Documente Academic
Documente Profesional
Documente Cultură
BY
A L A N GIUMELLI
A THESIS SUBMITTED IN P A R T I A L F U L F I L L M E N T OF
T H E REQUIREMENTS FOR T H E D E G R E E OF
M A S T E R S OF APPLIED SCIENCE
in
T H E F A C U L T Y OF G R A D U A T E STUDIES
( D E P A R T M E N T OF M E T A L S A N D M A T E R I A L S ENGINEERING)
T H E UNIVERSITY OF BRITISH C O L U M B I A
APRIL, 1995.
Date h/^ / .
DE-6 (2/88)
ABSTRACT
The aims of this work are to assess the kinetics of austenite grain growth, and to
estimate the three dimensional grain size distribution. Measurements are to be used for the
validation of a statistical model for the estimation of the kinetics of austenite grain growth
during hot strip rolling. A series of tests are performed with two industrial plain carbon
steels in the temperature range from 950C to 1150C. Grain size is measured by standard
manual procedures and by the use of image analysis equipment. Systematic errors in the
measurements are identified. A correction is proposed to account for the systematic error
in the measurement of the two dimensional grain size distribution by the use of image
analysis. The kinetics of austenite grain growth are discussed and the influence of second
phase particles is observed. Results are described well by a statistical model for grain
growth. A number of methods for the estimation of the three dimensional grain size
distribution are compared. It is concluded that the method proposed by Matsuura and Itoh
ii
T A B L E OF CONTENTS
Page
ABSTRACT ii
LIST OF T A B L E S vi
NOMENCLATURE x
ACKNOWLEDGMENTS xiii
C H A P T E R 1. INTRODUCTION 1
C H A P T E R 2. L I T E R A T U R E R E V I E W 4
C H A P T E R 3. SCOPE A N D OBJECTIVES 31
C H A P T E R 4. E X P E R I M E N T A L 32
C H A P T E R 5. RESULTS A N D DISCUSSION I.
T H E KINETICS OF AUSTENITE G R A I N G R O W T H 50
iv
5.10 Grain Growth During Hot Strip Rolling 89
E S T I M A T I O N OF T H E T H R E E D I M E N S I O N A L G R A I N SIZE DISTRIBUTION.... 91
6.0 Introduction 91
6.4 Comparison of the Methods of Takayama et al. and Matsuura and Itoh 110
6.5 Validation of the Method of Matsuura and Itoh With Measured 3-D
Results Ill
C H A P T E R 7. S U M M A R Y A N D CONCLUSIONS 114
BIBLIOGRAPHY 118
APPENDIX 124
v
LIST OF TABLES
Table Page
Table 2.1 - Grain growth exponents for the power law, for zone refined metals 7
Table 2.2 - Power law parameters reported for austenite grain growth 8
Table 5.1 - Correction of systematic errors for the mean equivalent area diameter. 63
Table 5.2 - Power law fitting parameters for the A36 and D Q S K steels 72
Table 6.2 - The 3-D grain size distribution using the image analyser results
Table 6.3 - The 3-D grain size distribution from the corrected 2-D distribution
Table 6.4 - The 3-D grain size distribution predicted using the method of
Table 6.5 - Comparison of methods for Al-Sn alloy and stainless steel 112
vii
LIST OF ILLUSTRATIONS
Figure Page
Figure 4.4 - Schematic diagram of vertical tube furnace and specimen holder 39
Figure 5.7 - Kinetics of austenite grain growth in A36 steel, heating rate=5C/sec. 66
Figure 5.10 - The statistical model, A36 steel, heating rate=5 C/sec 74
Figure 5.14 - Evolution of the A36 grain size distribution, normal grain growth.... 82
Figure 5.15 - Evolution of the grain structure, in the A36 steel, abnormal growth. 84
Figure 5.16 - Evolution of the A36 grain size distribution, abnormal grain growth. 85
Figure 5.17 - Austenite grain size in the as-received and solution treated A36 88
Figure 6.1 - Saltikov's and Huang and Form's 3-D grain size distribution 92
Figure 6.3 - Flow chart for applying the method of Matsuura and Itoh 101
ix
NOMENCLATURE
b Burger's vector
c lower limit of the smallest size class for the geometric scale
d grain diameter
d b average diameter
d m mean diameter
d R relative diameter
class i
class i
x
fa calculated fraction of grains with equivalent area diameter in size
class i
fpi fraction of grains in size class i, in contact with the edge of the
measurement field
F P pinning force
Npi number of grains of size class i which touch the edge of the field of
measurement
P pinning parameter
xi
Pp(d) probability that a grain is in contact with the edge of the field of
measurement
Psid) probability that the section diameter is less than d for a sphere
R gas constant
R 2
correlation coefficient
distribution
distribution
S 2
sum of the differences squared
t time
T temperature
V grain volume
xii
ACKNOWLEDGMENTS
This work is a part of a group project to model the industrial process of hot strip
rolling. In completing this thesis, I have worked closely with other members of the
research group, in order to make progress toward our common goal. The cooperative
atmosphere has been the catalyst for many group discussions regarding experimental
methods and the interpretation of results. I would like to thank all members of the hot
strip mill modelling group for their contributions to this work. In particular, Dr. Matthias
Militzer deserves a special mention for his guidance and support throughout the course of
this research. I am also grateful for the technical assistance provided by Mr. Rudy Cardeno
and Mr. Binh Chau and for the support and guidance of Dr. Bruce Hawbolt. I would like
to acknowledge the financial support provided by the American Iron and Steel Institute
I owe thanks also to my wife, Katrina, for the patience and support she has so
generously shown me, especially during the later stages of this work.
xiii
CHAPTER 1
INTRODUCTION
Steel is a versatile material which has found widespread use because of its
exceptional mechanical and magnetic properties, its relatively low cost, and the ease with
which it can be used in manufacturing processes such as forming, welding and machining.
The steel industry has experienced great successes as a result of these characteristics.
atmosphere in many industries, and steel producers have been forced to do battle with
each other, and with other materials producers for a share of a market which is demanding
higher quality products at lower cost. This has driven steel producers to improve their
process modelling, which links measurable processing parameters to the final properties of
the product. Process models are an excellent knowledge base for off line process
simulation, and ultimately, it is likely that models will be used for on line process control.
The main steps in the production of steel are refining of the material in the molten
state, followed by casting to form a solid product, then deformation of the solid to give
the specified mechanical properties, and a more useful product shape. If the desire is to
produce steel in form of sheet, the first stages of deformation will be performed by hot
rolling.
The process of hot strip rolling is shown schematically in Figure 1.1. A steel slab is
heated to temperature in a reheat furnace. Deformation begins in the roughing mill where
the slab is reduced in thickness by a series of rolling passes. The slab leaves the roughing
null with Utile change in width, but with significantly increased length and reduced
1
2
thickness; at this stage the steel is referred to as transfer bar. The next stage of
deformation is the finishing mill, at the end of which the material is referred to as strip.
Frequently, the aim is to complete all deformation while the steel is austenitic. The
rate, using water sprays on the runout table. At the end of the runout table the material is
Roughing mill
Reheat furnace
metallurgical phenomena which accompany hot rolling are grain growth, phase
product dictates its properties; the control of the metallurgical phenomena can ensure that
Process modelling of hot strip rolling is a sensible approach to achieving the goals of
improved product quality and reduced cost by the application of improved technology.
The modelhng task requires the luiking of the process parameters with the kinetics of the
experimentally, under conditions of time and temperature which are relevant to the
3
process of hot strip rolling. Measurements are to be used to develop and validate a model
to describe the kinetics of austenite grain growth during hot strip rolling. The model to be
used will have a fundamental basis, and the phenomenon of grain growth will be treated as
the evolution of a distribution of three dimensional grains. Grain size measurements are
generally performed on planar sections through the structure, and direct measurement of
the three dimensional grain size is a difficult task. However, procedures have been
reported to estimate the three dimensional grain size from two dimensional measurements.
The grain growth model is one component of the process model being developed by
an engineering research team at the University of British Columbia (UBC); the author of
this work is a member of that team. The U B C team is working in conjunction with United
States Steel and the National Institute of Standards and Technology. The aim of the U B C
group is to develop a versatile and fundamentally based model for the prediction and
control of the microstructure and mechanical properties of steel during hot rolling. The
model will be developed for a range of mill processing conditions and steel compositions,
including plain carbon, interstitial free and microalloyed grades. In this stage of
development of the process model, investigations have been directed toward the plain
carbon steels. The deliverable from this research will be a PC (personal computer) based
model, which will link the processing variables to the final microstructure and properties
of the product. Application of this model will allow the steel producer to deliver a higher
LITERATURE REVIEW
Several workers have developed process models for the prediction and control of
microstructure and mechanical properties during hot strip rolling of steel [1-5]. Process
models which have been developed are generally based on empirical relationships which
relate processing parameters to material structure and properties. A more desirable model
would have a fundamental basis. Such fundamental models will be capable of accurately
describing the thermal history of the strip, the rolling forces at each rolling pass, and the
and phase transformation, which accompany hot strip rolling. The kinetics of the
laboratory equipment which is able to approximate the thermal and mechanical conditions
in the mill. The individual events are combined using a model which describes the thermal
and mechanical history of the strip. Since the microstructural evolution can be quantified
and related to the stresses of deformation, the mechanical properties at each stage can be
anticipated; consequently, the mill forces can be predicted based on the properties. The
resulting model must then be validated by performing measurements on the mill under
industrial conditions to verify temperature and force predictions. The structure must also
be assessed in the hot rolled product, to verify microstructural predictions. This approach
is being taken by the U B C team. Here, one of the metallurgical events to be modelled is
4
2.2 Modelling of Grain Growth
Burke and Turnbull were among the first to model the kinetics of grain growth [6].
They assumed that the driving force for growth is the reduction in grain boundary area and
the subsequent release of boundary surface energy, 7^. The model was proposed for the
case of normal grain growth in a pure homogeneous material and the assumption was
made that the only forces acting on boundaries are those due to surface curvature. It was
also assumed that the average radius of curvature of a boundary is proportional to the
grain diameter, d, and that the rate of change of the diameter is proportional to the driving
^ = k.F
dt 2.1
Since the driving force for growth is assumed to be directly related to boundary area
per unit volume, it can be shown that F is proportional to the inverse of grain diameter, d.
By integration of equation 2.1, Burke and Turnbull were able to derive the parabolic
d -dQ=Ktexp(-Q/RT)
2
2 .2
where dg is the grain size at time equal to zero, t is time, AT is the growth constant, Q is the
activation energy for grain growth, R is the gas constant and T is absolute temperature.
The grain boundary energy, y , is included in the constant, K, and the exponential term
gb
Feltham in 1957 [7] and Hillert in 1965 [8] derived equations of the form of
equation 2.2 to describe the kinetics of grain growth. Feltham focused on the relationship
6
between driving force and boundary mobility, whereas Hillert concentrated on the effect of
The parabolic growth law has been found to apply to the increase in mean cell size
of a froth of soap bubbles [9]. This implies that the analysis is geometrically correct, since
in a network of soap bubbles, as in a network of grains, the driving force for growth is the
decrease in boundary surface area [6]. Monte-Carlo simulations of growth have also
yielded growth kinetics which approach those predicted by the parabolic law [10].
Unfortunately, experimental results often do not obey the parabolic equation. For
this reason, the original equation [6] is often modified and used in the form,
d -d(? = K t.exp(-Q/RT)
m
0 2.3
where m is the grain growth exponent and K is the fitting constant. The equation is
0
A short summary of experimental results from very pure zone refined materials,
which have been described using the power law, was compiled by Anderson [10] and is
shown in Table 2.1. The table demonstrates that m is rarely equal to two, even for
extremely pure metals. It is likely that this is due to the influence of imperfections such as
solute atoms and voids which limit boundary mobility, or have a boundary pinning effect
[6,7,10,16].
7
zone refined
Al 4.0 [11]
Fe 2.5 [12]
Pb 2.5 [13]
Pb 2.4 [14]
Sn 2.3 [14]
Sn 2.0 [15]
Table 2.1 - Grain growth exponents for the power law, for zone refined metals [10].
the assumption that the only forces which influence boundary motion are due to surface
curvature. In Burke and Turnbull's derivation [6], this assumption leads to the
proportional relationship between driving force and the inverse of grain size, the
integration of which leads to the exponent of two. Atkinson [16] suggests that the
mechanisms which influence the value of m include the effect of void size and distribution,
the presence of solute atoms, solute segregation, second phase particle size and
distribution, and texture effects. A study of these effects would be difficult to perform, due
to the small size of the voids and particles, and the low concentrations of solute which are
The power law has often been used to describe the kinetics of austenite grain
growth in process models for the hot rolling of steel [1,3-5]. The kinetics of austenite
grain growth are usually investigated by reheating and holding specimens of steel at a
constant temperature for varying times. A more elaborate cycle involving deformation as
8
well as heat treatment may be used to replicate grain growth during hot rolling. The
specimens are quenched and etched after the discrete holding time to reveal the grain
structure and allow the austenite grain size to be measured. A table of data relating grain
size to time at temperature is produced and models are developed by fitting the results to
the power law. The parameters which can be used for fitting are m, K and Q. Table 2.2 is
0
Table 2.2 - Power law parameters reported for austenite grain growth.
The use of the power law for ml is empirical. Therefore, the resulting equations
should only be used over the range of experimental conditions for which the model has
been validated. Since it is impossible to carry out experiments for all steels and all
conditions encountered in a hot strip mill, it is desirable to develop a more versatile model
which is based on fundamental principles, which can be validated with limited data.
grain growth models [19-23]. The factors which are known to influence grain growth
kinetics are the driving force for grain growth, the mobility of the grain boundary, and the
9
distribution of grain sizes in the structure. A statistical model is able to account for each of
these.
