P
¼ S
u,(i,J)
(5:95)
S
0
¼ S
(u,0),P
¼ S
(u,0),(i,J)
(5:96)
So far, only interior control volumes have been dealt with. The application of the boundary conditions
is now presented. Let us consider a scalar control volume located along the left boundary of the 2D
* Recall that the source term is a volumerelated term while the diffusive and convective terms are surfacerelated.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 211 7.11.2008 5:46pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 211
Cartesian domain (see Figure 5.9). Note that the west boundary of this control volume is at the domain
boundary and that there is no (physical) west node. To model a Dirichlet condition a very large
number is added to the source terms (S
p
¼ÀA, S
u
¼Af
ﬁx
, where Ais very large) such that the variable
at node P is set to a ﬁxed value (f
ﬁx
). Note that this procedure is identical to that described to impose
essential boundary conditions in the FEM. A Neumann boundary condition would be required to set
the west coefﬁcient to zero and to introduce the known ﬂux value through the source terms.
To obtain a reasonable convergence rate yet ensure numerically stable computations, under
relaxation factors are applied as follows:
p
new
¼ p* þa
P
p
0
(5:97)
u
new
¼ a
u
u þ 1 Àa
u
ð Þu
nÀ1
(5:98)
v
new
¼ a
v
v þ 1 Àa
v
ð Þv
nÀ1
(5:99)
where the underrelaxation factors are bounded: 0 <a
p
<1, 0 <a
u
<1, 0 <a
v
<1; and variables
u
nÀ1
and v
nÀ1
represent the values after the previous iteration.
The application of the controlvolume method to steadystate heat (or mass) transfer problems in
the presence of forced convection is based on the same formulation, with the thermal (or species)
ﬁeld computed after the velocity ﬁeld is calculated, i.e., the phenomena are uncoupled. The general
algorithm to solve this type of problems is
1. Initialize the guessed values for the pressure, velocity, and temperature ﬁelds
2. Solve the momentum equations to compute the guessed pressure and velocity ﬁelds
3. Compute the pressure correction ﬁeld
4. Compute the corrected pressure and velocity ﬁelds (with underrelaxation)
5. Solve the energy equation to compute the temperature ﬁeld (with underrelaxation)
6. If the convergence criteria has not been reached then update the guessed ﬁelds with the
corrected ones and go back to step 2
7. Otherwise, print the results and stop.
(a) (b)
u u
I, J −1
I, j +1
I, J +1
I, j
I, J
i, J i +1, J I +1, J
v
v
S
s
W
e
P
n
N
E
FIGURE 5.9 Scalar control volume for a 2D Cartesian domain at a boundary: (a) standard notation; (b) matrix
notation. Solid circles: scalar nodes; open circles: ucomponent of the velocity ﬁeld nodes; and open squares:
vcomponent of the velocity ﬁeld nodes.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 212 7.11.2008 5:46pm Compositor Name: BMani
212 Handbook of Thermal Process Modeling of Steels
Unsteadystate simulations of a scalar ﬁeld in a 2D Cartesian domain start with the corresponding
governing equation:
@
@x
G(~x)
@f
@x
þ
@
@y
G(~x)
@f
@y
À
@p
@x
¼
@
@t
r(~x)f ð Þ þ
@
@x
r(~x)uf ð Þ þ
@
@y
r(~x)vf ð Þ (5:100)
The rate of change term introduces a new independent variable: time. Thus, the solution of the time
varying ﬁeld involves the integration of the governing equation over the control volume and over
the time interval: t to t þDt. Let us proceed by integrating over the time interval the controlvolume
approximations already developed for the diffusive, source, and convective terms of the conserva
tion equation, as follows:
Int
diff
¼
ð
tþDt
t
G
e
A
e
f
E
Àf
P
dx
PE
ÀG
w
A
w
f
P
Àf
W
dx
WP
! &
þ G
n
A
n
f
N
Àf
P
dy
PN
ÀG
s
A
s
f
P
Àf
S
dy
SP
!'
dt (5:101)
Int
source
¼
ð
tþDt
t
S
u
dV f gdt (5:102)
Int
conv
¼
ð
tþDt
t
r ~x ð ÞufA
x
ð Þ
e
À r ~x ð ÞufA
x
ð Þ
w
Â Ã
þ r ~x ð ÞvfA
y
À Á
n
À r ~x ð ÞvfA
y
À Á
s
Â Ã È É
dt (5:103)
Since the order of integration may be interchanged, the accumulation (rate of change) term is
computed from
Int
accum
¼
ð
CV
ð
tþDt
t
@rf
@t
dt
2
4
3
5
dV (5:104)
Using the mean value theorem for integrals, assuming temporarily that the value of the scalar ﬁeld at
point P is representative of the whole control volume and applying a backward ﬁnitedifference
scheme,
Int
accum
¼ r
P
f
tþDt
P
Àf
t
P
À Á
DV (5:105)
The integrals of Equations 5.101 through 5.103 may be evaluated using explicit, fully implicit, or
Crank–Nicolson schemes, where the diffusive and convective terms are computed with the ﬁeld values
taken as their values at the previous time step, the next time step, or a linear combination of both,
respectively. Generalizing, the ﬁeld values in Equations 5.101 through 5.103 are computed with
f
i
¼ uf
tþDt
i
þ(1 Àu)f
t
i
, i ¼ E, W, N, S (5:106)
where 0 <u <1.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 213 7.11.2008 5:46pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 213
5.7 MONTE CARLO METHOD
Monte Carlo method is a numerical method that applies statistical procedures based on probabilistic
simulation and sampling to obtain approximate solutions to mathematical problems [28]. The ﬁxed
random walk variant of the Monte Carlo method can be illustrated by treating the case of steady
state diffusion without sources in a 2D Cartesian domain. Furthermore, constant properties will be
assumed. Then, the governing equation is
rf(~x) ¼
@
2
f(x, y)
@x
2
þ
@
2
f(x, y)
@y
2
¼ 0 (5:107)
The domain discretization is accomplished by distributing nodes throughout the system, includ
ing the domain boundaries. A control volume is associated with each node, as shown in Figure 5.10.
The EWSN notation of the FVM has been retained to take advantage of approximations already
developed.*
Integrating the governing equation over the control volume and applying a central difference
scheme to approximate the partial derivatives, one obtains
A
e
f
E
Àf
P
dx
PE
ÀA
w
f
P
Àf
W
dx
WP
!
þ A
N
f
N
Àf
P
dy
PN
ÀA
S
f
P
Àf
S
dy
SP
!
¼ 0 (5:108)
If the grid is uniform and the step size in both directions is identical (which means that all areas are
equal), the ﬁeld variable at node P is given by
f
P
¼
1
4
f
E
þf
W
þf
N
þf
S
ð Þ (5:109)
The Monte Carlo method is based on the probabilistic interpretation of Equation 5.109 as follows.
To estimate the ﬁeld variable at node P, an abstract particle (or random walk) is allowed to leave this
(b) (a)
S
s
e
n
N
w w e
n
s
I, J +1
I, J +1
I, J
I −1, J I +1, J
W P E
FIGURE 5.10 Monte Carlo domain discretization for a 2D diffusional problem in a Cartesian domain:
(a) EWSN notation; and (b) matrix notation. The arrows show the four possible future paths for a random
walker for the ﬁrst step of the estimation of the ﬁeld variable at nodal point P.
* The expressions that follow are usually derived in the Monte Carlo literature through ﬁnite differences; however, the results
are identical to those presented here.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 214 7.11.2008 5:47pm Compositor Name: BMani
214 Handbook of Thermal Process Modeling of Steels
point and wander through the mesh. The domain discretization shown in Figure 5.10a permits these
walks to occur in only one of four directions: E, W, N, and S. Furthermore, if the material is
isotropic, with constant properties and the mesh is uniform, the probability of a walk in any one of
those directions must be the same and equal to 1=4. Note that this is exactly the number obtained
through the discretization of the governing equation (see Equation 5.109). Therefore, the discretized
equation may be interpreted, for a single particle, as
f
P
¼ P
E
f
E
þP
W
f
W
þP
N
f
N
þP
S
f
S
(5:110)
where P
i
is the probability of a walk toward the direction i. Each probability must be positive (in this
case equal 1=4) and the sum of all probabilities must equal 1.
Consider now that N particles leave the nodal point P and may arrive at points E, N, W, or S.
Since the probability to move in any of those directions is identical, the estimation of the ﬁeld
variable at point P after the ﬁrst step of the Monte Carlo method would be
f
P
¼
1
N
NP
E
f
E
þNP
N
f
N
þNP
W
f
W
þNP
S
f
S
ð Þ
¼
1
N
N
4
f
E
þ
N
4
f
N
þ
N
4
f
W
þ
N
4
f
S
(5:111)
In matrix notation,
f
I,J
¼
1
N
N
4
f
Iþ1,J
þ
N
4
f
I,Jþ1
þ
N
4
f
IÀ1,J
þ
N
4
f
I,JÀ1
(5:112)
In the general case, none of the values on the RHS of Equation 5.112 are known and further steps
are required. For the particular case considered only N=4 particles would have arrived at each of
the neighboring points after the ﬁrst step. Accordingly, in a second step, only N=4 particles are
allowed to leave each one of the neighboring nodes. Referring to Figure 5.11, there are now 16
N
S
P E
W
FIGURE 5.11 The sixteen possible walks (shown by the arrows) of a random walker for the second step of
the estimation of the ﬁeld variable at nodal point P.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 215 7.11.2008 5:47pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 215
possible future positions. Also, note that there will be four random walks arriving at the nodal
point P, and so on.
The estimate of the dependent variable at the nodal point P after the second step is then
f
Ii,J
¼
1
N
N
4
f
I,J
þ
N
8
f
Iþ1,Jþ1
þ
N
8
f
IÀ1,Jþ1
þ
N
8
f
IÀ1,JÀ1
þ
N
8
f
Iþ1,JÀ1
þ
N
16
f
Iþ2,J
þ
N
16
f
I,Jþ2
þ
N
16
f
IÀ2,J
þ
N
16
f
I,JÀ2
(5:113)
If the value of the dependent variable at the point of its arrival is known, then the random walk is
stopped. Therefore, Equations 5.112 and 5.113 are developed for consecutive steps until all random
walks have stopped.
The random walk may be sampled by selecting the decision required at each step, i.e., which
direction to follow, based on a range of random numbers (RN) [29]. For example, to decide if a walk
in a given direction will actually take place a RN* between 0 and 1 is used as follows:
If 0 < RN < P
E
then the walk will be from P to E (5:114a)
If P
E
< RN < P
E
þP
N
then the walk will be from P to N (5:114b)
If P
E
þP
N
< RN < P
E
þP
N
þP
W
then the walk will be from P to W (5:114c)
If 1 ÀP
S
< RN < 1 then the walk will be from P to S (5:114d)
Therefore, it is possible to numerically build a random path for any given walker (or particle)
starting at the nodal point of interest and ending at a point on the domain boundary. If N random
walkers are allowed to move randomly through the mesh starting at point (I, J), each one of them
will eventually arrive at a nodal point on the boundary—where the dependent variable has a known
value. Denoting the value of the dependent variable at the boundary by f
B
(i) for the random walker
i, the Monte Carlo estimate for f
I,J
considering N random walkers is
f
I,J
¼
1
N
X
N
i
f
B
(i) (5:115)
Clearly, the procedure described above does not require the computation of the entire unknown ﬁeld
(as is the case for the ﬁniteelement, controlvolume, and ﬁnitedifference methods); thus, the Monte
Carlo method is very efﬁcient when one is interested in computing the dependent variable at
selected nodes in the mesh. In fact, due to this characteristic of the Monte Carlo method, it has
been coupled to more traditional numerical methods to improve computing performance in, for
example, virtual mold modeling [30].
When a constant source term is included in the analysis, the number of steps required for the
random walker to reach the boundary, m
i
, needs to be tallied. The Monte Carlo estimate for the node
(I, J) becomes
f
I,J
¼
1
N
X
N
i
f
B
(i) þ
S
f
Dx
2
4G
1
N
X
N
i¼1
m
i
(5:116)
* Most programming languages include either functions or routines to generate random or pseudorandom numbers.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 216 7.11.2008 5:47pm Compositor Name: BMani
216 Handbook of Thermal Process Modeling of Steels
where
m
i
is the number of steps required for the random walker i to reach the boundary
S
f
is a constant source term
G is a (constant) physical property such as the thermal conductivity or the diffusion coefﬁcient
The probability of the random walker to move in any given direction in the mesh may be related
to the ability of the medium to transport the property of interest. For example, if the properties are
allowed to vary within the ﬁeld, then the governing equation would be
rÁ G(~x)rf(~x) ¼
@
@x
G(~x)
@f(x, y)
@x
þ
@
@y
G(~x)
@f(x, y)
@y
¼ 0 (5:117)
Integrating over a control volume and using central differencing,
G
e
A
e
f
E
Àf
P
ð Þ
dx
PE
ÀG
w
A
w
f
P
Àf
W
ð Þ
dx
WP
!
þ G
n
A
n
f
N
Àf
P
ð Þ
dx
PN
ÀG
s
A
s
f
P
Àf
S
ð Þ
dx
SP
!
¼ 0 (5:118)
Considering again a uniform grid with equal steps in the x and ydirections (and, therefore, equal
areas for diffusion) and rearranging, one obtains
f
P
¼ P
E
f
E
þP
W
f
W
þP
N
f
N
þP
S
f
S
(5:119)
with
P
E
¼
G
e
=dx
PE
G
e
=dx
PE
ð Þ þ G
n
=dy
PN
ð Þ þ G
w
=dx
WP
ð Þ þ G
s
=dy
SP
ð Þ
(5:120a)
P
W
¼
G
w
=dx
WP
G
e
=dx
PE
ð Þ þ G
n
=dy
PN
ð Þ þ G
w
=dx
WP
ð Þ þ G
s
=dy
SP
ð Þ
(5:120b)
P
N
¼
G
n
=dy
PN
G
e
=dx
PE
ð Þ þ G
n
=dy
PN
ð Þ þ G
w
=dx
WP
ð Þ þ G
s
=dy
SP
ð Þ
(5:120c)
P
S
¼
G
s
=dy
SP
G
e
=dx
PE
ð Þ þ G
n
=dy
PN
ð Þ þ G
w
=dx
WP
ð Þ þ G
s
=dy
SP
ð Þ
(5:120d)
The numerator in each term of Equation 5.120 is the diffusion conductance per unit area in the given
ﬂow direction while the denominator is the sum of the diffusion conductance per unit area for all
four possible ﬂow directions. In a nonisotropic medium the probabilities will be higher for the
preferred directions of diffusive ﬂow. On the other hand, the diffusion conductance per unit area
diminishes as the distance between any two given nodes increases, which will result in a smaller
probability for the random walker to go into that particular direction. Similar observations would
apply regarding the effect of the controlvolume areas on the diffusion conductance if a nonuniform
mesh were adopted.
Let us now consider the implementation of the boundary conditions. A boundary condition of
the Dirichlet type is known as an absorbing barrier in the Monte Carlo nomenclature. As mentioned
earlier, if the random walker arrives at such a boundary the random walk is stopped. On the other
hand, a Neumann boundary condition is referred to as a reﬂecting barrier. Referring to Figure 5.12,
the discretized equation for a nodal point along the right boundary of a 2D Cartesian domain
(constant properties, unequal x and y step sizes) is
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 217 7.11.2008 5:47pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 217
f
P
¼ P
E
f
E
þP
W
f
W
þP
N
f
N
þP
S
f
S
þP
1
f
1
(5:121)
with
P
E
¼ 0 (5:122a)
P
W
¼
Dy=Dx
(Dy=Dx) þ(Dx=(2Dy)) þ(lDy=G)
(5:122b)
P
N
¼
Dx=(2Dy)
(Dy=Dx) þ(Dx=(2Dy)) þ(lDy=G)
(5:122c)
P
S
¼
Dx=(2Dy)
(Dy=Dx) þ(Dx=(2Dy)) þ(lDy=G)
(5:122d)
P
1
¼
lDy=G
(Dy=Dx) þ(Dx=(2Dy)) þ(lDy=G)
(5:122e)
where
l is a constant interface transfer coefﬁcient
P
1
is the probability of walking toward the surrounding medium
The probability functions for the random walk with constant step size for unsteadystate heat
transfer in a 2D Cartesian domain are summarized in Table 16.3 [28].
5.8 PHASEFIELD METHOD
The ﬁnal structural features after thermal processing of metallic materials deﬁne the mechanical
properties of the product. Therefore, mathematical modeling and computer simulation of the effect
of processing conditions on the evolution of structural features of the products of phase transitions in
thermal processing can be a powerful tool in process design and optimization. The evolution of
the thermal and concentration ﬁelds dictate the kinetics of the phase transitions and the structural
features in thermal processing. Thus, a suitable mathematical model requires the solution of the
(a) (b)
S
s
w
W
n
N
P ∞
∞
W Δy
S
P
N
Δx/2
FIGURE 5.12 Monte Carlo domain discretization for a 2D diffusional problem near a boundary in a Cartesian
domain: (a) EWSN notation; (b) sizes of the lateral sides of the control volume. The arrows show the four
possible future paths for a random walker.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 218 7.11.2008 5:47pm Compositor Name: BMani
218 Handbook of Thermal Process Modeling of Steels
governing equations for mass, energy, concentration, and (if convective effects are important)
momentum at both the parent and the product phases, as well as in the transition zone between
them. However, the structural features occur at very different length scales, which complicates the
algorithms. For example, in solidiﬁcation, the geometry of the pool proﬁle and the mushy zone is
described at a macroscopic (1 mm to 1 m) length scale; the dendritic morphology and microse
gregation patterns can be distinguished at a mesoscopic (10–1000 mm) length scale; and the grain
size and columnartoequiaxed transition occur at a microscopic (0.1–10 mm) scale [31]. Addition
ally, interfacial boundary conditions need to be satisﬁed and the position of the interface must be
monitored continuously. The need to track the microscopic sharp interface between the parent and
the product phases constitutes the fundamental difﬁculty faced by traditional mathematical models.
The phaseﬁeld method overcomes this problem by deﬁning a diffuse (as opposed to a sharp)
interface where a nonconserved structural parameter, known as the phaseﬁeld, varies smoothly
between the values of 0 and 1 in the parent and product bulk phases, respectively. The mathematical
problem is then changed from one involving tracking the advance of a sharp interface to another
where a system of stiff partial differential equations needs to be solved [32].
In contrast with other numerical methods, the domain considered in the phaseﬁeld method
includes only the diffuse interface and its neighborhood and is usually either a square (2D domains)
or a cube (3D domains). Accordingly, the domain discretization is accomplished using either
squares or cubes. To lower the number of computations the domain symmetry is fully exploited;
a ‘‘seed,’’ i.e., an initial phaseﬁeld distribution, is deﬁned to start the simulation. An example of a
mesh for a single quadrant is shown in Figure 5.13.
Phaseﬁeld models require a redeﬁnition of the governing equations of conservation laws in
terms of the phaseﬁeld and the formulation of a governing equation for the evolution of the phase
ﬁeld itself. The former may be accomplished using either an averaging or a variational method.
The averaging method consists in summing up the averaged conservation equations for the solid
and liquid phases [33]. Using this method, the following governing equations for mass, energy, and
species, considering both diffusive and convective transport, were derived:*
FIGURE 5.13 Domain discretization for a quadrant in a 2D Cartesian phaseﬁeld model.
* A phaseﬁeld modiﬁed momentum governing equation may also be found in Ref. [33].
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 219 7.11.2008 5:47pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 219
rÁ (1 Àf)~v
L
½ ¼ 0 (5:123)
@
@t
(r h) þrÁ (1 Àf)r ~v
L
h
L
½ ¼ rÁ fk
S
rT
S
þ(1 Àf)k
L
rT
L
½ (5:124)
@
@t
fC
S
þ(1 Àf)C
L
½ þrÁ (1 Àf)~v
L
rC
L
½ ¼ rÁ f D
S
rC
S
þ(1 Àf)D
L
rC
L
½ (5:125)
where
f is the phaseﬁeld (1 Àf¼f
L
)
~v
L
is the average intrinsic liquid velocity
r is the mixture density
h is the mixture enthalpy
k is the thermal conductivity
T is the temperature ﬁeld
C is the concentration ﬁeld
D is the diffusion coefﬁcient
L and S are the subindices that refer to the liquid and solid phases, respectively
The usual accumulation, convective ﬂow, and diffusive ﬂow terms may be identiﬁed in the previous
equations. Note that each equation contains the phaseﬁeld variable explicitly.
On the other hand, the variational form of, for example, the phaseﬁeld modiﬁed conservation
equation for specie i is [34]
@C
i
@t
¼ rÁ M
C
i
r
dF
dC
i
!
(5:126)
where
M
C
i
is a suitable mobility term
F is a local functional* used to deﬁne the free energy of the system
The coupling between the concentration ﬁeld and the phaseﬁeld is implicitly deﬁned through F.
It should be mentioned that although the phaseﬁeld variable is usually related to a fraction of
the product phase it might also represent the crystal orientation and other structural ﬁelds [34].
The phaseﬁeld is a nonconserved quantity, which implies that a governing equation cannot be
built from a phaseﬁeld balance. Based on the Gibbs–Thomson equation for an isotropic surface
energy, an equation for the evolution of the phaseﬁeld for a simple binary alloy is given by [33]
@f
@t
¼ m
k
G r
2
f À
(rfr)jrfj
jrfj
!
þm
k
T
m
ÀT þm
L
C
L
ð Þjrfj (5:127)
where
m
k
is a linear kinetic coefﬁcient
T
m
is the equilibrium melting temperature of the pure substance
m
L
is the slope of the liquidus line
G is the Gibbs–Thomson coefﬁcient
To complete the explicit deﬁnition of the phaseﬁeld equation, the phaseﬁeld proﬁle is assumed to
vary as
* A functional is a function of functions.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 220 7.11.2008 5:47pm Compositor Name: BMani
220 Handbook of Thermal Process Modeling of Steels
f ¼
1
2
1 Àtanh
~n
2d
(5:128)
where
~n is the coordinate normal to the interface
2d is the interface thickness
After substituting Equation 5.128 in Equation 5.127, the evolution of the phaseﬁeld is given by
@f
@t
¼ m
k
G r
2
f À
f(1 Àf)(1 À2f)
d
2
!
þm
k
T
m
ÀT þm
L
C
L
ð Þ
f(1 Àf)
d
!
(5:129)
Alternatively, the phaseﬁeld evolution equation model may be expressed as [32]
t(~n)
@f
@t
¼ À
dF
df
(5:130)
with the functional, for dendritic growth, deﬁned by
F ¼
ð
"
[W(~n)]
2
2
jrfj
2
þf (f) þbl
u
2
2
#
(5:131)
In Equations 5.130 and 5.131
t is the characteristic time of attachment of atoms at the interface
W is the interface thickness
b is the normalization constant
l is a dimensionless parameter related to the strength of the coupling between the phase and
diffusion ﬁelds
u is the dimensionless temperature ﬁeld
In this model
f (f) ¼ À
f
2
2
þ
f
4
4
(5:132)
The evolution of the phaseﬁeld together with the phaseﬁeld modiﬁed conservation equations
constitutes a system of partial differential equations, which is used to describe the evolution of the
structural features during phase transitions. Given that the phaseﬁeld parameter is present in all
partial differential equations, the system is fully coupled. On the other hand, the phaseﬁeld
approximation allows the solution of the evolution equations over the entire computational space
without referring to a speciﬁc phase [35]. The numerical solution of each of the partial differential
equations is usually carried out using an explicit ﬁnitedifference approximation (see Equations 5.12
through 5.14 for ﬁnitedifference formulae) although other numerical schemes have been imple
mented [36]. The calculations derived using ﬁnitedifferences may be computationally demanding
due to the fact that the maximum diffuse interface thickness should be of the order of the capillary
length (10
À10
m), which, coupled with a domain of 100 mmÂ100 mm, results in a very large
computational time in order to keep the numerical scheme explicit [35]. One way to overcome
this problem is to use a larger value for the interface thickness. Karma and Rappel [32] give a
detailed discussion of the computational limitations of the phaseﬁeld method and demonstrate the
convenience of using the socalled thininterface limit.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 221 7.11.2008 5:47pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 221
5.9 CONCLUDING REMARKS
Efﬁcient design and optimization of thermal processes rely on mathematical modeling and computer
simulation, which are based on governing equations and initial and boundary conditions represent
ing the evolution of the ﬁelds of interest. To model industrial systems, the mathematical problem is
highly nonlinear and requires the implementation of numerical methods. Four of those methods
(ﬁnite element, ﬁnite volume, Monte Carlo, and phaseﬁeld) have been succinctly presented in this
chapter. Due to space limitations, issues related to convergence and stability have not been
discussed.
The reader must be aware that the ﬁeld of numerical analysis related to the mathematical
modeling of thermal processes is not a static one. Driven by the need to produce efﬁcient numerical
solutions, a number of developments continuously occur in both the numerical methods and the
hardware used.
REFERENCES
1. Themelis, N.J., Techniques of process analysis in extractive metallurgy, Metall. Trans., 3, 2021, 1972.
2. Szekely, J., The mathematical modelling revolution in extractive metallurgy, Metall. Trans. B, 19B, 525,
1988.
3. Brimacombe, J.K., The extractive metallurgist in an emerging world of materials, Metall. Trans. B, 20B,
291, 1989.
4. RamosBanderas, A. et al., Mathematical simulation and modeling of steel ﬂow with gas bubbling in
through type tundishes, ISIJ Int., 43, 653, 2003.
5. Szekely, J., Evans, J.W., and Brimacombe, J.K., The Mathematical and Physical Modeling of Primary
Metals Processing Operations, John Wiley & Sons, New York, 1988, p. 16.
6. Emanuel, G., Analytical Fluid Dynamics, 2nd ed., CRC Press, Boca Raton, FL, 2001, p. 5.
7. Anderson, Jr. J.D., Computational Fluid Dynamics. The Basics with Applications, McGrawHill,
New York, 1995, p. 52.
8. Özis¸ ik, M.N., Heat Conduction, John Wiley & Sons, New York, 1980, p. 13.
9. Gerald, C.F. and Weatley, P.O., Applied Numerical Analysis, 7th ed., Pearson, Boston, MA, 2003,
Chapter 2.
10. Carnahan, B., Luther, H.A., and Wilkes, J.O., Applied Numerical Methods, John Wiley & Sons, New York,
1969, p. 429.
11. Ibid. ref. 7, Chapter 5.
12. Ibid. ref. 10, p. 343.
13. Ibid. ref. 6, p. 26.
14. Burnett, D.S., Finite Element Analysis. From Concepts to Applications, AddisonWesley, Reading, MA,
1987.
15. Stasa, F.L., Applied Finite Element Analysis for Engineers, Holt, Rinehart and Winston, New York, 1985.
16. Zienkiewicz, O.C. and Taylor, R.L., The Finite Element Method, 4th ed., McGrawHill, London, 1988.
17. Denis, S., Sjöström, S., and Simon, A., Coupled temperature, stress, phase transformation calculation
model numerical illustration of the internal stresses evolution during cooling of a eutectoid carbon steel
cylinder, Metall. Trans. A, 18A, 1203, 1987.
18. Ibid. ref. 14, p. 97.
19. Ibid. ref. 14, Chapter 13.
20. Ibid. ref. 14, p. 189.
21. Ibid. ref. 14, p. 652.
22. Ibid. ref. 15, p. 59.
23. Ibid. ref. 14, p. 461.
24. Versteeg, H.K., and Malalasekera, W., An Introduction to Computational Fluid Dynamics. The Finite
Volume Method, John Wiley & Sons, New York, 1985.
25. Gosman, A.D., Pun, W.M., Runchal, A.K., Spalding, D.B., and Wolfshtein, M., Heat and Mass Transfer in
Recirculating Flows, Academic Press, London, 1969.
26. Ibid. ref. 24, p. 120.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 222 7.11.2008 5:47pm Compositor Name: BMani
222 Handbook of Thermal Process Modeling of Steels
27. Patankar, S.V. and Spalding, D.B., A calculation procedure for heat, mass and momentum transfer in three
dimensional parabolic ﬂows, Int. J. Heat Mass Transfer, 15, 1787, 1972.
28. HajiSheikh, A., Monte Carlo Methods, in: Handbook of Numerical Heat Transfer, Minkowicz, W.J. et al.,
Eds., John Wiley & Sons, New York, 1988, Chapter 16.
29. Ames, W.F., Numerical Methods for Partial Differential Equations, 3rd ed., Academic Press, Boston, MA,
1992, p. 405.