To understand the influence of the width and shape of the grain size distribution, it is
have been directed towards this task [24-27]. A grain boundary has a surface energy and
the arrangement of the boundaries is such that the boundary area tends to be minimised
[6]. It is helpful to visualize this effect as being equivalent to surface tension. Atkinson
[16] has summarized the geometry of the structure and the trends which are a
consequence of the driving force for growth, as illustrated for a two dimensional structure
in Figure 2.1. Generally, in such an array, only three grain boundaries meet at a corner,
with the angle between them being 120. In a three dimensional array, three grain
boundaries meet at a grain edge at angles of 120. At a corner, four edges meet at
10928'. A polycrystalline metal is made up of grains which vary in size and shape. To
satisfy the geometrical considerations, grain boundaries must be curved surfaces. The
smallest grains will have boundaries with the lowest radius of curvature, with their centre
of curvature inside the grain. Larger grains will have boundaries with a higher radius of
curvature, with their centre of curvature outside the grain. The motion of a grain boundary
will reduce grain boundary area if the boundary is curved; the boundary moves toward the
centre of curvature. Since the driving force for growth is the liberation of energy by the
reduction of boundary area, grains which are large relative to the grains around them tend
to grow and grains which are small tend to shrink. Therefore, an important factor to
consider when modelling the kinetics of grain growth is the width and shape of the grain
size distribution.
10
Abbruzzese and Liicke [20,21] have demonstrated the use of a statistical model
which has been developed from the work of Hillert [8]. Abbruzzese and Liicke were able
to use the statistical model to describe grain growth under the influence of texture and in
the presence of second phase particles. For austenite grain growth in low carbon steels on
the hot strip mill, the pinning effect due to second phase particles is of primary
importance.
The statistical model for grain growth which accounts for the pinning effect of
The driving force for growth, F, is the liberation of grain boundary energy, y .
gb
Grains of size / can be described using which is the radius of a sphere of the same
volume as the grain. The relationship between the grain boundary surface area and the
The grain size distribution can be approximated by subdividing the distribution into
If a grain from size class i is larger than its neighbor from size class j, a driving force for
F
ij = Vgb
yj
R R
t )
F
ij F
ji 2.4
The growth rate for grains in size class i can be determined by assuming grains from
class i are surrounded statistically by grains from all other classes. Growth rate is
calculated using
dR
i
^Mg^WijFij
2.5
where is the probability that a grain from class i is in contact with a grain from class j
and M gb is the grain boundary mobility. Grain boundary mobility is a function of the grain
boundary diffusivity, the magnitude of the burgers vector, and the temperature.
12
The probability that a grain from class i is in contact with a grain from class j can be
fvj j
R
wa =
2.6
The fitting parameter for the model is P. The model is used by performing a
calculation for each time step to determine the number of grains entering or leaving each
size class. Smaller grains tend to srmnk and disappear while larger grains tend to grow, so
The model treats grains as three dimensional objects. The unique growth kinetics for
each grain size class are determined by considering the influence of grains from all
surrounding size classes. Therefore, the width and shape of the three dimensional grain
the three dimensional grain size distribution. This will be described in a later section.
Grain growth models are generally used to describe normal grain growth, which is
the continuous increase in mean grain size with time. Atkinson [16] describes normal grain
growth as scaling, since the grain structure can be made to appear identical at any point in
time by changing the magnification of the image. During normal grain growth, the width
and shape of the grain size distribution do not change significantly with time.
Abnormal grain growth can also occur, when discontinuous growth kinetics are
observed. During abnormal growth, the largest grains in the structure tend to grow much
faster than the grains around them, and the grain size distribution broadens. The growth
13
behaviour is unstable until the larger grains consume almost all of the smaller grains and a
It is generally accepted that abnormal grain growth is due to the pinning effects of
second phase particles, as discussed in recent review articles by Gladman [28], and
Worner and Hazzledine [29]. It is generally accepted that much of the progress in this field
has resulted from the work of Zener, as quoted in a classic paper on microstructure by
Smith [30]. As a result, the effect of particle pinning has become known as Zener pinning.
A moving boundary can be pinned by a second phase particle or pore in its path.
When a distribution of particles are present, a net pmning force will exist, the magnitude
j-,
3 v
v
Ar 2.7
where r is the particle radius, v is the second phase particle volume fraction and ygb is the
grain boundary energy [30]. Grain growth in the presence of particles will be affected if
the magnitude of the pinning force is significant compared to that of the driving force for
growth. The growth rate can approach zero as the grain radius tends toward a linnting
value, Rn . Zener proposed that Rn can be calculated from the pinning force. Since the
m m
pinning force is related to the pinning parameter, P, used in the statistical model for grain
_J_
Rlim ~ p 2.8
A grain size distribution which has a mean grain size approaching the lairing value
can become stagnant. However over time, second phase particles tend to either coarsen or
14
dissolve, depending on the equilibrium solubility of the second phase. As this happens, the
size and distribution of the second phase particles will change. Since the pinning force is
dependent on the size of the particles, the j^iting grain radius, Rn , will also change with
m
time. The grains which have the highest driving force for growth are those which are the
largest relative to the grains around them. They are the first to break free from the piruiing
particles and they consume the matrix of smaller grains which remain pinned. Gladman
6RQV (3 r RQ 1
-2
K
l 2 -Rmax -)
for the relationship between the size of the pinning particles, r, the volume fraction of the
second phase, v, the average grain radius, R and the radius of the largest grains present,
0
appropriate ratio of RQ to R ,
MAX Gladman was able to predict the time and temperature for
the onset of abnormal grain growth for a number of steels. In plain carbon steels, Gladman
showed that the likely identity of the second phase particles which cause abnormal grain
growth is aluminum nitride. This result has been reported also by other workers [33,34].
A grain is a three dimensional object with a size which can be characterized by its
volume. However, for the sake of simplicity, grain size tends to be described using units of
length. Most measurements of grain size are obtained from a planar section through the
three dimensional structure. The measured value, which is a two dimensional result, is not
equal to the true three dimensional grain size. In this work, grain size measurements are to
15
be used to validate a statistical model for grain growth. It is the three dimensional grain
size distribution that is of interest since the statistical model is fundamentally based.
The three dimensional grain size can be described by the equivalent volume
diameter, d , which is the diameter of a sphere with the same volume as the grain [35], as
v
defined by
2.10
The three dimensional grain size distribution, f(d ), can be obtained by measuring
v
material is not easy to measure. Individual grains would have to be observed by complete
most investigations of grain growth kinetics, these procedures are impractical. As a result,
the three dimensional grain size distribution, f(d ), is often estimated from the two
v
Measurement of the two dimensional grain size can be made by determining the area
of each grain visible in the planar section through the structure. The two dimensional grain
2.11
dimensional grain size distribution,/^), which can be used directly to estimate the three
16
dimensional distribution [36,37,39]. The two dimensional arithmetic mean equivalent area
diameter, d , is defined as the sum of all grain diameters divided by the number of grains.
Ab
The mean grain area, A , is defined as the sum of the area of all grains, divided by
m
the total number of grains. By the substitution of A m for A in equation 2.10, the geometric
mean equivalent area diameter, d , Am can be calculated. The mean area can be determined
by using quantitative image analysis equipment to measure the area of each grain, as
Mean grain area can also be determined using Jeffries' method, as described by
A S T M E l 12 [41]. A simple count of the number of grains is all that is required, and the
mean area is calculated by dividing the image area by the number of grains. The major
difference between the measurement of mean area by image analysis and by Jeffries'
method is the treatment of grains which touch the edge of the field of view. Using the
image analyser, only whole grains are measured. Therefore grains which touch the edge of
the frame are ignored. Using Jeffries' method, each grain which touches the edge of the
method. Again, measurements can be made using image analysis equipment, as described
methods require that lines of known length are superimposed on the planar section of
grain boundaries, and each intersection of a line and a boundary is counted as one
intercept. The mean linear intercept length, l , is the total line length divided by the
m
number of intercepts. A number of methods have been proposed which use the linear
intercept diameter for the estimation of the three dimensional grain size distribution
[35,38].
17
random planar section through a three dimensional structure. The number of grains which
are measured must be significant so that errors due to the randomness of the section are
avoided. A S T M E1382 [40] and E l 12 [41] suggest that a minimum of five separate fields
are measured, and that at least fifty grains are visible in each field [40,41].
The methods previously described are satisfactory for cases where the grain
structure is relatively uniform. For structures with a grain size distribution which is
bimodal (composed of fine and coarse grains), A S T M E l 181 contains some additional
suggestions [42]. For structures where the fine and coarse grains are distinctly separate, a
mean grain size can be determined in each region by measuring the grain size in each
region using standard techniques already described. The area fraction of each region can
The overall mean grain size can be calculated by combining the mean and area fraction of
each region.
It is generally thought that during normal grain growth, the three dimensional grain
can be described by the peak grain size, d, and the standard deviation, s, [23,38] using
2.12
18
i=l 2.1.3
For a three dimensional distribution, Equation 2.12 and 2.13 are modified by the
addition of the subscript, V. For example, d becomes d . The mean equivalent volume
v
diameter, d , is given by
Vm
2.14
The two dimensional grain size distribution which exists during normal grain growth
equations 2.12 and 2.13. Other distributions have been proposed [16,35] which describe
experimental data with a limited improvement in accuracy, but the log normal distribution
class, i=l
class, i=k
discrete size classes. The class limits are usually arranged using an arithmetic or a
geometric scale, as described by Table 2.3 [35]. The arithmetic scale is described by the
class width, A, and the number of classes, k. The geometric scale is described by the lower
limit of the smallest size class, c, the geometric multiplying factor, a, and the number of
classes, k.
A number of mathematical conversions have been proposed for estimating the three
dimensional grain size distribution from two dimensional measurements. Some methods
are based solely on the measurement of the two dimensional linear intercept length
20
[44,45], while others are based on the measurements of area [36,37,39], or on the
Saltikov [36] and Huang and Form [37,46] assumed the grain shape to be spherical.
Saltikov proposed a method which requires the measurement of the two dimensional grain
size distribution, f{dA). The method is used to apply a correction to the number of grains
observed in each size class. Huang and Form proposed a method which is similar to the
Use of Saltikov's method requires that the following assumptions are made. Firstly,
grains are assumed to be spherical in shape. Secondly, the measured distribution can be
represented by dividing the data into a number of discrete size classes. And finally, the
three dimensional grain size for each size class is not a range of grain sizes, but it is the
upper limit of that size class. That is, the number of grains in the size class in the two
dimensional distribution is equal to the number of grains which have a diameter which is
between the upper and lower hmit of the size class. But, in the calculated three
dimensional distribution, all grains in a size class are assumed to have a diameter equal to
The method is based on the probable diameters which result from intersection of a
plane and a sphere, as illustrated by Figure 2.2. The shape of the intersection between a
plane and a sphere of diameter, d , is a circle. The diameter of the circle, d , will be less
v A
than or equal to dy. By definition, d is the equivalent volume diameter, while d is the
v A
It can be easily shown that the probability, Ps(d ), that the measured diameter is
A
2.15
Hence the probability that the measured diameter is between d Ai and d ^ will be the
A
For a distribution of spherical grains of different sizes, the problem becomes more
complex. A grain section with measured diameter, d , must be from a grain which has a
A
true diameter, d , that is greater than or equal to the measured one. Considering the
v
largest size class, the true number of grains in this class will be greater than the observed
number, since some of the grain section diameters measured in lower size classes are due
to truncation of grains from the largest size class. The number of grains per unit volume
1-Hdvk-l) 2.16
where dy^ is the upper limit of the size class for the measured data [36].
Only one correction is applied to the largest size class, since it is assumed that there
is no chance that grain sections observed in that size class actually belong to spherical
grains from a larger size class. For all size classes smaller that the largest one, other
corrections must be applied. The number of grains observed in classes smaller than the
class of interest, which actually belong in the class of interest, is accounted for. This
increases the number of grains in the size class. Then, the number of grains observed in the
23
class, which belong in size classes larger than the class of interest, is determined using
equation 2.15. This number is subtracted from the result. A general equation can be
written,
2.17
where C is the coefficient for the nth term and k is the number of size classes.
n
The work of Saltikov is not unique in its approach to estimating the three
dimensional distribution, as others have used a similar theory [47,48]. Methods were
developed in the earlier part of this century at a time when advanced means of calculation
were not available, and tables of coefficients were generally reported. Each method can
example, Saltikov provided coefficients for twelve size classes which are geometrically
Saltikov's method applies a correction only for the probability that the planar section
reveals a diameter which is less than the diameter of the grain; this is called truncation
[35]. A second correction should also be applied to account for the fact that it is more
likely that a large grain will be intersected by a planar section than a small grain; this a
correction for sampling [35]. Huang and Form [37,46] proposed a method which is
identical to Saltikov's, except that it accounts for sampling as well as truncation. If a cube
of unit volume is considered, the correction for the sampling is estimated by dividing the
2.18
24
The main problem reported for methods based on the assumption of spherical grain
shape is the prediction of a negative number of grains for the smaller size classes. Huang
and Form reported such a result [46]. Aaron et al. suggested that negative counts can
occur if some of the smallest grains are not detected when the measurements are being
performed [49]. The smallest classes are most affected, since errors are compounded by
the nature of the corrections applied. The biggest source of error is likely to be the
assumption that the grain shape is spherical, since an array of spherical grains cannot fill
space [49]. A number of methods have been proposed which make use of more realistic
A method has been proposed by Takayama et al. to estimate the three dimensional
grain size distribution from the measured mean area , A , and the mean linear intercept
m
length, l , [38,51]. This method assumes that all grains are tetrakaidecahedral in shape,
m
and therefore the grains are space filling. In addition, the grain size distribution is assumed
to be log normal and grains of different size are distributed randomly in space.