30. Nastac, L., A Monte Carlo approach for simulation of heat ﬂow in sand mold and metal mold castings
(virtual mold modeling), Met. Trans B., 29B, 495, 1998.
31. Nastac, L., Modeling and Simulation of Microstructure Evolution in Solidifying Alloys, Kluwer Academic,
Boston, MA, 2004, Chapter 1.
32. Karma, A. and Rappel, W.J., Quantitative phaseﬁeld modelling of dendritic growth in two and three
dimensions, Phys. Rev. E, 57(4), 4323, 1998.
33. Beckermann, C. et al., Modeling melt convection in phaseﬁeld simulations of solidiﬁcation, J. Comput.
Phys., 154, 468, 1999.
34. Gránásy, L., Pusztai, L., and Warren, J.A., Modelling polycrystalline solidiﬁcation using phaseﬁeld
theory, J. Phys. Condens. Matter, 16, R1205, 2004.
35. Grujicic, M., Cao, G., and Miller, R.S., Computer modeling of the evolution of dendrite microstructure in
binary alloys during nonisothermal solidiﬁcation, J. Mater. Synth. Process., 10(4), 191, 2002.
36. Raisin, I., Miller, W., and Succi, S., Phaseﬁeld lattice kinetic scheme for the numerical simulation of
dendritic growth, Phys. Rev. E, 72, 066705–1, 2005.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 223 7.11.2008 5:47pm Compositor Name: BMani
Modeling Approaches and Fundamental Considerations 223
Gur/Handbook of Thermal Process Modeling of Steels 190X_C005 Final Proof page 224 7.11.2008 5:47pm Compositor Name: BMani
6
Modeling of Hot and Warm
Working of Steels
Peter Hodgson, John J. Jonas, and Chris H.J. Davies
CONTENTS
6.1 General Introduction and Outline of Important Microstructure Processes........................ 225
6.2 Flow Curve Modeling—Single Pass ................................................................................. 226
6.2.1 General Hot Working Behavior .............................................................................. 226
6.2.2 Constitutive Models for Hot Working.................................................................... 228
6.2.3 Onset of Dynamic Recrystallization ....................................................................... 230
6.2.4 Advanced Constitutive Models Incorporating Dynamic Recrystallization ............ 231
6.3 Static Recrystallization....................................................................................................... 233
6.3.1 T
nr
Approach ........................................................................................................... 234
6.3.2 Models for Static Recrystallization......................................................................... 237
6.3.3 StrainInduced Precipitation Models....................................................................... 241
6.3.3.1 Alternative Approach to Deformation and Recrystallization Modeling........ 243
6.3.3.2 Application of the Normalization Method to Softening between Passes ...... 243
6.3.4 Advanced Constitutive Model Incorporating Dynamic Recrystallization.............. 245
6.4 Grain Growth ..................................................................................................................... 247
6.5 Transformation and Final Mechanical Properties.............................................................. 248
6.6 Warm Working .................................................................................................................. 249
6.7 Stress Strain and Microstructure Modeling for Multipass Conditions .............................. 250
6.8 Advanced Simulation and Modeling Techniques.............................................................. 252
6.8.1 Cellular Automaton Simulations............................................................................. 253
6.8.2 Monte Carlo Potts Simulations ............................................................................... 258
6.8.3 Phase Field Simulations .......................................................................................... 261
6.8.4 Coupled Simulations ............................................................................................... 261
6.8.5 Common Problems.................................................................................................. 262
References ..................................................................................................................................... 262
6.1 GENERAL INTRODUCTION AND OUTLINE OF IMPORTANT
MICROSTRUCTURE PROCESSES
The hot working of steel involves a wide range of microstructural reactions, including both static
and dynamic effects. The modeling of these processes has been at both a fundamental and applied
level. The models have been developed to the stage where they have been used for design and
control of rolling mills, the prediction of mechanical properties, and the support of fundamental
research. One of the major trends in recent years has been to combine models that work over
different scale lengths (i.e., from continuum down to molecular level) to provide greater insight and
improved accuracy.
Within the steel industry, the most common models are those that represent the average state of
the microstructure and the effect of this on the mechanical response. This includes the response
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 225 3.11.2008 1:55pm Compositor Name: JGanesan
225
during hot and warm working processes and the ﬁnal room temperature mechanical properties. For
these types of models, the important microstructural reactions that need to be considered include
.
Work hardening
.
Dynamic and static recovery
.
Dynamic, static, and metadynamic recrystallization
.
Dynamic and static precipitation
.
Grain growth
.
Transformation
The modeling of these processes generally involves following the events that occur at the micron
scale (e.g., change in grain size and fraction recrystallized) although many models also include some
measure of the dislocation density. However, the latter models are more concerned with an average
dislocation density due to its effect on the hot strength and driving force for recrystallization, rather
than following individual dislocations and their interactions.
The advantage of these relatively simple microstructural models is that they can be easily
integrated into thermal mechanical models of the process. Most online rolling mill models combine
a simple ﬁnite difference model for temperature calculation with mechanical equations for force and
torque. By following the microstructure evolution over the entire process and during cooling to
room temperature, using this approach, it is then possible to predict the mechanical properties.
To obtain more accurate understanding of the temperature–strain–strain rate ﬁelds over the cross
section of the steel, it is necessary to use ﬁnite element models for the thermomechanical aspects.
This is particularly the case for shape rolling where the simple mechanics analysis is inadequate.
As shown at the end of this chapter it is also now possible to combine ﬁnite element models with
more advanced microstructure simulation techniques, leading to socalled multiscale models.
The focus of the current chapter is the micron scale models as they demonstrate the general
principles and the important aspects of data generation and analysis. The other modeling approaches
outlined above will also be covered with an emphasis on their current abilities and limitations.
6.2 FLOW CURVE MODELING—SINGLE PASS
6.2.1 GENERAL HOT WORKING BEHAVIOR
The stress–strain behavior of austenite is characterized by work hardening to a peak, followed by a
decrease in stress with increasing strain until a steady state is reached (Figures 6.1 and 6.2). The
peak stress and associated strain increase with increasing strain rate and decreasing temperature. As
shown by Figure 6.2, the stress and strain to the peak stress also increase for more highly alloyed
steel, such as the stainless steel shown here.
This shape arises from the competing inﬂuences of work hardening, dynamic recovery, and
dynamic recrystallization. Austenite undergoes much less dynamic recovery than ferrite (this will be
covered in Section 6.6) and the buildup of strain in the matrix can lead to the initiation of
recrystallization during deformation, termed dynamic recrystallization [1]. The true onset of
dynamic recrystallization is difﬁcult to assess and most use the strain for an easily distinguishable
volume fraction—often 5% recrystallization. Metallography of the quenched microstructure would
provide deﬁnitive evidence for the onset of dynamic recrystallization but it is very difﬁcult from a
practical perspective. This is particularly the case for commercial lowcarbon and microalloyed
steels. In these steels, it is necessary to be able to quench to room temperature quickly enough to
avoid any change in the austenite microstructure before transformation to martensite. If the
microstructure is fully martensitic, then the prior austenite grain boundaries can be etched, although
this is difﬁcult for modern clean steels. In addition, steels with lower hardenability are difﬁcult to
quench to a fully martensitic microstructure. Therefore, the most common method is to analyze the
stress–strain curves, as discussed below.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 226 3.11.2008 1:55pm Compositor Name: JGanesan
226 Handbook of Thermal Process Modeling of Steels
The initiation of dynamic recrystallization for most steels occurs at between 50% and 80% of
the strain to the peak in the stress–strain curve. At peak stress the work hardening of the austenite is
balanced by the work softening due to the introduction of strainfree grains. The amount of dynamic
recrystallization at the peak varies with the alloy and deformation conditions, but is often around
20%. Beyond the peak the rate of generation of more strainfree grains leads to ﬂow softening. The
dynamically recrystallized grains also deform and work harden, and at the steady state there is full
dynamic recrystallization and the rate of softening and hardening are balanced.
This microstructural evolution is shown in Figure 6.3 for the stainless steel deformed in Figure
6.2. This clearly shows the early nucleation of new equiaxed grains on the grain boundaries of the
original larger austenite grains. As deformation progresses, ﬁrst the grain boundaries become
completely covered and this is followed by an increase in the volume fraction of grains nucleated
inside the original grains. It is also apparent that the grain boundaries are completely covered in new
recrystallized grains whereas for others there is little evidence of recrystallization.
0
20
40
60
80
100
120
140
160
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Strain
S
t
r
e
s
s
(
M
P
a
)
900ЊC
925ЊC
950ЊC
975ЊC
1050ЊC
FIGURE 6.1 Flow curves for a CMn steel deformed at 0.1 s
À1
and at different temperatures.
0
20
40
60
80
100
0 0.5 1 1.5 2 2.5 3 3.5
Strain
S
t
r
e
s
s
(
M
P
a
)
1000ЊC
1100ЊC
1200ЊC
FIGURE 6.2 Flow curves for a 304 stainless steel deformed at 0.01 s
À1
and at different temperatures.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 227 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 227
The modeling approaches that will be considered here are the empirical models that link the
stress to the deformation parameters: strain, strain rate, and temperature. The other approach is the
internal variable one where single parameter models will be considered.
6.2.2 CONSTITUTIVE MODELS FOR HOT WORKING
The empirical models for hot working attempt to directly link stress to the deformation parameters:
strain, strain rate, and temperature. One of the most important parameters in hot working is the
Zener Hollomon parameter, Z, also termed the temperature compensated strain rate:
100 µm 100 μm
(a)
ε e =0.3
100 μm
(b)
e =0.5
= 1
100 mm
(c)
e = 1.0
100 μm
(d)
e =2.0
100 μm
e = 3.0
(e)
FIGURE 6.3 Microstructural evolution of the 304 stainless steel deformed at 9008C and a strain rate of
0.01 s
À1
to different strains.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 228 3.11.2008 1:55pm Compositor Name: JGanesan
228 Handbook of Thermal Process Modeling of Steels
Z ¼ _ « exp (Q
def
=RT) (6:1)
where
_ « is the strain rate (s
À1
)
Q
def
is the apparent activation energy for deformation
R is the gas constant
T is the deformation temperature (K)
The basic concept is that if a metal is deformed under two different strain rates and temperatures, but
the same Z, then the ﬂow curves will be the same.
The stress at a given microstructural state, for example the steadystate stress where there is full
dynamic recrystallization, can then be related to Z through various equations. In general the steady
state stress is rarely obtained and the peak stress is a more common measure. The three equations
that have been proposed are
1. The low stress law
s ¼ AZ
a
(6:2)
2. The high stress law
s ¼ A
0
Ln
Z
b
_ _
(6:3)
3. The uniﬁed law
s ¼ A
00
sinh
À1
(Z=b)
m
(6:4)
The low stress law was developed for creep conditions, but has also been applied in many empirical
models for hot working. The high stress law is rarely used, although it is suitable for the ﬁnishing
trains of hot strip mills. There are now cases where the uniﬁed law has been applied, usually by the
equivalence to the inverse hyperbolic sine expression:
sinh
À1
x ¼ ln x þ
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
x
2
þ1
_
_ _
(6:5)
The application of these laws also requires strain hardening to be incorporated. The simplest model
is power law hardening, which is suitable for low to moderate strains (e.g., up to a strain of 0.5).
Considering the low stress law with strain hardening gives
s ¼ s
0
«
a
Z
a
(6:6)
or on expansion
s ¼ s
0
«
a
_ «
a
exp(aQ
def
=RT) (6:7)
This form of the equation has found wide application in online modeling of hot working. One of the
advantages is that it is easy to obtain the model parameters and also it can be linearized by taking the
logarithms of both sides of the equation. However, because it only applies to a limited range of
conditions it is generally necessary to have different model coefﬁcients for roughing mills (which
are closer to the low stress conditions) and ﬁnishing mills (which are closer to the high stress
conditions).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 229 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 229
More complex work hardening laws can be incorporated in these approaches. It is also possible
to allow for the strain softening effects of dynamic recrystallization. First, though, a discussion on
how to determine the onset of dynamic recrystallization will be taken into account.
6.2.3 ONSET OF DYNAMIC RECRYSTALLIZATION
The direct measurement of the start of dynamic recrystallization is extremely difﬁcult for the reasons
given above. Therefore, it is more common to examine the ﬂow curves, and more importantly the
rate of work hardening (ds=d«) as the strain increases.
What is known as ‘‘linear work hardening’’ is only observed at relatively low temperatures, well
below those employed in hot (or even warm) working. At the latter temperatures, the ﬂow curves are
always concave downward. This departure from linearity is associated with the operation of
dynamic recovery, which acts to remove dislocations and thereby reduce the rate of work hardening.
When dynamic recrystallization begins to play a role, the general shape of the curve does not depart
from its downward concavity. Instead, it is the curvature itself that begins to change as a result of the
operation of an additional softening mechanism [2]. A method for detecting the initiation of
dynamic recrystallization via the change in the rate of work hardening (the effect of the operation
of an additional softening mechanism) will now be described.
Some typical ﬂow curves determined on an aluminumkilled (AK) steel at various temperatures
and strain rates are illustrated in Figure 6.4 [3]. For the present purposes, these curves must be
differentiated (to derive the work hardening rate) and so must be smoothed numerically. This can be
done using a ninthorder polynomial, for example. The derivative, u ¼ds=d« can then be calculated
and plotted against stress, as illustrated in Figure 6.5. The points of inﬂection in the three plots
correspond to the moment when the additional softening mechanism (dynamic recrystallization)
begins to play a role. Otherwise, the curves would simply continue their downward trends, without
changing their curvatures, as is normally observed when dynamic recovery is the sole softening
mechanism.
The stress levels (and therefore the strains) at the point of inﬂection can be determined more
precisely by performing a second differentiation. The values of –du=ds are shown plotted against
s in the inset. The stresses and strains identiﬁed in this way are displayed in Figure 6.6 and are
200
180
950ЊC, 10 s
−1
950ЊC, 1 s
−1
1050ЊC, 10 s
−1
1050ЊC, 1 s
−1
160
140
120
100
80
60
40
20
0
0 0.2 0.4
True strain
AK
T
r
u
e
s
t
r
e
s
s
(
M
P
a
)
0.6 0.8 1
FIGURE 6.4 Effect of temperature and strain rate on the ﬂow curves of an AK steel. (From Poliak, E.I. and
Jonas, J.J., ISIJ Int., 43, 684, 2003.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 230 3.11.2008 1:55pm Compositor Name: JGanesan
230 Handbook of Thermal Process Modeling of Steels
associated with the onset of recrystallization, as outlined above. In this diagram, data for
Nbmicroalloyed steel are also presented, as are the values of the peak stresses and strains. It
can be seen that the critical strains associated with the onset of dynamic recrystallization are
approximately half the strains associated with the peak stresses.
6.2.4 ADVANCED CONSTITUTIVE MODELS INCORPORATING DYNAMIC RECRYSTALLIZATION
While it is possible to incorporate dynamic recrystallization into the simple models, it is more
common to use phenomenological constitutive models that consider an internal structural parameter
(i.e., average dislocation density) for the work hardening term. In these approaches, it is assumed
that the stress during straining tends toward a saturation stress as a result of a dynamic equilibrium
between athermal work hardening (dislocation storage) and thermally activated work softening
(annihilation of dislocations). This is more realistic for the high strain behavior where dynamic
recrystallization is likely to occur.
60
600
500
400
300
200
950ЊC, 1 s
1
950ЊC, 10 s
1
S
t
r
a
i
n
h
a
r
d
e
n
i
n
g
r
a
t
e
q
(
M
P
a
)
100
0
80
1050ЊC, 1 s
1
100 120 140
True stress s (MPa)
160 180
10
6
4
2
0
60 110

d
q
/
d
s
s (MPa)
160
1050ЊC, 1 s
1
950ЊC, 10 s
1
AK 950ЊC, 1 s
1
210
8
200
FIGURE 6.5 Strain hardening plot for the AK steel.
250
200
AK
s
p
s
p
s
c
s
c
e
p
e
p
e
c
ε
c
Nb
Nb
AK
150
50
0
11 12 13
log Z log Z
14 15 16 11 12 13 14 15 16
100
T
r
u
e
s
t
r
e
s
s
(
M
P
a
)
1
0.8
0.6
0.2
0
0.4
T
r
u
e
s
t
r
a
i
n
FIGURE 6.6 Relation between the critical and peak stresses and strains for the AK and Nb steel.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 231 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 231
In this approach the ﬂow stress s
wh
at any time during strain hardening is assumed to be related
to the average dislocation density r by
s
wh
¼ s
0
þGba
ﬃﬃﬃ
r
p
(6:8)
where
s
0
and a are material constants
G is the shear modulus
b is the magnitude of the Burgers vector
The two favored approaches for modeling the evolution of the dislocation density are the
Mecking and Kocks [4] and Estrin and Mecking [5] models. While a number of authors favor the
Kocks and Mecking approach because the physical basis for this equation appears more appropriate
for hot working, the present authors have found that the Estrin and Mecking model tends to give a
better ﬁt over a wide range of conditions for many steels.
In the Estrin and Mecking approach, the evolution of the average dislocation density during
deformation is given by
dr
d«
¼ k
0
Àk
00
r (6:9)
where
« is the strain
k
0
is a coefﬁcient related to the athermal storage of moving dislocations, which become
immobilized after traveling a constant distance
k
00
is a coefﬁcient associated with dynamic recovery, which is thermally activated and,
therefore, temperature and strain rate dependent
Combining Equations 6.8 and 6.9, gives
s
wh
u ¼ k
0
Àk
00
s
2
(6:10)
where
u ¼ds
wh
=d« represents the work hardening rate
k
0
and k
00
are material constants
Therefore, a plot of s
wh
u vs. s
2
should be linear. Figure 6.7 shows this for stainless steel. There are
some departures from linearity, but to a ﬁrst approximation it is reasonable and a signiﬁcant
improvement over the Kocks and Mecking formulation, where a u–s plot should be linear but
shows pronounced curvature for all stresses.
Integration of Equation 6.10 leads to
s
wh
¼ s
ss
*
2
þ s
2
0
Às
ss
*
2
_ _
exp Àb« ð Þ
_ _
1=2
(6:11)
where
s
wh
is the ﬂow stress during the work hardening regime
s
ss
*
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
k
0
=k
00
_
is the hypothetical saturation stress
s
0
is the yield stress
b¼2k
00
is a material constant
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 232 3.11.2008 1:55pm Compositor Name: JGanesan
232 Handbook of Thermal Process Modeling of Steels
The hypothetical saturation stress s
ss
*
corresponds to the saturation stress attained during work
hardening and dynamic recovery. This can be modeled using the hyperbolic sine law (Equation
6.12). In the s
wh
u – s
2
examples this is obtained by extrapolating to s
wh
u ¼0.
Z ¼ A
0
sinh as
ss
*
_ _ _ ¸
n
(6:12)
Combining Equations 6.11 and 6.12 gives the overall work hardening curve. The ﬂow softening by
dynamic recrystallization is then obtained by incorporating a softening term. Here the rate of ﬂow
softening is given by an Avramitype equation using strain rather than time:
X
dyn
¼ 1 Àexp Àk (« À«
c
)=(«
ss
À«
c
) ð Þ ½
n
(6:13)
where
«
c
is the critical strain for the start of dynamic recrystallization
«
ss
is the strain where the steadystate region is reached
To generate the form of this equation, a softening expression is used:
X
dyn
¼ s
wh
Às ð Þ= s
wh
Às
ss,DRX
ð Þ (6:14)
where s
ss,DRX
is the steadystate stress at full dynamic recrystallization and s
wh
is the stress
expected in the absence of dynamic recrystallization. Therefore, the following equation is obtained
s ¼ s
ss
*
X
dyn
s
ss
*
Às
ss,DRX
_ _
(6:15)
with X
dyn
given by Equation 6.13.
6.3 STATIC RECRYSTALLIZATION
After deformation, the stored energy in the steel provides a driving force for recovery and
recrystallization. Due to the high stacking fault energy of austenite, the amount of recovery is
very small and most softening of the microstructure occurs by static recrystallization. As for
10,000
15,000
20,000
25,000
30,000
35,000
40,000
45,000
50,000
0 20,000 40,000 60,000 80,000 100,000
s
2
s
w
h
q
900ЊC, 0.01 s
−1
900ЊC, 0.03 s
−1
900ЊC, 0.1 s
−1
FIGURE 6.7 Value of s
wh
u as a function of s
2
for stainless steel under different deformation conditions.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 233 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 233
dynamic recrystallization the nucleation happens predominantly on the original austenite grain
boundaries (Figure 6.8) and the recrystallization front grows into the deformed grain.
There are two types of static recrystallization processes. The process of nucleation and growth
of new grains is termed conventional static recrystallization, whereas metadynamic recrystallization
refers to the growth of grains that were nucleated during dynamic recrystallization. The recrystal
lization behavior in early studies showed a clear change in behavior during unloading when a small
amount of dynamic recrystallization had occurred. For example, the rate of recrystallization did not
increase with increasing strain beyond this point and also the sensitivity to strain rate and tempera
ture changed [6]. This supported a clear change in mechanism from conventional static recrystal
lization. However, as will be shown in the following sections, more recent work has suggested that
there is a much more complex change in recrystallization behavior. A more convenient description
used now is straindependent and strainindependent recrystallization.
In the following sections, the modeling of recrystallization between deformations will be
followed using this deﬁnition. First, a simple experimental approach to develop an understanding
of interpass recrystallization is presented.
6.3.1 T
nr
APPROACH
During steel rolling at high temperatures, the workpiece usually recrystallizes during the interpass
intervals between successive reductions. As the workpiece cools, a temperature is generally reached
at which recrystallization stops. This temperature is known as the recrystallizationstop temperature
or norecrystallization temperature and is conventionally referred to as the T
nr
. This temperature is
of considerable importance in the design of rolling schedules. For this reason, a convenient method
has been devised to determine this temperature based on the use of torsion testing to simulate the
time–temperature–deformation schedule.
Before describing this test, it must be pointed out that the T
nr
is not a ﬁxed temperature for a
given steel, such as a melting point or phase transition temperature. It depends instead on the details
of the schedule, including the cooling rate and interpass time. Thus, in processes such as plate
rolling, where the interpass times are rather long (e.g., 10–20 s), the T
nr
may not in fact be reached at
all unless another mechanism, such as carbonitride precipitation, intervenes (see below). Con
versely, when the interpass times are very short, as in rod rolling (e.g., less than 0.1 s), recrystal
lization will cease at relatively high temperatures, even in the absence of carbonitride precipitation.
In such cases, the presence of solutes plays an important role in retarding recrystallization [7].
An example of a torsion simulation of the rolling of a Nbmicroalloyed steel in a Steckel mill is
presented in Figure 6.9 [8]. Here, the sample was cooled at 18C=s, where 30 s interpass times were
50 μm
FIGURE 6.8 Nucleation of recrystallization on the original austenite grain boundaries.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 234 3.11.2008 1:55pm Compositor Name: JGanesan
234 Handbook of Thermal Process Modeling of Steels
used, and the pass strain was 0.3. Simulated rolling began at about 12008C and the material was
308C cooler when each successive pass was applied. Several features of this diagram are of interest
at this point. The ﬁrst is that considerable work hardening is displayed in the ﬁrst seven passes, even
though the temperatures range from 12008C down to 10208C. That is, the ﬂow stress doubles
following yielding (and can even triple if larger pass strains are used). The second is that there is
considerable softening following unloading. This is of course caused by static recrystallization.
Finally, the gradual increase in the approximate level of each curve, from about 50 MPa to over 100
MPa, is entirely associated with the cooling of the workpiece.
Turning now to passes 8–12, it is clear that the shapes of the curves have changed. No longer
does softening (recrystallization) take place between passes. Thus, the workpiece has passed
through the T
nr
between passes 7 and 8 (i.e., between 10208C and 9908C). Another difference is
that the stress level (to which the rolling load is proportional) now increases more rapidly with
decreasing temperature than above the T
nr
. This is because of the work hardening that is retained in
the absence of recrystallization. The retained strain adds a second component of ﬂow stress increase
to that associated with temperature decrease.
On cooling beyond pass 12, it can be seen that the ﬂow stress decreases. This decrease is
associated with deformation in the twophase, austenite þferrite region. The direct cause of the
decrease is that ferrite is softer than austenite at the same deformation temperature. Passes 13 and 14
involve deformation in the intercritical region (between Ar
3
and Ar
1
). Once the transformation is
complete, e.g., at pass 16, only the ferrite phase is present during further deformation.
One way of determining the T
nr
, Ar
3
, and Ar
1
as measured in torsion simulations is illustrated in
Figure 6.10 [8]. This type of diagram is constructed by ﬁrst calculating the mean ﬂow stress (MFS)
associated with each rolling pass. This is deﬁned by
MFS ¼
1
«
Â
_
s d« (6:16)
Strain
S
t
r
e
s
s
(
M
P
a
)
S
t
r
e
s
s
(
k
s
i
)
300
250
200
150
100
50
0
1 2 3 4 5 0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
FIGURE 6.9 Stress–strain curves of torsional simulation of average schedule (ﬁnal pass temperature ¼7458C).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 235 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 235
Here « represents the pass strain over which the integration of the ﬂow stress s is carried out. The
value of MFS associated with each rolling pass is then plotted against the inverse absolute
temperature of the particular pass. To produce a smoother curve, the results of several simulations
are generally combined in a single plot (e.g., four trials in Figure 6.9).
On this diagram, the T
nr
is deﬁned as the point of intersection of the two straight lines ﬁtted to
the data points associated with deformation above (in region I to the left) and below (in region II to
the right) the T
nr
. Because the actual transition from full recrystallization to no recrystallization is a
gradual one, involving a temperature interval of partial recrystallization, the T
nr
is not an exact
temperature. The Ar
3
is usually deﬁned as the point on the curve where the cooling behavior departs
signiﬁcantly from that established in region II. In a similar manner, the Ar
1
is deﬁned as the point at
which the intercritical (region III) behavior departs signiﬁcantly from the extrapolated temperature
increase behavior of region IV.
The T
nr
s associated with various steels have been measured by numerous researchers. Because
of the importance of this parameter in the design of rolling schedules, particularly for plate rolling,
correlations have been derived relating the experimental values to the chemical compositions of the
steels. One such correlation is [8]
T
nr
¼887 þ464 Cþ890 Ti þ363 Al À357 Si þ6445 Nb À644
ﬃﬃﬃﬃﬃﬃ
Nb
p
þ732 VÀ230
ﬃﬃﬃﬃ
V
p
(6:17)
The effects of Nb, Ti, Al, and V addition on the T
nr
can also be compared graphically, as indicated in
Figure 6.11 [10,11]. Here it can be seen that Nb addition has by far the most powerful effect on the
T
nr
, followed by Ti and Al, and that V has the least inﬂuence.
One word of caution with the T
nr
test and using the analysis in Figure 6.10 is that this type of
plot is essentially based on the high stress law (s / (1=T)) and it is possible to see a change in slope
even for full recrystallization in some tests. However, by combining this analysis with the careful
examination of the actual multihit ﬂow curves it is possible to remove this problem.
0.65
300
1265
(2300)
1060
(1940)
903
(1660)
Temperature (ЊC (ЊF))
780
(1435)
680
(1255)
250
200
150
100
M
e
a
n
f
l
o
w
s
t
r
e
s
s
(
M
P
a
)
M
e
a
n
f
l
o
w
s
t
r
e
s
s
(
k
s
i
)
50
0
44
36
29
22
15
7
0
0.7
I II III IV
0.75 0.8 0.85
1000/T (1/K)
T
nr
Ar
3
Ar
1
0.9 0.95 1.0 1.05
FIGURE 6.10 MFS as a function of 1000=T.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 236 3.11.2008 1:55pm Compositor Name: JGanesan
236 Handbook of Thermal Process Modeling of Steels
6.3.2 MODELS FOR STATIC RECRYSTALLIZATION
The kinetics of static recrystallization follows the typical Avramitype relationship with time, in
both the straindependent and independent regions:
X ¼ 1 Àexp (Àkt
n
) (6:18)
where
X is the fraction recrystallized
t is the holding time
k and n are constants
An example of the ﬁt between the data and this model is shown in Figure 6.12.
In this case, k is a complex function of the strain, strain rate, temperature, and composition.
Sellars introduced [11] the concept of a time constant into Equation 6.18. This time constant is
generally the time to 50% recrystallization. Others have used the time for the start (e.g., 5%) or
ﬁnish (e.g., 95%) recrystallization as these are often the times you want to know. However, these are
very difﬁcult to determine experimentally. Besides the fact that 50% recrystallization is easier to
measure, any error in this measurement has a small error in the time. This is clearly shown in Figure
6.12 where an error of Æ10% in the amount of recrystallization gives a much smaller error than for
the other two times.
If the data for a wide range of steels are considered, then the common feature for t is that t
0.5
shows two distinct regions (Figure 6.13). As noted above it was originally proposed that the lower
Nb
Ti
1000
900
T
n
r
(
Њ
C
)
800
0 0.10
Initial solute content (at %)
0.20
Al
V
FIGURE 6.11 Effect of different alloying elements on T
nr
temperature. (From Cuddy, L.J., International
Conference on Thermomechanical Processing of Microalloyed Austenite, DeArdo A.J., Ratz G.A., and Wray
P.J., Eds., The Metallurgical Society of AIME, Pittsburgh, PA, 1981, 129.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 237 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 237
strain behavior where the rate of recrystallization increases with strain was conventional static
recrystallization, whereas the higher strain behavior was metadynamic recrystallization. The early
deﬁnition of metadynamic recrystallization was the growth of nuclei, or small grains that formed
during deformation (i.e., by dynamic recrystallization). The reason for this was that the change in
behavior (marked by «* in Figure 6.13) was close to the peak strain. At low Z values, typical of
those used in many laboratory tests, this is the case. However, at high Z and higher alloy contents it
is now clear that «* can be greater than «
p
. Therefore, the straindependent region must contain a
mix of conventional static and metadynamic recrystallization and the exact behavior above «* is
uncertain. Therefore, it is now proposed to call these two regions straindependent and strain
independent rather than to assign a single mechanism to each behavior. One recent interesting
observation in austenitic stainless steel was that this change occurred at $50% of dynamic
recrystallization [12]; however, more work is required to develop an overall understanding.
0
10
20
30
40
50
60
70
80
90
100
10 100 1000
Time (s)
S
o
f
t
e
n
i
n
g
(
%
)
t
5
± 10%
t
50
±10%
t
90
± 10%
FIGURE 6.12 Softening fraction as a function of time for a 304 stainless steel deformed at 9008C and 0.1 s
À1
to « ¼1.
0.2 0.4 0.6 0.8 1 2 4 6
10
100
1000
e
∗
e
∗
e
∗
e
∗
0.01 s
−1
0.03 s
−1
0.1 s
−1
1.0 s
−1
e
p
e
c
T
i
m
e
f
o
r
5
0
%
s
o
f
t
e
n
i
n
g
,
t
5
0
(
s
)
Strain
FIGURE 6.13 Effect of strain rate on time for 50% softening at a constant temperature of 9008C in a 304
stainless steel.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 238 3.11.2008 1:55pm Compositor Name: JGanesan
238 Handbook of Thermal Process Modeling of Steels
Using the time to 50% recrystallization as the time constant, Equation 6.18 becomes
X ¼ 1 Àexp À0:693
t
t
0:5
_ _
n
_ _
(6:19)
where
t
50
¼ A«
Àp
_ «
Àq
d
r
0
exp
Q
rex
RT
_ _
(6:20)
This can be applied to both the straindependent and independent regimes. For example, the
coefﬁcients in Table 6.1 have been obtained for the steels shown using hot torsion tests. The
important points to note from this table are that the straindependent region has a strong sensitivity
to strain, and temperature and is a weak function of strain rate, whereas in the strainindependent
region the opposite is the case.
These equations assume that there is an independent effect of strain rate and temperature.
However, it has been shown in an earlier section that for hot working the strain rate effect could
act through the Zener Hollomon parameter, Z. If this is included in Equation 6.20 then the following
equation and constants in Table 6.2 are obtained:
t
0:5
¼ A
0
«
Àp
Z
Àq
d
r
0
exp
Q
rex
RT
_ _
(6:21)
TABLE 6.1
t
50
Equations for Postdeformation Recrystallization in Different Steels
Strain Dependent Strain Independent
Grade p q Q
app
(kJ=mol) p q Q
app
(kJ=mol) Ref.
SS 304 1.48 0.42 207 0 0.61 48 [12]
CMn 2 0.34 215 0 0.8 83
HSLA 2 0.37 330 0 0.83 97 [13]
IF 2 0.37 250 0 0.8 70
IF 1.9 0.4 192 0 0.8 89 [14]
IF (B) 1.9 0.4 293 0 1 89
X65 2.5 0.3 390 [15]
TABLE 6.2
t
50
Equations for Postdeformation Recrystallization Using Z in Different Steels
Strain Dependent Strain Independent
Grade p q Q
rex
(kJ=mol) p q Q
rex
(kJ=mol) Ref.
SS 304 1.48 0.42 375 0 0.61 292 [12]
CMn 2 0.34 354 0 0.8 222
HSLA 2 0.37 567 0 0.83 334 [13]
IF 2 0.37 436 0 0.8 255
IF 1.9 0.4 372 0 0.8 269 [14]
IF(B) 1.9 0.4 497 0 1 293
X65 2.5 0.3 606 [15]
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 239 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 239
The Q
rex
in this case is likely to be closer to the real Q
rex
. However, to obtain the true Q
rex
requires
a more complicated experimental program. To only obtain the thermal effects would require the
samples to be deformed at a given deformation temperature and then recrystallized at the tempera
ture of interest. This is the approach used for recrystallization after cold rolling. This is not practical
in hot working where the deformation and annealing temperatures are generally the same. There
fore, Q
rex
in these types of equation is only an apparent activation energy.
Using this approach, though, leads to some interesting changes between the straindependent
and independent regimes. Now the two apparent activation energies for recrystallization are much
closer together, although not always the same as was originally proposed for CMn steels [16]. It is
still not clear what the signiﬁcance of these differences is.
The solute drag effect from elements such as Nb, Mo, and Ti acts both through the constant A
and the apparent activation energy. There are no completely comprehensive models to account for
all of the solute drag effects and the general preference is to essentially divide the model coefﬁcients
into similar steel grades. One model that has been developed for Nb steels with 0.01–0.03 wt% Nb
is [17]
t
50
¼ (À5:24 þ550[Nb]) Â10
À18
«
À4:0þ77[Nb]
d
2
0
exp
330,000
RT
(6:22)
In this case, Nb is also acting through strain sensitivity. The reason for this is that solute drag affects
the work hardening and dynamic recovery rates and therefore this would be expected to alter the
strain sensitivity.
The role of solute drag in retarding recrystallization is clearly demonstrated in Figure 6.14,
where the Ti IF steel with 0.083% Ti is about two times slower than the plain C steel, while the
Nbmicroalloyed steel with 0.024% Nb is a further three times slower. Another interesting feature is
how the sensitivity of recrystallization to composition is much stronger for the straindependent than
strainindependent regimes. As will be shown in the following section, it is very important to be able
to model the effect of solute elements on the recrystallization behavior as this also has an indirect
effect on the precipitation behavior.
Recrystallization also leads to a change in the grain size of the microstructure. Again there are
two basic equations for the straindependent and strainindependent recrystallized grain sizes [18]:
0.1 0.2 0.4 0.6 0.8 1 2 4
0.1
0.2
0.4
0.6
0.8
1
2
4
6
8
10
20
40
IF
CMn
HSLA
T
i
m
e
t
o
5
0
%
s
o
f
t
e
n
i
n
g
,
t
0
.
5
(
s
)
e
FIGURE 6.14 Role of solute drag in retarding recrystallization kinetics.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 240 3.11.2008 1:55pm Compositor Name: JGanesan
240 Handbook of Thermal Process Modeling of Steels
d
rex
¼ «
À0:5
d
0:4
0
exp
45,000
RT
_ _
Straindependent (6:23)
d
md
¼ 2:6 Â10
4
Z
À0:23
Strainindependent (6:24)
It is still unclear whether the change from a straindependent to independent recrystallized grain
size occurs at the same strain as for the recrystallization kinetics, as there is limited data available.
One of the interesting features of this is that the grain size in the strainindependent region is
between 1.5 and 2 times the steadystate dynamic grain size and only depends upon Z as in the case
of the dynamically recrystallized grain size. This suggests that this mechanism has the potential to
lead to very ﬁne grains at high strain rates.
The effect of composition also has not been extensively studied although it would appear that
elements that favor solute drag, such as Nb, tend to lead to a ﬁner recrystallized grain size in the
straindependent region. It is difﬁcult to exactly compare in the strainindependent region as the Z
parameter will have a different activation energy. This again highlights one of the difﬁculties when
comparing across steel grades with these equations; any effect that changes Q
def
will have an
indirect effect on the other terms in these equations.
6.3.3 STRAININDUCED PRECIPITATION MODELS
It is now generally accepted that the complete retardation of recrystallization between deformations
is due to straininduced precipitation. Here ﬁne precipitates form on the dislocation structure and
effectively stop the nucleation and growth of new grains. There is still debate regarding the exact
mechanism(s) through which this occurs. For some time the accepted view was that the precipitates
form on the grain boundaries or subgrain boundaries and exert a pinning force sufﬁcient to stop the
migration of the bulges formed at the original grain boundary. However, Sellars and coworkers [19]
have proposed that the main cause of retardation is that the precipitates form on the dislocation
network within the grains, which reduces the amount of recovery that is a necessary precursor to
recrystallization. They have used advanced electron microscopy techniques to show the presence of
these very ﬁne precipitates on dislocations.
The exact mechanism is not that important for most of the modeling approaches to date as they
are based on predicting the driving force for precipitation. However, for some of the more advanced
models being developed for the hot working of steels the actual mechanism will become more
important.
The Dutta and Sellars [20] model to predict the time for the start of straininduced precipitation,
t
ps
(or time to 5% precipitation, t
0.05ps
) as a function of the process variables and composition, is
t
ps
¼ t
0:05
¼ 3 Â10
À6
[Nb]
À1
«
À1
Z
À0:5
exp
270,000
RT
_ _
exp
2:5 Â10
10
T
3
( ln k
s
)
2
_ _
(6:25)
This was based on an inversion of the standard equation for steadystate nucleation rate, J, of the
form
J ¼ NZbexp (ÀDG*=kT) (6:26)
where
N is the density of nucleation sites per unit volume
Z is the Zeldovich nonequilibrium factor
b is the rate of atomic attachment to the critical nucleus
DG* is the critical free energy for nucleation
k is Boltzmann’s constant
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 241 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 241
This expression can be reduced to
J ¼ (N=a
2
)[Nb]D
0
exp (ÀQ
diff
=RT) exp (ÀDG*=kT) (6:27)
where
[Nb] is the molar concentration of Nb
D
0
is the diffusion coefﬁcient for Nb in austenite
Q
diff
is the activation energy for diffusion of Nb in austenite
a is the lattice parameter
The critical free energy for the formation of the nucleus can be divided into the chemical driving
force for nucleation and the volume strain energy due to nucleus formation. Dutta and Sellars
proposed that the latter could be ignored, as Nb(C,N) is incoherent with the matrix. In this case,
DG* is then inversely proportional to the square of the chemical driving force for nucleation, DG
v
.
This, in turn, can be related to the solubility product:
DG
v
¼ À(RT=V) ln k
s
(6:28)
k
s
¼ [Nb] [C þ12N=14]
sol
=10
2:26À6770=T
(6:29)
where
V is the molar volume
k
s
is the solubility product as deﬁned above
In another work by one of the present authors in the development of Equation 6.22, the constant at
the front of Equation 6.25 was doubled, which is in the range of data by Dutta and Sellars.
Figure 6.15 shows the predicted precipitation start curves for a strain of 0.3 for a typical micro
alloyed steel with 0.03 Nb along with the softening curve for the hightemperature recrystallization
1.E+05
1.E+04
1.E+03
1.E+02
1.E+01
1.E+00
1.E−01
1.E−02
1.E−03
6 7 8 9 10 11 12
(1/T (K)) 10,000
t
(
s
)
t
20
Nb steel
t
20
CMn steel
t
Precipitation
FIGURE 6.15 Precipitation and recrystallization start curves at an applied strain of 0.3.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 242 3.11.2008 1:55pm Compositor Name: JGanesan
242 Handbook of Thermal Process Modeling of Steels
behavior of both this steel and a simple CMn steel. The intersection between the precipitation and
recrystallization start lines corresponds to the temperature above which recrystallization is controlled by
solute drag (signiﬁcant recrystallization occurs before precipitation), and below which it is controlled
by precipitationand precipitate coarsening (i.e., precipitation occurs before signiﬁcant recrystallization).
The important feature is that if there was no solute drag (i.e., if the steel recrystallized at
a similar rate to the CMn steel as shown by the t
20
CMn curve) then the intersection would be
$8408C whereas it is $9508C for the Nb steel. In part this explains why V and Ti are not as
effective in stopping recrystallization during rolling, as they have a very low solute drag effect
compared to Nb. Therefore, even if these elements had a driving force for precipitation similar to the
Nb(CN), recrystallization would occur rapidly and remove the deformation driving force before
formation; this has been demonstrated for a V containing steel [21].
6.3.3.1 Alternative Approach to Deformation and Recrystallization Modeling
It is frequently useful to be able to predict stress values at arbitrary strains and at temperatures and
strain rates of industrial interest. This can be done using the normalization procedure described
below. First, it should be noted that the peak stresses and associated strains in sets of stress=strain
curves determined over a series of temperatures and strain rates depend only on the Zener–
Hollomon parameter, Z ¼ _ « exp(Q=RT), where _ « is the strain rate, Q the activation energy of the
deformation, T the temperature, and R the universal gas constant. Thus, the peak stress s
p
is given
by s
p
¼F(Z) and the peak strain «
p
by «
p
¼C(Z). Normalized stress–strain curves can now be
produced by dividing given values of s by s
p
and of « by «
p
.
Two normalization parameters are now introduced, u ¼s=s
p
and w¼«=«
p
. These cover the
range from 0 to 1 and are independent of Z. Thus u ¼s(Z, w)=s
p
¼f(w) describes the ﬂow curve in
this approach. Crossmultiplication and substitution leads to the relation s(Z, w) ¼F(Z) Á f(w),
which is applicable to any stress before the peak. Expressed in this way, the ﬂow stress at any
given moment is a product of a thermal component, F(Z), which speciﬁes the rate of the controlling
mechanism under the selected conditions of temperature and strain rate, and an athermal contribu
tion to the ﬂow stress, f(w), which describes the temperatureindependent hardening behavior.
Some normalized stress–strain curves for the AK steel of Figure 6.16 prepared with the aid
of this formalism are presented in Figure 6.17. Such u vs. w plots can be differentiated to obtain the
Zindependent work hardening rate:
Q ¼ {{@u=@w}
«
¼ {@u=@w}
w
where the two derivatives are identical because w¼«=«
p
at a constant strain rate. In this way, the
u vs. s plots illustrated above can be replaced by Q vs. u plots, their normalized equivalents. Here,
the values of Q at the inﬂection point, Q
c
, can again be identiﬁed with the initiation of dynamic
recrystallization. Such normalized descriptions of work hardening are of particular utility in the
modeling of strip rolling, when the temperature and strain rate are changing continuously, even
within a single roll bite. Further details of the application of this method to industrial rolling can be
found in Ref. [22].
6.3.3.2 Application of the Normalization Method to Softening between Passes
The use of normalized stress–strain curves is not only applicable to work hardening at different
temperatures and strain rates, but can be extended to the occurrence of softening between passes.
For this purpose, the effects of static recovery must be neglected. Under rolling conditions, this is
largely justiﬁed because the MFS determines the rolling load and not the yield stress. While static
recovery affects the yield stress, it has little inﬂuence on the MFS because the loss of dislocations
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 243 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 243
attributable to recovery during an interpass interval is quickly made up during the ﬁrst 2% or 3% of
straining in the next pass.
The effect of softening by recrystallization during increasing interpass times is depicted in
Figure 6.18. Here it can be seen that as the holding time is increased, the ﬂow stress of the material
regresses further and further back toward the origin of the curves. Determination of the softening
rates that apply during rolling can thus be performed in the laboratory, under simpliﬁed conditions,
and algebraic expressions ﬁtted to the observations. The resulting equations can then be employed
in mill simulations.
The effect of the increasing speed of successive mill stands is illustrated in Figure 6.19, where
an experiment was carried out at 9508C, but at strain rates of 1 s
À1
(ﬁrst pass) and 5 s
À1
(second
pass). It can be seen that the use of a suitable strain rate normalization procedure brings the 5 s
À1
200
180
160
140
120
T
r
u
e
s
t
r
e
s
s
(
M
P
a
)
100
80
60
40
20
0
0 0.2 0.4 0.6
True strain
0.8
950ЊC, 10 s
−1
1050ЊC, 10 s
−1
1050ЊC, 1 s
−
−1
950ЊC, 1 s
−1
AK
1
FIGURE 6.16 Effect of strain rate and temperature on the ﬂow curves of AK steel.
1.2
AK, 950ЊC–1050ЊC, 1–10 s
−1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6
e/e
p
s
/
s
p
0.8 1 1.2
a
1
FIGURE 6.17 u–w (normalized stressstrain) plots for AK steel.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 244 3.11.2008 1:55pm Compositor Name: JGanesan
244 Handbook of Thermal Process Modeling of Steels
curve into coincidence with the ﬁrst pass curve determined at 1 s
À1
. Thus, the true (normalized)
softening applicable to the pass interval is actually 53% and not 37%. Further details of this
procedure, which is particularly applicable to static recrystallization, can be found in Ref. [23].
6.3.4 ADVANCED CONSTITUTIVE MODEL INCORPORATING DYNAMIC RECRYSTALLIZATION
A number of researchers are developing models for recrystallization that are more closely based on
physical phenomena. In this case, the rate of recrystallization will be related to factors such as the
dislocation density, the subgrain misorientation, the nucleation density, and the boundary mobility.
Pietrzyk and coworkers have developed a number of models along these lines with differing degrees
of complexity.
The model is based on an approach by Sandstrom and Lagneborg [24], who developed an
equation describing the evolution of the dislocation density as a function of the deformation
conditions. As described in the following section, aspects of this have been modiﬁed in the light
of recent developments. The model and calculation structure have been described in detail else
where [25] and the following summarizes the key features.
1.2
0.8
0.4
0
0 0.4
Nb steel
AK
steel
t
3
t
2
t
1
t
1
< t
2
< t
3
0.8
e/e
p
s
/
s
p
1.2
FIGURE 6.18 Fractional softening in terms of the normalized ﬂow curves.
0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6
Normalized strain e/e
p
N
o
r
m
a
l
i
z
e
d
s
t
r
e
s
s
s
/
s
p
0.8 1 1.2
950ЊC
1.4
True strain
X = 53%
180
160
140
120
100
T
r
u
e
s
t
r
e
s
s
(
M
P
a
)
80
60
40
20
0
0 0.1 0.2
1 s
−1
x = 37%
x = 53%
5 s
−1
1 s
−1
0.3 0.4 0.5
950ЊC
0.6
FIGURE 6.19 Use of normalized curves to determine fractional softening in AK steel when the strain rate
changes from one pass to another.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 245 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 245
The overall equation describing the evolution of the dislocation population is
dG(r, t)
dt
¼ f(D«) Àg(«) À
vy
D
mtp(r,t) (6:30)
where
f(D«) represents the athermal storage (hardening) of dislocations
g(«) is the thermally activated softening (recovery), while the ﬁnal term represents the changes
caused by recrystallization, with t ¼mb
2
=2
There are a range of equations used to describe each of the physical phenomena [25] incorporated in
this model. The ﬂuctuations in the dislocation density due to storage, recovery, and recrystallization
are described by the evolution of the volume distributions of dislocations G(r, t)dr [25], which has a
dislocation density between r and r þdr at time t, where G(r, t) is the distribution function. In the
evolutionary model, each fraction of the dislocation population is considered; there are also variants
where only the average fraction is considered as this provides more rapid calculation times.
The behavior during deformation is handled in a way similar to the single parameter
(i.e., dislocation density) models described above such as Estrin and Mecking. Once deformation
is interrupted the storage of dislocations ceases and the evolution of the dislocation density is
controlled by recovery and recrystallization. The equations for dislocation population, grain
reﬁnement, and grain growth are then solved as a function of time for each interval of dislocation
density.
The use of a dislocation population rather than a single average density also introduces
complications into the order of the calculations. For example, during work hardening it is necessary
to ﬁrst calculate the hardening of the dislocation population with the highest dislocation density;
otherwise, there will be positive feedback in the calculation. For recrystallization, the order of
dislocation densities in which the calculation is performed is unimportant as this process moves
the dislocation density from its current value to zero and so will not impact on neighboring
dislocation density intervals.
The difﬁculty with these models is in obtaining data to allow the tuning of the model
coefﬁcients. One of the reasons for this is that in a number of cases certain parameters can be
coupled. For example, in most cases the dislocation density will increase while the subgrain size
decreases during a single deformation test.
At present this model still has a number of limitations. Some of these include
1. Use of a single dislocation type to describe the internal state of the material during
deformation and recrystallization. Others [24] have suggested that there are two major
dislocation populations with different rates of evolution and effects on the microstructure.
2. No information regarding misorientations across grain and subgrain boundaries. The
amount of stored energy and the driving force for recrystallization are also dependent
on the angle of misorientation across boundaries. The heterogeneity in the distribution
of stored energy and the spread in misorientation increases with the development of
subboundaries and is expected to inﬂuence the nucleation and growth of recrystallization.
3. Assumption that the subgrain structure is established at an early stage of deformation. The
subgrain size is initially dependent on
ﬃﬃﬃ
r
p
and then reaches a steadystate size given by an
empirical equation related to the Zener–Hollomon parameter [14,15].
4. The model does not take into account the different orientations within the material and their
consequences on the mechanical (i.e., stress–strain) behavior. It assumes that all grains
follow the same hardening law and that one grain size is associated with one dislocation
density.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 246 3.11.2008 1:55pm Compositor Name: JGanesan
246 Handbook of Thermal Process Modeling of Steels
Importantly, the model assumes that all evolution equations of dislocation population and grain size
are the same during and after deformation. Hence, the mechanisms that take place during dynamic
and static recovery or during dynamic, static, or metadynamic recrystallization are assumed to be the
same. However, it is interesting to note that the model was able to ﬁt both the dynamic and static
microstructural changes, including the transition from straindependent to independent recrystal
lization with a single set of parameters, where these constants were obtained from a combination of
literature values and ﬁtting of ﬂow curves and straindependent recrystallization data (Figure 6.20).
This important result highlights the power of more physically reasonable models; the ability to both
interrogate the models to assess which factors are important and further the ability to predict
behaviors beyond the limited laboratory data.
6.4 GRAIN GROWTH
Following complete recrystallization there is normal or abnormal grain growth. Abnormal grain
growth only occurs under rare deformation conditions in thermomechanical processing. This is the
case where a few grains rapidly grow through the microstructure to an extremely large size (e.g.,
100 þmm) leading to a mixed microstructure. There are two conditions where this seems to occur:
during the reheating of steels containing precipitates and after very light reductions. The latter can
occur in section rolling where light sizing passes are used during ﬁnishing. However, in general this
is very rare and has not been adequately modeled for steel.
Therefore, the focus of this section will be on normal grain growth after complete recrystal
lization. The classical approach to grain growth is from the work of Burke and Turnbull [26] where
the grain size is related to the holding time after recrystallization by
d ¼ kt
n
(6:31)
Under ideal growth conditions, it is proposed that n equals 0, 5. This equation is more generally
written as
d
2
¼ d
2
0
þkt (6:32)
6
5
4
3
2
1
0
(a) (b)
−1
−2
−3
−2.5 −2.0
0.2 s
−1
2.0 s
−1
10.0 s
−1
−1.5 0.0
l
n
(
t
0
.
5
)
6
5
4
3
2
1
−1
−2
0
l
n
(
t
0
.
5
)
ln (strain)
Strain rate
0.5
850ЊC
Temperature
900ЊC
950ЊC
1000ЊC
1.0 −1.0 −0.5
−2.5 −2.0 −1.5 0.0
ln (strain)
0.5 1.0 −1.0 −0.5
FIGURE 6.20 Evolution of the time for 50% recrystallization with strain at different (a) strain rates and
(b) temperatures. (From Roucoules, C., Pietrzyk, M., and Hodgson, P.D., Mater. Sci. Eng., A339, 1, 2003.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 247 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 247
After full recrystallization the initial grain size in this equation is the recrystallized grain size and the
time is then the time from the end of full recrystallization to the start of the next deformation, or the
start of transformation.
However, steels contain numerous impurities and particles that provide a drag on the moving
boundaries. If the effect of particles is ignored then it has been observed that the exponent in
Equation 6.31 is much greater than 2. However, there needs to be some care in this approach. The
general method to obtain the exponent is to use Equation 6.32, effectively assuming d
m
)d
m
0
, and
plot log (d) vs. log (t). Unfortunately, with high m values this is easily satisﬁed. A more appropriate
test is whether d
2
)d
2
0
. If this is satisﬁed then there is greater conﬁdence in the value of m. Values
of m in the literature range from 6 to 10.
The temperature sensitivity for growth is obtained by expanding Equation 6.32 to
d
m
¼ d
m
0
þk
s
t exp
ÀQ
g
RT
_ _
(6:33)
Typical values for commercial grades are given in Table 6.3. However, these models are really only
suitable when the interpass times are of the order of seconds. For shorter times, there is an unrealistic
prediction of the growth behavior. This arises from the mathematical behavior of Equation 6.33 with
high exponents. One approach, outlined in [27], is to use power lawhardening for short growth times.
Another word of caution with all of these models is that to satisfy d
2
)d
2
0
experimentally.
Generally long growth times are required, which could mean that the nature of the boundaries being
studied is very different, due to the sweeping up of impurities, from that which exists during
the short interpass times. In essence there is a tendency to use data where there is very little growth
to develop models that with ﬁne recrystallized grain sizes and short times predict high growth rates.
6.5 TRANSFORMATION AND FINAL MECHANICAL PROPERTIES
Modeling the transformation from austenite to ferrite, pearlite, bainite, and martensite is very
complex and beyond the scope of this chapter. There are a variety of approaches for this ranging
from the simple Avramitype method to more complex thermodynamic and kinetic models. The
group at the University of British Columbia has developed a number of models over the past 20
years for rod, bar, and hot strip mills and readers are directed to their work for the most compre
hensive industrial models at present [28].
Increasingly, transformation studies are involving the advanced modeling techniques discussed
toward the end of this chapter. This is because they have the ability to combine the thermodynamic
and geometric aspects. However, for all aspects of modeling this is one area that requires signiﬁ
cantly more research before the models can be used in a routine predictive way.
Simple empirical models have been developed for many of the key features of the austenite to
ferrite transformation. For example, the temperature for the start of transformation (Ar
3
) in plate
rolling has been given by [29].
TABLE 6.3
Constants for Grain Growth Equation
for a Variety of Steel Types
Steel m k
s
Q
g
(kJ=mol)
CMn(V) 7 1.45 Â10
27
400
CMnTi 10 2.6 Â10
28
437
CMnNb 4.5 4.1 Â10
23
435
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 248 3.11.2008 1:55pm Compositor Name: JGanesan
248 Handbook of Thermal Process Modeling of Steels
Ar
3
C ð Þ ¼ 910 À310 C À80 Mn À20 Cu À15 Cr À55 Ni À80 Mo (6:34)
(C, Mn, Cu, Cr, Ni, and Mo representing the weight percent of these alloying elements in
composition) while the ferrite grain size after transformation can be estimated from the following
equations for CMn and Nb microalloyed steels, respectively [18]:
d
a
¼ d
a0
1 À0:45
ﬃﬃﬃﬃ
«
r
p
ð Þ (6:35)
where «
r
is the retained strain and
d
a0
¼ b
0
þb
1
C
eq
_ _
þ b
2
þb
3
C
eq
_ _
_
T
À0:5
þb
4
1 Àexp b
5
d
g
_ _ _ _
(6:36)
where
C
eq
is equal to CþMn=6
_
T is the cooling rate (8C=s)
d
g
is austenite grain size (mm)
d
a
¼ d
alim
þA 1:6 À«
r
ð Þ þB
_
T
À0:5
(6:37)
in which d
alim
is a limiting ferrite grain size.