The first step taken by Takayama et al. [38,51] in the development of the method
was the determination of the probability distribution of the linear intercept length. The
intersection of a random line with a solid object of volume, V, will produce an intercept of
number of random lines are constructed, a probability distribution which relates / to dy can
determine the probability distribution for the tetrakaidecahedral grain shape and for a log
normal distribution of grain sizes [52]. The mean linear intercept length was determined to
be a function of the width of the grain size distribution. A n equation was fit to the
25
simulation results relating the standard deviation of the log normal distribution, s , and the v
peak volumetric grain size, d , to the mean linear intercept, l . The mean area, A , and
Vg m m
mean linear intercept, l , were related to the arithmetic mean volume diameter, d , using
m vt)
it was possible to relate the mean area, A , and the mean linear intercept, l , directly to
m m
= 0.6066W ^exp(5(lnsv) / 2 ) 2
"m y
2.19
A = 0.48610 i ^ exp(4(ln^)
m L y
2
2.20
These equations can be rearranged and combined with equation 2.14 to give
(
s =exp 21rJ 1.14935
v
2.21
2^
0.82575A2.5 3(hu ) v
-exp
li
2 . 2 2
which can be applied to the measured results, to estimate the three dimensional grain size
distribution.
Jeffries' method for measuring mean area, A , and Heyn's method for measuring
m
mean linear intercept, l , are used [41]. These parameters are easily determined with or
m
without image analysis equipment, making it possible to obtain a quick estimate of the
three dimensional grain size distribution. However, it should be emphasised that this
26
method is limited by the assumption that the grain size distribution is log normal, and that
all grains are tetrakaidecahedral in shape. In fact, a range of grain shapes have been
observed [24-27,43,54].
An alternative method for estimating the three dimensional grain size distribution,
based on a range of grain shapes, has been proposed by Matsuura and Itoh [25,39]. They
used twelve different types of regular, equiaxed polyhedra to represent the range of grain
shapes observed in a polycrystaUine solid [39]. By examining the work of Rhines and
Patterson [43], they were able to determine a simple relationship between the number of
grain faces, / , the equivalent volume diameter for the grain, dy, and the mean equivalent
2.23
Matsuura and Itoh [39] used a similar approach to that taken by Takayama et al.
[52] to determine a probability distribution for the equivalent area diameter. The
intersection of a random plane with a solid object having volume, V, will result in a
polygon with area, A, and appropriate values of dy and d can be easily calculated. If a
A
large number of random planes are generated, a probability distribution for the equivalent
area diameter can be constructed. Matsuura and Itoh chose to describe the probability
performed computer simulations using the twelve grain shapes to determine the probability
distribution curves were characterised by the position and magnitude of the peak
27
probability and the maxirnum value of the relative diameter, d . These parameters were
RM
The method is applied by measuring the two dimensional grain size distribution,
f{d ), using quantitative image analysis. The measured distribution is then divided into a
A
number of discrete size classes. Each size class, /, is made up of grains having a measured
diameter between d Ai and d ^.iy A l l grains in size class i for the three dimensional
A
distribution are represented by the upper limit of the size class, d , which is equal to d .
Vi Ai
It is assumed that a three dimensional size class contains grains of only one shape, and the
approximation for d . There is a finite probability that the relative diameter is equal to
vb
any value between zero and d RMi . The measured fraction of grains in each size class, f i, is
A
assumed to be equal to the sum of the probability distributions for each size class, / Q . The
method is illustrated for three size classes in Figure 2.3 [39]. In the figure, the hatched
region under each curve is representative of the integration of the probability function
from each size class, over the range of diameters in the second size class. The results from
each integration can be multiplied by the true three dimensional fraction of each size class
(fvbfv2 and/yj), then summed, to determine the fraction of grains observed in the two
dimensional distribution, in the second size class, f 2- In reality, it is the two dimensional
A
distribution that is most easily measured, and the true three dimensional fractions are
2.24
where k is the number of size classes, and fyj is the true fraction of grains in each size
class. The fraction of grains which should be observed in the two dimensional distribution
in size class i will be equal to f . Ci Note that the last term in the equation accounts for the
probability that a grain belonging to size class (/-l) has produced a section area with an
equivalent area diameter that belongs in size class i. This is illustrated in Figure 2.3, where
the tail of the probability function from size class 1 is integrated from 1.0 to d j, to RM
account for the section diameters which will be observed in the two dimensional
The three dimensional grain size distribution is known when values of f Vi are
determined for size classes from i=l to i=k. Equation 2.24 is used to create a matrix of k
equations with k unknowns. A solution is determined for/y; to fy^ by nrinmising the sum
2.25
Values of f Vi which are determined by solving the matrix of equations are used to
calculate a revised mean volumetric grain size, dy . New values of Jj are determined using
b
equation 2.23 and probability functions are also redefined. The new functions are used in
30
equation 2.24 to solve again for f . The calculations are repeated until the values of f
Vi Vi
The method of Matsuura and Itoh does not rely on unrealistic assumptions regarding
the type of grain size distribution or the grain shape. To apply the method, the user needs
to create a routine using computer code, making the method less convenient than the
The goal of this research is the measurement of austenite grain growth kinetics in
plain, low carbon steels. Relevant time and temperature conditions are selected to reflect
the industrial process of hot strip rolling. Area and linear intercept grain size
measurements are made and used for the validation of a statistical model for grain growth.
This study has two objectives. Firstly, austenite grain size will be measured in two
temperature and time conditions which are relevant to hot strip rolling. The measured
Secondly, an appropriate method for the estimation of the three dimensional grain
size distribution will be determined. The methods proposed by Saltikov [36], Huang and
Form [37], Takayama et al. [38] and Matsuura and Itoh [39] will be compared. The most
suitable methods will be used to estimate the three dimensional grain size distribution, to
be applied to the statistical model for grain growth described by Abbruzzese and Liicke
[20,21].
31
CHAPTER 4
EXPERIMENTAL
Samples of two plain carbon steels, denoted as A36 and DQSK, were obtained from
United States Steel (US Steel) Gary Works. Specimens were cut from the samples, to be
thermally cycled in a previous study [55]. Measurements from the 1080 steel were used
for comparing the methods for estimating the three dimensional grain size distribution.
The chemical compositions of the three steels used in this study, in weight percent,
are reported in Table 4.1. The A36 and DQSK steels differ primarily in carbon content and
manganese content, with the A36 being more highly alloyed. A low alloy content makes
the austenitic microstructure difficult to reveal, affecting the specimen shape and the
design of experiments. The 1080 steel was chosen for its higher alloy content since the
austenitic microstructure in this material can be more easily revealed. A l l steels are
aluminum killed. The chemical analysis of the 1080 steel is included in Table 4.1.
Steel C Mn P S Si Cu Ni Cr Al N
A36 0.17 0.74 .009 .008 .012 .016 .010 .019 .040 .0047
DQSK 0.038 0.30 .010 .008 .009 .015 .025 .033 .040 .0052
32
33
Inspection of the Fe-Fe3C phase diagram shows that for these carbon levels, the
carbide [56], these products resulting from the decomposition of high temperature
austenite. In order to reveal the location of the prior austenite boundaries, it is necessary
to quench at a high rate so that the decomposition of austenite does not produce a
nucleates at the austenite grain boundaries. If the amount of ferrite can be limited to
between 5% and 15%, it will outline the prior austenite boundaries. Alternatively, if a
higher quench rate is achieved, the diffusional transformation of austenite can be averted
and a martensitic structure will result. Certain etchants have been used to reveal the prior
Samples of the A36 and DQSK steels were received in the form of transfer bar,
taken at the end of rough rolling on a hot strip mill. While many specimens were thermally
cycled in the as-received condition, some of the A36 specimens were solution treated in an
air furnace for 3 hours at 1200C, prior to being tested. These specimens were sealed in
quartz tubes under a partial vacuum to avoid excessive oxidation and decarburization. The
specimen temperature was not monitored directly, but furnace temperature was
maintained within 10C of the set point. After holding for 3 hours, specimens were
The size and shape of specimens machined from the transfer bar were determined by
the equipment to be used to apply the thermal cycle. The A36 tests and a few D Q S K tests
so that a quenching fluid could be passed down the axis after the thermal cycle. The
marketed by Dynamic Systems Inc., Troy, N Y . The minimum tubular wall thickness was
limited to the diameter of ten austenite grains, in order to rninimize the effects of the free
34
surface on the grain growth kinetics [6]. The maximum wall thickness was timited by the
quench rate required to reveal the structure. Two tubular geometries were chosen, each
20mm in length with an inner diameter of 6mm. The wall thickness was either 1mm or
A vertical tube furnace was also used, to perform many of the tests for the D Q S K
steel. The specimen thickness was again limited to a rmnimum of ten grain diameters and a
maximum determined by the required quench rate. Specimens were rectangular in shape,
15mm wide, 25mm high and either 1.5mm or 3mm thick. A 2mm diameter hole was
drilled through the thickness at the top of each specimen so that it could be hung in the
furnace. Figure 4.2 illustrates the specimen design for the vertical tube furnace.
35
15 mm
E
LO
temperatures in the range of 900C to 1200C, using soak times extending to 1200
seconds. It was possible to impose carefully controlled thermal cycles using computer
controlled resistance heating capabilities of the Gleeble. During heating, the specimens
were held in a vacuum of less than 10~ Torr to minimize decarburization and oxidation.
4
Figure 4.3 is a schematic diagram of the specimen support in the Gleeble test
chamber. Each thermal cycle, including the quench, was programmed into the Gleeble
control system. Specimens were prepared by cleaning and spot welding the thermocouple
wires. Wires were welded at the specimen mid-length with each wire aligned with the tube
Figure 4.3 - Schematic diagram of the specimen support in the Gleeble test chamber.
37
During the thermal cycle, the temperature was controlled by a feedback system
operating at a sampling rate of 100Hz. Temperature was monitored using the output of
the Pt/Pt-10%Rh, Type S, thermocouple intrinsically spot welded to the specimen surface.
heating was excellent for the lower heating rates of 5C/second and 50C/second.
However, reproducible temperature overshoot was observed for the higher heating rates,
since a time delay exists between the application of power and the resultant temperature
change. The magnitude of the temperature overshoot was dependent on the heating rate,
Temperature control during the isothermal holding period was excellent, always
Most specimens were quenched by passing hehum down the tube axis. Higher
quench rates of up to 250C/second were achieved by also directing hehum onto the outer
specimen surface. Water was used as an alternative to hehum when the maximum quench
rate was required; rates as high as 600C/second were achieved. However water
quenching was not a favorable option since water vapor contaminated the Gleeble vacuum
system.
The DQSK steel required an extremely high quench rate to retain the outline of the
prior austenite microstructure. The vertical tube furnace was used for most of the D Q S K
tests to avoid the problems associated with water quenching in the Gleeble, and to achieve
the highest possible quench rates (approaching 1500C/second). The heating rate and
thermal history of specimens heated in the tube furnace was dictated by the laws of heat
38
transfer. During the thermal cycle, oxidation was minimized by directing argon past the
The furnace consists of a vertical alumina tube, 900mm in length, with an inner
diameter of 38 mm and an outer diameter of 44 mm. The tube is electrically heated using
thermocouple feed back, the position of the control thermocouple being outside of the
alumina tube. The furnace was set to 22C above the desired specimen temperature, this
offset being experimentally determined. The furnace temperature was allowed to stabilize
for a period of at least 8 hours at each temperature before the tests could be performed.
The furnace rheostat was set so that the controller was on a heating cycle and on a cooling
cycle for equal periods of time, the control limits being within 5C.
The specimen was hung in the furnace from a chromel wire 1.2mm thick. The wire
was also used to support a Type S thermocouple, electrically insulated by a twin bore
alumina sheath. The chromel wire and thermocouple arrangement were housed in an open
ended quartz tube with an inner diameter of 12mm and an outer diameter of 14mm. The
quartz tube allowed for the motion of the apparatus within the alumina furnace tube
during quenching. Above the furnace, a support arrangement was used to locate the
Thermocouple, type S
Figure 4.4 - Schematic diagram of the vertical tube furnace and specimen holder.
40
thermocouple mtrinsically spot welded on the centre of the broad face, 3mm up from the
specimen's base. The thermocouple wire diameter was 0.25mm. Temperature was
monitored using a digital display with cold junction compensation. The thermocouple
output was recorded by plotting the millivolt signal directly on a Kipp and Zonan, having a
pen speed of 2mm/second and operated on a 20mV full scale. The specimen thermal
history was monitored during heating, soaking and quenching, the chart recorder plots
Heating rates were a function of the furnace temperature and specimen thickness.
The vertical position of the specimen was kept constant for all tests, at a location 25mm
from the furnace temperature control thermocouple. Soak temperatures were 1000C,
1050C, 1100C and 1150C. Due to the expected grain size, specimens 3mm thick were
used at the highest temperature. At lower temperatures, 1.5mm thick specimens were
used. The surface of the specimens generally reached the desired temperature within 3
After reaching temperature, specimens were held for periods of up to 420 seconds.
The reproducibility of the thermal history for each specimen was verified by comparing the
chart recorder plots in the temperature range above 900C. Each specimen was assumed
to have reached temperature when the temperature interpreted from the plot was within
5C of the desired soak temperature. Figure 4.5 illustrates a typical furnace thermal cycle.
solution of iced salt water, maintained at a temperature of approximately -10C. The salt
water bath was placed at the bottom of the vertical tube, as illustrated in Figure 4.4, so
that the transport time from the furnace to the quenchant was minimized.
41
Most thermal cycles involved simply heating specimens to the desired temperature,
holding, then quenching. However, some Gleeble tests used more elaborate treatments,
such as, multiple heating rates and hold times, and periods of controlled cooling prior to
quenching. The heat treatments can be divided into four groups: 1. Standard Gleeble
thermal cycles, 2. Furnace tests, 3. Stepped Gleeble thermal cycles, and 4. Continuous
cooling tests.
Figure 4.6 shows a schematic time versus temperature plot for a standard Gleeble
thermal cycle. Table 4.2 gives an overview of all standard Gleeble tests which were
performed.
thick., Rate
mm C/sec.