The ﬁnal room temperature mechanical properties (LYS: Lower Yield Stress, TS: Tensile
Strength) for hotrolled steels are predicted from Hall–Petchtype equations that combine the
composition and ferrite grain size effects [18]:
LYS (MPa) ¼ 62:6 þ26:1[Mn] þ60:2[Si] þ759[P] þ212:9[Cu] þ3286[N] þ19:7d
À0:5
a
(6:38)
TS (MPa) ¼ 164:9 þ634:7[C] þ53:6[Mn] þ99:7[Si] þ651:9[P] þ472:6[Ni]
þ3339:4[N] þ11d
À0:5
a
(6:39)
6.6 WARM WORKING
While the majority of hot working processes occur in the fully austenitic region there are some cases
where deformation takes place in the twophase austenite þferrite or predominantly ferrite phase;
the former is generally referred to as intercritical rolling, while the latter is referred to as warm
working. Intercritical rolling has been used to impart added strength to plate steels by working the
ferrite phase and retaining this deformation to room temperature. Warm working can be used to
either produce a worked structure or a fully recrystallized structure at room temperature.
Intercritical rolling is extremely complex and difﬁcult to model. Figure 6.9 showed the stress–
strain behavior of a steel deformed below the Ar
3
. Here there was signiﬁcant ﬂow softening once the
austenite to ferrite transformation had commenced. This behavior has been modeled using a simple
law of mixtures approach, where the overall strength is predicted by modeling the strength of the
austenite and ferrite phases:
s ¼ s
a
V
a
þs
g
(1 ÀV
a
) (6:40)
where
s
a
is the ferrite strength
V
a
is the ferrite volume fraction
s
g
is the austenite strength
Here the strength of the austenite must incorporate any work hardening from previous passes. After
the ﬁrst intercritical deformation, the situation becomes more complex, though. As the temperature
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 249 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 249
is decreasing there will be further transformation of austenite to ferrite; therefore, there will be a
volume fraction of ferrite that is unworked and a volume fraction deformed in the previous pass. The
worked ferrite also has the potential to recrystallize. This will become even more complicated in
subsequent passes.
This level of complexity is almost impossible to model, particularly as there is also the potential
for strain partitioning between the harder and softer phases. Most rolling mills would tend to
approach this problem by using a simple empirical model linking the ﬂow stress to temperature
and ignoring strain hardening effects. This is not unreasonable as the austenite phase should be
heavily deformed and so will not have a strong strain effect and the ferrite phase shows much lower
rates of work hardening than austenite. The data in Figure 6.10 highlight how the mean ﬂow stress
(MFS) appears to vary in a simple way with temperature below the Ar
3
, and it is possible to develop
simple empirical equations for this behavior.
For warm working, there is a range of strategies. Recently rolling mills have been processing
low C steels at temperatures where the ﬁnishing operation is below the Ar
3
[30]. In these steels, the
transformation from austenite to ferrite happens over a very short temperature range and the
behavior can be predicted by just modeling the deformation behavior of the ferrite phase. For
higher C steels, such as medium C steels that are warm forged, the modeling again becomes
extremely complex. However, in this case the aim is to forge at temperatures where there is little
microstructural change and hence empirical models can be developed to predict the ﬂow stress.
6.7 STRESS STRAIN AND MICROSTRUCTURE MODELING
FOR MULTIPASS CONDITIONS
The above modeling has dealt with models for the processes occurring during a single deform
ation or between two deformations. Rolling is a multideformation process that can involve a
number of deformations in the full recrystallization, partial recrystallization, nonrecrystallization,
and warm working regimes. The T
nr
test is a convenient way to examine this behavior but it is
limited to conditions that typically apply to plate mills. Table 6.4 outlines the normal range of
strain, strain rates, and temperatures for different types of rolling mills. Also given is the range
of interdeformation times. For continuous rolling mills such as rod, bar, and the ﬁnishing train of
a hot strip mill, the strain rate and interdeformation time are coupled; each sequential pass has a
higher strain rate and a shorter interpass time. For reversing mills, such as plate mills, the interpass
time will vary along the length of the plate because the head at one pass becomes the tail end at the
next. As the plate also gets longer, but the roll speed does not vary, the interpass times may increase
during rolling.
TABLE 6.4
Typical Range of Processing Conditions for Conventional Hot Rolling Mills
Mill Type Strain per Pass Total Strain Strain Rate (s
À1
) Temperatures (8C) Interpass Time (s)
Plate 0.05–0.4 1–4.5 0.3–30 1300–700 5–20
Strip
Finishing 0.2–0.5 $2.5 0.5–10 1300–900 $10
Roughing 0.1–1.3 1.5–3.2 10–100 1050–700 1–5
Total 4–5.7
Structural 0.05–0.8 1–4.5 0.3–30 1250–800 $10
Bar 0.3–0.6 2–5 0.5–300 1200–900 0.07–3
Rod 0.3–0.6 4–7 Up to 2000 1200–900 0.01–3
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 250 3.11.2008 1:55pm Compositor Name: JGanesan
250 Handbook of Thermal Process Modeling of Steels
When developing total mill models it is important to understand the objective of the activity.
There are signiﬁcant differences between (a) building up fundamental understanding, (b) obtaining
predictions of forces and properties, and (c) online models for control. With the ongoing increase
in computing power it would be tempting to suggest that all should be able to be accommodated
within a single model. However, this is still not the case. Online models must be extremely robust
and in many cases need to be linearized. This is why many rolling mills still use versions of the low
stress law with simple power law hardening in their mill setup and online control models.
If the objective is only to predict the ﬁnal mechanical properties, then it is generally not
necessary to couple the microstructure and thermomechanical models. Rather the temperature–
deformation models are run for the entire rolling process and the temperatures–time history, average
pass strains, and strain rates are captured in a ﬁle that then becomes the input ﬁle for the
microstructure model. This is particularly the case for strip and plate mills. For mills with more
inhomogeneous strain distributions over the cross section, it may be necessary to write more
information for certain different regions; this approach has been used in multipass modeling of
bar rolling [31].
One of the major areas of debate in multipass models is how to handle partial recrystallization.
Most modelers still take the same approach as the ﬁrst models developed by Sellars [32]. In this
approach if there is partial recrystallization between two passes then two fractions are created: one
completely recrystallized with a grain size given by Equation 6.23, the other completely unrecrys
tallized with a retained strain equivalent to the applied strain. This process continues for subsequent
passes and it is possible to generate a large number of fractions of microstructure with different
retained strains and austenite grain sizes, and these are then used to calculate different fractions of
ferrite grain sizes, which are then averaged by volume and used to calculate the strength from the
Hall–Petchtype equations.
However, there are problems with this approach. The ﬁrst is that the prediction of recrystal
lization kinetics using this simple law of mixtures approach does not match those observed. Instead
the rate of recrystallization is most accurately predicted using the retained strain (which is the
average strain over the microstructure) and the next strain—again though it does not appear to
always be the case and some work has shown that only a fraction of the retained strain contributes to
the recrystallization processes after the next pass [18]. The simple explanation for this complexity is
that the real factors that are contributing to recrystallization are not modelled. During the ﬁrst
interdeformation period, the most potent sites for recrystallization will have been activated. For the
interval following the next pass, new sites need to be activated and these could be in the original
grains, which now have two deformations, or the new grains, or at the boundaries between the two.
More work is still required to model this condition accurately. It is probably fortunate that most
rolling scenarios do not lead to this complex microstructure state; in controlled rolling the addition
of Nb is used to stop recrystallization and so partial recrystallization tends to be restricted to low
temperature plate rolling of CMn steels and hot strip mills.
For online force prediction models it is also necessary to model recrystallization for accurate force
predictions but to the authors’ knowledge none of these models uses the above approach of breaking
the microstructure down into different fractions. In this case most models use the strain averaging
approach, also mentioned here. If the example of the simple power law model is considered with
power law hardening, then the following equation is obtained
s ¼ s
0
_ «
a
(« þ«
ret
)
a
exp (Q=RT) (6:41)
where «
ret
is the retained strain from previous passes. For Nbmicroalloyed steels deformed below
the T
nr
this is just the sum of previous passes below this temperature. For CMn or other steels where
there is a potential for recrystallization, this is more complex. In this case, the retained strain is
modeled using an Avramitype equation. Here, the fraction of strain retained (i.e., not recrystallized)
is important. Unless there is a full microstructure model running in the background, it tends to be
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 251 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 251
reduced to a t
50
equation that does not contain a grain size term. One of the interesting issues,
though, is that the fraction retained strain is not the same as the fraction of recrystallization, and this
difference becomes greater, the larger the strain. This is clearly demonstrated in Figure 6.21 for three
levels of recrystallization; for example 80% softening leads to less than 50% retained strain («
t50
).
One approach is to develop an empirical relationship between the two; another is to develop models
for retained strain. These approaches are rather limited; in many mills a simple method has been to
use the T
nr
approach and assume full recrystallization above and none below. This is a potentially
more robust method, particularly for plate mills.
For hot strip mills, there are also a range of strategies that are used. One of the advantages of
these mills over plate mills is that the rolling schedule for a given grade, ﬁnal thickness, and width is
ﬁxed unless schedule free rolling is used. This allows standbystand adaptation to counter the
unknowns in the models. However, even here there is an increasing trend for greater reliance on mill
models rather than the adaptation strategy.
Finally, for very accurate understanding of the processes occurring, it is necessary to fully
couple the various thermomechanical and microstructure aspects. This is because the different
microstructure events occurring at different areas can affect the overall deformation behavior. In
addition, if the extra work through strain accumulation is not incorporated then the heat of
deformation will be underestimated. Interestingly, though, while there have been some attempts
to combine these types of models with ﬁnite element models the general strategy now is to move to
multiscale models rather than introducing average microstructural states at given nodes and follow
ing their evolution. The recent developments in these multiscale approaches are outlined in the
following section.
6.8 ADVANCED SIMULATION AND MODELING TECHNIQUES
The above sections have dealt with the modeling of physical phenomena without actually following
their real evolution. The equations represent an average state of the microstructure; even when
considering different volume fractions for a given state it is still the average of that state. To fully
understand what is happening it is necessary to be able to predict the shapes of grains and to follow
the nucleation and growth of individual grains and how they interact with each other. This provides
much more information related to the physical processes that are occurring and the features that
control their evolution.
While there is no doubt that simulations have beneﬁted from fast computers and efﬁcient
programming environments, the evolution of experimental techniques has also contributed to
the evolution of simulations. Electronbackscattered diffraction (EBSD) in the scanning electron
s
e
t20
e
t50
e
t90
e
FIGURE 6.21 Schematic of retained strain at different recrystallization fractions.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 252 3.11.2008 1:55pm Compositor Name: JGanesan
252 Handbook of Thermal Process Modeling of Steels
microscope (SEM) has changed the face of metallography, delivering topological and crystallo
graphic information to the researcher. In an extension of this approach, focused ion beam SEMs are
coupled with EBSD to extract threedimensional (3D) serial section data sets. Experimental
facilities that allow insitu testing—either at elevated temperature, or under load, or both—have
been designed around neutron, synchrotron, and conventional xray sources and allow time series
capture of deformation, recrystallization, grain growth, and transformation events. The rich data sets
resulting from these experimental techniques provide a wealth of input data and parameters for
complex simulations. In turn, simulations suggest new experiments that must be conducted to obtain
the necessary physical parameters to drive the simulation.
Despite the advent of more complex simulations and everimproving experimental techniques,
the risk inherent in all of this is that it ends up with a mapping exercise, rather than advancing our
understanding of a phenomenon. Thus in all of our efforts to reproduce the physical world we need a
reality check: do we understand any more about the phenomenon than we did when we started the
exercise? This question is probably more important to ask of simulations than of empirical or
semiempirical models because by and large simulations are expected to be used in an explanatory
manner, rather than the data driving the model. And this brings us to the difference between models
and simulations. In the context of this chapter, the term ‘‘simulation’’ is used for a computer
program that attempts to reproduce some feature of the microstructural behavior, usually as it
evolves, whereas ‘‘model’’ is used more generally to indicate either the framework of a simulation or
a predictive equation.
There are several techniques that may be used to simulate the microstructural evolution of steels
during hot and warm working, but not all of the elevated temperature phenomena have been
simulated by all of the techniques. The three techniques that have been used regularly to simulate
microstructural evolution are the cellularautomaton (CA) technique, the Monte Carlo (MC)–Potts
method, and the phase ﬁeld method. Of these, the MC–Potts method is probably the most widely
used and was the earliest to be applied to iron [33], in a simulation of grain growth. MC–Potts
simulations have been used to simulate recrystallization and grain growth following the cold
working of steels [34,35] and austenite ferrite transformation [36]. The CA technique has been
used to simulate static [37] and dynamic recrystallization [38] during elevated temperature working,
static recrystallization following cold work of interstitial free steel [39], and austenite–ferrite
transformation [40]. Hybrid CA=MC approaches have been successful in simulating recrystalliza
tion [41]. The phase ﬁeld technique is applied to austenite decomposition [42–45] and has been
applied to grain growth in other systems. A fourth approach, the front tracking or vertex simulation
[46,47], has been applied in a limited manner to the recrystallization of ferritic stainless steels [48],
and more generally to grain growth [49], but the coupling of these to experimental data remains a
signiﬁcant hurdle. Vertex models have principally been used to validate topological theories of grain
growth and Zener pinning. Readers wanting to explore vertex models should obtain a copy of the
Surface Evolver software [50].
The CA, MC–Potts, and phase ﬁeld approaches are each be considered below. There are issues
that are common to all—such as how the simulation is initialized with experimental data, and how
the rules governing evolution are obtained in the warm and hot working regime—and these are
discussed. As can be seen that the trend is for coupled simulations, either the coupling of different
physical phenomena using the same technique, or exploiting the advantages of each technique and
coupling two or more techniques to obtain a more complete ‘‘throughprocess’’ simulation. This
exciting and fairly new aspect of simulation is also discussed.
6.8.1 CELLULAR AUTOMATON SIMULATIONS
Cellular automata have a history dating back to the 1940s, but were ﬁrst applied in materials
engineering in the 1990s to simulate recrystallization [51–54] and solidiﬁcation [55,56]. Since then,
there has been a fairly rapid development of codes and methodologies applicable to metals, but the
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 253 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 253
basic premise of the CA remains the same: the system to be modeled is represented by local rules
governing the interaction of cells arranged in an array and the global behavior is an emergent
phenomenon. This approach has the attraction that it mirrors our perception of physical systems; a
recrystallizing metal does not know that Avrami kinetics exist, but rather the Avrami (JMAK)
kinetics emerge as a result of the local interactions occurring across the moving front of a
recrystallizing grain. At the local level in the simulation, the rules are deterministic—repeating
the same step in a simulation on the same cell will give the same result—but the application of these
rules can be governed by probability, giving the outcome of the simulation the appearance of the
physical system, in which we do not expect the evolution of microstructure to be deterministic
except in terms of average measurements. The CA is typically a mesoscale simulation, by which we
mean that we scale the automaton to cover from several tens to a few hundred grains.
Formally, a CA is an array of ﬁnitestate machines in one, two, or three dimensions, governed
by rules for changing the state of the cells in the array. A simple example of a CA can be set up by
arranging 20 (or more) coins on a table in a line or a square grid. If the coins are placed randomly on
the table, some will fall as heads, some as tails. We can now invent rules by which coins are turned
over. For example, if two neighbors of a coin are heads ﬂip the coin to a head. Immediately a few
questions arise: how do we deﬁne the neighbors on a square grid? What happens at the end of the
line, or the edge of the square grid? Do we ﬂip the coins as we go, or do we somehow store the ﬂips
and only make the changes once each coin in the array has been examined? We address these
questions with reference to the materials phenomena we want to simulate.
In materials science and engineering, we are mostly interested in 2D or 3D cellular automata. A
cell may carry one or more attributes as its state and these can be changed independently or in a
coupled manner. Binary cellular automata (our coin ﬂip automaton) include phase transformation
CA (solid=liquid, ferrite=austenite) and recrystallization CA, whereas more complex CA might
incorporate crystallographic orientation selected from a range of possible states. The scale of each
cell is in the order of fractions of a grain and needs to be scaled such that the curvature of a growing
boundary can be adequately approximated by a stepped structure. (This can mean that the CA must
be rescaled during the course of a simulation.) The local dislocation density, temperature, and solute
content can also be captured as a cell’s attributes, but these variables are not always strictly states of
the automaton in that they are often governed by continuous constitutive relations and heat or mass
transfer equations: they do act to control the state of the cell. This also gives us a hint that we ought
to be able to couple CA to other types of models that can provide the coarserscale information, such
as ﬁnite element simulations.
In order for the CA to operate, two sets of rules are required: the neighborhood of each cell must
be speciﬁed—the extent of interaction between cells in the array—and the rules for transition of the
state of cells must be encoded.
There are three common cell neighborhoods in 2D cellular automata and each produces a
characteristic grain shape and error between simulation and experiment, both of which may require
correction. The von Neumann neighborhood [51] is the simplest and considers only the nearest
neighbors (Figure 6.22a); uncorrected grains are diamond shaped before impingement [55]. The
Moore neighborhood examines the eight neighbors surrounding a cell (Figure 6.22b) and produces
facetted hexagonal grains. A variation on the Moore neighborhood alternately examines the six
diagonal cells and produces octagonal grains before impingement [51,53,57]. This variation may
also be implemented in 3D simulations with appropriate increase in the number of cells and
diagonals examined in any step [58]. More recently the concept of the random grid CA has been
introduced to model metallurgical phenomena [59], and this appears to overcome the morphology
problems inherent in gridbased neighborhoods (Figure 6.23).
In addition to the type of neighborhood, the boundary conditions at the edges of the CA must be
speciﬁed. In two or three dimensions these are usually periodic, but under circumstances of a
temperature, strain, or solute gradient, at least one boundary pair (in 2D) must be a mirror boundary,
whereat the cells at the edge mirror those in the adjacent layer. If the CA is not sufﬁciently large
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 254 3.11.2008 1:55pm Compositor Name: JGanesan
254 Handbook of Thermal Process Modeling of Steels
compared to the scale of the object that is modeled, this boundary type can lead to signiﬁcant errors
in the simulation output. In a simulation of hot rolling, the variation between surface and centerline
requires mirror boundaries, but mirror boundaries at the front and back of the rolled material may
also have to be used.
The rules for transition are those that govern nucleation and growth, but prior to the application
of these rules, the CA must be initialized by assigning each cell a state corresponding to a
microstructural feature. In two dimensions the CA may be initialized by reading in a digitized
micrograph and mapping each pixel from the micrograph to a cell in the CA. This has the desirable
effect of establishing the length scale of the CA. However, under these circumstances, the model is a
blind model—characteristics other than those that can be photographically imaged are not included
in the simulation. This may be solved by using pixel data output from EBSD software, which can
produce orientation data coupled to spatial coordinates. A further problem is that periodic boundary
conditions cannot be applied to these realworld images. An alternative method to initialize a
simulation is to use a seeding algorithm based on microstructural and texture distribution data,
and this approach is more likely to be useful in the case of simulations of hightemperature
deformation of low carbon steels where the texture and microstructural data will in any case be
derived rather than measured directly.
Where data can be obtained by EBSD, the rules for nucleation and growth of individual texture
classes may be determined and used in the simulation. Again, at hot working temperatures, and
especially where a phase transformation intervenes between the hightemperature test and the room
temperature acquisition of data, this information may not be available directly and the austenite
(a) (b)
FIGURE 6.22 Twodimensional CA neighborhoods: (a) von Neumann and (b) Moore.
FIGURE 6.23 Twodimensional CA simulation of recrystallization using the von Neumann neighborhood.
(From Yazdipour, N., Davies, C.H.J., and Hodgson, P.D., Comp. Methods Mater. Sci., 7, 168, 2007.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 255 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 255
texture must be computed from the ferrite using a known crystallographic relationship for the
transformation; the austenite grain size data must also be obtained indirectly. Textures can be
measured in the austenite phase ﬁeld [61], but this is not common, as it requires a hightemperature
texture goniometer.
Nucleation of a new phase or recrystallized grain is dealt with by comparison with experiment
ally determined nucleation rates. Under warm working conditions, these can be directly related to
the texture and microstructure of the steel [30,62]. Under isothermal conditions, the state of a cell in
the CA is altered to a nucleus based on the probability of occurrence of the event:
P
N
¼
_
Ndt
N
CA
(6:42)
for a nucleation rate,
_
N, in a time interval dt over a CA with a total of N
CA
cells. Under
nonisothermal conditions, the nucleation rate is temperaturedependent and the Scheil approach is
applied.
Growth of the newly formed grain is governed by the probability:
P
G
¼
_
t
2
t
1
v dt
s
CA
(6:43)
where
v is the interface velocity
s
CA
is the size of a CA cell
The velocity can be texture componentdependent and anisotropic, if the experimental data are
available for such calibration. Growth velocities are calculated using the Cahn–Hagel approach
[63,64], originally developed for austenite decomposition and taken further by Jensen for individual
texture classes [65], or by measuring the largest grain during interrupted experiments. The interface
migration rate, v
i
, of the recrystallizing grains (for each deﬁned texture class, i) is given by
v
i
¼
1
S
i
dX
i
dt
(6:44)
where
S
i
is the interfacial area density separating the new recrystallized grains from the unrecrys
tallized grains
i represents the texture class under consideration
The interfacial area S
i
can be deduced from line intercept measurements using the standard Saltykov
stereological relation such that
S
i
¼ 2N
i
L
(6:45)
where N
i
L
is the number of times per unit length a test line intersects a recrystallized=unrecrystallized
boundary. For greater accuracy, dX
i
=dt is determined from
dX
i
dt
¼
(1 ÀX
i
) ln
1
1 ÀX
i
_ _
t
m
0
(6:46)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 256 3.11.2008 1:55pm Compositor Name: JGanesan
256 Handbook of Thermal Process Modeling of Steels
where m
0
is the instantaneous slope of the log{ln(1=[1ÀX
i
])} vs. log(t) plot [64,66]. As a
consequence the nucleation rate is also determined.
Cells in the neighborhood of a nucleus are captured based on the probability of growth. The
state of each cell is stored so that only the state of cells in the previous time step (t À1) inﬂuences
the state of cells in the current time step, t.
A more broadly applicable schema requires independent calculation of the two components of
the boundary velocity—the boundary mobility, M, and the driving pressure, P—and from these,
calculate the velocity:
v ¼ MP (6:47)
Boundary mobility follows the common relationship for diffusioncontrolled processes:
M ¼ M
0
exp
ÀQ
RT
_ _
(6:48)
where
M
0
is a mobility constant
Q is an apparent activation energy for boundary migration
R and T are the gas constant and temperature, respectively
The average boundary mobility may be determined fromknown values of M
0
and Q(see, for example,
Ref. [67]), but this neglects the effect of orientation on mobility and the sensitivity of these parameters
to composition. However, there is the potential to use simulations in an inverse manner in order to
test the orientation dependence of boundary mobility for a limited number of orientation classes.
The driving pressure is determined from the dislocation density, r, the shear modulus, m, and
the Burger’s vector, b:
P ¼
1
2
rmb
2
(6:49)
The temperature dependence of shear modulus and the Burger’s vector for both austenite and ferrite
are collected in Ref. [68]. Including the dislocation density in the simulation results in a more
ﬂexible simulation, the local stored energy can be computed (Figure 6.24) via a dislocation density
constitutive equation, and included in the nucleation criterion. As with boundary mobility, the
simulation can be used to test nucleation criteria by comparing the resultant simulated textures with
x

a
x
i
s
y

a
x
i
s
D
i
s
l
o
c
a
t
i
o
n
d
e
n
s
i
t
y
y

a
x
i
s
D
i
s
l
o
c
a
t
i
o
n
d
e
n
s
i
t
y
x

a
x
i
s
D
i
s
l
o
c
a
t
i
o
n
d
e
n
s
i
t
y
x

a
x
i
s
y

a
x
i
s
FIGURE 6.24 Evolution of dislocation density within grains. (From Yazdipour, N., Davies, C.H.J., and
Hodgson, P.D., Comp. Methods Mater. Sci., 7, 168, 2007.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 257 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 257
experimental data. Furthermore, it is via dislocation density that the simulations can be coupled to
deformation simulations [69]. Functions for particle nucleation and particle growth or dissolution
can also be included in the CA provided that empirical relationships are available.
To our knowledge, there is no commercial CA software for solidstate transformations, whereas
a commercial package for solidiﬁcation exists [70], and this presents a limitation for the interested
researcher. In order to get to the ﬁrst working simulation, most researchers will have to start coding
from scratch. Using a programming environment such as Matlab* or Octave
y
will generally reduce
this time, and have the advantage of many builtin visualization functions for the output and many
graphical user interface functions for the input. The CA and MC simulations in this chapter were
coded using Matlab.
6.8.2 MONTE CARLO POTTS SIMULATIONS
CA simulations are reasonably efﬁcient because once a cell has been transformed, and as long as the
thermomechanical history is monotonic, it is no longer involved in the simulation. Cells at the center
of recrystallized grains or transformed phases are tagged as such, and their neighborhood need not
be inspected. In contrast to this approach, wherein calculations are only required at the boundary of
a growing grain or second phase, the basic MC approach selects sites at random from a lattice
without reference to the microstructure that is being represented. Thus, as grains grow and the ratio
of calculation points at the grain centers to those at grain boundaries increases, the CA examines
fewer sites per simulation time step, whereas the MC simulation requires more calculations per
successful orientation change. In practice, the decrease in efﬁciency can be mitigated by a variation
to the algorithm that weights attempted reorientations in favor of boundaries [71].
Although the terminology is slightly different, the lattice onto which the microstructure is
superimposed is established in much the same manner as the CA array. A triangular lattice (in 2D)
is most commonly employed (see Figure 6.25), but square lattices are sometimes used (see Figure
6.26). Using a triangular lattice results in a six cell neighborhood, and has the advantage that cell
centers are equidistant. However, whereas for a square lattice the storage array dimensions can be
used to deﬁne the grid points of the lattice, the coordinates of the triangular lattice must be
computed. For an array of lattice points of unit centertocenter spacing, on a Cartesian (x, y)
grid, for unit spacing in the x axis, the y coordinate spacing is
ﬃﬃﬃ
3
p
=2. Successive grid points in the
x direction are displaced by onehalf. The lattice has a strong effect on the outcome of the
simulation, with the potential for undesirable (and nonrepresentative) simulation artifacts such as
the pinning of grains [72] (and see Figure 6.26d) and growth anisotropy [71]. The basic algorithm
runs as follows.
A site is selected at random from the lattice, and the state of this site is changed to a new state
selected from among all possible states allowed in the simulation. The number of possible states
from which the selection is made can be varied, but a number greater than 36 is preferred [73]; most
commonly used is 48 orientations, as this represents a reasonable tradeoff between the need for like
grains to be separated by a grain of different orientation and the number of successful ﬂips per
iteration (which affects the rate at which the simulation progresses). Following this provisional state
change, the change in energy associated with the change in state is computed. There are several
forms of equations used for the energy of a site on the lattice [72,74]. One form of the interaction
equation is
E ¼ ÀJ
z
j¼1
d
s
i
s
j
À1
_ _
þH
s
i
(6:50)
* Matlab is a trademark of The Mathworks Inc; 1994–2007 The Mathworks, Inc.
y
Octave is distributed under the GNU General Public License; www.octave.org.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 258 3.11.2008 1:55pm Compositor Name: JGanesan
258 Handbook of Thermal Process Modeling of Steels
where
J is the boundary energy
z is the coordination number of the ith site
s
i
is the orientation of the site being examined
s
j
is the orientation of a neighboring site
H
s
i
is the volumetric energy
The d
s
i
s
j
term is the Kronecker delta, which takes a value of 1 for s
i
¼s
j
and 0 for s
i
6¼ s
j
. The effect
of this is that for sites whose orientation is the same (i.e., sites that lie within the same grain or
phase) the summation is not incremented.
An alternative equation that is perhaps easier to conceptualize is
E ¼
z
j¼1
g
s
i
s
j
þH
s
i
(6:51)
where g
s
i
s
j
is the scalable boundary energy between neighboring sites on the lattice, and the other
terms have the same meaning as before. For adjacent sites within a grain (i.e., sites of the same
orientation) g
s
i
s
j
is zero, whereas for sites of different orientation, g
s
i
s
j
is a unit of boundary energy.
More detailed explanatory examples of the energy calculations can be found elsewhere [71,72].
The volumetric energy component can be either the stored energy of deformation [74] or the
chemical free energy of the lattice site [36,72]. These parameters can be computed using analytical
equations and stored with the lattice site information.
(a)
(c)
(b)
FIGURE 6.25 Microstructures for single runs of MC simulation initialized with 300 grains on a 200 Â200
triangular lattice: (a) 100 MCS; (b) 10,000 MCS; and (c) origin of grains at 10,000 MCS.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 259 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 259
A decrease in the total boundary energy is required for the change of state to be assured, and the
probabilities that a change will be accepted are generally given as [71]
p(DE) ¼
1 DE 0
exp (ÀDE=kT) DE > 0
_
(6:52)
where for DE>0 a random number is generated and compared to the expression to determine
whether or not the state change is accepted. It is observed here that the temperature, T, is the lattice
temperature and is not the same as a measured temperature. It is not readily scaled a priori to the
physical temperature of the system. In practice it is often the case (e.g., Refs. [71,75,76]) that the
probabilities are reduced to 1 for DE 0 and 0 for DE>0.
After each evaluation, the simulation time is incremented. One unit of time in the simulation
(MCS: Monte Carlo step) has elapsed when the number of attempted state changes is equal to the
number of lattice sites. The resultant simulations produce realistic microstructures (Figure 6.25) and
can be interrogated for grain size distributions (Figure 6.27), topological information, and kinetics.
MC simulations have been shown to reproduce the topological phenomena observed experimentally
[73] and have been used to verify related phenomena such as particle effects [49].
Despite the ﬁdelity of the topological characteristics of the simulations, one of the main
limitations of MC simulations is the difﬁculty in connecting the length and timescale of the
simulation to experimental data. Saito and Enomoto [33] scaled MC time with the boundary
(a) (b)
(c) (d)
FIGURE 6.26 Microstructures for single runs of MC simulation initialized with 300 grains on a 200 Â200
square lattice: (a) 1,000 MCS; (b) 10,000 MCS; (c) origin of grains at 10,000 MCS; and (d) immobile
boundary.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 260 3.11.2008 1:55pm Compositor Name: JGanesan
260 Handbook of Thermal Process Modeling of Steels
diffusion coefﬁcient for iron, and in simulations of recrystallization Radhakrishnan et al. [74]
directly linked physical boundary energies to the boundary energy term in Equation 6.49 and the
probabilities to boundary misorientation in Equation 6.50. Baudin et al. [77] used experimental data
to initialize their simulation of recrystallization in Fe3%Si sheet. Raabe [78] also shows that MC
simulations can be scaled by boundary mobilities. However, calibration of one or both of timescale
[74] or energy [77] has generally proved elusive. Hybrid CA=MC models [41] would seem to
capture the best features of both approaches. While some very sophisticated MC packages exist in
the research domain [79], as with CA packages there is a dearth of commercially available software.
6.8.3 PHASE FIELD SIMULATIONS
In contrast to the sharp interfaces enforced by the CA and MC approaches, the interfaces in a phase
ﬁeld simulation are diffuse. This is because instead of an integer spin (=orientation=state) variable
the phase ﬁeld approach ascribes an order parameter that is, for example, varied between À1
and þ1, and which can vary continuously across an interface [80,81]. In modeling the austenite–
ferrite transformation, the austenite might be represented by an order parameter of À1, the ferrite
by þ1, and the interface by a continuously varying parameter across the thickness. While in most
CA and MC simulations the boundary is deﬁned by a change in orientation, and as a consequence is
not explicitly modeled, the effect of the order parameter is to elevate the importance of the boundary
in the problem being solved. This has advantages where soft impingement is to be modeled. Phase
ﬁeld simulations are becoming more common with the advent of commercial software [82] that
signiﬁcantly reduces the leadtime to the ﬁrst simulation compared to CA and MC simulations. An
added advantage is that phase ﬁeld software can be readily coupled with common commercial
thermodynamic software [44,83].
6.8.4 COUPLED SIMULATIONS
Most metallurgical processing involves more than one physical phenomenon occurring either
concurrently or sequentially. It would be ideal to be able to simulate thermomechanical processing
10,000 MCS
0.14
0.12
0.1
0.08
0.06
0.04
0.02
−1 −0.5 0
log (R/R
m
)
F
r
e
q
u
e
n
c
y
0.5 1
0
100 MCS
FIGURE 6.27 Grain size distributions for single runs of MC simulation initialized with 300 grains on a
200 Â200 triangular lattice.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 261 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 261
where deformation, recovery, recrystallization, and phase transformation are all coupled. Simula
tions that couple ﬁnite element or crystal plasticity output with either MC or CA simulations of
recrystallization are common [74,78,84], and some have been extended to include neural network
routines [85], though none of these is applied to steel metallurgy. Similarly, deformation models
have been coupled to phase transformation simulations [86]. But while in principle the component
models should be able to be coupled, the exercise is controlled as much by software design as by the
models themselves and would beneﬁt from the interdisciplinary approach exempliﬁed by Abbod
et al. [85], and the forethought apparent in the work of the group at Aachen [69]. The objectoriented
approach to programming seems to have advantages here.
6.8.5 COMMON PROBLEMS
Similar problems are encountered whichever simulation approach is adopted. For recrystallization
and phase transformations, current models for the rate of nucleation rely heavily on empirically
determined relationships, but for the purposes of the mesoscale simulations discussed here, these
relationships may be sufﬁcient. In the case of warm working, the nuclei of recrystallized grains can
be distributed based on the texture of the deformed structure [30]. More detailed nucleation models
require atomistic simulations, but implementation of atomistic simulation is not straightforward, and
the time and cost of this must be balanced against the beneﬁt of improved information to warrant
their inclusion in our mesoscale simulations.
Boundary mobility data are not readily available at hot and warm working temperatures, and
average values must be relied on. Again, these averages may be sufﬁcient for the simulation, but in
this case it is more likely that improvements will be made with additional detail. As has been
determined for aluminum and copper, the variation of mobility with boundary misorientation is not
continuous and mobility data based on classes of orientation relationships—high mobility bound
aries, low mobility boundaries, and average mobility boundaries—are likely to prove sufﬁcient.
Initializing the model also requires characterization of the elevated temperature microstructure,
and different approaches may have to be taken in the different temperature regimes. Whichever
approach is used–direct input of 2D or 3D experimental data or the construction of a virtual
microstructure using statistical data—there are problems that have yet to be fully resolved. Direct
input of data enforces nonperiodic boundary conditions on the simulation, and this may mean that
larger areas (and arrays) must be considered with attendant increases in simulation times. It is also
assumed that the data input into the simulation are clean, but in reality it is known that not all pixels
are indexed in EBSD, and that the proportion of unindexed or misindexed points increases with
deformation. Thus, the data must be cleaned either prior to input or by the simulation code. On the
other hand, it is difﬁcult to approximate the correct texture and boundary misorientation distribution
using the statistical approach. The effect of each of these shortcomings must be quantiﬁed in terms
of a sensitivity analysis.
Finally, as is the case with all statistical simulations, several runs are required if the result of the
simulation is to be used quantitatively to describe the microstructural evolution or qualitatively to
verify microstructural laws.
REFERENCES
1. Doherty R.D., Hughes D.A., Humphreys F.J., Jonas J.J., Jensen D.J., Kassner M.E., King W.E.,
McNelley T.R., McQueen H.J., and Rollett A.D. Materials Science and Engineering A 1997; A238:219.
2. Poliak E.I. and Jonas J.J. Acta Metallurgica et Materialia 1996; 44:127.
3. Poliak E.I. and Jonas J.J. ISIJ International 2003; 43:684.
4. Mecking H. and Kocks F. Acta Metallurgica 1981; 29:1865.
5. Estrin Y. and Mecking H. Acta Metallurgica 1984; 32:57.
6. Petkovic R.A., Luton M.J., and Jonas J.J. Canadian Metallurgical Quarterly 1975; 14:137.
7. Jonas J.J. Steel Research International 2005; 76:392.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 262 3.11.2008 1:55pm Compositor Name: JGanesan
262 Handbook of Thermal Process Modeling of Steels
8. Boratto F., Barbosa R., Yue S., and Jonas J.J. International Conference on Physical Metallurgy of
Thermomechanical Processing of Steels and Other Metals, Tamura I., Ed., vol. 1. Tokyo, Japan: The
Iron and Steel Institute of Japan, 1988, p. 383.
9. Pickering F.B. Physical Metallurgy and the Design of Steels. London, UK: Applied Science Publishers
Ltd., 1978, p. 275.
10. Cuddy L.J. International Conference on Thermomechanical Processing of Microalloyed Austenite,
DeArdo A.J., Ratz G.A., and Wray P.J., Eds. Pittsburgh, P.A.: The Metallurgical Society of AIME,
1981, p. 129.
11. Sellars C.M. Hot Working and Forming Processes, Davies G, and Sellars CM, Eds. London: The Metals
Society, 1980, p. 3.
12. DoghanManshadi A., Jonas J.J., Hodgson P.D., and Barnett M.R. ISTJ International 2007; 48:1478.
13. Cartmill M.R., Barnett M.R., Zahiri S., and Hodgson P.D. ISIJ International 2005; 45:1903.
14. Zahiri S.H., Byon S.M., Kim S.I., Lee Y., and Hodgson P.D. ISIJ International 2004; 44:1918.
15. Hodgson P.D., Zahiri S.H., and Whale J.J. ISIJ International 2004; 44:1224.
16. Hodgson P.D., Jonas J.J., and Yue S. Materials Science Forum 1993; 113–115:473.
17. Hodgson P.D. Materials Forum 1993; 17:403.
18. Hodgson P.D. and Gibbs R.K. ISIJ International 1992; 32:1329.
19. Dutta B., Valdes E., and Sellars C.M. Acta Metallurgica and Materialia 1992; 40:653.
20. Dutta B. and Sellars C.M. Materials Science and Technology 1987; 3:197.
21. Pussegoda L.N, Hodgson P.D., and Jonas J.J. Materials Science and Technology 1991; 7:129.
22. Poliak E.I and Jonas J.J. ISIJ International 2003; 43:692.
23. Poliak E.I and Jonas J.J. ISIJ International 2004; 44:1874.
24. Sandstrom R. and Lagneborg R. Acta Metallurgica 1975; 23:387.
25. Roucoules C., Pietrzyk M., and Hodgson P.D. Materials Science and Engineering 2003; A339:1.
26. Burke J.E and Turnbull D. Progress in Metal Physics 1952; 3:220.
27. Maccagno T., Jonas J.J, and Hodgson P.D. (1996) II, pp. 36:720–728, The Iron and Steel Institute of Japan,
Japan [CN]. ISIJ International 1996; 36:720.
28. Militzer M. ISIJ International 2007; 47:1.
29. Ouchi C., Sampei T., and Kozasu I. Transactions ISIJ 1982; 22:214.
30. Barnett M.R. and Jonas J.J. ISIJ International 1999; 39:856.
31. Lee H.W., Kwon H.C., Im Y.T., Hodgson P.D., and Zahiri S.H. ISIJ International 2005; 45:706.
32. Sellars C.M. Materials Science and Technology 1990; 6:1072.
33. Saito Y. and Enomoto M. ISIJ International 1992; 32:267.
34. Choi S.H. and Cho J.H. Materials Science and Engineering A 2005; 405:86.
35. Okuda K. and Rollett A.D. Computational Material Science 2005; 34:264.
36. Tong M., Li D, and Li Y. Acta Materialia 2004; 52:1155.
37. Das S., Palmiere E.J., and Howard I.C. Materials Science Forum 2004; 467–470:623.
38. Qian M. and Guo Z.X. Materials Science and Engineering A 2004; 365:180.
39. Raabe D. and Hantcherli L. Computational Material Science 2005; 34:299.
40. Lan Y.J., Li D.Z, and Li Y.Y. Computational Material Science 2005; 32:147.
41. Rollett A.D. and Raabe D. Computational Material Science 2001; 21:69.
42. Meccozzi M.G., Sietsma J., Zwaag S.vd., Apel M., Schaffnit P., and Steinbach J. Metallurgical and
Materials Transactions A 2005; 36:2327.
43. Huang C.J., Browne D.J., and McFadden S. Acta Materialia 2006; 54:11.
44. Nakajima K., Apel M., and Steinbach I. Acta Materialia 2006; 54:3665.
45. Militzer M., Mecozzi M.G., Sietsma J., Zwaag Svd. Acta Materialia 2006; 54:3961.
46. Humphreys F.J. Materials Science and Technology 1992; 8:135.
47. Nakashima N., Nagai T., and Kawasaki K. Journal of Statistical Physics 1989; 57:759–787.
48. Sinclair C.W., Weygand D., Lepinoux J., and Brechet Y. Materials Science Forum 2004; 467–470:671.
49. Harun A., Holm E.A., Clode M.P., and Miodownik M. Acta Materialia 2006; 54:3261.
50. http:==www.susqu.edu=facstaff=b=brakke=evolver= (accessed March 2007).
51. Hesselbarth H.W. and Gobel I.R. Acta Metallurgica and Materialia 1991; 39:2135.
52. Pezzee C.F. and Dunand D.C. Acta Metallurgica and Materialia 1994; 42:1509.
53. Davies C.H.J. Scripta Metallurgica and Materialia 1995; 33:1139.
54. Davies C.H.J. and Hong L. Scripta Materialia 1999; 40:1145.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 263 3.11.2008 1:55pm Compositor Name: JGanesan
Modeling of Hot and Warm Working of Steels 263
55. Rappaz M. and Gandin CA. Acta Metallurgica and Materialia 1993; 41:345.
56. Spittle J.A. and Brown S.G.R. Acta Metallurgica and Materialia 1994; 42:1811.
57. Goetz R.L. and Seetharaman V. Scripta Materialia 1998; 38:405.
58. Davies C.H.J. Scripta Metallurgica and Materialia 1997; 36:35.
59. Janssens K.F. Modelling and Simulation in Materials Science and Engineering 2003; 11:157.
60. Yazdipour N., Davies C.H.J, and Hodgson P.D. Computer Methods in Materials Science 2007; 7:168.
61. Lischewski I. and Gottstein G. SolidtoSolid Phase Transformations in Inorganic Materials, Howe JM,
Ed., vol. 2. Warrendale, PA: TMS, 2005, p. 577.
62. Barnett M.R. ISIJ International 1998; 38:78.
63. Cahn J.W. and Hagel W.C. Decomposition of Austenite by Diffusional Processes, Zackey Z.D. and
Aaronson H.I., Ed., vol. 1. New York: Interscience Publication, 1960, p. 131.
64. Cahn J.W. and Hagel W.C. Acta Metallurgica 1963; 11:561.
65. Jensen D.J. Scripta Metallurgica and Materialia 1992; 27:533.
66. Gokhale A.M. and DeHoff R.T. Metallurgical Transactions A 1985; 16:559.
67. Sun W.P. and Hawbolt E.B. ISIJ International 1995; 35:909.
68. Militzer M., Sun W.P., and Jonas J.J. Acta Metallurgica and Materialia 1994; 42:133.
69. Mukhopadhyay P., Loeck M., and Gottstein G. Acta Materialia 2007; 55:551.
70. http:==www.esigroup.com=SimulationSoftware=Die_Casting_Solution=Products=Continuous_Casting=
index_html (accessed March 2007).
71. Rollett A.D., Manohar P., Roters, F., Barlat, F., and Chen, L.Q. Continuum Scale Simulations of
Engineering Materials: Fundamentals, Microstructures, Process Applications, Raabe D., Roters F., Barlat
F, Chen L.Q, Eds. Weinheim: V.C.H. Verlag, GmbH & Co. Wiley, 2004, p. 77.
72. Holm E.A. and Battaile C.B. JOM 2001; 53:20.
73. Srolovitz D.J., Anderson M.P., Sahni P.S., and Grest G.S. Acta Metallurgica 1984; 32:793.
74. Radhakrishnan B., Sarma G.B., and Zacharia T. Acta Materialia 1998; 46:4415.
75. Gao J. and Thompson R.G. Acta Materialia 1996; 44:4565.
76. Srolovitz D.J., Grest G.S., Anderson M.P., and Rollett A.D. Acta Metallurgica 1988; 36:2115.
77. Baudin T., Juillard F., Paillard P., and Penelle R. Scripta Materialia 2000; 43:63.
78. Raabe D. Acta Materialia 2000; 48:1617.
79. Ivasishin O.M., Shevchenko S.V, Vasiliev N.L., and Semiatin S.L. Materials Science and Engineering A
2006; 433:216.
80. Chen Q. and Wang Y. JOM 1996; 48:13.
81. Mecozzi M.G., Sietsma J, Zwaag Svd, Apel M, Schaffnit P, and Steinbach I. Metallurgical and Materials
Transactions A 2005; 36.
82. http:==www.micress.de.
83. Grafe U., Böttger B., Tiaden J., and Fries S.G. Scripta Materialia 2000; 42:1179.
84. Raabe D. and Becker R. Modelling and Simulation in Materials Science and Engineering 2000; 8:445.
85. Abbod M.F., Howard I.C., Linkens D.A., and Mahfouf M. Sixth International Conference on Computa
tional Science, Heidelberg, Germany: Springer Verlag, 2006, p. 993.
86. Lan Y.J., Xiao N.M., Li D.Z., and Li Y.Y. Acta Materialia 2000; 53:991.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C006 Final Proof page 264 3.11.2008 1:55pm Compositor Name: JGanesan
264 Handbook of Thermal Process Modeling of Steels
7
Modeling of Casting
Mario Rosso
CONTENTS
7.1 Introduction........................................................................................................................ 265
7.1.1 Simulation of Casting Processes: General Considerations ..................................... 265
7.1.2 Reasons for and Potential of Simulation of Casting Processes .............................. 268
7.1.3 Analytical Modeling and Numerical Methods........................................................ 269
7.1.3.1 Analytical Modeling.................................................................................. 269
7.1.3.2 Numerical Methods ................................................................................... 270
7.2 Solidiﬁcation Mechanisms................................................................................................. 273
7.2.1 Homogeneous Nucleation....................................................................................... 273
7.2.2 Heterogeneous Nucleation ...................................................................................... 275
7.2.3 Growth from the Melt ............................................................................................. 275
7.2.4 Dendrite Growth ..................................................................................................... 277
7.2.5 Structure and Defects in Ingots............................................................................... 280
7.2.6 Structure and Defects in Continuous Castings ....................................................... 285
7.2.7 Approach to Filling and Risers ............................................................................... 287
7.3 Industrial Casting of Steel Ingots and Billets .................................................................... 297
7.3.1 Continuous Casting................................................................................................. 297
7.3.2 Ingot Casting........................................................................................................... 300
7.4 Outlook for Modeling of Casting ...................................................................................... 301
7.4.1 MoldFilling Calculations ....................................................................................... 302
7.4.2 Boundary Conditions .............................................................................................. 305
7.4.3 Dimensionless Analysis .......................................................................................... 305
7.4.4 Predicting Stress Distribution ................................................................................. 307
7.4.5 Modeling of Microstructure .................................................................................... 307
References ..................................................................................................................................... 310
Further Reading ............................................................................................................................ 311
7.1 INTRODUCTION
7.1.1 SIMULATION OF CASTING PROCESSES: GENERAL CONSIDERATIONS
The properties of castings are determined by a number of metallurgical and microstructural
characteristics that have their origin along the whole chain of processes that are involved in
manufacturing. They start from alloy formation to its treatments, handling and transportation,
pouring and solidiﬁcation, and the ﬁnal heat treatments.
Initially, the attention of numerical computation has focused toward solidiﬁcation mechanisms
and the development of algorithms for microstructure prediction. With time the tendency to model
the insurgence and growth of numerous types of defects has been grown.
As an example, it is known that thermomechanical treatments, such as hot isostatic pressing,
by substantially reducing or deleting microcavities in castings, determine a general increase in fatigue
resistance of the components, without any variation in microstructure. Moreover, the better mechanical
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 265 7.11.2008 6:21pm Compositor Name: BMani
265
properties of forged components are to be put in relation to the partial or full deactivation, during the
plastic deformation, of the defects generated in the primary casting processes of those components.
The aim of the numerical simulation of a casting process will be that of evaluating
the generation of solidiﬁcation structures, as they will be determined in the cooling history, and
at the same time to take into consideration all those factors of danger linked, at each point of the
casting, with the probability of presence of different kinds of defects. Only this way it will be
possible to evaluate (more qualitatively than quantitatively) the ﬁnal mechanical property.
The following is a list of defects that can be present in castings:
.
Surface oxidation of the melting alloy and entrapment of these oxide ﬁlms under the free
surface; this problem is particularly vast, including the process chain beginning from alloy
preparation, to holding and treatments, to pouring.
.
Entrapment of air and gases during pouring; critical points for this problem are the
turbulence of the melt front during falling from the ladle, the mold ﬁlling, and the formation
of gas bubbles from the cores after they are submerged by the liquid.
.
Generation of cold laps and cold joints; in other terms, the discontinuity of the metal
matrix by the formation of large surfaces of oxides. These problems are generally
connected with the occurrence of stationary phases in the advance of the melt front
during the mold ﬁlling or in the joining of two highly oxidized melt fronts (and at low
temperature) behind a core.
.
Generation of micro and macro porosity and nucleation of solute gases. These defects
are generally due to many factors, among which are the solidiﬁcation shrinkage and the
presence of sufﬁciently high levels of hydrogen in solution (where its concentration can be
much higher than the mean by the presence of segregation phenomena).
.
Concentration of defects such as coarse microstructure and shrinkage porosity owing
to convective ﬂows during solidiﬁcation; this problem is essentially related to the unstable
condition of the melt during cooling if the colder metal is in the upper part of the mold
and the hotter in the lower one (this often being determined by the particular casting
system adopted).
.
Generation of hot tears, due to stresses coming from excessive boundary restraints during
cooling.
.
Generation of residual stresses from solidiﬁcation and heat treatments.
For each one of these types of potential defects, the numerical simulation helps in deﬁning the risk
level and designing processes that can avoid such defects as far as possible.
On the other hand, not every one of the underlined phenomena have been developed with the
same level of accuracy in the commercial codes: this is determined by the inherent complexity of
some of these, such as the oxidation of the free surface and the entrapment of oxides.
Another remarkable reason for the complexity of the matter is the correlation existing between
the presence of defects (of different shape, size, and distribution) and the resulting decrease in
mechanical properties. As an example, the yield strength generally decreases in direct proportion
with the presence of the volumetric presence of porosity of any shape, while fatigue strength is
sensitive also to shape and dimension of the defects present in a particular region of the casting. This
correlation is still highly investigated for all the classes of materials both at an experimental level
and by means of a special area of the numerical simulations, this being the simulation of plastic
strains at a microstructure level.
Once the myth of the predictability of the mechanical property has been put aside (with some
exception, maybe, for hardness, which on the other hand is the less interesting property), it remains a
fact that the numerical simulation can reproduce a large spectrum of phenomena, both physical and
chemical, that would be hard to deal with or even to understand in the casting design process.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 266 7.11.2008 6:21pm Compositor Name: BMani
266 Handbook of Thermal Process Modeling of Steels
But before analyzing the reason for which the numerical simulation is ever more adopted in the
analysis of the casting processes, it is necessary to clearly set a very simple issue, so simple as
critical, for the success of this decision: the quality of information obtained from simulations is
directly determined by the quality of the input data. The input of the simulation is composed by the
following categories:
1. Geometry of casting and molds
2. Physical characteristics of the alloy and the molds
3. Kinematic boundary conditions
4. Thermal boundary conditions
The ﬁrst category of information is not particularly critical, because the mold geometry is known
a priori and often it is transferred directly from the solidmodeling codes used for component
design. On the other hand, a very delicate problem is that with increasing the geometrical
complexity, the number of elements on which the computation is done correspondingly increases
and, as a consequence, the dimension of the solution matrix and the time for the solution. It is not
rare that a simulation running on a performant’s workstation can last more than a week. However,
these situations are normally avoided, because the time needed to obtain useful information is
much less. In any case, the strategies for the simulation run must be adapted in relation to the
level of accuracy desired for the solution. The second category of information that must be given
is that of the physical characteristics of the alloy and of the molds. To this aim, the code databases
have all the necessary values for many industrial standards, so that this phase of data entry is not
usually seen as a problem.
Moreover, it must be noted that the characteristics adopted for the alloy and the mold can be
quite different from the real ones. Often, when the alloy is missing from the database, there is the
tendency to adopt the nearest alloy to the one effectively used in the production, thus generating a
systematic error of unknown proportion on the ﬁnal results. Quite curiously, there is the tendency
to forget this error, most probably because the code fairly gives a solution anyway, so that the time
needed to ﬁnd better input data is delayed to the next project.
Another parameter is the critical fraction of solid. It represents the fraction of solid within
the solidifying alloy after which the material loses its feeding ability. This number, of vital
importance for determining the shrinkage porosity, is normally suggested by code developers for
classes of material, but the reader must be aware of the fact that it can considerably vary as a
function of the alloying elements, of the particular treatment performed on the melt, and in some
way also of the casting section. For this reason it is always better to experimentally determine
this critical factor, and may be others, at least by adjusting it effectively on the shrinkage porosity
found in the casting.
The third category of information is that of the particular mode in which the melt is transferred into
the mold. The complex shape of the metal as it falls from the ladle into the basin is always, for sake of
simplicity, replaced by a simple vertical ﬂow of constant section and with a ﬂowrate equal to the mean
ﬂow rate throughout the ﬁlling phase. The simplifying assumptions determined by this procedure are
numerous and would need a separate section. Here it is only mentioned that phenomena such as free
surface oxidation and air entrapment can be analyzed only with accurate ﬁlling analysis, by reprodu
cing the inlet behavior in the most precise way. Thus, the widely adopted simplifying assumptions
mentioned above prevent the process designer from any critical evaluation on these issues.
This does not mean, however, that the ﬁlling analysis is not of use in the way it is performed
usually; the uneven temperature distribution of the alloy within the mold at the end of the ﬁlling is a
highly valuable information with respect to the constant temperature boundary condition one has to
set when the ﬁlling phase is avoided. Thus, the accuracy of the shrinkage porosity prediction is
much higher if the ﬁlling analysis is performed. The fourth relevant factor for precision of the
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 267 7.11.2008 6:21pm Compositor Name: BMani
Modeling of Casting 267
numerical simulation is that of the boundary conditions of heat transfer. Essentially, the problems
are the mold temperature and the heat transfer coefﬁcients (HTC).
As concerns the temperature distribution within the mold, it must always be kept in mind that
in permanent mold castings there is a considerable change from the initial preheating condition and
the actual temperature distribution after some parts have been produced. This different temperature
distribution determines a different cooling history within the casting and, as a consequence, a
different location of the shrinkage porosity. Thus, many casting and solidiﬁcation cycles have to
be performed in the numerical simulation before the tool system reaches a nearly steadystate
condition, which is assumed to be the real one in the production. Only at that point the numerical
simulation can provide some realistic image of what happens inside the mold. It is normal to see
some feeders beginning to work properly only after 10–15 simulation runs.
Once the mold temperature is deﬁned, the HTC have to be set. In this case, it is normal to adopt
the HTC present in the commercial code database. These are always independent from the interface
temperatures and represent a mean behavior of what is really going on: the HTC are very high at the
beginning of cooling, and then they decrease rapidly with time. The enormous difference between
the two facts (the drastic simplicity of constant HTC of the simulation on one side, the extremely
rapid HTC change in reality on the other) could discourage from performing any simulation at all.
On the contrary, the predictions that are made on the localization of shrinkage porosity are most
often surprisingly accurate. This means that an average HTC is normally effective in determining
the heat extraction law in the inner part of the casting, while it may fail in determining the local
cooling conditions on the casting surface, where solidiﬁcation times are underestimated. In general,
the user of a numerical simulator must be aware of the degree of approximation necessarily involved
in a numerical analysis.
If the particular process under analysis is for some reason different from those available in the
database, such as copper inserts, or aluminum chills, or others, an experimental determination of
the HTC is the only way to get speciﬁc data that no commercial code can give and guaranteeing the
precision of the simulation.
Particular attention should be given to the speciﬁc contraction mode of the casting within the
mold (in particular in permanent molds). In fact, the casting contraction determines the formation of
an air gap between casting and mold, which deeply alters the HTC. The uneven distribution of this
air gap determines preferential ways of heat extraction that can be very different from what is
expected by the classical constant HTC throughout the casting surface.
To this aim, some late releases of commercial codes can deal with such a problem, by modeling
the HTC as a function of the local interface pressure and the formation and size of the air gap
between the casting and the mold. This enhances considerably the precision of the simulation, in
particular in the highpressure die casting processes, where HTC are very high at the beginning due
to the applied pressure, falling drastically near to zero in those regions of the casting–tool interface
where the air gap opens.
7.1.2 REASONS FOR AND THE POTENTIAL OF SIMULATION OF CASTING PROCESSES
Summarizing, the most critical factors for numerical simulation of castings are two:
1. Not all the types of defects of castings are analyzed with the same degree of reliability, so
far. Some types of defects are not dealt with at all, such as oxide formation. This potentially
troublesome situation is likely to become progressively better in the future, by means of
new software developments and the increase in computational speed.