A36 as-received 2 5 900 120
A36 as-received 2 5 950 10, 120, 600
A3 6 as-received 2 5 1000 10, 60,120, 600, 900
A36 as-received 2 5 1050 1,10, 60,120, 300, 750
A36 as-received 2 5 1100 1,10, 30, 60, 120, 300, 600
A3 6 as-received 2 5 1150 1,10, 30, 60,120, 300, 450
A36 as-received 2 5 1200 300, 600
A3 6 as-received 2 50 950,1050,1150 120
A36 solution treated 2 50 950,1050,1150 120
DQSK as-received 1 5 1050 120
DQSK as-received 1 5 1100 60, 120
DQSK as-received 1 5 1150 60
temperature plot shown in Figure 4.5. Table 4.3 summarizes the standard furnace cycles
seconds.
Stepped thermal cycles applied to A36 specimens on the Gleeble are described in
time
Specimens to be used for grain size measurement were cut at the thermocouple
location, to ensure a known thermal history. Specimens were mounted in a cold setting
aery he resin and subsequently ground and polished to a l[im diamond finish. A number of
different etchants were used, depending on the condition of the specimen. These etching
metallograph system using black and white Polaroid Type 55 film at magnifications from
x80 to x900. The revealed austenite grain boundaries were traced on transparent plastic
From the traced austenite structure, the mean grain area, A , was determined by
m
Jeffries' method [41]. Each whole grain was counted once and each partial grain, cut by
the edge of the field of measurement, was counted as a half grain, as described by A S T M
standards [41]. From the mean area, the mean equivalent area diameter, d , Am was
cooling, 2C/sec.
time
H 0 quenched A3 6
2 Saturated aqueous picric acid or [56]
480 by 640 colour pixel points. The area of each whole grain.was measured individually
from the traced microstructure using standard A S T M methods [40]. An image processing
routine was developed so that the width of grain boundaries detected by the system was
49
on average, 2 pixels. The area of a grain was defined as its net internal area plus half the
area of its boundary, so that the pixels taken up by the boundary could be accounted for. It
was verified that the sum of the area of each grain was equal the total area of the image.
The image analyzer did not measure the area of partial grains. That is any grains which
touched or were cut by the edge of the field of measurement were ignored. From the mean
grain area, A, an equivalent area diameter, d , was also determined, and the two
A
rectangular grid of 12 lines was placed randomly over the tracing. Each intersection of a
grain boundary with a line from the grid was counted as 1. A line which touched a
boundary but did not cross it was counted as a half, and a line which exactly intercepted a
grain corner, where three boundaries were seen to meet, was counted as 1.5. The mean
linear intercept length was determine by dividing the total grid length by the number of
counted intercepts.
The mean linear intercept was also determined for some specimens from the traced
structure using the image analyzer and standard A S T M methods [40]. The analyzer used a
grid made up of all of the pixels on the screen. An image processing routine was used to
reduce the width of each boundary to 1 pixel. The mean intercept length was calculated as
the total number of pixels in the screen divided by the number of pixels which made up the
boundaries.
CHAPTER 5
Specimens of the A36 steel were heat treated on the Gleeble then quenched using
helium, and etched in 4% picral. Appropriate quench rates were determined to be between
heated to 1150C at a rate of 5C/second and held for 60 seconds before being quenched
The structure is approximately 10% ferrite, with the remainder being a combination
of pearlite, bainite and martensite. During the transformation, ferrite, which is white in the
print, has nucleated on the prior austenite grain boundaries in the form of grain boundary
allotriomorphs; almost all boundaries appear to be decorated in this way. Due to the high
cooling rate, secondary Widmanstatten side plates have grown out from the grain
boundary ferrite, creating a feathery appearance. At the boundary between the growing
ferrite and the austenite, pearlite has nucleated, and it is visible as the dark structure inside
the austenite grains. Within some grains, acicular regions are also visible and are thought
unimportant for the purposes of grain size measurement, as long as the prior austenite
boundaries are decorated with ferrite. Near the top left corner of the print, a lightly etched
region within a larger grain is visible, and laths of martensite are easily distinguished.
50
51
500pjn
Figure 5.1 - A36 Steel, Gleeble heated at a rate of 5C/sec to 1150C and held for
60 seconds. He quenched. Magnification is approximately x85.
Figure 5.1 shows features typical of all of the hehum quenched A36 specimens.
Larger grained specimens generally contained more regions of martensite within the
austenite grains, while smaller grained specimens had higher proportions of ferrite and
pearlite. The rate of formation of ferrite increased with decreasing austenite grain size,
requiring higher quench rates to obtain the necessary austenite grain boundary
identification. Specimens with an austenite grain size (mean equivalent area diameter),
d ,
Am of less than 20(im transformed at such a high rate that the formation of excessive
Fine grained specimens of A36 were water quenched in the Gleeble attaining cooling
A36 specimen heated to 1000C at a rate of 5C/second and held for 10 seconds. The
specimen was water quenched and etched in aqueous picric acid. The mean equivalent
50um
Figure 5.2 - A36 Steel, Gleeble heated at a rate of 5C/sec to 1000C and held for
lOseconds. H 0 quenched. Magnification is approximately x820.
2
The water quenched structure is almost entirely martensitic. The prior austenite
boundaries appear to be dark since they have been preferentially attacked by the etchant.
growth has been prevented by the high quench rate. The action of the etchant is likely to
be due to the segregation of solute to the austenite boundaries, and the lack of solute
redistribution during the martensitic transformation [57]. Residual strains from the
D Q S K specimens were water quenched after heat treatment in the Gleeble or in the
vertical tube furnace. Specimens with mean austenite grain size, d , Am of less than 40(i.m
were etched using a 4% picral solution, while those with larger grains were etched in nital,
specimen heated to 1050C in the tube furnace, held for 420 seconds before water
quenching. The specimen is etched in 4% picral to give mean equivalent area diameter,
d ,
Am of 36.2|im.
The prior austenite boundaries are decorated by ferrite, but the ferrite distribution is
non-uniform along the boundaries. Within the austenite grains, the darker structures of
pearlite and bainite are visible. The rate of the transformation from austenite to ferrite is
limited by the rate of redistribution of alloying elements (carbon and manganese). Since
the alloy content of D Q S K is relatively low, the transformation occurs very quickly. The
low alloy content also affects the quality of the etch. The quench rate was achieved by
immersing the specimen in an agitated, iced solution of saturated salt water at -10C.
Although the D Q S K steel did not etch as well as the A36 steel, some signs of the location
of the prior austenite boundaries are present. However, a more ideal structure would be
preferred.
54
l(X)|im
Figure 5.3 - D Q S K Steel, furnace heated to 1050C and held for 420seconds. H 0
2
The rate of the transformation for more coarsely grained austenite was lower and a
specimen heated to 1150C in the vertical tube furnace and held for 227 seconds before
being quenched in the iced salt water solution. The specimen was etched using nital then
The structure is similar to that of the helium quenched A36, in that ferrite has
side plates have grown. Adjacent to the decorated austenite boundaries, regions of darker
etching pearlite can be seen. Also, grains of polygonal ferrite have formed within some of
the smaller grains, nucleating at the interface between the grain boundary ferrite and the
55
receding austenite. As observed in the A36 steel, some of the intermediate regions may be
upper bainite.
50f)Lim
Figure 5.4 - D Q S K Steel, furnace heated to 1150C and held for 227seconds. H 0
2
The 1080 steel was used in a previous grain growth study [55] because it could be
quenched to martensite and etched to reveal prior austenite boundaries. The structure is
shown in Figure 5.5, for a specimen heated to 1100C and immediately water quenched.
500|jm
Figure 5.5 - 1080 Steel, furnace heated to 1100C and immediately H 0 quenched.
2
A number of dark, parallel streaks can be seen. These are thought to be MnS
A transparent tracing of the prior austenite grain structure was produced from each
a guide, the following assumptions were made to assist in the interpretation. Firstly, grains
do not exist within the boundaries of another grain. Secondly, grain corners are at the
57
The water quenched A36 and 1080 specimens were relatively easy to interpret since
the structure was predominantly martensitic and the appropriate etchant produced an
The D Q S K and the hehum quenched A36 steels were examined on the assumption
that the grain boundaries were highhghted by the nucleation and growth of ferrite
complicated the structure. In regions where little ferrite was visible, the orientation of the
pearlite provided some clues as to the prior austenite structure. In all cases, when the
visible structure was discontinuous, the typical geometry of a network of grains was
considered.
The results of the grain size measurements are summarized in the Appendix. Most
specimens were measured using A S T M standard methods, which recommend that at least
fifty grains be visible in each field [40,41]. For some specimens, high magnification was
required in order to accurately interpret the structure and as a result, fewer grains were
visible in the field of view. The magnification was chosen so that a compromise was
reached between the number of grains visible and the practical interpretation of the
structure; both factors contribute to errors in the measurement. In situations where the
total number of grains from all fields was less than 100, the results were considered to be
only an estimate of the grain size rather than a statistically valid measurement.
Some specimens had a structure of distinctly separated fine and coarse grains, as
illustrated in Figure 5.6. The area fraction of each region was measured using the image
Mean equivalent area diameter, d ,Am and mean linear intercept length, l , were
m
determined by the use of manual methods and by the use of image analysis [40,41]. A
more complete description of the measurement methods used is found in Section 4.5.
250|jm
Figure 5.6 - A36 Steel, Gleeble heated to 1000C and held for 120seconds. He
quenched. Magnification is approximately x410.
59
For each of the measured parameters, a 95% confidence interval was calculated. The
measured value in each field was assumed to be a variable distributed normally about a
true mean value. The student's ^-distribution was used for the calculation of the confidence
interval, as described by any general statistics text [65]. In most cases, five measurement
The 95% confidence intervals for the manually measured values of linear intercept
length, l had a maximum of 14.0%, a minimum of 2.2% and a mean value of 8.2%.
m
Similarly, the 95% confidence intervals for the manual measurement of the mean
grain area, A , had a maximum of 23.8% and a nidnimum of 1.4%, with the mean value
m
being 10.6%. From the mean area, an equivalent area diameter, d , Am was calculated; it is
this value that is actually of interest. Since the equivalent area diameter is a function of the
square root of the area, the confidence intervals become smaller and are a maximum of
12.3% and a minimum of 0.7%, with the mean value being 5.3%.
The mean equivalent area diameter measured on the image analyser had a 95%
confidence interval with a maximum value of 12.3%, a minimum of 1.0% and a mean
value of 5.5%.
A complete list of the 95% confidence intervals are reported with the results in
Tables A.1 to A.7 in the Appendix. The confidence intervals are a reflection of the
variability of the structure since they are calculated from measurements made on a number
of different fields from the same specimen. A large interval indicates that the grain size in a
specimen has some degree of non-uniformity while a small value indicates that the
specimens where the austenite grain structure was more difficult to reveal.
60
Comparison of the equivalent area diameter measured manually and that obtained on
the image analyser reveals that the manual measurement was consistently higher, by an
average of 7%. Calibration checks showed that the area measured for a single object by
each method was the same. The differences between the mean equivalent area diameters
were found to be a result of the treatment of grains which touch the edge of a field. Using
Jeffries' method [41], grains which touched the edge of the frame were counted as half
grains. Using the image analyser, grains which touched the edge were ignored. This
implies that in order to get the higher result, the structure intersected by the edge of the
field consistently had larger grains than the structure in the centre of the image.
In order to explain the larger grains at the edge of the image, the following
intersected by a random line than a single small grain. In fact, the probability of
intersection is directly proportional to the grain diameter. If it is assumed that the spatial
distribution of grains in a structure is random, it follows that the probability that a grain
will be intersected by the edge of a field is proportional to the diameter of the grain, since
the location of the edge of the field is essentially a random line. Obviously, the probability
of intersection is also proportional to the number of grains of that diameter which are
present in the distribution. Therefore, large grains are intersected by the edge of the field
more often than small grains. The grain size distribution measured by image analysis
ignores the edge grains and therefore skews the result towards a distribution with a lower
mean.
This argument applies to all equiaxed structures and it can account for almost all of
the discrepancy between the two mean equivalent area diameter results. Less grains are
measured by the analyser than by the manual method. The size distribution of the grains
which are ignored by the analyser can be estimated by assuming that the probability that
61
they are in contact with the edge of the field, Pp{d), is proportional to the equivalent area
grain diameter, d , and to the fraction of grains in the two dimensional distribution, f , as
Ai Ai
described by
k
^fAj Aj
d
j =1
5.1
As an example, the specimen of 1080 steel heated to 1050C and quenched had a
mean equivalent area diameter, d , Am of 50.3|im, measured on 1142 grains using the image
analyser. The mean equivalent area diameter measured by Jeffries' method [41] was
53.6(im from 1319 grains. Note that the grains cut by the edge of the field are counted as
half grains by Jeffries' method. Therefore, 354 grains were ignored in the image analyser
measurement. The grain size distribution of the missing edge grains can be calculated by
354
^Fi ~ k fAAAI
^fAjd jA
j=1
5.2
To remain consistent with Jeffries' method, half of this number can be added to the
number of grains measured in each size class, N , by the image analyser, to obtain the
Ai
N Fi
N ci=-r+N
A 2 Ai 5.3
The correction which is proposed here is consistent with Jeffries' method, since only
half of the edge grains are added. Rejection of all edge grains, as is done during the image
62
analysis, skews the distribution to a lower mean since more large grains are rejected than
small ones. On the other hand, consideration of all of the edge grains would skew the
distribution toward a higher mean. The intermediate approach of considering half of the
edge grains is the most appropriate because it will give the best approximation of the true
The corrected fraction of grains in each size class can then be determined by
r _ NAa
JACi ~ k
1 ACJ
N
i=1
5.4
The true fraction of grains in each size class, f , Ai is initially not known, but the
fraction measured by the image analyser can be used as a first approximation. The
corrected fraction, f a,
A
w u
^ b e c
l s e r
to the true fraction, and so equations 5.2, 5.3 and
5.4 can be applied iteratively, by substituting f ACi for f Ai in equation 5.2. The corrected
grain size distribution converges to a final solution in approximately ten iterations. The
final distribution will be closer to the true two dimensional grain size distribution than the
measured result obtained from the image analyser, since the skewing effect of the edge of
calculated which compares well with the Jeffries value of 53.6|im. Without the applied
correction, the discrepancy between the two results is 6.0%. With the correction applied,
the discrepancy is reduced to 1.5%, which in view of the required accuracy is negligible.