2. Good precision of the solution can be obtained only by means of an equal precision of the input
data; these have to be always considered as a simpliﬁcation of the reality, either for the
difﬁculty in knowing exactly the processing condition, or for the need to simplify the cal
culation when the physics of the problem is far too complicated, such as in HTC evaluation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 268 7.11.2008 6:21pm Compositor Name: BMani
268 Handbook of Thermal Process Modeling of Steels
Nevertheless, the reasons for the numerical simulation remain the same, because the importance of
the solution depends not only on the particular response that is given (the piece is good or not) but
on the degree of consciousness that is determined in the casting designer about the problems
potentially involved in his or her work. In this sense, the numerical analysis, even when it cannot
foresee a problem (being unable to foresee a problem that is later found at the production stage),
gives an essential reference guide for its solution.
This particular usefulness of the numerical simulation is widely accepted by the users, simply on
the base of their experience. Besides, it is known that as an average, 9 simulations on 10 are realistic.
The major risk of the beginner is to ask for perfect solutions, which would never be asked by
any expert.
Once these minimal conditions are accepted for the adoption of a numerical simulation tool, the
advantages and the potential they can offer will clearly emerge in large numbers. It can be summarized
as follows:
.
Production time and costs of the production tools are greatly reduced by the reduction of
setup, scraps, corrections.
.
Yield of the casting is increased by avoiding ineffective risers and by reducing the effective
ones.
.
Production rates are increased, owing to the optimization of the related critical factors such
as maximum casting rates and solidiﬁcation times (by accurately locating chills and cool
ing channels) and by precisely deﬁning the extraction times.
.
Castings of high geometrical complexity can be dealt with, by reducing the time for
optimization and setup.
.
By coupling the solidiﬁcation analysis with the structural analysis of the component,
it is possible to evaluate how critical a defect can be, once it has been found in a particular
region of the casting. Sometimes a defect is accepted if it can be shifted in rather uncritical
zones.
.
If the critical regions of a casting are known, it is possible to develop easier and more
effective check lists at the production control level.
.
More complex castings can be produced; this can be obtained by better knowledge of the
process and of its critical points: The numerical analysis opens new process areas to the
manufacturer by reducing the time and cost of the development.
.
It gives a straightforward way of interpretation of the causes for a scrap. This determines
the general ability to maintain the quality of the production factors under control and to
actuate politics of growing quality.
.
It is a tool that forces the design and production functions to think in quantitative terms on
the quality and on its critical factors; this results in a much greater ﬂux of information
between designers and manufacturers. Thus, designers are aware of the consequences of
their geometrical choices on the product quality. In other words, the numerical simulation
is a powerful tool for developing concurrent engineering processes aimed at product
enhancement and simpliﬁcation.
.
It represents a tool for increasing the culture for quality and helps in training the staff.
7.1.3 ANALYTICAL MODELING AND NUMERICAL METHODS
7.1.3.1 Analytical Modeling
The term analytical modeling refers to the description of a physical phenomenon that is performed
on a certain domain without any approximation of the involved variables. The integration of some
differential equations describing the nature of the phenomenon is done by classical techniques by
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 269 7.11.2008 6:21pm Compositor Name: BMani
Modeling of Casting 269
applying speciﬁc boundary conditions. This approach is limited by the ability to solve such a system
by increasing the complexity of the domain where the solution is looked for and by the complexity
of the boundary conditions.
Typically, the analytical modeling of a solidiﬁcation problem can be easily performed on
elementary domains such as a cylinder or a thin strip, or an inﬁnite halfspace. Moreover, even for
such elementary geometries, the boundary condition must be set in the easiest way, such as constant
interface temperature, constant HTC, predetermined heat transfer laws, and so on.
Thus, analytical modeling is unable to deal with complexity, which is the common feature of
any industrial production process; there, the geometry of the part to be manufactured is always
complex and rich in details and the interaction of the casting with the whole system of tools has
necessarily to be taken into account to achieve convincing results.
Once this intrinsic limitation of analytical modeling has been pointed out, its advantages can
be easily deﬁned. First of all, it is easy and fast and it can give highlevel responses; in other
words, it can predict if a certain phenomenon can occur in a given condition. Secondly,
surprisingly accurate analytical results can be obtained by using just a few terms, and sometimes
only one. Moreover, the analytical solution has the advantage that it will reveal the inﬂuence
of the various experimental and physical parameters directly, whereas a more accurate but
numerical method will not. The analytical solution is extremely ﬂexible and can be applied to
any problem.
Particularly useful analytical descriptions, as collected and developed by Kurz and Fisher [1],
are available for pointing out the effect of the principal processing parameters in the following
cases:
.
Macroscopic heat ﬂux modeling
.
Solute and heat ﬂux calculation related to microstructure formation
.
Local equilibrium at the solid=liquid interface
.
Nucleation kinetics in a pure substance
.
Atomic structure of the solid=liquid interface
.
Thermodynamics of rapid solidiﬁcation
.
Interface stability analysis
.
Diffusion at the dendrite tip
.
Dendrite tip radius and spacing
.
Eutectic growth
.
Transient in solute diffusion
.
Homogenization of interdendritic segregation in the solid state
It can be said, with some approximation, that analytical models can be very useful to students and
researchers, allowing the best comprehension of the physic of the process and to foresee the
effect of changing important variables such as the processing materials or the heat extraction rate;
on the other hand, they fail in giving a quantitative description of complex geometries and, for
this reason, are less used by casting engineers, who must found their work on a more pragmatic
approach, even if it is based on volume discretization and convergence methods, which means
approximation.
7.1.3.2 Numerical Methods
The solution to industrial problems always comes from numerical methods, which can be classiﬁed
into the following categories: Finite difference methods (FDM), ﬁnite volume methods (FVM), and
ﬁnite element methods (FEM).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 270 7.11.2008 6:21pm Compositor Name: BMani
270 Handbook of Thermal Process Modeling of Steels
The FDM uses a structured grid in which the derivatives in the differential equations are
replaced by ﬁnite difference approximations. The accuracy of the expressions is of the second
order. With this method a truncation error arises as the difference between the partial derivative and
its ﬁnite difference representation. The discretized equation is written for each cell or grid point in
the mesh. The resulting set of algebraic equations is then solved for the unknown values of the
dependent variable at each cell or grid point. These unknowns are typically associated with the cells
and are considered to represent the cell average value rather than the value at a particular point.
By integrating the model equation over the control volume, a similar concept can be applied in
the FVM, and the resulting discretization is perfectly similar to the results given by the FDM and the
two methods can be considered as the same. The key feature of the FVM approach is that the FVM
is based on ﬂux integration over the control volume surfaces. The method is implemented in a
manner that ensures local ﬂux conservation, regardless of the grid structure.
FDM and FVM methods both adopt structured cells, which means that the cells have a simple
rectangular shape, they have the same size, and also have the same orientation in the volume domain
of the casting. This fact generally determines a really straightforward and quick discretization of the
geometry under consideration, but, on the other hand, has a lot of consequences:
.
There is some difﬁculty in representing curved surfaces, unless the number of elements is
greatly increased (see Figure 7.1); the more advanced commercial codes, for this reason,
overcome such limitation by deﬁning boundary planes between different meshes (one
coarse, the other ﬁne), and then providing speciﬁc integration routines between the two.
.
Boundary between the casting and the mold is made of cells not parallel to the boundary
(except where the orientation of the boundary coincides with one of the orientation planes
of the mesh). This determines a surface area greater than the effective one (which is
corrected by appropriate routines) and the insurgence, on those locations, of speed vectors
nonparallel to the boundary (see Figure 7.1).
.
Structured mesh somehow affects the accuracy of the solution of the ﬂuid ﬂow during
the moldﬁlling analysis: if a stream of melt arrives orthogonally onto a horizontal plane,
then expanding on it freely in all directions, a noncircular melt pool expansion will be
easily observed.
FDM
Mould
FIGURE 7.1 Example of application of FDM.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 271 7.11.2008 6:21pm Compositor Name: BMani
Modeling of Casting 271
.
Moreover, stress calculations are not supported at all. In order to perform a stress
calculation on the casting, an entirely new ﬁnite element mesh has to be built on the
volume domain.
When applied to a simple onedimensional diffusion equation with constant properties on a regular
grid, the FDM, FVM, and FEM produce almost identical discretized equations. However, when
applied to multidimensional problems with variable properties and grids, the methods in general
give different algebraic equations.
In particular, the following can be considered in the FEM approach:
.
It is somewhat more complex than the FVM in generating the mesh on the volume
domain. This depends to a great extent on the particular commercial code adopted and
the efﬁciency of its mesh generator. Some of them are very effective in building the mesh
on complex geometries. Moreover, often the mesh has already been built by designing
engineers in order to perform structural analysis of the component (except of course the
casting system).
.
It generally better describes the product geometry: this determines a better description of
thin section and complex shapes, a smaller number of elements and nodes (see Figure 7.2)
and, as a consequence, less memory, disk space, and shorter run times.
.
It does not suffer from mesh nonparallel to the mold boundary (see Figure 7.2).
Irrespective of the ﬁnite volume or ﬁnite element formulation of the numerical code used, it must
be kept in mind that a direct correlation exists between the number of nodes and the run time of
the simulation; at the same time, an accurate and convergent solution to a casting problem greatly
depends on the node number. As an example, it is a common rule to adopt at least three elements
(or cells) along the thickness of a thin casting element in order to have a reliable solution in that
particular region. Applied to the common complex geometries of the casting industry, this
determines a high number of nodes and consequently high run simulation times. It is quite
ordinary to observe simulation runs in the order of 2–3 days for ﬁlling and solidiﬁcation analysis
on complex geometries such as an engine head. If stress analysis is also required, the CPU time
will increase further.
Mould
FEM
FIGURE 7.2 Example of application of FEM.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 272 7.11.2008 6:21pm Compositor Name: BMani
272 Handbook of Thermal Process Modeling of Steels
7.2 SOLIDIFICATION MECHANISMS
Solids can be classiﬁed by the placement of the atoms in the space and, particularly, by atoms
periodicity and arrangement. A material can be crystalline, amorphous, or a mixture of them. Short
range order considers the conﬁguration of a single atom and the group of the nearest atoms; if the
basic structure of shortrange order can be translated through some set of welldeﬁned vectors to
construct the entire solid, it is possible to deﬁne also longrange order, typical atoms conﬁguration
for metals. Glassy materials are characterized by the presence of an atomic shortrange order and the
absence of longrange order. There are also materials, the quasicrystals, which possess a periodic
longrange positional order.
Solidiﬁcation is a phase transformation that leads to the formation of crystalline metals and
alloys from liquid upon cooling; it is a thermodynamic process possible in undercooled state and the
driving force is the free energy change. This liquid–solid transition plays an important role for
casting, because it largely determines microstructure.
When solidiﬁcation occurs following the sequence of nucleation and growth. Nucleation during
solidiﬁcation is the formation of crystal from the melt which represents the ﬁrst appearance of the
new phase. It inﬂuences many microstructer properties like grain size, morphology, size of segre
gation, and compositional homogeneity. Crystal growth occurs through the transport of crystal
species from the ﬂuid phase to the crystal interface followed by incorporation of these species into
the crystal. A ﬁrst classiﬁcation may be done between homogeneous and heterogeneous nucleation.
Homogeneous nucleation occurs in a pure liquid without the presence of foreign nucleating
particles, whereas in heterogeneous nucleation the process begins on foreign substances.
7.2.1 HOMOGENEOUS NUCLEATION
Considering a pure metal and deﬁning T
m
as the melting point, the free energy change per unit
volume at fusion temperature is given by
DG
v
¼ G
S
v
ÀG
L
v
¼ H
S
ÀH
L
ÀT
m
(S
S
ÀS
L
) ¼ 0 (7:1)
DG
v
¼ DH
f
ÀT
m
DS ¼ 0 (7:2)
DH
f
¼
DS
T
m
(7:3)
where G
S
v
and G
L
v
, H
S
and H
L
, S
S
and S
L
are the free energy, the enthalpy, and the entropy per unit
volume, of the solid and of the liquid respectively. The enthalpy change DH
f
is the latent heat of
fusion L
v
per unit volume.
Equations 7.1 and 7.2 can be combined:
DG
v
¼
DH
f
(T
m
ÀT)
T
m
¼
DH
f
(DT)
T
m
¼
L
v
DT
T
m
¼ DS
f
DT (7:4)
Therefore, the driving force is proportional to the undercooling DT, provided that the latent heat and
the entropy of fusion do not vary much with the temperature.
The free energy change DG per unit volume to form a solid spherical embryo of radius r from a
pure liquid is given by
DG ¼ DG
v
þDG
i
¼
4pr
3
DH
f
DT
3T
m
þ4pr
2
g
SL
(7:5)
where DG
v
and DG
i
are the change in volume of free energy (which has a negative contribution) and
the interfacial free energy (Figure 7.3) [2]. g
SL
is the solid–liquid interfacial free energy (Figure 7.3).
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 273 7.11.2008 6:21pm Compositor Name: BMani
Modeling of Casting 273
When DG has a maximum, the critical radius r* occurs:
r* ¼
2g
SL
DG
v
¼
2g
SL
T
m
L
v
DT
(7:6)
Figure 7.15 shows the behavior of free energy change of an embryo as a function of the radius, for
T <T
m
; DG
v
is negative due to the metastable liquid, whereas DG
i
increases monotonically.
Clusters larger than r* (nuclei) are stable because they are connected to a free energy decrease
and thus they grow spontaneously, whereas particles smaller than r* (clusters or embryo) lower their
free energy by dissolution of the solid and are unstable.
The activation energy for nucleation of a spherical nucleus of radius r* in a pure melt is given by
DG* ¼ DG
hom
*
¼
16pg
3
SL
3DG
2
v
¼
16pg
3
SL
T
2
m
3L
2
v
DT
2
(7:7)
For metals, the homogeneous nucleation rate, described as the number of nuclei per unit volume per
unit time, is expressed as
_
N ¼ k
2
exp À
k
3
T(DT)
2
_ _
(7:8)
where k
2
(typically 10
42
m
À3
s
À1
) and k
3
are constants. From Equation 7.8 it is clear that
_
N increases
sharply when supercooling reaches a critical range between 0.2T
m
and 0.4T
m
.
Surface
energy
Total
energy
r
*
+
−
Volume
energy
F
r
e
e
e
n
e
r
g
y
Δ
G
0
FIGURE 7.3 Free energy change at T<T
m
. (From Campbell, J., Castings, Butterworth Heinemann Ltd.,
Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 274 7.11.2008 6:21pm Compositor Name: BMani
274 Handbook of Thermal Process Modeling of Steels
Assuming that the equilibrium concentration of critical clusters will not vary, it is possible to
deﬁne the nucleation rate as
_
N ¼ k
4
N
L
exp
DG*
kT
_ _
(7:9)
where
k
4
is a constant that represents a preexponential factor for diffusion
N
L
is the total number of atoms per unit volume in the liquid
7.2.2 HETEROGENEOUS NUCLEATION
Heterogeneous nucleation is the most common crystal formation process, because it requires less
activation energy; in fact the presence of preexisting surfaces or foreign substances such as
impurities, inclusions, and oxide ﬁlms aids nucleation, lowering DG*.
The free energy change during heterogeneous nucleation can be expressed as
DG
het
*
¼ DG
hom
*
f (u) (7:10)
where f(u) is a shape factor, which is connected to the wetting angle between a particle and the
liquid. A good nucleating agent must have a small contact angle between the nucleating particle and
the growing solid.
The heterogeneous nucleation rate can be expressed as
_
N
het
¼ k exp À
f (u)DG
hom
*
þDG
D
kT
_ _
(7:11)
where DG
D
is the activation energy for diffusion in the liquid toward the nuclei across the
solid=liquid interface.
For typical metals
_
N ¼ 10
30
exp
16pg
3
SL
T
2
m
V
2
m
f (u)
3kDH
2
f
TDT
2
_ _
(7:12)
7.2.3 GROWTH FROM THE MELT
This process may generally happen in three different mechanisms:
1. Interface structurecontrolled mechanisms, typically for nonmetallic phases
2. Diffusioncontrolled mechanisms, characteristics of metallic phases
3. Combination of the above mechanisms
In diffusioncontrolled mechanisms, the rate of growth is determined by the ﬂux of atoms. For
metallic materials it may be useful to distinguish between the growth of a pure substance or an alloy;
in a pure metal the solid=liquid interface moves normal to itself by a continuous random atom
attachment (diffuse and rough interfaces), whereas the interface of a metallic alloy appears typically
ﬂat and sharp, with preferentially a lateral growth. Lateral growth takes place in speciﬁc zones, like
twodimensional nucleation, screw dislocations, reentrant corner due to the presence of a twin
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 275 7.11.2008 6:22pm Compositor Name: BMani
Modeling of Casting 275
boundary, and stacking faults intersecting the faceted interface. Lateral growth requires more
supercooling than continuous growth.
Actually, solidiﬁcation of pure metals goes through a series of transitions: initially the solidi
ﬁcation front is planar, then, at higher rates, it prosecutes in cellular mode, and for higher velocities
the cells grow into projections, which are called dendrites.
For the alloys, the three growth forms are still present. Cellular and dendritic growth are
promoted by preferential rejections or incorporation of solute, temperature ﬂuctuation, and mech
anical shocks; these events cause local small morphological perturbation which result in instability
of the S=L interface (Figure 7.4).
The driving force for instability is undercooling, produced by the segregation of alloying
elements ahead of the front (Figure 7.5). This occurrence causes a reduction of the melting point
and, if the reduction is sufﬁcient to reduce the melting point below the actual temperature at that
point, the liquid become locally constitutionally supercooled.
In Figure 7.5 the effects of segregations during freezing are schematized. When the liquidus
temperature reaches T
L
, the melt starts to freeze; the ﬁrst solid has the composition kC
0
, where k
is the partition coefﬁcient and deﬁnes how the solute alloy partitions between the solid and
liquid phases.
k ¼
C
S
C
L
(7:13)
.
When k $1, there is a little tendency toward segregation
.
When k ( 1, there is a strongly partitioning alloying element
If the ﬁrst solid has the composition kC
0
, the alloy remaining in the liquid must be rejected ahead of
the advancing front, and the liquid changes its composition from C
0
to C
0
=k.
Front Side
Solid Liquid
Temperature
T
mp
T
mp
T
mp
T
T
Temperature regime
Plane
growth
Dendrite
growth
Cell
growth
FIGURE 7.4 Transition of growth morphology from planar, to cellular, to dendritic. (From Campbell, J.,
Castings, Butterworth Heinemann Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 276 7.11.2008 6:22pm Compositor Name: BMani
276 Handbook of Thermal Process Modeling of Steels
It is possible to estimate the extension of the solute layer ahead of the front:
d ¼
D
R
(7:14)
where
D is the coefﬁcient of diffusion of the solute in the liquid
R is the rate of the front
Therefore, constitutional undercooling will occur when the temperature gradient G in the liquid is
G À
T
L
ÀT
S
D=R
(7:15)
where T
L
and T
S
are the temperature of the liquid and the temperature of the solid, respectively.
T
L
and T
S
can be eliminated, assuming straight lines on the equilibrium diagram; substituting
C
0
, k, and m, where m is the slope of the liquidus line, an equivalent statement can be obtained:
G À
mC
0
(1 Àk)
kD
(7:16)
The degree of stability of the growth conditions can be estimated by the ratio G=R, which is the
factor that measures the degree of stability of the growth and controls the type of growth front.
7.2.4 DENDRITE GROWTH
Dendritic growth represents one of the three kinds of nonplanar solidiﬁcation fronts that issue from
the increase in the degree of morphological instability; the other kinds of growth are cellular growth
and a mixture of the two.
T
e
m
p
e
r
a
t
u
r
e
kC
0
C
0
C
0
C
0
/k
C
0
/k
T
e
m
p
e
r
a
t
u
r
e
T
L
T
S
Composition
C
o
m
p
o
s
i
t
i
o
n
Distance from freezing point
Effectively
undercooled
zone
Distance
of melt
F
r
e
e
z
i
n
g
p
o
i
n
t
A
c
t
u
a
l
t
e
m
p
e
r
a
t
u
r
e
o
f
m
e
l
t
Soluterich zone built up
ahead of advancing front
as a result of solute rejection
(k < 1) on solidification
FIGURE 7.5 Effect of segregation during freezing of a binary alloy. (From Campbell, J., Castings,
Butterworth Heinemann Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 277 7.11.2008 6:22pm Compositor Name: BMani
Modeling of Casting 277
Fixing an original melt composition C
0
and temperature gradient G
L
, and increasing growth
velocity V, it is possible to show a transition from planar to cellular, from cellular to dendritic, from
dendritic to cellular, and from cellular to planar. Cellular structures have been found at low and at
high growth velocity; the temperature gradient at low velocities and the capillarity effect at high
velocities are responsible for cellular growth.
A dendrite consists of a variable number of series of branches. Its formation is typical for metals
that solidify under small thermal or concentration gradients. Metal ingots, alloy castings, and
weldments commonly show dendritic structures.
Dendritic growth involves two main processes:
1. Steadystate propagation of the tip region, which leads to the formation of the main branch
2. Timedependent crystallization of the other series of side branches
Figure 7.6 shows the nucleation and growth of dendrites inside a casting; a columnar dendrite
nucleated on the mold wall grows forward and sideways, generating secondary arms from the
primary; during solidiﬁcation these arms will bond together forming grains. A grain may be formed
by thousands of dendrites, or, in the extreme, only of a single primary arm.
Grain boundaries are originated from groups of dendrites with different orientation, origin
ated from different nucleation events (Figure 7.7). The formation of a dendrite begins with the
breakdown of an unstable planar solid=liquid interface; the perturbation is ampliﬁed until a
difference in the growth of the tips and depressions. The tip tends to grow more rapidly than
a depression, because it can reject the solute in lateral direction, which tends to be accumulated in
the depressions.
The perturbation adopts the forms of cells, and if the conditions for growth are such that they
lead to dendrite formation, the cell will rapidly change to dendrites. All dendrites grow with a
preferred growth direction, such as (001) axis in cubic materials; this characteristic of anisotropy has
been used in some industrial processes to produce single crystals for solar cell substrates or the
columnar region of cast alloys.
Nucleation
site
Grain size
Primary dendrite arm spacing
Secondary debtrite arm spacing
Mould
wall
FIGURE 7.6 Schematic illustration of the formation of a raft of dendrites to make grains. (FromHiggins, R.A.,
Engineering Metallurgy, 6th edn., Edward Arnold, London, 1998.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 278 7.11.2008 6:22pm Compositor Name: BMani
278 Handbook of Thermal Process Modeling of Steels
An increase in growth rate V decreases the primary dendrite spacing (l
1
) value. For a given
value of temperature gradient in the liquid G
L
, l
1
goes through a maximum as a function of growth
rate or velocity V.
For a given V, l
1
decreases as G
L
is increased at the L–S front. Since l
1
increases linearly
with the square root of the inverse cooling rate 1=j
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
G
L
V
p
j, a relation between l
1
and mass transport
in the liquid is very likely.
There is more than one theoretical model that has been proposed to describe l
1
as a function of
V, G
L
, jG
L
V j, and the alloy characteristics. According to the Hunt model [4], l
1
can be predicted by
l
1
¼
64D
L
G[mC
0
(k
0
À1) Àk
0
G
L
D
L
V
À1
G
2
L
V
_ _
1=4
(7:17)
where
G is Gibbs–Thomson coefﬁcient
D is the solute diffusivity in the liquid
In castings, the grain size, which is related to the primary dendrite arm spacing, is sometimes
important. However, more often the secondary dendrite arm spacing (SDAS or DAS) is the most
important parameter.
As DAS decreases, so the ultimate strength, ductility, and elongation increase. This parameter is
also important for some heat treatments; cast material with ﬁner DAS shows a better response to
homogenizing, with better properties or faster treatments.
DAS is controlled mainly by solidiﬁcation time. Secondary arms form initially very near to the
dendrite tip, presumably due to the presence of noise at the dendrite tip. The actual secondary spacing
developed in a fully solidiﬁed material is coarser than that of the initial spacing; this arises frommelting
or dissolution of the smaller arm at the expense of the larger arms and then from the minimization of
surface energy during the coming closer of side branches to the neighboring dendrites.
(a) (b) (c)
(d)
(e)
FIGURE 7.7 Dendritic growth of metallic crystals from the liquid state [3]. A solid pure metal (d) gives no
hint of its dendritic origin since all atoms are identical, but an impure metal (e) carries the impurities between
the dendritic arms, thus revealing the initial skeleton.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 279 7.11.2008 6:22pm Compositor Name: BMani
Modeling of Casting 279
For all except the most dilute alloys, the timedependent law of coarsening of secondary spacing
l
2
has been developed for directional solidiﬁcation conditions, which is given by
l
2
¼ 5:5(Mt
f
)
1=3
(7:18)
where t
f
is the solidiﬁcation time and M is
M ¼
GD
L
ln
C
f
C
0
_ _
k
0
DT
0
C
f
C
0
À1
(7:19)
where C
f
is the ﬁnal composition of the liquid at the base of the dendrite. For directional
solidiﬁcation, a relation between the local solidiﬁcation time and the magnitude of the cooling
rate jG
L
Vj can be expressed by
t
f
¼
DT
f
jG
L
Vj
(7:20)
where DT
f
is the temperature difference between the tip and the base of the dendrite.
A reduction of l
2
can be obtained by increasing the cooling rate of cast materials; this results in
a reduction in microsegregation and will produce a more easily homogenized casting.
7.2.5 STRUCTURE AND DEFECTS IN INGOTS
Ingots have a larger thickness in at least one transverse direction; the depth of the liquid core is small
and solidiﬁcation occurs in a stationary space (Figure 7.8).
The solidiﬁcation structures observed in conventional ingots are more complicated than in con
tinuously cast products, due to the absence of a steadystate regime. The solidiﬁcation time depends on
the ingot size and can vary from 1 h to several days. The time taken for a given point to cool from the
liquidus temperature to the solidus can be for example 35 h at the center of a 180 t ingot.
It is possible to distinguish three characteristic zones in terms of the size, shape, and orientation
of the grains, corresponding to
1. Chill layer
2. Columnar growth
3. Equiaxed regions
The chill zone is composed of numerous ﬁne, equiaxed grains with random orientations; the
columnar (dendritic) zone is perpendicular to the surface below the chill zone and shows a variable
thickness from the top to the bottom of the ingot. In large steel the columnar grains can reach lengths
of 200 mm or more. The equiaxed region has a great variety of grain sizes and morphologies.
Equiaxed grains accumulate at the bottom of the ingot because of the difference in density between
the liquid and the solid. The SZ zone in Figure 7.8 is the zone with the oldest dendrites, which lost
their structure and become spheroidized; CEZ and H zones are the areas where the grains come in
contact with one other, when the last liquid is ﬁnished.
In metals and alloys that are single phase from the melting point down to room temperature
(austenitic stainless steels, ferritic stainless steels), it is possible to obtain a microstructure of the
ﬁnished casting not too different from that which was present during solidiﬁcation.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 280 7.11.2008 6:22pm Compositor Name: BMani
280 Handbook of Thermal Process Modeling of Steels
Contrarywise lowcarbon and lowalloy steels are materials that show phase changes during
cooling, and the situation can be more complicated, as schematized in Figure 7.9 [2].
Segregation may be deﬁned as any deviation from uniform distribution of the chemical elements
in the alloy. All castings are segregated to different degrees. Macrosegregations develop on the scale
of the entire casting product and are due to convection, whereas microsegregations occur on a
microscopic scale between dendrite arms.
Carbon, nitrogen, and oxygen diffuse readily in the solid already formed and it is easy not to
reach thermodynamic equilibrium. Substitutional alloying elements with lower diffusivities, such as
manganese, chromium, and nickel, show little tendency to segregate, whereas silicon, sulfur, and
phosphorus have low partition coefﬁcients and tend to concentrate in the liquid.
For solutes whose partition coefﬁcients are less than 1 (S and C), their concentrations in the
dendrites formed at ﬁrst are lower than in the alloy as a whole.
In Figure 7.10 the regions in an ingot where the concentration of carbon is less than and greater
than that in the initial liquid are schematically displayed; the sedimentation cone, which is the zone
where the ﬁrst equiaxed dendrites have been arranged, contains less carbon than the theoretical
composition (negative segregation) [5].
Microsegregation can be reduced by a homogenizing heat treatment, but macrosegregations
cannot be removed so simply. Ingots can also show segregations on an intermediate scale, and these
are called mesosegregations. These segregated regions have an elongated shape, with a narrow
dimension of the order of some millimeters.