Results from other examples are reported in Table 5.1. The difference between the results
Steel Temp time Jeffries Jeff. Image LA. diff. corrected diff.
d , urn grains
Am d ,\im
Am grains d , |lm
Am
1080 1050 120 59.8 1055 55.3 885 -7.6 58.8 -1.6
1080 1050 420 70.7 756 64.7 629 -8.5 67.8 -4.2
A36 1000 600 92.3 229 81.5 185 -11.7 87.4 -5.3
DQSK 1000 120 32.3 356 31.3 271 -2.9 32.4 +0.3
DQSK 1150 420 178.1 298 162.1 231 -9.0 174.6 -2.0
Table 5.1 - Correction of systematic errors for the mean equivalent area diameter
The corrected distribution can be assumed to be closest to the true two dimensional
grain size distribution. The mean equivalent area diameter, d , from Jeffries' method [41]
Am
is always closer to the true mean equivalent area diameter than the result obtained directly
from the image analyser. The corrected mean result is generally within the 95% confidence
hmits of the Jeffries result. A l l results are affected by the statistics of the measurement and
Table 5.1 demonstrates that the agreement between Jeffries' results and the corrected
mean equivalent area diameter improves as the measured number of grains increases. The
scatter in the relative magnitude of the difference between Jeffries' results and the true
manual methods also show that the manual result is larger, by an average amount of
approximately 15%. The difference can be attributed to the way that each method
determines the number of intercepts. Manually, the intersection between the grid and a
grain boundary is counted as one intersection. Using the image analyser, the grid length is
considered to be the total number of pixels on the screen, and that is 640 by 480. The
number of intercepts is counted as the number of pixels which make up the area of the
grain boundaries. The matrix of pixels in the screen can be assumed to be a grid of 480
horizontal lines that are 640 pixels in length. Using Heyn's method, a grain boundary or a
portion thereof, that is horizontal on the screen will only constitute a single intercept.
Using the image analyser, a horizontal boundary will be counted by the analyser as several
intercepts, with the number of intercepts being equal to the number of pixels in the
horizontal portion of the boundary. Therefore the image analyser counts more intercepts
than Heyn's method. The mean linear intercept length is equal to the total line length
divided by the number of intercepts. Therefore, the analyser result is lower than the Heyn's
measurement. The true mean linear intercept length, l , is closest to the value obtained
m
manually, using Heyn's method. Linear intercept measurements performed with the image
analyser should thus be viewed with caution. The argument above could also be applied by
assuming that the image analyser screen is a grid of 640 vertical lines, 480 pixels in length.
Mean equivalent area diameter measured by Jeffries' method [41] is plotted against
time in Figure 5.6, for as-received A36 steel specimens heated at a rate of 5C/sec. The
symbols are the data points and the lines show the fit obtained using the power law, to be
65
described in Section 5.5. The scatter of results is reasonable when the magnitude of the
errors is considered.
At all temperatures, the grain size is seen to increase with time and the growth rate
increases with rising temperature. At 1000C for 60 seconds and 120 seconds and at
1050C for 10 seconds and 60 seconds, two grain sizes are reported to exist
simultaneously, for each specimen. Under these conditions, abnormal grain growth was
observed and separate, measurable groups of grains were present. The grain size
The data appears to be divided into two sets, with one group of results being less
than 50|im and the other being greater than 80(im. The finer structures were observed
after heat treatment at lower temperatures and shorter times. To explain the discontinuous
growth between 50|im and 80|nm, the presence of a second phase precipitate, having a
boundary pinning effect is assumed. At 950C the pinned structure is stable for times up to
600 seconds. A t 1000C and 1050C the fine structure is unstable and with time, abnormal
grain growth begins as the pinning particles dissolve or coarsen leading to a reduction of
In the coarse structure, the data shows, that the rate of grain growth slows down
substantially with increasing time; at times greater than 450 seconds at all temperatures, a
hmiting grain size is being approached. At 1000C and 1050C there is little growth of the
coarse structure, once abnormal grain growth has ceased. In the specimens held at 1100C
and 1150C, a substantial amount of grain growth has occurred during heating, since the
initial grain sizes are greater than 80(J.m. It is likely that abnormal grain growth has
Figure 5.7 - Kinetics of austenite grain gowth in A36 steel, heating rate=5C/second.
Lines are the power law, for ^ > 8 0 L i m , / -d ' =1.51E47xexp(-840/i?r)
22
0
8 22
for d <80Lim,
Am d - -d =5.46E54.*.exp(-1291/i?r)
3 37
0
337
67
Results from the DQSK specimens heated in the vertical tube furnace are plotted in
Figure 5.8. The mean equivalent area diameters were measured manually using Jeffries'
method and curves were fit using the power law, to be discussed in section 5.5.
At 1000C and 1050C significant grain growth was not observed. At 1100C
abnormal growth was observed and a single curve does not fit well to the stepped nature
of the data. Unlike the A36 steel, measurements for the DQSK could not be made on the
fine and coarse structure separately. At 1150C, significant growth has occurred during
heating since the initial grain size is greater than 120(im. At all temperatures, the grain size
is approaching a limiting value, although at 1100C and 1150C, after 450 seconds, grain
workers show similar results to those reported here [31,55,60]. Austenite grain size is
generally reported in the range of 10|im to 250|Lim and abnormal growth is often observed
when samples are heated in the temperature range of 900C to 1150C. Results differ
primarily in the time dependence of growth. For both steels in this investigation, the
growth rate is seen to decrease significantly with time and a limiting grain size is quickly
approached. Some other investigations have reported similar growth behaviour [31,55],
Abnormal growth and a limiting grain size are phenomena which are typically
observed when second phase particles are present. In a steel similar in composition to the
A36, Gladman investigated abnormal growth [31] and was able to quantitatively describe
the phenomena by considering the pinning effects of aluminium nitride. His calculations
cannot be applied to this study because the particle size and volume fraction of the A1N in
the as-received condition in the A36 and DQSK steels is not known.
68
200
20 H
o -h 1 1 1 1 1 1
0 100 200 300 400 500 600
Time, seconds
Figure 5.8 - Kinetics of austenite grain growth in D Q S K steel, heated in the tube furnace.
Lines are the power law, d -d - =5.02E56.texp(-1271/fl7)
5M 5 66
69
Variability in the growth kinetics of austenite in the DQSK and the A36 steel can be
partially accounted for by differences in the puuiing effects. The influence of pinning
particles is determined by the amount of aluminium and nitrogen in the steel, the
precipitation kinetics of A1N and the size distribution of the austenite grains, as described
by equations 2.4, 2.7, and 2.8. Each of these factors is affected by the thermal and
mechanical history of the material. Further, the content of alloying elements such as
carbon and manganese has an influence on the kinetics of growth since the boundary
Differences in the heating cycle can have a significant influence on the growth
kinetics. Figure 5.9 is a comparison of austenite grain growth in the as-received A36 steel,
heated to temperature by different heating cycles. The slowly heated specimens were
heated at a constant rate of 5C/second to 1100C and held. Rapidly heated specimens
underwent a stepped heating cycle, and were heated initially at 5C/second to 900C and
held for 120 seconds, then heated again at 100C/second to 1100C. The curves in the
The initial grain size in specimens heated at the 100C/second was 47|i.m while in
those heated at 5C/second the initial grain size was 88p:m. The apparent j ^ i t i n g grain
size was lOOiim in the quickly heated specimen and 140|Lim in the slowly heated one.
These different sizes can be explained by grain growth during heating, and the influence of
A1N particles. Nucleation effects during the transformation to austenite are likely to be
^significant due to the identical thermal cycles up to 900C. The difference in initial grain
size can be explained by grain growth occurring during heating since the rapidly heated
specimen spent less time at high temperature, and therefore less growth occurred. The
effect of holding time at 900C can be ignored with respect to grain growth since at this
temperature, limited grain coarsening has been observed. The different limiting grain size
70
is evidence that the dispersion of particles is affected by the heating cycle, since Zener has
described the limiting grain size as function of particle volume fraction and radius [30]. If
equilibrium conditions exist, particle volume fraction is unchanging, and the smaller
j ^ i t i n g grain size must be due to a finer dispersion of closely spaced particles. The
influence of second phase particles will be discussed in greater detail in Section 5.6.
The power law and the statistical model for grain growth were fit to data from the
A36 and the D Q S K steels. Measured values of mean equivalent area diameter, d Am were
used to determine the constants for the fit of the power law. Estimated values of the
average equivalent volume diameter, d , vb were used to fit the statistical model. The
statistical model of Abbruzzese and Liicke [20,21] was applied by Militzer et al. [61,62].
The power law for grain growth was fit to the results of mean equivalent area
diameter measured for the D Q S K steel heated in the tube furnace and the A36 steel heated
in the Gleeble at a rate of 5C/second and 100C/second. In each case the parameters of
m, K and Q were used to fit the power law equation by a least squares procedure. The
0
sum of the squares was minimised using a solving routine on a Microsoft Excel
spreadsheet.
Separate curves were needed to fit to fine and coarse austenite grain growth A3 6
180
160 H
I 140
120
100
80
60
5C/sec(1100C)
cf -d
22
0
822
=1.51 E47.f.exp(-84CW?7)
40
100C/sec (1100C)
20 du
9
-L1 14
=1.94E68.f.exp(-1089/flT)
9
0 ~I 1 1 1 1 1 1
0 100 200 300 400 500 600 700 800
time, seconds
Figure 5.9 - Effect of heating rate on austenite grain growth in A36 at 1100C.
72
Table 5.2 - Power law fitting parameters for the A36 and D Q S K steels, for the
equation, d -d =K(pxTp(-Q/RT).
m
0
m
Use of the power law with m not equal to 2 is empirical. When m is used as a fitting
parameter, the equation becomes a curve fitting empirical expression for describing the
austenite growth kinetics, as shown in Figure 5.7, 5.8 and 5.9. However, the reported
values obtained for m, K and Q are widely scattered. The values obtained for the A36
0
steel heated to temperature using two different heating rates are entirely unrelated, despite
the fact that results were obtained using the same steel. The equations cannot be used
outside of the experimental conditions for which they were determined. For these reasons,
The statistical model was applied by Militzer et al. [61] to the results for the A36
steel heated at 5C/second and 100C/second, and the DQSK steel. The calculations are
based on the average equivalent volume diameter, dy^, estimated from the measured mean
described by Gjostein [63]. Consequently, a value of 7^=0.7Jhr is used for the A36 and
2
73
7g=0.8Jrrr is used for the DQSK steel. The grain boundary mobihty, M ,
2
gb is
approximated by that of pure austenitic iron and is calculated using the grain boundary
diffusivity, D , and the Burgers vector, b, from Frost and Ashby [64],
gb
D b ah
2
kT 5.5
Mobihty decreases quickly as the temperature decreases. The pimiing parameter, P, was
estimated, in order to fit the model to the data of each isothermal test series, for each
steel.
The fit of the model is shown in Figure 5.10 for the A36 steel heated at 5C/second
in the Gleeble, and in Figure 5.11 for the D Q S K steel heated in the tube furnace.
The model fits well to the growth kinetics of the coarse A36. A t this stage, no
attempt to fit the model to the abnormal growth condition has been reported for the A36
steel. For the DQSK, all data, including the abnormal growth results were desrcibed using
the model. Since during abnormal growth the pinning force changes with time, at 1100C,
two values of the pmning parameter were used, resulting in a good fit for the abnormal
grain growth data. The discontinuity in the curve is a result of the sudden change in P,
related to the changing pinning behaviour of the A1N particles in the material. For the A36
and the D Q S K steel results, the statistical model is more effective than the power law in
According to equation 2.7, Rn is the reciprocal of the pinning parameter, P , so that the
m
limiting grain radius can be predicted from the pinning force estimations made with the
statistical model. Figure 5.12 is a comparison of the experimental and predicted values of
240
Figure 5.10 - The statistical model and the results for the A36 steel
for a heating rate of 5C/second. Lines are generated by the model [61,62].
75
Figure 5.11 - The statistical model and the results for D Q S K steel.
Lines are generated by the model [61].
0 20 40 60 80 100 120 140
R lim (experimental), (im
The results indicate that the statistical model can be applied to describe the grain
growth kinetics using a single fitting parameter, P. Furthermore, the value of P can be
used to predict the jjiniting grain radius with reasonable accuracy, demonstrating a
consistency between the experimental evidence and the model's theoretical basis. Since the
describe grain growth kinetics under a wide range of conditions, as long as reasonable
The aim in applying the statistical model is to describe the kinetics of grain growth
during hot strip rolling, under conditions which are not isothermal. A test series was
performed with the A36 steel to measure the kinetics of grain growth during continuous
cooling conditions. During hot rolling, between the roughing mill and the finishing mill, a
typical temperature change from 1120C to 1000C can occur over a period of
stepped thermal cycle to 1120C and a continuous cooling cycle of 2C/second was
imposed. The statistical model was applied by Militzer et al. [61], employing the
isothermal tests. Figure 5.13 illustrates the results. Good agreement is obtained,
demonstrating that the statistical model can be used to model the kinetics of grain growth
Most of the grain growth kinetics observed in this investigation were influenced by
the presence of second phase particles, since at all temperatures, a tendency toward a
limiting grain size was observed. The A36 and DQSK steels contained similar proportions
75
Figure 5.13 - Application of the statistical model to describe continuous cooling conditions [61].
79
The temperature at which both constituents were completely soluble in both steels is
- 5.6
reported by Gladman [32], where [Al] is weight percent aluminum, [N] is weight percent
In the as-received condition, the dispersion of A1N in each of the samples, water
quenched off the hot strip mill, is not known. Gladman has reported that A1N precipitates
slowly during cooling, but more quickly during heating [32]. Even if no precipitation
occurred in the steels during the initial quench on the mill, it is certain that a dispersion of
particles exists during the isothermal holding period in the furnace heated D Q S K and in
the slowly heated A36, due to the low heating rates employed. Even specimens of A36
heated at 100C/second were first taken to 900C at a rate of 5C/second and held for 120
seconds. It is likely that these also had a dispersion of A1N particles present during
The observed abnormal growth kinetics and the tendency to approach a jj^ting
grain size are evidence of the presence of A1N particles. Since the initial particle size
possible. However Gladman's expression [32] can be used to estimate the size of the
particles which are present when abnormal growth is observed. In the A36 steel heated at
5C/second and held at 1000C for 60 seconds, if it is assumed that the initial ratio of the
largest to the average grain radius is 2.0, and A1N is in equilibrium, then the critical
required to observe a particle of this size; such an investigation is difficult to perform and
It seems likely that both of the as-received steels are in a supersaturated condition
with respect to soluble aluminium and nitrogen, consistent with water quenching of the
transfer bar on the hot strip mill; but the degree of supersaturation is not known. The rate
at which solutes can precipitate and coarsen, and the rate at which precipitates can
dissolve, is diffusion limited. In either case, the distribution of particles is changing with
time, as is the pinning force. The observed phenomena of abnormal grain growth is due to
the dynamic nature of the pinning force. The limiting grain sizes are also likely to change
with time, as the distribution of pmning particles tends to dissolve or coarsen. Further, the
effect of heating rate on the Ijmiting grain size in the A36 steel can be attributed to a
change in the A1N particle distribution during heating. Additional studies are required to
The two dimensional grain size distribution was measured using the image analyser.
Changes in the distribution during grain growth will be discussed for the case of normal
grain growth, as occurs in the A36 steel heated to 1150C at 5C/sec, and for abnormal
grain growth as observed in the A36 steel heated to 1000C at 5C/second. As a measure
i(z,-X f c )(i--Fj
i=i
R2
=
where X ; is the measured fraction of size class i and 7; is the calculated fraction of size
class /, determined from the log normal distribution described by equation 2.11, the peak,
d , and the standard deviation, s . X is the average of all X and Y is the average of all
Ag A b t b
81
Y . The log normal distribution is determined by solving for the best fit of the log normal
t
curve to the measured results. A least squares solution, similar to that used to fit the
power law, is applied, where the sum of the difference between calculated and measured
s = exp 2/i(l^)-lnG/))'
i=l
5.8
Figure 5.14 shows the evolution of the measured two dimensional grain size
distribution during normal grain growth at 1150C, at times of 1 second, 60 second and
450seconds. The curves shown are the best fit log normal distributions. Table 5.3
o 0.15 H
+
o
CO
Lt 0.10
0.05 H
0.00 i r- ' i' ' ' ' i' ' ' ' i' ' 1
' i' r
20.0 31.3 48.8 76.3 119.2 186.3 291.0 454.8
Grain size, EQAD, |nm
Figure 5.14 - Evolution of the A36 2-D grain size distribution at 1150C.
83
During normal grain growth at 1150C, the relative width of the distribution does
not change greatly with time, but the peak steadily increases. The shape of the distribution
is closely approximated by a log normal distribution at all times. The size distribution
evolves by the shrinkage and disappearance of the smaller grains and the coarsening of
large grains. The distribution observed at 1 second is the widest, possibly due to the after
effects of abnormal grain growth during heating. However, the distribution width is
observed to be stable and constant at 60 seconds and 450 seconds. Therefore, after a short
Atkinson [16].
Photomicrographs showing the structure before, during and after abnormal grain
growth in the A36 steel held at 1000C, are shown in Figure 5.15. At 10 seconds, the
specimen was water quenched so that the fine grains could be revealed. At 120 seconds
and 600 seconds, a helium quench was employed. The development of the structure over
time is visibly dramatic, being uniform and fine at 10 seconds, bimodal at 120 seconds and
The evolution of the two dimensional grain size distribution during abnormal growth
at 1000C is shown in Figure 5.16. Initially, at 10 seconds, the structure is uniform and
fine. A t the onset of abnormal growth, the largest grains in the structure break away from
the changing dispersion of pinning particles, since they have the greatest driving force for
growth. This is evident in the widening of the distribution and the appearance of the
bimodal structure at 120 seconds. The plot at 120 seconds suggests the existence of two
Figure 5.15 - Evolution of the grain structure, in the A36 steel held at 1000C,
showing abnormal grain growth. Magnification is approximately x410.
85
0.25
10 sec.
0.20 H
c
0.15
o
cd
Lt 0.10 H
0.05 H
0.00
i 1 1 i i i i i r
4.0 6.3 9.8 15.3 23.8 37.3 58.2 91.0 142.1 222.0
Grain size, EQAD, (xm
Figure 5.16 - Evolution of the A36 2-D grain size distribution at 1000C.
86
The reduced correlation coefficient in Table 5.3 demonstrates that the distribution
deviates from log normality. When the pinning particles are sufficiently dissolved or
coarsened, all grains are able to grow, and the fine grains quickly disappear. The structure
regains its uniform appearance as the distribution narrows and the distribution shape
Thus, having found a significant effect of A1N precipitates on the grain growth
solution treatment. A number of A36 specimens were soaked in a furnace at 1200C for 3
hours to dissolve the A1N precipitates, then water quenched to prevent their precipitation
during cooling. A test series was performed to compare the kinetics of austenite grain
growth in the heat treated and the as-received specimens. The results of these tests are
950C, 1050C and 1150C, and held at each temperature for 120 seconds. A n
J
intermediate heating rate was chosen to reduce the precipitation of A1N during heating and
The solution treated specimens displayed a grain size which was larger than that
obtained in the as-received specimens for all thermal cycles employed. Since the solution
treatment was designed with the intent of minimising the pinning effect of the second
phase particles, it is not surprising that the solution treated specimens have a larger grain
size; a higher rate of grain growth would be expected with the reduced influence of second
phase particles.
Differences in the structure of the specimens prior to the thermal cycle may also
have some influence on the observed differences in grain size. Prior to thermal cycling, the
structure of the as-received A36 steel was mostly acicular ferrite and pearlite, while the
87
solution treated specimens were almost all martensite. The prior austenite grain size of the
as-received structure also appeared to be much finer than that of the solution treated
material. During the transformation back to austenite on heating, the initial structure of the
as-received specimen would be expected to have a refining effect on the nucleation of the
new phase, since the as-received structure would be expected to offer more nucleation
sites for transforming to austenite. A larger number of nuclei could explain the finer grain
sizes. Further tests would be required in order to conclusively state whether nucleation
effects, or the effect of second phase particle dominate the grain growth kinetics. It is
The structure of the as-received specimen after heating to 950C was bimodal,
showing signs of abnormal grain growth, whereas, the structure in the solution treated
specimen heated to 950C was uniform. This result can only be explained by an initial
difference in the dispersion of second phase particles. Therefore, from these results, it can
be concluded that the solution treatment reduced the effect of second phase particles.
Both sets of results show an inflection in the plot at 1050C. The inflection indicates
that, despite the anticipated difference in the initial particle distribution and structure, a
similar trend with respect to the kinetics of grain growth is occurring in each type of
specimen. Few specific statements can be made regarding the kinetics of isothermal
growth at each temperature, since only one holding time has been used.
88
Figure 5.17 - Comparison of austenite grain size in as received and solution treated
A36 steel, heated to 950C, 1050C and 1150C and held for 120 seconds.
89
At 1150C, the grain size is much larger than in the specimens held at lower
temperatures. This result can be explained by a the combined effect of increased A1N
solubility and increased atomic mobility. The solution temperature of A1N in the A36 steel
Therefore at 1150C, much of the A1N will be already in solution, or in the process of
dissolving. With fewer second phase particles present, the rate of grain growth is higher
since the pinning force is reduced. Also with increasing temperature, grain boundary
In conventional process models the power law is used to describe grain growth
kinetics. It is usually fit first to results similar to those reported here, and then extrapolated
to describe grain growth occurring under the conditions of hot strip rolling.
In this work however, it has been observed that the kinetics of grain growth differ
when the heating cycle is changed or when the as-received specimens are solution treated
prior to being used for grain growth tests. The power law has been fit with different
parameters, to the results obtained even for the same material, heated to temperature by
different thermal cycles. This scatter in the fitting parameters cannot be described from any
fundamental perspective.
During conventional grain growth tests, rapid growth occurs during heating and the
initial grain size is large. A smaller initial grain size can be achieved by deformation and
recrystalhzation of the specimen. However, quenching the structure after the test, to
reveal the prior austenite grain size, then becomes difficult. Therefore, the early stages of
grain growth, which are relevant to the industrial process of hot strip rolling, are rarely
measured. In laboratory tests, the specimen must be held at temperature for extended
periods so that significant growth is observed and a reasonable time dependence for the
Under hot strip rolling conditions, deformation occurs in each of the rolling passes,
and grain growth occurs in the time between stages of deformation, when recrystallization
is complete. In a typical rolling mill, interpass times range from 1 second to 20 seconds
and the delay between the roughing mill and finishing mill is typically 60 seconds. Hold
times for laboratory studies of grain growth are usually in the range of 500 seconds to
1000 seconds [3,4,18], or longer [5]. Thus an extrapolation of the empirical power law to
Differences in the grain growth kinetics have been described with respect to the
influence of second phase particles. Abnormal grain growth and the attainment of a
limiting grain size are rarely observed during hot rolling of plain carbon steels, providing
evidence that second phase particles have little influence. Before rolling, the steel is
soaked at between 1200C and 1300C in a reheat furnace. At this temperature, ALN is
usually completely soluble. During rolling, the temperature of the material, and the
solubility of A1N drop, but Gladman states that precipitation of A1N during cooling is slow
[31]. Therefore it would be expected that A1N particles have limited effect until the later
stages of rolling, at lower temperatures where austenite grain growth is not significant.
Militzer et al. have described the application of a statistical model to the prediction of
austenite grain growth kinetics in A36 and DQSK steels during hot strip rolling [61,62]. It
is reported that a pjjming parameter of P=0 is required to model grain growth occurring
between the roughing and the finishing mills, which corresponds to unpinned grain
growth. Such a situation is realistic, since in the hot strip rnill, it is likely that little or no
ALN particles will precipitate between the roughing and fMshing mills, in the
corresponding temperature range from 1120C to 10Q0C. The kinetics of austenite grain
growth can then be described by Burke and Turnbull's parabolic growth kinetics [6], as
6.0 Introduction
Methods for the estimation of the three dimensional grain size distribution were
applied to two dimensional results obtained from the 1080, the A36 and the D Q S K steel.
The 1080 steel was used for much of the work since the microstructure was easily
interpreted and a large number of grains could be measured. The three dimensional grain
size distribution was described using the mean equivalent volume diameter, d Vm and the
The methods of Saltikov [36] and Huang and Form [37,46] were applied to the two
dimensional grain size distributions measured on the image analyzer for the 1080 steel,
heated to 1050C and held for 0 seconds, 120 seconds and 420 seconds. Figure 6.1 shows
the measured distribution and the distribution as predicted by each method, for the
specimen held for a time of 0 seconds. The correlation coefficient, R , for the fit of the
2
measured distribution to a log normal distribution is 0.99. The results from each method
are quite different, with the mean equivalent volume diameter, dy , from Saltikov's
m
method [36] being equal to 70.2|im and that obtained from Huang and Form's method
91
92
0.24
0.22 -
0.20 -
0.18 -
0.16
0.14
0.12
0.10
0.08
0.06 -
0.04 -
0.02 -
0.00
-0.02 Measured 2-D data
-0.04 Saltikov's method
Huang and Form's method
-0.06 H
~l 1 1 1 1 1 1 1 1
6.3 9.8 15.3 23.8 37.3 58.2 91.0 142.1 222.0
Figure 6.1 - Saltikov's [36] and Huang and Form's [37] 3-D grain size distribution
compared to the measured 2-D distribution for 1080 steel heated to 1050C and quenched.
93
In Saltikov's original publication [36], his method required that the grain size
a, of 1 0 0 1
(1.2589) since application of the method required the use of reported tables of
coefficients. In this case, a equal to 1.25 has been used. The mathematics of the procedure
were set up using a Microsoft Excel spreadsheet so that any multiplying factor could be
chosen, since the principles of the calculation are independent of a. The major difference
between the two methods is that Saltikov [36] corrects for sampling only, while Huang
and Form [37,46] correct for both sampling and truncation. Up to the point where the
truncation correction is applied, the two methods are identical. The truncation correction
increases the number of smaller grains and decreases the number of large grains, since it
accounts for the probability that a grain is sectioned, based on the grain diameter. It is
therefore expected that the mean predicted by Saltikov is greater than that predicted by
Huang and Form; this is seen to be the case. The width of the estimated distribution
cannot be assessed by calculation of the log normal standard deviation because the
negative number of grains in the smallest size classes makes such a calculation impossible.
one of the major problems with both of these methods. Even when an artificially
generated, perfect log normal distribution was used as input for the methods, a negative
number of grains was predicted in the smaller size classes. This is obviously an unrealistic
result. Huang and Form [46] and Aaron et al. [49] have reported similar negative counts
Both methods operate by correcting the number of grains in each size class, by
terms in the equation is equivalent to the number of corrections applied, and is a function
of the relative magnitude of the size class, with the largest class having a single correction
and the nth largest class having n corrections. Therefore, any errors which result from
94
assumptions inherent in each method are compounded for the smallest size classes. The
assumption that grains are spherical in shape is incorrect, as stated by Aaron et al. [49],
because an array of spherical grains cannot fill space. The methods are based on the
probability that an object is sectioned to give a plane of lesser diameter. It has been shown
that such probabilities are a function of object shape [36,39]. It is the probability
calculation resulting from the assumption of spherical grain shape that directly introduces
The three dimensional grain size distribution was estimated from the measured
results of mean linear intercept length, l , and mean area, A , using the method of
m m
Takayama et al. [38]. This method was applied to all of the A36 specimens heated at
5C/second and 100C/second, and to the 1080 steel specimens held at 1050C.