H
CEZ
FEZ
SZ
(a) (b)
CZ
1
2
3
FIGURE 7.8 Macrostructures and schematic longitudinal section of a 330 kg steel ingot. The total height is
2.05 m and the bottom diameter 0.51 m [4]. (a) Simpliﬁed longitudinal halfsection. The light bands represent
the regions 1, 2, and 3 illustrated by the macrostructures in b. (b) Macrostructures of regions 1, 2, and 3 in A. H
is the head or hot top, CZ is the columnar zone, CEZ is the coarse equiaxed zone, FEZ is the ﬁne equiaxed
zone, and SZ is the region of equiaxed grains in which the dendrites have become spheroidized. (From
DurandCharre, M., Microstructure of Steels and Cast Irons, Springer, New York, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 281 7.11.2008 6:22pm Compositor Name: BMani
Modeling of Casting 281
Solidification of the liquid to
δ ferrite
1
2
3
4
5
6
7
8
End of solidification; sketch of
grain boundaries, segregations
and impurities
Creation of dislocation during
cooling, due to plastic strain
connected to differential
contraction of the casting
Nucleation of austenite grains
on reaching the temperature
for the formation of the γ phase
Start of conversion from austenite to
ferrite, usually from grain
boundaries and corners
Final ferritic microstructure, with
ghost boundaries of the previous
austenite grains
Formation of a new subgrain
structure due to further
cooling strains
End of the conversion from
ferrite to austenite; δ grains are
still discernible because the lines
of segregations
FIGURE 7.9 Phase change during cooling of Molten lowcarbon and lowalloy steels [2].
FIGURE 7.10 Carbon segregation in a steel ingot; the signs ‘‘þ’’, ‘‘À,’’ and ‘‘þþ’’ indicate positive and
negative differences compared to the nominal composition. (From DurandCharre, M. Microstructure of Steels
and Cast Irons, Springer, New York, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 282 7.11.2008 6:22pm Compositor Name: BMani
282 Handbook of Thermal Process Modeling of Steels
Mesosegregations can be distinguished in three classes:
.
Atype segregations (beyond columnar zones)
.
Vtype segregations (in the central equiaxed zone)
.
Axial segregations (along symmetry axis of the cast product)
Channel segregates can be controlled by
.
Increasing the rate of solidiﬁcation, therefore decreasing the time available for their
formation
.
Adjusting the chemical composition of the alloy, to have a soluterich liquid with more
buoyancy at the temperature within the freezing zone
The term inclusion is used to describe an undesired nonmetallic particle in a solidiﬁed product.
Inclusions can be classiﬁed as exogenous and endogenous. Exogenous inclusions are ones that
are simply entrained in the liquid, originating from refractory linings or slag layers. They are
usually avoided by appropriate measures during casting (special pouring conﬁgurations, dams,
skimmers, ﬁlters, etc.). Endogenous inclusions are ones formed by precipitation from the liquid,
due to the presence of impurities, particularly oxygen, but also sulfur, phosphorus, and nitrogen.
Large inclusions can have an extremely detrimental inﬂuence on fatigue strength. Inclusion
contents can be controlled by good melting practice to reduce impurity levels (Figure 7.11).
Pool of soluterich
buoyant liquid
Primary
pipe
Secondary
pipe
Positive
segregation
“A”
channel
segregates
“V”
channel
segregates
Cone of
negative
segregation
Rising plumes of
buoyant liquid
Channels dissolved
through columnar
dendrite zone
Dendrite fragments
falling like snow from
emerging streams
Heap of heavy
equiaxed crystals
(a) (b) Partially solidified ingot Partially solidified ingot
FIGURE 7.11 Development of segregation in a killed steel ingot during (a) solidiﬁcation and in the (b) ﬁnal
ingot. (From Campbell, J., Castings, Butterworth Heinemann Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 283 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 283
Most inclusion phases are less dense than the metal and ﬂoated off into a slag before teeming.
The majority of inclusions are oxides; generally, to reduce the oxygen content, some reducing
agents (aluminum and silicon) are used to form their respective oxides, which are removed
thereafter with slag. Another important impurity is sulfur, which can be controlled by the addition
of elements such as manganese, calcium, and magnesium.
The ductility of plain carbon steel casting is sensitive to the type of sulﬁde inclusions that form.
Generally it is possible to classify this type of inclusions in
.
Type 1 inclusions, with a globular form
.
Type 2 Sulﬁdes, which take the form of thin grain boundary ﬁlms; they seriously embrittle
the steel
.
Type 3 sulﬁdes, with a compact form; they do not seriously damage the properties of the steel
It seems that the change from type 1 to type 2 is brought about by a lowering of the oxygen content.
The major cause of porosity formation is the decrease in volume when liquid transforms to
solid. Holes are formed when pockets of liquid are isolated inside the solid, for example, in the
interdendritic spaces. Small pores, less than about 1 mm in size, are generally harmless. Larger ones
may subsequently be closed during hot working of the cast product.
Many pores are gas bubbles formed by the rejection of certain solute elements from the
liquid during solidiﬁcation. The gases most commonly involved are carbon monoxide, nitrogen,
and hydrogen. The presence of these gases depends on the melting and reﬁning route followed.
The ﬁnal oxygen content depends principally on the additions made during the ﬁnal reﬁning
steps (e.g., Si, Al). Nitrogen is introduced into steel mainly by reaction with air, and its content
can be controlled to a certain extent by liquid reﬁning steps such as gas scavenging, vacuum
treatments, etc. However, nitrogen is sometimes a deliberate alloying addition in a number of
special steel grades. Hydrogen generally has a low solubility in steels, but can be introduced into
the liquid at high temperature from moisture. The general hygrometry of the charge materials,
refractories, and atmosphere is therefore an important factor. All these elements have much lower
solubility in the solid than in the liquid, corresponding to a partition coefﬁcient signiﬁcantly less
than one, so that they are rejected into the liquid during solidiﬁcation. Their concentration in the
liquid rises, and if the saturation point is reached, gas bubbles are formed. In fact, a certain degree
of supersaturation is required for bubble nucleation. The bubble formation and release process can
affect the morphology of the columnar solidiﬁcation front. The bubbles may coalesce, and most of
them eventually rise to the liquid surface. However, some may become trapped in the mushy zone,
or in places where two solidiﬁcation fronts meet, especially in complex castings, leading to
cavities in the solid.
Niyama et al. [6] have correlated porosity to thermal parameters and found that castings were
sound only over a certain limit value of the following function:
N ¼
G
s
ﬃﬃﬃﬃ
T
p
where
N is the value of Niyama function (equals to 1 for steel and 0.75 for cast iron)
G
s
is the thermal gradient (8C=cm)
T is the cooling rate (8C=s)
Conduction of a great number of experiments on steel has been demonstrated that this model
works well on steel.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 284 7.11.2008 6:23pm Compositor Name: BMani
284 Handbook of Thermal Process Modeling of Steels
7.2.6 STRUCTURE AND DEFECTS IN CONTINUOUS CASTINGS
The traditional method of casting ingots is a batch process; piping and microstructural and chemical
variations are present throughout the ingot. Continuous casting allows reducing these problems,
producing higher quality steels for less cost.
The molten metal is teemed onto a starting block in a watercooled bottomless mold, generally
made of copper. In contact with the mold and starting block, the steel forms a solid skin, which
serves as a container for the liquid. The starting block and solidifying ingot are withdrawn from the
mold at a controlled rate. In the early stages of solidiﬁcation the thin skin is supported by a series of
rollers and is rapidly cooled by water sprays.
In Figure 7.12 an axial section of the shape of the liquid and solid þliquid mushy zones is
schematized. The latter region, shown in dark gray, consists of a cohesive dendritic skeleton
impregnated with liquid. The liquid pool contains suspended solid crystals, except in the super
heated top zone.
From Figure 7.13, which shows the macrostructure of a continuously cast stainless steel bloom,
it is possible to see two distinct zones:
.
Central region of ﬁne equiaxed grains
.
Outer region of columnar grains, elongated normal to the ingot surface; their width
increases from the surface to the center
There should be a third zone, corresponding to very ﬁne chill crystals, which is present at the
extreme skin, but cannot be clearly distinguished.
Bottom of
the liquid
pool
Bottom of
the mushy
zone
Superheated
liquid
Liquid
M
o
u
l
d
Solid
Displacement
Liquid + solid
mushy zone
FIGURE 7.12 Axial section through a continuously cast ingot (From DurandCharre, M., Microstructure of
Steel and Cast Irons, Springer, New York, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 285 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 285
During the solidiﬁcation the ﬁrst solid crystals form in contact with the mold wall, producing a
thin layer of ﬁne grains, called the chill zone. In the columnar zone, the grains nucleate on those in
the chill layer and they grow perpendicular to the solidiﬁcation front. Equiaxed crystals form with
random orientations and no preferred macroscopic growth direction; the growth of equiaxed grains
eventually stops the extension of the columnar zone. The proportion between the depth of the
columnar and equiaxed zones depends on several factors:
.
Temperature range between the liquidus and solidus
.
Degree of superheating
.
Efﬁciency of cooling
.
Hydrodynamic conditions (natural convection, stirring, etc.)
In general, the columnar zone is thicker the greater the amount of superheating above the liquidus.
Lowering and homogenizing the temperature of the liquid pool are used to reduce the thickness of
the columnar zone, and this is achieved by electromagnetic stirring (EMS).
Defects crack, segregation (including porosity), and inclusions are the major problems, as in
ingot casting. Longitudinal surface cracks is one of the major subjects that makes it difﬁcult
to develop highspeed casting technology. Surface cracks derive probably from the roughness
of the solidiﬁed shell; thus, the homogeneity of initial solidiﬁcation is of considerable importance.
A high cooling rate corresponds to high productivity, but also to high unevenness, and so to a
less uniform shell.
Continuously cast billets solidify mainly in a dendritic structure that can lead to heavy
centerline segregation and cavities. However, axial segregation, center unsoundness, and V
FIGURE 7.13 Transverse section of a square continuously cast stainless steel bloom. The sharp separation
between the outer columnar zone and the central equiaxed region is due to the use of EMS during the casting
process. (From DurandCharre, M., Microstructure of Steel and Cast Iron, Springer, New York, 2004.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 286 7.11.2008 6:23pm Compositor Name: BMani
286 Handbook of Thermal Process Modeling of Steels
segregates are less predominant in continuously cast steel. The axial segregation and center
unsoundness in billets are often found periodically along the strand, which may pose serious
quality problems. The factors responsible for severe axial segregation and porosity are
.
Minimal equiaxed structure
.
Large superheat
.
Predominant columnar structure
.
No EMS application in the upper part of the strands
.
Increased section size
.
Higher aspect ratio
.
Carbon content
The carbon content inﬂuences both the width of the central segregation zone and the size of the
axial voids. For very high carbon contents, the centerline consists of a tubular segregated region.
The severity of V segregates increases with carbon content. The mechanism of formation of V
segregates involves the settling of dendrites, volume shrinkage, and interdendritic ﬂow in the very
long mushy zone.
In continuous casting, the production of good quality billets for lowcarbon steels depends on a
very low oxygen activity in the mold to avoid pinhole occurrence and low total oxygen contents to
minimize the amount of inclusions. This last requirement becomes important in the case of billets
casters with small radius where the inclusions accumulate near the upper side of the billets.
7.2.7 APPROACH TO FILLING AND RISERS
The design of the running system used to guide the metal from the ladle into the mold is crucial. The
simplest and quickest way to make a casting consists in pouring the liquid metal into the opening of
the running system under the action of gravity (e.g., gravity sand casting and gravity die casting).
When a liquid metal enters into a mold, there is a repulsive force, due to surface tension g,
which resists the entry of the metal and, even if the metal enters, it is subjected to the resistance of
surface tension, which tends to reverse the metal ﬂow and to empty it out if there is a reduction in
the ﬁlling pressure.
The Laplace equation deﬁnes the pressure difference between the pressure inside the metal p
i
and the external pressure p
e
, when the pressure difference across the interface is in balance with the
effective pressure due to surface tension g,
Dp ¼ p
i
Àp
e
¼ g
1
r
1
þ
1
r
2
_ _
(7:21)
where r
1
and r
2
deﬁne the curvature of the meniscus.
When the ﬁlling pressures exceed the resistance of surface tension, for example, when ﬁlling
a circular section tube of radius r, the relation becomes
Dp ¼ p
i
Àp
e
>
2g
r
(7:22)
When ﬁlling a narrow plate of thickness 2r, the radius at the right angles becomes inﬁnite, and the
relation becomes
Dp ¼ p
i
Àp
e
>
g
r
(7:23)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 287 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 287
For a gravity casting made under normal atmospheric pressure the equation becomes
Dp ¼ p
a
þrgh ð Þ À p
a
þp
m
ð Þ >
g
r
(7:24)
where
h is the height of the liquid
r is the density of the liquid
g is the gravity acceleration
p
a
is the atmospheric pressure
p
m
is the pressure of the mold gases
Therefore, rgh is the hydrostatic pressure at the depth h. The ﬁrst term of the equation is the total
pressure inside the metal (hydrostatic þatmospheric pressure), whereas the second term is the
pressure in the mold (atmospheric þpressure of the mold gases).
Simplifying the equation
rgh Àp
m
>
g
r
(7:25)
The above equation leads to the following conclusions:
.
For narrow sections it is necessary to use a vent, to reduce the back pressure due to the
outgassing.
.
Atmospheric pressure plays no part in the ﬁlling behavior of thin sections in air.
Vacuum is therefore not helpful in overcoming the resistance to ﬁlling, although it may help by
reducing the p
m
term by outgassing the mold prior to casting. A good running system should yield
repeatable casting with a low scrap rate. Another important element of a good system is the
economy of size, which increases the ratio of ﬁnished casting weight to total cast weight and allows
the production of more castings from the existing melt supply.
The mold should be ﬁlled at the required speed. To avoid air entrapment the whole running
system must be designed so that the velocity of the metal in the gates is below a certain critical
value. This value varies from one alloy to another.
Another important factor that deﬁnes the quality of a runner system is connected to the delivery
of only liquid metal into the mold cavity, without other phases such as slag, oxide, sand, or
undesirable gases. If gas bubbles remain entrapped, they rise and burst at the liquid surface, causing
surface turbulence and pores in the casting, specially at the upper surfaces above ingates.
Dross or slag should not enter into the mold cavity.
Figure 7.14 shows a system with the runner placed in the cope and the gates in the drag; in this
solution the metal has insufﬁcient time to settle down and to organize itself free from dross, oxide,
and bubbles, and for these reasons this system is not the best way to reduce defects in casting.
A more satisfactory system is illustrated in Figure 7.15. Here the runner is in the drag and the
gates in the cope. In this system, the runner has to ﬁll ﬁrst before the gates are reached. Thus,
the metal has a short but valuable time to rid itself of bubbles and dross, most of which can be
trapped in the dross trap or against the upper surface of the runner.
Surface turbulence should be avoided by the running system; it should control the fragmentation
of the stream, gathering it together, so that it can enter the mold as a compact spreading front at a
velocity sufﬁciently low that the danger of any breakup of the front is eliminated.
Topgated systems are always (with the exception of very thin wall castings where surface
tension takes over the control of surface turbulence) poor running systems, because the metal enters
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 288 7.11.2008 6:23pm Compositor Name: BMani
288 Handbook of Thermal Process Modeling of Steels
from either top or side gates and so suffers by falling freely and splashing into the mold cavity.
Bottomgated systems are therefore always required if surface turbulence is to be eliminated.
Bottomgated systems displayed with the base of the runner turned so that metal directly enters the
mold (Figure 7.16) yield a high percentage of scrap and therefore are not an economical solution. The
high velocity of the liquid metal on its impact at its base is not contained; the bulk of the metal follows
in an untidy fashion, mixed with air and mold gases, and ricochets from the far wall. The burst of the
bubbles when they rise to the surface of the melt causes additional droplets to be created by splashing.
The droplets, in contact with air form oxides, which pollute the melt when back into the bath.
The system given in Figure 7.17 has a downrunner designed to ﬁll quickly, excluding air.
Furthermore, its base is designed to eliminate the initial splash.
This point in the running system should be the lowest point in the system: having reached here,
all subsequent ﬂow of the molten metal should be uphill, displacing the air ahead. The molten metal
should ﬁnd itself in an expanding system, slowing as it goes, experiencing as much opportunity as
possible to become quiescent before entering the mold.
As the least, the runners should be easy to remove; preferably, the system should break off. As a
next best option, it should be removable with a single stroke of a clipping press, or a straight cut
with a saw or abrasive wheel.
Conical
cup
Reverse
taper sprue
Runner
in cope
Gates
in drag
Casting
in drag
FIGURE 7.14 Poor running system. (From Campbell, J., Castings, Butterworth Heinemann Ltd., Oxford,
UK, 1991.)
Weir
bush
Correct
taper sprue
Well
Stepped runner
in drag
Gates
in cope
Casting
in cope
FIGURE 7.15 More satisfactory running system. (From Campbell, J., Castings, Butterworth Heinemann
Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 289 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 289
The sprue (or downrunner) must be designed with the right size; if a sprue is oversized it will
take more time to ﬁll and air will be taken down with metal, causing turbulence in the downrunner.
If the sprue is correctly sized, the metal ﬁlls quickly, excluding air. The area of the sprue exit is an
important parameter, because it controls the rate at which metal ﬁlls the mold. To calculate the
required sprue dimensions it is ﬁrst necessary to deﬁne an average ﬁlling rate for the casting; this
can be done checking the thinnest section on the casting and, from this, ﬁnding its corresponding
solidiﬁcation time. Another solution able to determine the ﬁlling rate is to measure the rate of rise of
the metal in the mold; After that, the next step consists in calculating the average rate of ﬁlling,
which is the ratio between the ﬁlling time and the weight of the casting (Figure 7.18).
FIGURE 7.16 Poor bottomgated system. (From Campbell, J., Castings, Butterworth Heinemann Ltd.,
Oxford, UK, 1991.)
FIGURE 7.17 Improved bottomgated system. (From Campbell, J., Castings, Butterworth Heinemann Ltd.,
Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 290 7.11.2008 6:23pm Compositor Name: BMani
290 Handbook of Thermal Process Modeling of Steels
The Bernoulli’s relation allows calculating the velocity of a falling liquid at a height h:
v
2
¼ 2gh (7:26)
The above equation, for a stream considered to pour from a ladle, with an initial velocity as zero.
Entering a sprue after a distance h
1
, and ﬁnally reaching the base of the sprue, with a total falling h
2
,
the equation becomes
v
2
2
Àv
2
1
¼ 2g h
2
Àh
1
ð Þ (7:27)
The lowvelocity v
1
is associated with the biggest area A
1
, whereas the highvelocity v
2
to the
smaller area A
1
. Considering a continuous stream, v
1
A
1
¼v
2
A
2
, then
h
2
h
1
_ _
¼
A
1
A
2
_ _
2
(7:28)
To summarize, it is possible to modify the rate of ﬁlling of a casting by adjusting the size of the
sprue exit.
A properly designed downrunner base is essential for reducing turbulences due to the splash of
the ﬁrst molten metal at the base of the sprue (Figure 7.19). The well is one of the most widely used
sprue base; it prevents the splash of the melt along the runner when the ﬁrst molten metal arrives at
the space of the sprue. A properly designed well should have a diameter twice the sprue exit, with a
depth twice the depth of the runner.
In general, the runner should have an area at least twice the sprue exit area and half the gate
area; to be ﬁlled completely it should be placed in the lower half of the mold. This localization gives
to the metal the time necessary to settle and to remove gas bubbles. In the case of multiple gates, the
runner should be stepped, to give a slight pressurization, which helps to balance the ﬁlling of the
gates (Figure 7.20). The gates guide the liquid metal from the runner to the mold cavity. They
Area A
1
=
Area A
2
h
1
h
2
h
2
h
1
A
1
2
A
2
FIGURE 7.18 Geometry of the falling stream. (From Campbell, J., Castings, Butterworth Heinemann Ltd.,
Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 291 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 291
should not be placed directly at the base of the downrunner, because the rate of entry would be too
high, causing molding defects. The incorporation of some rightangle changes of direction and the
use of a well could help the metal to reorganize itself from a mix of liquid and gases into a coherent
mass of liquid.
An important rule about the design of the gates is that their area should not be too small to
increase the metal velocity during mold cavity ﬁlling. In fact, the running system should be designed
to reduce the metal velocity during its progress below dangerous levels. To avoid the formation of
hot spot at the point at which the gate contacts the casting, it is necessary that the gate modulus
should be only half the local casting modulus; in this manner the gate freezes ﬁrst and no pocket of
liquid remains trapped at a later stage of freezing.
The best design approach to avoid the carrying through of pockets of air inside the mold cavity
is to place the gates at the lowest point of the casting, and the runner below that, to ensure the runner
ﬁlls completely, then the gates, and only ﬁnally can the casting start to ﬁll.
In direct gating, the gates are connected directly into the mold cavity; sometimes it causes
problems because of the effect it has on the solidiﬁcation pattern of the casting.
In the ﬁlling of the plate of Figure 7.21, the metal ﬂows sideways from the gate into the sides of
the plate, cooling during the movement and freezing in layers away from the fresh hot metal, which
continues to arrive through the gate. The ﬁnal freezing of the ﬂow path is slow because of the
preheated mold around the path, and so its structure is coarse and porous.
2L
L
D
2D
FIGURE 7.19 Proper downrunner base. (From Campbell, J., Castings, Butterworth Heinemann Ltd.,
Oxford, UK, 1991.)
FIGURE 7.20 Balanced multiple gating system. (From Campbell, J., Castings, Butterworth Heinemann Ltd.,
Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 292 7.11.2008 6:23pm Compositor Name: BMani
292 Handbook of Thermal Process Modeling of Steels
The formation of porosity is encouraged by the gas precipitation under the conditions of slow
cooling, and shrinkage may contribute if local feeding is poor because the ﬂow path is too distant
from a source of feed metal. Bottom gating into bosses can take advantage of the boss as a useful
feed path (Figure 7.22), even if this action increases the problems with slower cooling, leading to
enhanced gas porosity and coarse structure. Such ﬂow paths are a nuisance in large castings, which
take longer times to ﬁll, but the problem can be reduced by ﬁlling faster, where possible.
The other major problem of bottom gating is that the coldest metal is at the top and the hottest at
the bottom. When feeders are placed on the top of the casting, in heavy section castings the
convection has time to reverse the unfavorable temperature gradient, but in castings of intermediate
sections, convection appears to be not effective enough to completely reverse the temperature ﬁeld
to make top feeders really effective. A not suitable thermal regime may cause defects associated
with porosity and coarse structure in random pockets throughout the casting.
FIGURE 7.21 Direct gating with boss. (From Campbell, J., Castings, Butterworth Heinemann Ltd., Oxford,
UK, 1991.)
FIGURE 7.22 Direct gating. (From Campbell, J., Castings, Butterworth Heinemann Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 293 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 293
The riser is a gate, which allows solving the ﬂow path problem that arises because the hot metal
that is required to ﬁll the casting is gated directly into the casting and has to travel through the
casting to reach all parts. With a riser the main ﬂow path is created outside the casting and not inside
(Figure 7.23).
Metal is therefore diverted initially away from the casting, through the riser, only entering
subsequently by displacement sideways from the riser as fresh supplies of hot metal arrive. The
fresh supplies ﬂood up into the top of the riser, ensuring that the riser remains hot, and that
the hottest metal is delivered to the top of the mold cavity. The system has the special property
that the riser and slot gate combination acts not only to ﬁll but also to feed. One of the disadvantages
of the riser and slot gate system is the considerable cutoff and ﬁnishing problem, since the gate
often has to be sited on an exterior surface of the casting.
In the normal gate where it is required to freeze before the casting section to avoid the hotspot
problem in the junction, the thickness of the gate is held to half of the casting thickness or less; in
this case the gate is equivalent to a feeder neck, through which feed metal is required to ﬂow until
the casting has solidiﬁed. Whereas a thickness of double the casting thickness would then be
predicted to avoid the junction hot spot under conditions of uniform starting temperature, the
preheating effect of the gate due to the ﬂow of metal through it might mean that a gate as narrow
as half of the casting section may be good enough to continue feeding effectively.
If ﬁlling is reasonably fast then the resistance to the ﬂow provided by a slot gate that is too
narrow will cause the riser to ﬁll up to a high level before much metal has had a chance to ﬁll the
mold cavity. The metal will therefore spill through the gate into the casting from its elevated level.
It is desirable, therefore, to ﬁll slowly, or to have a gate with minimal resistance to metal ﬂow. In this
way the system can work properly, with the molten metal in the riser and casting rising substantially
together. The metal will then enter the mold gently. It is also important for the gate not to be thinner
than the casting when the casting wall thickness reduces to approximately 4 mm. A gate 2 mm thick
would hold back the metal because of the effect of surface tension allowing the metal head to build
up in the riser. When the metal eventually breaks through, the molten metal will emerge as a jet,
splashing into the mold cavity. For casting sections of 4 mm or less the gate should be at least as
thick as the wall. For casting less than 2 mm thickness made under normal ﬁlling pressure, the
feeding of thinsection castings can probably be neglected and any hotspot problems can also be
disregarded, with the result that the ingate can be equal to the casting thickness. Surface tension
controls the entry through the gate and the further progress of the metal through the mold cavity,
reducing the problems of surface turbulence. Fill speed can therefore be increased.
L
X
FIGURE 7.23 Riser and slot gate to both gate and feed at a vertical plate via its end. (From Campbell, J.,
Castings, Butterworth Heinemann Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 294 7.11.2008 6:23pm Compositor Name: BMani
294 Handbook of Thermal Process Modeling of Steels
To have a ﬁrst idea about the rates necessary to make the system work best it is useful to
schematize the ﬁrst molten metal that ﬂows through the gate and along the base of the cavity as a
stream; it shall be assumed that the height of the stream corresponds to the height of the meniscus
that can be supported by surface tension.
If L is the distance to be run from the gate, the solidiﬁcation time of the metal is t
f
in that section
thickness, then the limiting condition where the metal just freezes at the limit of ﬂow can be
deﬁned as
t
f
¼ L=V (7:29)
where V is the velocity of ﬂow of the metal stream. The corresponding rate of ﬂow Q in, for
instance, kilogram per second for metal of density r is easily shown to be
Q ¼ VHr ¼ LHr=t
f
(7:30)
and the time t to ﬁll a casting of height H
C
is given by
t ¼ t
f
(H
C
=H) (7:31)
Thin cast sections do not necessarily require fast ﬁlling rates to avoid premature solidiﬁcation; the
most important thing is the steady, continuous advance of the meniscus. Of course, if the rate of
ﬁlling is made to be too slow, then the rate of advance of the molten metal front will become
unstable because of ﬁlm problems in some materials. Filmfree systems will not suffer this problem,
and vacuum casting may also assist.
Up to now, the term risers has been considered as a special form of feeders, which act as an
uprunner, in which the metal rises up the height of the casting. It is also common to refer to
conventional feeders placed on the top of castings as risers; this kind of risers works as reservoirs of
feed metal, so that undesirable shrinkage cavities in the casting are eliminated or moved to locations
where they are acceptable for the intended application of the casting.
In Figure 7.24, the contribution of each one of the three distinct stages of volume contraction is
shown: molten metal shrinkage, solidiﬁcation shrinkage, and solid contraction [7].
Solid shrinkage is accommodated by making the mold cavity somewhat larger than the desired
dimensions of the ﬁnal casting, whereas molten metal and solidiﬁcation shrinkage are the concern of
risering practice.
Risers supply molten metal feed metal to compensate for the solidiﬁcation shrinkage within the
casting, reducing defects as internal shrinkage voids, surface deformation or dishing, and surface
L
0
Original liquid
metal level
L
1
L
2
Solidification
shrinkage roughly 3%
Original liquid interface, L
3
Liquid shrinkage
about 1.6%/100ЊC
superheat
Solid contraction
L
3
–L
4
Final form of solid, L
4
FIGURE 7.24 Contribution of the three distinct stages of volume contraction in the shrinkage of lowcarbon
steel. (From Stefanescu, D.M. and ASM, ASM Metals Handbook, Vol. 15: Casting, ASM International, Metals
Park, OH, 1988.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 295 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 295
puncture; therefore, the shrinkage in the riser=casting system is concentrated in the riser, which will
then be removed from the ﬁnished casting.
The riser must often be larger than the casting it feeds, because it must supply feed metal for as
long as the casting is solidifying; to reduce the size of the riser, it is possible to chill the casting,
reducing the solidiﬁcation time, or insulating the riser.