Some of the results for the A36 steel heated at 5C/second are shown in Table 6.1.
Mean linear intercept, l , was determined by manual measurement using Heyn's linear
m
intercept method [41]. The measurement involved the placement of a grid over a
photomicrograph of the structure so that number of intercepts per unit length could be
determined. Mean area was measured by a manual procedure using Jeffries' method [41].
Photomicrographs of the structure were used to determine the number of grains per unit
area. The correlation coefficient, R , for the fit of the measured distribution to a log
2
normal one is reported for specimens where the two dimensional distribution was also
determined.
Values of the standard deviation of the log normal distribution, s , were calculated
v
directly from the measured results by using equations 2.21 and 2.22. Calculated values of
s are reported in Table 6.1, as are overall mean values. The calculated and mean values
v
|im (im 2
|_im (im
A36 1000 600 0.97 78.8 6687 105.6 1.57 112.7 1.33
A36 1150 450 0.97 169 33506 247.1 1.42 252.2 1.33
Table 6.1 - The three dimensional grain size distribution estimated using the method
of Takayama et al. [38].
There is significant scatter in the s results, as reported in Table 6.1. The scatter
v
predicted by the method is not consistent with experimental observations of the two
dimensional grain size distribution. It is reasonable to assume that the width of the two
dimensional distribution is a strong function of the width of the true three dimensional
distribution. During normal grain growth in the A36 at 1150C, the two dimensional
distribution width was seen to be fairly constant, s being 1.71 at 1 second, 1.58 at 60
A
seconds and 1.62 at 450 seconds (refer to the discussion in Section 5.7). From Table 6.1,
The log normal standard deviation, s , is dependent on the ratio of l to the square
v m
intervals for l were calculated to be approximately 10%. From Figure 6.2, it is apparent
m
Accuracy in the determination of l and A would improve with an increase in the number
m m
of grains measured.
the standard deviation was used. Table 6.1 reports d Vm results obtained for the A36
deviation of the log normal distribution is estimated to be equal to 1.33. Equation 2.22
leads to
^Vm = 1
-22^Am 6.1
The complete list of results from the A36 steel heated at 5C/second shows that the
average of the standard deviation is approximately 1.33 for each of the five different
1100C it is 1.32 and at 1150C it is 1.26. This indicates that the scatter in s v can be
measurements are used together, a statistically stable result for s can be determined; this
v
value approaches 1.33. It is unfortunate that the required number of grains seems to be of
the order of 1000, since it is not practical to propose that every measurement of grain size
2.4
I 2.2
2.0
1.8 H
1.6 H
1.4 H
1.2
1.0
0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25
/ divided by the square root of A
m
1/2
Figure 6.2 - The relationship between s and the ratio, / /(A )
y m m , using the method of
Takayama et al. [38].
98
The variability of the l m and A m measurement was smaller for the 1080 steel
specimens, since the structure was more clearly revealed, and a much larger number of
grains were measured for each of these specimens. Linear intercept length and mean area
were determined manually [41] and by using image analysis equipment [40]. The method
The systematic error in the measurement of mean linear intercept, l , and mean m
area, A ,
m by the image analyzer was discussed in Section 5.2. The linear intercept length
measured on the image analyzer is much lower than that measured manually, leading to
smaller values of s , as explained by Figure 6.2. Mean area is also lower but the difference
v
is not as great. If the image analyzer results are used then a value of s equal to 1.15 is
v
calculated values of d Vm and reveals that there is only a minor difference in the mean
volume diameters. Since the estimated three dimensional distribution is very sensitive to
cannot be used with the method of Takayama et al. [38] to estimate the three dimensional
The method of Takayama et al. [38] applied to the 1080 steel using the manual
with a standard deviation of 1.41; in a specimen held at 1050C for 120 seconds, d Vm was
71.2(j,m and s v was 1.44 and in a specimen held at 1050C for 420 seconds, d Vm was
83.6|im and s v was 1.46. As would be expected during normal grain growth, the
distribution width is estimated to change little, with the mean grain size steadily increasing.
99
If the mean value of s , equal to 1.43, is used with the 1080 steel manually
v
d Vm = U9d Am 6 2
From equation 6.1 and 6.2, it can be said that the mean equivalent volume diameter,
d ,
Vm predicted by the method of Takayama et al. [38] is approximately 20% greater than
the mean equivalent area diameter, d , Am determined by Jeffries' method [41], for the cases
can be written.
In summary, there are two main problems with the method of Takayama et al. [38].
Firstly, the method requires that the grain size distribution is log normal. Examples
discussed in this section were observed to have log normal grain size distributions in two
dimensions; for these cases, the method is therefore apphcable. However, several
specimens were observed to have grain size distributions which were not log normal
(abnormal grain growth was observed) and in these cases, use of the method of Takayama
Secondly, the method is very sensitive to the ratio of l m to the square root of A . m
During measurement of grain size in the A36 and DQSK steels, errors in the order of 10%
were encountered because only a hmited number of grains could be measured. Errors of
method works reasonably well with microstructures which are easily measured, and with
100
other structures when a large number of grains are analyzed. Thus, successful application
The three dimensional grain size distribution was also estimated using the method of
Matsuura and Itoh [39]. The method was applied to results obtained from the 1080 steel,
the A36 steel heated at 5C/second, and the DQSK steel heated in the furnace, as
summarized in Table 6.2. The mean equivalent volume diameter, d , Vm and the standard
deviation of the log normal distribution, s , v were calculated from the estimated
distribution. A flow chart shown in Figure 6.3 demonstrates how the method is applied,
As a first assessment of the method of Matsuura and Itoh [39], the two dimensional
grain size distribution measured directly by the image analyzer was used as input. The
method consistently predicted that the three dimensional grain size distribution has a
higher mean than the two dimensional grain size distribution. The width is predicted to
the ratio of the mean equivalent area diameter (obtained from the image analyzer), d ,Am to
the mean equivalent volume diameter, d . Vm From Table 6.2, it is obvious that there is
some variability in the results, and for the wider distributions, a higher ratio has been
calculated. A simple conversion factor can be used for cases where the width of the grain
size distribution does not vary substantially; this is the case, where grain growth is normal
<^STOP^)
Figure 6.3 - Flow chart of the procedure for applying the method of Matsuura and Itoh [39].
102
Table 6.2 - 3-D distribution from the image analyzer results, for 1080 steel, A36
steel and DQSK steel, after Matsuura and Itoh [39].
Figure 6.4 shows the relationship between the distribution width, s , and the ratio of
A
dAm to d .Vm From the plot, it appears that a linear relationship exits between s A and
dimensional grain size distribution for specimens having a two dimensional log normal
103
standard deviation, s , in the range from 1.8 to 2.4. However, none of the distributions
A
assessed in this work had those values of s . A n alternative route to assessment of the
A
computer simulation.
The method of Matsuura and Itoh [39] should actually be applied to the true two
dimensional grain size distribution, rather than to the distribution measured on the image
analyzer. As discussed in Section 5.2, a systematic error is made when the two
dimensional grain size distribution is measured. A correction can be applied to account for
the error, thus obtaining the true two dimensional grain size distribution. The method of
Matsuura and Itoh [39] was applied to a number of cases where this correction was
Table 6.3 - 3-D distribution from corrected 2-D distribution, after Matsuura and
Itoh.
104
1.55
Comparison of Table 6.2 with Table 6.3 reveals that the ratio of d Vm to dAm for
corresponding specimens is almost the same, with the average ratio from Table 6.3 being
slightly higher at 1.25. For the same specimens from Table 6.3, the ratio is 1.24. This
difference can be explained by the relationship between the distribution width and the ratio
of dVm to d ,
Am as described by Figure 6.4. The corrected two dimensional distributions
are slightly wider than the measured distributions; this leads to the wider three dimensional
the conversion factor of 1.23 from all of the results in Table 6.2 is still approximately true.
This result compares well with the result of approximately 1.2, found using the method of
Takayama et al. [38]. A more detailed comparison of the method of Takayama et al. [38]
and the method of Matsuura and Itoh [39] will be made in Section 6.4.
The distribution estimated by the method of Matsuura and Itoh [39] was generally
similar in shape to the measured distribution. Distributions which were measured as log
normal in two dimensions, remained so after the method was applied. Figure 6.5 shows
the three dimensional grain size distribution in the A36 steel held at 1000C for 10
seconds, 120 seconds and 600 seconds, as representative examples. The log normal
distribution (shown) was fit by the method of least squares, by rninimizing the value of S 2
calculated using equation 2.21. The measured two dimensional grain size distribution and
its corresponding log normal curve are also shown; the measured results are plotted as
The two dimensional and three dimensional distributions at 10 seconds and 600
seconds are reasonably well approximated by the log normal curve. At 10 seconds, R for 2
106
the two dimensional distribution is 0.97 and 0.94 for the three dimensional distribution. At
600 seconds, R for the two dimensional distribution is 0.97 and 0.88 for the three
2
dimensional distribution. Both examples show that the estimated distribution tends to be
more scattered than the measured one. The method seems to exaggerate deviations from
log normality. For the case of abnormal grain growth at 120 seconds, the initial
estimated three dimensional distribution deviates even further from log normality, with R 2
equal to 0.62. This result is not unreasonable since the grain size distribution during
abnormal grain growth is not expected to be log normal; a bimodal grain structure was
actually observed.
inherent in the calculation method, and inaccuracies due to the measurement statistics.
One error in the calculation method is the determination of the number of faces of a
grain. In reality, the number of faces of a grain, / , must be an integer whereas the
calculation permits / to be equal to any decimal value greater than 4. A lower limit of 4 is
chosen since it is impossible for a polyhedron to have less than 4 sides. Furthermore, in
order to smplify the calculation of the probability distribution function for each grain
shape, Matsuura and Itoh assumed that the grain shape could be approximated by planar
faced, regular polyhedra. In reality, grains have curved surfaces and are unlikely to be
entirely regular. The assumption that each three dimensional size class can be represented
by a single grain shape is also an over simplification, which could be the source of some
error.
initial measured scatter. The two dimensional grain size distributions were determined
from a limited number of grains and the initial scatter in the measured distributions is a
result of this. It follows that if the two dimensional distribution could be more accurately
107
determined, the scatter in the three dimensional distribution could be minimized. In order
more size classes should be measured, with, on average, 50 to 100 grains for each size
class. Therefore, an impractical number of grains, between 1500 and 3000, would need to
the grain size distribution, have performed measurements of the distribution with the
number of grains in this range [25,43,46,66]. In this study, approximations of the two
dimensional grain size distribution were usually obtained from 200 to 300 grains.
The stability of the method of Matsuura and Itoh [39] was assessed by varying the
width and number of size classes used to predict the mean equivalent volume diameter in
the 1080 steel specimen, held at 1050C for 0 seconds. Both arithmetically scaled and
geometrically scaled size classes were assessed, with a total of 1142 grains being measured
in the two dimensional distribution. Figure 6.6 shows the convergence of the mean
diameter. The arithmetically scaled and geometrically scaled size classes produced similar
results. For the number of size classes greater than 20, the method predicted a mean
geometrically scaled distribution for less than 20 size classes were less stable than those
obtained from the arithmetically scaled distribution; but the differences were not
significant. Both types of distributions provided mean values which were within 3% of
each other, when the number of size classes were greater than 10. The log normal
4.0 6.3 9.8 15.3 23.8 37.3 58.2 91.0 142.1 222.0
Grain size, EQAD, Lim
Figure 6.5 - Evolution of the 3-D grain size distribution for A36 steel held at 1000C
from the method of Matsuura and Itoh [39].
109
Figure 6.6 - Effect of the number of size classes on the method of Matsuura and Itoh [39].
110
6.4 Comparison of the Methods of Takayama et al. and Matsuura and Itoh
predicted by each of the methods for the estimation of the three dimensional grain size
distribution are compared in Table 6.4 using the results from the 1080 steel held at
1050C. The results measured manually by Heyn's method and by Jeffries' method [41]
were used with the method of Takayama et al. [38]. The true, corrected two dimensional
grain size distribution was used with the method of Matsuura and Itoh [39]. The mean
values estimated by the method of Takayama et al. [38] and the method of Matsuura and
Itoh [39] are in very close agreement, with the difference being on average, 3%.
The estimation of the log normal standard deviation is a different matter. The
considering the sensitivity of the method of Takayama et al. [38], and by comparison of
the estimated value of s v to the measured value of s A for each method, the standard
deviation estimated using the method of Matsuura and Itoh [39] seems to be the most
reasonable.
Table 6.4 - Three dimensional grain size distribution predicted using the method of
Takayama et al. [38] and Matsuura and Itoh [39] for 1080 steel held at 1050C.
Ill
6.5 Validation of the Method of Matsuura and Itoh With Measured 3-D Results
As seen from the above discussion, the method of Matsuura and Itoh [39] is the
most reasonable so far reported in the literature. Matsuura et al. [25] confirmed this by
validation of the method with three dimensional measurements. The three dimensional
disintegration, and individual measurement of each grain volume. The two dimensional
grain size distribution was also measured on a section through the specimen, by the use of
image analysis equipment. Approximately 1600 grains were measured. The results were
reported in the form of eleven arithmetically scaled size classes. The measured distribution
compares well to that predicted by the method of Matsuura et al. [39], as shown in Figure
6.7.
Very little literature is available which reports the measurement of the three
dimensional grain size distribution, probably because such a task is difficult to perform.