The riser must be designed to satisfy the liquid and solidiﬁcation shrinkage requirements of the
casting.
The total feeding requirement will depend on
.
Speciﬁc alloy
.
Amount of superheat
.
Casting geometry
.
Molding medium
The shape of a casting will affect the size of the riser (Table 7.1), because the longer the casting
takes to solidify, the longer the riser must maintain a reservoir of liquid metal.
For thinsection castings, a portion of the liquid and solidiﬁcation shrinkages will be fed by
liquid metal entering the mold from the gating system, and feed metal requirements may be smaller
than what would ordinarily be calculated.
The correctly localized risers should avoid shrinkage cavities in the casting; to obtain this,
solidiﬁcation should proceed directionally from those parts of the casting farthest from the riser,
through the intermediate portions of the casting, and ﬁnally into the riser itself, where the ﬁnal
solidiﬁcation will occur.
Feeding distances in sections of uniform thickness depend on alloy characteristics and section
conﬁguration to determine how far directional solidiﬁcation can be sustained. If the walls of the
progressively solidifying intermediate sections begin to come together, disrupting and constricting
the feeding channels through which feed metal can move, centerline shrinkage will occur. Once the
feeding distance of a section has been exceeded, the development of centerline shrinkage will not be
overcome by increasing the size of the riser.
In castings consisting of sections of varying thickness and conﬁguration, the thicker sections,
which solidify slowly, are separated from each other by thinner, rapidly solidifying sections. The
heavier sections supply the feed metal demands of the lighter sections, which act as risers.
The risering methods must divide a casting into sections requiring risers by determining the
feeding paths by which solidiﬁcation will move directionally from early to latesolidifying sections.
These feed paths can often be manipulated by proper application of chilling or insulating materials
to minimize risering needs.
TABLE 7.1
Minimum Volume Requirements for Risers on Steel Castings
Minimum Riser Volume=Casting Volume (V
r
=V
c
), %
Insulated Risers Sand Risers
Type of Casting H=D¼1:1 H=D¼2:1 H=D¼1:1 H=D¼2:1
Very chunky (cubes, etc.): dimensions in ratio 1:1.33:2 32 40 140 198
Chunky: dimensions in ratio 1:24 26 32 106 140
Average: dimensions in ratio 1:3:9 19 22 58 75
Fairly rangy: dimensions in ratio 1:10:10 14 16 30 38
Rangy: dimensions in ratio 1:15:30 9 10 13 15
Very rangy: dimensions in ratio 1: >15: >30 8 8 11 13
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 296 7.11.2008 6:23pm Compositor Name: BMani
296 Handbook of Thermal Process Modeling of Steels
Figure 7.25 shows a casting with two heavy parts connected by a thinner section; in Figure 7.25a
there is no riser, and shrinkage develops in the two heavy sections. In Figure 7.25b, shrinkage remains
in the section without a riser, because the connection freezes at ﬁrst. No shrinkage cavities form using
two risers, as shown in Figure 7.25c. A method alternative to the use of two runners could be the
application of a chill (Figure 7.25d), to reduce the solidiﬁcation time of the heavy section belowthat of
the connection. Another solution could be the use of an insulator pad on the connection.
A riser should ensure that liquid feed metal is available for the whole solidiﬁcation time of the
casting. There are several methods to calculate the optimal riser size:
.
Shape factor method
.
Geometric method
.
Modulus method
.
Computerized methods
7.3 INDUSTRIAL CASTING OF STEEL INGOTS AND BILLETS
7.3.1 CONTINUOUS CASTING
Continuous casting is the established and relatively modern method of production of semiﬁnished
steel products such as billets, blooms, and slabs. Continuous casting offers the following advantages
over the ingot casting process:
1. It yields a semiﬁnished shape (billet, bloom, or slab) or nearnet shape casting in a nearly
continuous fashion, unlike ingot casting, which produces unﬁnished shape of a large cross
section.
Shrinkage
(a) Workpiece with no risers
(c) Risers located on both ends
(b) Riser added to one side
(d) Chill applied to one end and riser to other end
Insulator or exothermic pad
(e) Riser used on one end and insulator
or exothermic pad on opposite end
FIGURE 7.25 Risering of isolated heavy sections joined by a thinner section to minimize shrinkage and
number of risers. (From Stefanescu, D.M. and ASM, ASM Metals Handbook, Vol. 15: Casting, ASM
International, Metals Park, OH, 1988.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 297 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 297
2. It offers increased productivity and yield, reduced energy and manpower, improved
product quality, and consistency of quality.
3. It is the preferred casting process and provides lower emissions harmful to the environment
and plant operator, reduces stock levels and enables shorter delivery times, and offers
reductions in capital costs for new steel plants.
Additional advantages of this process are the total elimination of primary mills, linking of the
continuous caster directly with the reheating furnaces and rolling mills, and directing of the casting
sections to the ﬁnal product dimensions, thereby leading to further reduction of operations and
investment costs.
In contrast to ingot castings, in continuous casting, the depth of the molten metal core may easily
reach 20 m and above, which is far greater than in an ingot casting; as a result, the solidiﬁcation of the
center of a continuously cast section is largely removed from the mold. Solidiﬁcation occurs in
a moving section, and time is viewed from the standpoint of a reference frame that moves into
the strand. In addition, the continuously cast strand usually changes its orientation from vertical
to horizontal during its movement through the casting machine. Submerged pouring tubes and
EMS are normally used in continuous casting rather than ingot casting. Most of these differences
affect ﬂuid ﬂow conditions in the molten metal core, which signiﬁcantly affects the structures and
segregation.
Historically, the use of shorter molds (700–1500 mm) in continuous casting has a clear cost
advantage at speeds of up to 5 m=min. However, with the current trend toward high, thick slabs,
longer molds seem to be required. Longer molds can improve operational safety during transient
conditions and allow a longer heating period in case of sticker alarms.
In a continuous process, the molten metal steel is ﬁrst poured from a ladle or melting furnace
into a refractorylined tundish (which works as a reserve of metal), which then steadily directs its
ﬂow into a stationary, vertically oscillating watercooled copper mold (contained by steel backing
plates) controlled by a stopper nozzle or slide gate valve arrangement (Figure 7.26). Synthetic oxide
slags or inertgas blanketing are used to protect the surface of the molten steel pool.
Regarding the design and material requirements for molds, different designs have been used;
molds made from plate provide excellent mold life and avoid the necessity to manufacture the molds
from solid copper blocks. The most suitable material for mold production is nearly pure copper, with
small additions of elements that promote precipitation hardening or raise the recrystallization
temperature to increase the mold life.
Prior to startup of a casting, a movable starter or dummy bar is introduced into the bottom of the
mold and sealed in order to restrict the initial ﬂowof steel fromthe mold and to forma solid skin. After
commencement of pouring the mold, the dummy bar is gradually withdrawn at a rate proportional to the
addition of molten steel into the mold. The strand of steel that follows immediately is semisolid
(a molten metal core encased in a solidiﬁed thick shell) over much of the length of the casting
machine.
Parameters like the length of the mold and the casting speed are such that the shell thickness is
capable of sustaining the pressures of the molten metal core upon exiting from the copper mold. To
prevent sticking usually a reciprocating motion is superimposed toward the downward travel and a
lubricant is added to the mold.
Usually the oscillation is sinusoidal, with a frequency from 50 to 300 cycles per minute. During
each oscillation a fairly short stroke is used, to provide a short negative strip during each oscillation,
in which the mean downward velocity of the mold movement is greater than the speed of
withdrawal of the casting strand in the casting direction.
Below the mold some rolls support the steel strand, where a secondary cooling sprays cooling
water onto the strand surface to complete the solidiﬁcation process. During secondary cooling the
maximum cooling should be accomplished; however, overcooling and large temperature increases
must be avoided.
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 298 7.11.2008 6:23pm Compositor Name: BMani
298 Handbook of Thermal Process Modeling of Steels
The amount of heat removed by water sprays depends on various parameters, including the
volume of the water, its temperature, and spray pressure. After the completion of the solidiﬁcation
stage, the steel strand is cut by a welding torch into the desired lengths.
In Figure 7.27 several arrangements in use for the continuous casting of steel for producing
slabs are shown; the types of continuous casting machines in use include vertical (V), vertical with
bending, vertical with progressive bending, circular arc with straight mold, circular arc with curved
mold, progressive bending with curved mold, and horizontal (H).
Melting furnace
Ladle
Tundish
Mold
Secondary
cooling
Withdrawal
rolls
FIGURE 7.26 Schematic representation of the continuous casting process of steel. (From Stefanescu, D.M.
and ASM, ASM Metals Handbook, Vol. 15: Casting, ASM International, Metals Park, OH, 1988.)
33 m
V VB VPB CAS CAC PBC
H
25 m
20 m
12 m
10 m
5 m
3 m
FIGURE 7.27 Principal types of continuous casting. (From Stefanescu, D.M. and ASM, ASM Metals
Handbook, Vol. 15: Casting, ASM International, Metals Park, OH, 1988.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 299 7.11.2008 6:23pm Compositor Name: BMani
Modeling of Casting 299
7.3.2 INGOT CASTING
Ingots are cast shapes produced with a cross section that is suitable for rolling, extrusion, and
forging operations. They are commonly used in steel and copper industries. In ingot castings, steel is
poured into a top or bottom of cast iron molds. The molds are often equipped with hot tops
comprising insulating boards and exothermic compounds to greatly reduce the depth of the
shrinkage cavity formed during the solidiﬁcation of ingot. After a required period, the molds are
removed from ingots and charged into soaking pits for rolling into ﬁnished products.
Ingots are processed by three separated operations.
1. Teeming: it consists of pouring the molten steel into the ingot mold. The molten metal ﬂow
from a hole in the bottom of the ladle. The ingots are usually square or rectangular shaped,
and tapered with rounded corners.
2. Stripping: as soon as the ingots are partially cooled. A crane separates the ingots from
the molds.
3. Soaking: after the stripping, the ingots are placed into a soaking pit at 12008C for 6–8 h, to
reach a uniform temperature throughout the ingot. The purpose of the operation is
to prevent the outer surface of the ingot from solidifying before the inside; carbon,
phosphorous, and sulfur tend in fact to solidify last and to congregate around the center
portion of the steel. After the soaking is complete, the ingots are removed from the soaking
pit and carried to the rolling mill.
Unlike continuously cast semiﬁnished products, ingots have larger thickness in at least one
transverse direction, the depth of liquid core is small, solidiﬁcation occurs in a stationary space,
and time is measured from the viewpoint of a ﬁxed observer.
Steel ingot can be classiﬁed according to the deoxidation practice used or by the amount of gas
evolved during solidiﬁcation (Figure 7.28). In fact, signiﬁcant amounts of oxygen and other gases
can dissolve in the molten metal during steelmaking and the same gases, during solidiﬁcation, are
rejected, because the solubility limit decreases as the temperature decreases. These types of steel are
called killed, semikilled, capped, and rimmed steels.
The gas is generated primarily by the reaction between carbon and oxygen dissolved in the steel,
which is favored thermodynamically at lower temperatures.
The steel is termed killed if no gas is evolved because it lies quietly in the molds. Killed steel
ingots are fully deoxidized so that there is only a slight or practically no evolution of gas during
solidiﬁcation. These ingots are normally cast in ‘‘hottopped, bigendup molds’’ to reduce the depth
of the shrinkage cavity. Killed steel is characterized by a homogeneous structure, even distribution
of chemical composition and properties, and formation of pipe in the upper central portion of the
1
2
3
4
5
6
7
8
FIGURE 7.28 Classiﬁcation of steel ingot according to the deoxidation practice used. (From Campbell, J.,
Castings, Butterworth Heinemann Ltd., Oxford, UK, 1991.)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 300 7.11.2008 6:23pm Compositor Name: BMani
300 Handbook of Thermal Process Modeling of Steels
ingot, which is later cut off and discarded. All steels that have more than 0.3% carbon are killed.
Alloy steels, forging steels, and carburizing grades of steels are usually killed.
The most common deoxidizing elements used for killed steel are Al and ferroalloys of Mn and
Si; however, calcium silicide and other special strong deoxidizers such as V, Ti, and Zr are
sometimes used.
Semikilled steel ingots are partially deoxidized so that gas evolution is not completely sup
pressed during solidiﬁcation by deoxidizing additions. There is a greater degree of gas evolution
than in the killed steel, but less than in the capped or rimmed steel. An ingot skin of considerable
thickness is formed prior to the beginning of gas evolution. A correctly deoxidized semikilled steel
ingot does not have a pipe, but does have wellscattered large blowholes in the topcenter half of the
ingot. However, they weld shut during rolling of the ingot. Semikilled steels, generally, have a
carbon content in the range of 0.15%–0.30%.
The main features of semikilled steels are
1. Uniformity in composition, which is intermediate between those of killed and rimmed
steels, and less segregation than rimmed steel
2. Pronounced tendency for positive chemical segregation at the top center of the ingot
Rimmed steel ingots are characterized by only a small amount of deoxidation and a greater degree of
gas evolution during solidiﬁcation in the mold and a marked difference in chemical composition
across the section and from top to bottom of the ingot. This results in the formation of an outer ingot
skin or rim of relatively pure iron and an inner liquid portion of the ingot with higher segregation=
concentration of alloying elements, especially carbon, nitrogen, sulfur, and phosphorus, having
lower melting temperatures. Most lowcarbon steels (<0.25% C) are rimming steels. Rimmed
ingots are best suited for the manufacture of ﬂatrolled products (plates, sheets) as well as wires,
rods, and other products in the coldrolled or subcritically annealed condition, where good surface
or ductility and better ingot surfaces are required.
Capped steel ingots have similar characteristics to those of a rimmed steel ingot, but to a degree
intermediate between those of rimmed and semikilled steels. Less deoxidizer is used to produce a
capped ingot than to produce a semikilled one. This induces a controlled rimming action when the
ingot is cast, i.e., thinner zone and less segregation in the core. The gas entrapped during
solidiﬁcation is in excess of that required to counteract normal shrinkage, resulting in a tendency
for the steel to rise in the mold. The capped ingot practice is usually applied to steels with carbon
contents greater than 0.15%, which are used for the production of sheet, strip, tin plate, skelp, wire,
and bars.
7.4 OUTLOOK FOR MODELING OF CASTING
During design of a new casting, performing numerous trials and tests till obtaining optimum design
is the general practice in the foundry. Since this procedure is expensive and causes long lead times
before production, the use of computer simulation techniques in the foundry environment is very
important. In simulation of casting processes the following main parameters have to be addressed:
1. Analysis of ﬂuid ﬂow (mold ﬁlling and convectiondriven currents in molten metal)
2. Thermal analysis of solidiﬁcation (latent heat)
3. Stress analysis (residual stress state and ﬁnal geometry of casting)
4. Macroscopic porosity formation
Solidiﬁcation is an illustration of the manifold lengthscales that intervene in materials science.
Macroscopic models are primarily concerned with macroscopic entities such as temperature, stress,
strain, or velocity ﬁelds. In modeling of solidiﬁcation processes, these entities can be related to the
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 301 7.11.2008 6:24pm Compositor Name: BMani
Modeling of Casting 301
process parameters. However, there is also a need for prediction of microstructure=defects, and ﬁnal
mechanical=functional properties of the castproduct; i.e., process scale (cm, m), grain size (mm),
dendrites=eutectic spacings (mm), and atomistic scale (<nm) [8].
Process simulation capabilities have been extended beyond thermal and ﬂow modeling for
casting. Coupled modules for the calculation of grain structure, porosity, hot tearing, and solidstate
transformation are commercially available such as MAGMASOFT, ProCAST, NovaCast,
MAVIS2000, FLOW3D, CAPCAST, PAMCASTSIMULOR, SOLIDCast, ConiferCAST, PAS
SAGE=PowerCAST, CalcoSOFT, SUTCAST, dieCAS, ADSTEFAN, MICRESS, etc. The time
consuming modeling of the casting geometry can be avoided by using the existing models in
commercial CAD programs like AutoCAD, DUCT, and IDEAS as an interface which transfers the
casting geometry to the software.
7.4.1 MOLDFILLING CALCULATIONS
The normal equations assumed in order to perform moldﬁlling calculations are the following:
.
Fluid is incompressible, which means that the density of the ﬂuid does not change in time
or space under pressure or temperature. This assumption is not true for molten metaly but
for the purpose of the mathematical model it is necessary that the density within each ﬂuid
element remains constant, as it moves along a streamline. That is, the substantial derivative
of the material density is zero
dr
dt
¼
@r
@t
þv Á rr ¼ 0 (7:32)
.
Fluid obeys Newton’s law of viscosity. For a Newtonian ﬂuid, the viscosity, by deﬁnition,
depends only on temperature and pressure (and also the chemical composition of the ﬂuid
if the ﬂuid is not a pure substance), not on the forces acting upon it. In this case, the shear
force per unit area is proportional to the negative of the local velocity gradient. In one
dimension, this can be expressed with the following equation:
t
yx
¼ Àm
@v
x
@y
(7:33)
where
t
xy
is the shear stress exerted in the xdirection along a plane of constant y
v
x
is the velocity of the ﬂuid in the xdirection
m is the viscosity of the ﬂuid
This can be usually checked in the material database list. Sometimes, such as in rheocast
ing, the ﬂuid is sensitive to the local shear rate. In this case, several viscosity models can be
available, depending on the commercial code, such as the Carreau–Yasuda [9]:
m ¼ m
1
þ(m
0
Àm
1
) 1 þ(l _ g)
a
½
(nÀ1)=a
where
_ g is the strain rate
m
0
is the viscosity at zero strain rate (Pa s)
m
1
is the viscosity at inﬁnite strain rate (Pa s)
l is the relaxation time (s)
a, and n are coefﬁcients relative to the casting material and its processing conditions
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 302 7.11.2008 6:24pm Compositor Name: BMani
302 Handbook of Thermal Process Modeling of Steels
.
Fluid obeys Fourier’s law of heat conduction, which states that the heat ﬂux by conduction
is proportional to the temperature gradient. In one dimension, this can be expressed with
the following equation, where q
x
is the heat ﬂux in the xdirection, T is the temperature, and
k is deﬁned as the conductivity.
q
x
¼ Àk
@T
@x
(7:34)
.
Fluid is isotropic, which means that the conductivity of the ﬂuid is the same in all
directions. With this assumption, Fourier’s law can be expressed in generalized coordin
ates, where q is the heat ﬂux vector and r is the gradient operator:
q ¼ ÀkrT (7:35)
.
Equation of continuity is developed by applying the law of conservation of mass. It states
that the rate of increase in density within a small volume element in space is equal to the
net rate of mass inﬂux, where r is the density, and v is the velocity of the ﬂuid.
@r
@t
¼ ÀrÁ (rv) (7:36)
The divergence on the righthand side can be expanded as
@r
@t
þv Á rr ¼ Àr Á rv (7:37)
so that the continuity equation can be written as a substantial derivative:
dr
dt
¼ Àr(rÁ v) (7:38)
If it is assumed that the ﬂuid is incompressible, the substantial as it is done, the substantial
derivative becomes zero and the continuity equation can be reduced to the following velocity
divergence.
(rÁ v) ¼ 0 (7:39)
.
Equation of motion is developed by applying Newton’s second law. It states that a small
volume element moving with the ﬂuid is accelerated by the body force, minus the pressure
and viscous forces acting upon it:
r
dv
dt
¼ Àrp À(rÁ T) þf (7:40)
The right side of the equation is a summation of body forces. In particular, rp is a pressure
gradient and arises from normal stresses that turn up in any ﬂow, rÁt is representative of
shear forces in the ﬂuid (normally viscous effects) and f represents other forces, such as
gravity. It is assumed that
.
Fluid has constant viscosity (mr¼rm)
.
Body force is gravitational ( f ¼rg)
.
Fluid is incompressible (see Equation 7.32)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 303 7.11.2008 6:24pm Compositor Name: BMani
Modeling of Casting 303
.
Fluid is Newtonian (see Equation 7.33)
as is normal for castings, then the equation of motion can be written in this form:
r
dv
dt
¼ rg Àrp þmr
2
v (7:41)
which is known as the Navier–Stokes equation and strictly represents a statement of the
conservation of momentum.
.
Equation of energy is developed by applying the conservation of energy. It states that the
rate of gain in energy per unit volume equals the energy gained by any source term, minus
the energy lost by conduction, minus the rate of work done on the ﬂuid by pressure and the
viscous forces, per unit time:
r
dE
dt
¼ S À(rÁ q) À(rÁ pv) À(rÁ jt Á vj) (7:42)
where
E is the total energy
S is any source term
q is the heat ﬂux
P is the pressure
t is the viscous stress tensor
v is the velocity of the ﬂuid
Since the total energy of the ﬂuid is simply the sum of mechanical (K) and internal energy
(U), the mechanical energy can be subtracted to give the equation of internal energy:
r
dU
dt
¼ S À(rÁ q) ÀP(rÁ v) À(: rv) (7:43)
If the following are assumed
.
Fluid is isotropic and obeys Fourier’s law (4),
.
Fluid is incompressible and obeys the continuity Equation 7.39,
.
Fluid conductivity is constant (rÁ krT¼kr
2
T)
.
Viscous heating is negligible (t: rv ¼0)
then the equation of internal energy can be simpliﬁed to the following:
dU
dt
¼ Àkr
2
T þS (7:44)
If the internal energy is expressed as a function of volume and temperature,
dU ¼
@U
@V
_ _
T
dV þ
@U
@T
_ _
V
dT (7:45)
and given the assumption that the ﬂuid is incompressible, the change in volume must be
zero, and the speciﬁc heat capacity of the ﬂuid is deﬁned by
C
V
¼
@U
@T
_ _
V
(7:46)
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 304 7.11.2008 6:24pm Compositor Name: BMani
304 Handbook of Thermal Process Modeling of Steels
Since the heat capacity of a liquid, at constant volume, is approximately equal to the heat
capacity at constant pressure, the internal energy Equation 3.13 is reduced to the familiar
heat equation:
rC
p
dT
dt
¼ kr
2
T þS (7:47)
7.4.2 BOUNDARY CONDITIONS
Boundary conditions are required to solve the energy and momentum equations. The importance of
these boundary conditions cannot be underestimated. It is the boundary conditions that drive the
ﬂuid, and it is the boundary conditions that can inﬂuence stability and convergence.
There are three types of boundary conditions for the momentum equations:
.
Velocity of the ﬂuid can be speciﬁed (essential or Dirichlet).
.
Pressure of the ﬂuid can be speciﬁed (natural or Neuman).
.
Surface force at the boundary is speciﬁed.
There are also three types of boundary conditions for the energy equation:
.
Temperature at the interface is speciﬁed.
.
Heat ﬂux at the interface is speciﬁed.
.
Heat ﬂux at the interface is related to the temperature difference.
The particular way these conditions are assigned depends on the particular commercial code. Other
assignments must be given regarding the presence of gas in the mold, the storage frequency of the
ﬂuid results, the reference pressure, the pressuredriven inlet, the ﬁnal ﬁll fraction, and many others
depending on the code.
The typical results that can be obtained in numerical calculations are the following:
.
Filling behavior
.
Free surface evolution
.
Natural and forced convection currents
.
Dynamic pressure of the liquid
.
Entrapped gas
.
Filter behavior
In the case of lowpressure die casting, it is possible to model the liquid bath of metal and to apply a
pressure on it (in order to drive the metal in the mold) via injected air. Some kind of inject boundary
is sometimes available in order to specify the amount of injected air. Often these methods may be
delicate to converge.
7.4.3 DIMENSIONLESS ANALYSIS
The power of dimensionless analysis lies in the fact that these dimensionless numbers are able to
characterize a system. This can be demonstrated in Table 7.2, where several dimensionless numbers
have been calculated for the ﬂow of molten aluminum, through a narrow pipe (5 cm in diameter)
with a velocity of 1.0 m=s.
The Reynolds number represents the ratio between the inertial and viscous forces. In this case,
where the Reynolds number is particularly high, the inertial forces will dominate the momentum
equation. This means that any minor change to the viscosity will have little effect upon the ﬂow
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 305 7.11.2008 6:24pm Compositor Name: BMani
Modeling of Casting 305
proﬁle. So even if the viscosity does change, the assumption that the viscosity is constant has little
effect on the solution. Of course this assumption is not valid if the viscosity happens to change by
several orders of magnitude.
The Froude number represents the ratio between inertial and gravitational forces. In this case,
where the Froude number is close to unity the inertial and gravitational forces balance. This would
suggest that small changes in the density would have a signiﬁcant effect on the ﬂow proﬁle, so that if
the density did change, the assumption that it was constant would not be valid. Unfortunately, this is
the case with the solidiﬁcation of molten metal, where the density can change from 5% to 10%. To
correct this invalid assumption, rather than rewriting the original mathematical model, it is simpler
to add a buoyancy source term to the momentum equation. The validity of using this correction can
be evaluated after the simulation.
The Peclet number represents the ratio between the heat transport by convection and the heat
transport by conduction. In this case, where the Peclet number is high, the heat transfer is dominated
by convection. This means that a minor change to conductivity has little effect upon the temperature
proﬁle, and thus, it is reasonably safe to assume that the conductivity is constant, even when it is not.
The Brinkman number is a measure of the importance of viscous heating relative to the heat
ﬂow resulting from conduction. In this case, where the Brinkman number is very low it is safe to
assume that the effect of viscous heating is negligible.
Dimensionless analysis can also be used to make comparisons between independent systems.
For example, if two systems have the same Froude and Reynolds number, then both systems are
described by identical dimensionless momentum equations. If in addition, the two systems have the
same initial and boundary conditions, then the two systems are mathematically identical. Such
systems are said to be dynamically similar.
This situation arises between water and molten metal, which means that water can be used as a
dynamic analogue to predict the ﬂow characteristics of molten metal. Since it has been found that
the ﬂow of water through a pipe is turbulent when the Reynolds number exceeds 2200, and that the
majority of casting systems have a Reynolds number that exceeds this value, then it can be
concluded, that for the majority of casting systems, the ﬂow of molten metal will be turbulent. In
other words, when the Reynolds number is very high, the walls of the mold are too far away from
the bulk of the molten metal to provide effective constraint that would normally prevent its breakup.
TABLE 7.2
Dimensionless Numbers of Molten Aluminum
through Narrow Pipe at 1 m=s
Density 2690 kg=m3 Reynolds 32,000
Viscosity 4.5 Â10
À3
Pa s Froude 2
Conductivity 95 W=mK Peclet 1.5 Â10
3
Speciﬁc heat 1060 J=kg K Brinkman 9.0 Â10
À9
Reynolds number Re ¼
rDV
m
Froude number Fr ¼
V
2
gD
Peclet number Pe ¼
DVC
p
r
k
Brinkman number Br ¼
mV
2
kDT
Prandtl number Pr ¼
C
p
m
k
Gur/Handbook of Thermal Process Modeling of Steels 190X_C007 Final Proof page 306 7.11.2008 6:24pm Compositor Name: BMani
306 Handbook of Thermal Process Modeling of Steels
Oxidation takes place whenever liquid metal is exposed to air. During the handling and transfer
of this liquid, the oxide layer may be broken and dispersed into the liquid by the action of pouring,
splashing, and stirring. Such an abundance of oxides can create severe problems. Although ﬁlters
can be placed in the running or gating system to remove such oxides, it cannot remove oxides that
are formed after the ﬂuid has passed the ﬁlter.
The entrapment and dispersion of such oxides are caused mainly by the folding of the liquid
surface upon itself. Such action is related to the surface turbulence of the ﬂuid. It is important to note
that the turbulence associated with a high Reynolds number only refers to the bulk of the ﬂuid and
that it is quite possible for the surface to remain undisturbed. To predict the entrapment of oxides it
is necessary to predict the onset of surface turbulence. This is achieved with the Weber number
(We), which is deﬁned as the ratio between the inertial pressure of the ﬂuid and the action of surface
tension, which can prevent any disturbance,
We ¼ v
2
rR=s (7:48)
where
s is the surface tension
R is the surface curvature
7.4.4 PREDICTING STRESS DISTRIBUTION
Stresses evolved at the component and mold are due to nonuniform cooling, phase transformations,
and thermomechanical interactions during casting processes. Prediction of stress distributions is
critical for design and durability of the mold, and quality of the cast product.
Thomas et al. [10] have developed a twodimensional FE model to predict the internal stresses
generated in an ingot. Grill et al. [11] have calculated heat ﬂow by the FDM, and evaluated
deformation by FEM in planestress state. Tszeng and Kobayashi [12] have performed a two
dimensional stress analysis of continuous casting of a lowC steel. They calculated transient
temperature distributions during solidiﬁcation by using temperature