However, three dimensional measurements have also been made by Williams and Smith
[26] using stereo radiography on an aluminum-tin alloy. Two dimensional data was
obtained on the same specimen by Aaron et al. [49], who compared a number of
estimation methods based on the assumption of spherical grain shape. The actual measured
two dimensional data is not reported, but it can be estimated by back calculation using the
results from Saltikov's method [36]. Therefore, it is possible to obtain two dimensional
and three dimensional measurements for the alummum-tin specimen; that data can be used
The data from the aluminum tin alloy is far from ideal, since the two dimensional
results are in the form of back calculated values, and only nine size classes are given. With
only nine size classes, it is likely from Figure 6.6 that the method of Matsuura and Itoh
[39] has not converged to a stable solution. The three dimensional results are also
somewhat dubious, since the distribution was measured in three dimensions by x-ray
112
stereo radiography, with the size of each grain being determined by comparison with a
sphere of known diameter. The accuracy of such a method is not likely to be high and only
seven measured size classes were reported. Even so, an estimation of the three
dimensional grain size distribution obtained using the method of Matsuura and Itoh [39]
Material d ,\im
vh *v d ,\im
vh s v
Table 6.5 - Comparison of methods for Al-Sn alloy and stainless steel.
113
0.30
0.25 H
Figure 6.7 - Comparison of the measured 3-D grain size distribution with
the estimated distribution, for stainless steel, using the method of
Matsuura and Itoh [39], from Matsuura et al. [25].
CHAPTER 7
The objectives of this work were firstly, to quantify the kinetics of austenite grain
growth in two plain carbon steels, and secondly, to determine an appropriate method for
For the purpose of investigating the kinetics of grain growth, the grain size was
measured in specimens of A36 and DQSK steel, heated to temperatures between 950C
and 1150C, and held for periods up to 900 seconds. Linear intercept, l , and mean area,
m
A , were measured using the Heyn and Jeffries manual methods and by the use of image
m
diameter, d , was calculated from mean area, A . Standard statistical methods were used
Am m
to determine 95% confidence intervals of approximately 10% for the mean linear
measurement methods.
Systematic errors in the determination of the linear intercept length and the mean
equivalent area diameter were identified and discussed. The true linear intercept is closest
to that determined manually by Heyn's method [41]. The true mean area can be
image analysis [40]. The number of grains which touch the edge of the image being
measured must be known. Mean area determined by Jeffries' method [41] is a good first
dimensional grain size distribution by the use of the image analyzer, since the true two
dimensional grain size distribution can then be determined from this result. Such a
114
115
measurement is not difficult to perform and the necessary image analysis equipment is
The kinetics of austenite grain growth were observed in the A36 steel and in the
D Q S K steel. In both materials abnormal growth and a limiting grain size were observed.
Growth kinetics in the A36 steel were influenced by heating rate and by solution treatment
of the specimens. The kinetics of grain growth were adequately interpreted with respect to
the influence of second phase particles, which were concluded to be aluminum nitride,
from the work of Gladman [28]. The phenomenon of abnormal grain growth and a hndting
grain size were also explained by the presence of aluminum nitride particles.
The kinetics of grain growth observed under laboratory conditions are expected to
be quite different from those observed in a hot strip mill, because the thermal and
mechanical history obtained during hot rolling is difficult to reproduce. Both the
dispersion of second phase particles and the width of the grain size distribution influence
the kinetics of grain growth and both are influenced by the thermal and mechanical
processing history.
The power law was fit to results of mean equivalent area diameter obtained for the
A36 and the DQSK steels. The fitting constants used were found to be dependent on the
thermal history. Because the exponent, m was greater than two, no fundamental basis for
the power law equation was found. It must be concluded that the power law is an
empirical relationship, which should not be extrapolated to model grain growth for
conditions outside of those for which the results have been obtained.
The statistical model described by Abbruzzese and Lticke [20,21] was applied to the
results from the A36 and the DQSK steels by Militzer et al. [61,62]. The pmiing
parameter, P, was used to fit the model and the parameter could be interpreted from a
theoretical perspective [61,62]. The pirating parameter, P, was also estimated for hot strip
116
rolling conditions [61,62]. The results obtained in this work were therefore used to
Thus, as anticipated in the first objective, the kinetics of austenite grain growth have
been quantified. The results have been used to validate a statistical model, which can be
used to model the kinetics of austenite grain growth during hot strip rolling.
The evolution of the measured two dimensional grain size distribution was described
for a case of normal growth and abnormal growth in the A36 steel specimens heated at
5C/second. Under conditions of normal grain growth at 1150C, the distribution width
and shape were seen to change little over time, with the distribution being closely
approximated by a log normal curve. The mean of the distribution was observed to
increase by scaling.
At 1000C and after 10 seconds, the distribution shape was also apparently log
normal. Under conditions of abnormal growth at 120 seconds, the distribution was seen to
widen dramatically, and then at 600 seconds, to return to a width and shape similar to that
initially observed.
Methods for estimation of the three dimensional grain size distribution were
compared by applying each to results obtained from the 1080, the A36 and the D Q S K
steels. It was determined that the methods of Saltikov [36] and Huang and Form [37],
based on the assumption that the grain shape is spherical, were unacceptable; both
methods predicted a negative number of grains for the smallest size classes observed.
The method of Takayama et al. [38] was found to be sensitive to the ratio of linear
intercept to the square root of the mean grain area. When the method's sensitivity was
dampened by the use of a mean value for the log normal standard deviation, s , reasonable
v
estimations were obtained for the equivalent volume diameter, d . Vm The method was
found to be unsuitable for determining the width of the distribution and could not be
The method of Matsuura and Itoh [39] was found to be appropriate for calculating
the mean equivalent volume diameter and the width of the distribution. The true, corrected
two dimensional grain size distribution must be used as input. Some scatter was observed
in the histograms predicted by the method, but the distribution shape generally appeared to
be reasonable. The scatter was a results of the statistics of the measurements, which were
performed using a limited number of grains, due practical considerations. Some scatter
procedure. The method was applied to measured distributions where the shape of the
distribution was log normal, and also to distributions where the shape was not log normal.
It was found that a stable solution was predicted when greater than ten size classes were
employed. The method was originally validated using measured three dimensional results
was reported in the literature [49] and this was also used to validate the method.
The methods of Takayama et al. [38] and Matsuura and Itoh [39] produced values
for the mean equivalent volume diameter, d , Vm which were almost equal. Differences
between the two results could be accounted for by the statistics of the measurements. For
cases where normal grain growth was observed, both methods confirm that the
approximate relationship,
dvm =
L2dAm = l.5lm
It must be concluded that the method of Matsuura and Itoh [39] was the most
suitable of those compared, for the estimation of the three dimensional grain size
distribution. Therefore, the second objective, to determine an appropriate method for the
estimation of the three dimensional grain size distribution, was also met.
BIBLIOGRAPHY
of Steel, Winnipeg, Canada, Edited by G.E. Ruddle and A.F. Crawley, Aug. 1987,
Properties in Plate and Sheet Steel Production, The Sumitomo Search, No.44, Dec.
1990, ppl59-168.
[3] C. Devadas, I.V. Samarasekera and E.B. Hawbolt, The Thermal and Metallurgical
State of Steel Strip during Hot Rolling: Part III. Microstructural Evolution, Met.
[4] P.D. Hodgson and R . K . Gibbs, A Mathematical Model to Predict the Mechanical
Properties of Hot Rolled C-Mn and Microalloyed Steels, ISIJ International, Vol. 32,
[5] C M . Sellars and J.A. Whiteman, Recrystalfization and Grain Growth in Hot Rolling,
[6] J.E. Burke and D. Turnbull, Recrystallization and Grain Growth, Progress in Metal
[7] P. Feltham, Grain Growth in Metals, Acta.Metall., Vol.5, Feb. 1957, pp97-105.
[8] M . Hillert, On the Theory of Normal and Abnormal Grain Growth, Acta MetalL, V o l .
13,Mar.l965,pp227-238.
[9] J.C. Fisher and R. Fullman, Private communcation to Burke and Turnbull, [6.].
[10] M . P . Anderson, G.J. Srolovitz, G.S. Grest and P.S. Sahni, Computer Simulation of
[11] P. Gordon and T.A. El-Bassyouni, The Effect of Purity on Grain Growth in
[12] H . Hu, Grain Growth in Zone Refined Iron, Can.Metall.Q., Vol.13, 1974, pp275-
286.
[13] G.F. Boiling and W.C. Winegard, Grain Growth in Zone Refined Lead, Acta
[14] J.P. Drolet and A . Galibois, The Impurity Drag Effect on Grain Growth, Acta
[15] E.L. Holmes, and W.C. Winegard, Grain Growth in Zone Refined Tin, Acta Metall.,
[16] H . V . Atkinson, Theories of Normal Grain Growth in Pure Single Phase Systems,
Multiple Hot Working, Proc. 7th Risty Int.Symp.on Metallurgy and Materials
[18] P.C. Campbell, E.B. Hawbolt and J.K Brimacombe, Microstructural Engineering
Applied to the Controlled Cooling of Steel Wire Rod: Part II. Microstructural
pp2779-2790.
Derivation and General Discussion of the Model, Acta Metall, V o l . 34, 1986,
pp905-914.
[21] G . Abbruzzese and K . Lticke, Theory of Grain Growth in the Presence of Second
pp2671-2678.
[23] S.K. Kurtz and F . M . A . Carpay, Microstructure and Normal Grain Growth in Metals
[24] F.N. Rhines and K.R. Craig, Mechanism of Steady State Grain Growth in Aluminum,
[25] K . Matsuura, Y . Itoh, M . Kudoh, T. Ohmi and K. Ishii, Three-dimensional Grain Size
Distribution in SUS304 Stainless Steel, ISIJInt., Vol. 34, No. 2, 1994, ppl86-190.
[26] W . M . Williams and C.S. Smith, A Study of Grain Shape in an Aluminum Alloy And
[27] C . H . Desch, The Solidification of Metals from the Liquid State, Journ.Inst.Metals,
[28] T. Gladman, The Theory and Inhibition of Abnormal Grain Growth in Steels, JOM,
[29] C.H. Worner and P . M . Hazzledine, Grain Growth Stagnation by Inclusions or Pores,
[30] C. Zener as quoted by C.S. Smith, Grains, Phases and Interfaces: A n Interpretation of
[31] T. Gladman, On the Theory of the Effect of Precipitate Particles on Grain Growth in
[34] P.T. Mazarre, S.W. Thompson and G. Krauss, Microalloying and Austenite Grain
1991,pp497-509.
[35] H.E. Exner, Analysis of Grain and Particle size Distributions in Metallic Materials,
[36] S.A. Saltikov, The Determination of the Size Distribution of Particles in an Opaque
ppl63-169.
[37] K.P. Huang and W. Form, A Simplified Method to Reconstruct a Spatial Grain Size
1990, pp332-340.
for the Estimation of the Grain Size of Polyciystalline Materials, Mat.Trans., JIM,
[43] F . N . Rhines and B.R. Patterson, Effect of the Degree of Prior Cold Work on the
[45] R . G . Cortes, A.O. Sepulveda and W.O. Busch, On the Determination of Grain Size
[46] K.P. Huang and W. Form, A Simplified Method to Reconstruct a Spatial Grain Size
[49] H.B. Aaron, R.D. Smith and E.E. Underwood, Spatial Grain-Size Distribution from
paper 16.
Shape Assumption. Part 1: Theory and Testing on Spherical Particles, Mat. Charact.,
[51] N . Furushiro, Y . Takayama and S. Hori, Estimation of the Mean Grain Diameter of
[53] R.T. DeHoff, The Estimation of Particle-Size Distributions from Simple Counting
1965, pp25-29.
[54] S.K. Kurtz and F.M.A. Carpay, Microstructure and Normal Grain Growth in Metals
[56] Metals Handbook Desk Edition, Edited by H.E. Boyer and T.L. Gall, American
[57] L . E . Samuels, Optical Microscopy of Carbon Steels, American Society for Metals,
Ohio, 1980.
1984.
[60] O.O. Miller, Influence of Austenitizing Time and Temperature on Austenite Grain
[61] M . Militzer, A . Giumelh, E.B. Hawbolt and T.R. Meadowcroft, University of British
[62] M . Militzer, A . Giumelli, E.B. Hawbolt and T.R. Meadowcroft, Austenite and Ferrite
Grain Size Evolution in Plain Carbon Steel, 36th Metal Working and Steel
[63] N . A . Gjostein, H . A . Domain, H.I. Aaronson and E. Eichen, Acta metall, Vol.14,
1966, ppl673.
Oxford, U K , 1982.
1970, pp383-390.
APPENDIX
(%)
Table A.1 - A36 austenite grain size results obtained by standard Gleeble thermal
treatment and manual measurements, using Heyn's and Jeffries' procedures [41].
124
125
Table A.l(cont) - A36 austenite grain size results obtained by standard Gleeble thermal
treatment and manual measurements, using Heyn's and Jeffries' procedures [41].
126
(%)
Table A.l(cont.) - A36 austenite grain size results obtained by standard Gleeble thermal
treatment and manual measurements, using Heyn's and Jeffries' procedures [41].
127
c/ conf
Table A.2 - A36 austenite grain size results obtained by standard Gleeble thermal
treatment and measurement using the image analyzer [40].
Table A.3 - A36 austenite grain size results obtained by continuous cooling on the
Gleeble, and measurement by Heyn's and Jeffries' procedures [41].
128
(%)
1050 AR 1 371 46.4 9.2 55.9 2455.1 15.8
Table A.4 - A36 austenite grain size results obtained by stepped Gleeble thermal
treatment and manual measurement by Heyn's and Jeffries' procedures [41].
129
A.5 - DQSK austenite grain size results obtained by tube furnace thermal treatment
and measurement using Jeffries' procedure [41] and the image analyzer [40].
130
Table A.6 - D Q S K austenite grain size results from standard Gleeble thermal
treatment and measurement by Jeffries' procedure [41].
1050 0 1313 43.8 2.2 53.6 2258.2 6.6 50.3 1953.7 7.5
1050 120 1055 49.3 4.0 59.8 2810.4 6.2 55.3 2405.6 7.4
1050 420 756 58.5 5.8 70.7 3922.0 9.2 64.7 3273.9 8.2
Table A.7 - 1080 steel austenite grain size results from furnace thermal treatment
applied in previous work [55], and measurement by Heyn's and Jeffries' procedure
[41], and by the image analyzer [40].