Metallography
161
Determination
of Volume
Distribution
Function by Lineal Analysis
K. G. STJERNBERG Department
of Physics, Chalmers
University
of Technology,
Gothenburg,
Sweden
A linear method for calculating the volume frequency distribution function is described. The factors affecting lineal analysis are defined, and a diagrammatic method of obtaining the size frequency distribution function is presented.
Introduction Information about grain size and grain shape in metals is often obtained by studying a polished and etched section in a metallographic microscope. It is usually sufficient to determine a mean grain size by counting either the number of grains per unit area (Jeffries’ method’) or the number of grains per unit length (Heyn’s method2). The latter method has the advantage that the mean grain size is directly related to the specific grain surface through a simple relation.3 To obtain more information
about the structure,
the sectioned areas of the
individual grains are measured (area1 analysis), or a straight line is drawn intersecting many grains and the distances between the pairs of points (intercepts) in which this line cuts each grain are measured (lineal analysis). In the former case the area1 distribution distribution function. Using the assumption mathematical
function
is determined,
and in the latter the lineal
of spherical grains, Bockstiege14 has derived a simple
relation between the spatial grain size distribution function and the
lineal distribution function. In this paper a method is described to determine by lineal analysis the volume frequency function, g(Z), defined by g(Z) dZ equals the fraction of volume occupied by grains with diameters between 1 and 1 + dl. It is also shown that corrections General
must be made when it is impossible to measure along one single line. Theoretical
Considerations
In the following presentation it is assumed that thematerial grains. Starting from Bockstiegel’s
consists of spherical
equation, some quantities will be calculated. Metallography,
Copyright
0
1969 by American
Elsevier Publishing
2 (1969)
161-170
Company,
Inc.
K. G. Stjernberg
162 The following
notation
d(h) dA = number
is used:
of intercepts
per unit
length
with
lengths
between
h and
X + dh. N’
zzz
total number
of intercepts
silnV) ~
= probability
F(Z) =
Nl distribution
n3(h) dA = number
per unit length. that an intercept
is less than
1 (the lineal
function). of grains
per unit
volume
with
diameters
between
h and
X + dh. jjl3
=
total number s:, n3(h) dh
F3(1) =
N3 (grain
of grains
= probability
size distribution
G(1) = fraction
of volume
(volume
distribution
occupied
of a grain is less than 1
by grains
with
diameters
less than
I
function).
l/N1
= mean
TX
l/N3
= mean volume
intercept
length. of the grains.
we introduce
f”(Z) = $F3(1);
f’(Z) = &F’(Z);
According
that the diameter
function).
I=
In addition,
per unit volume.
g(l) = $
G(j)
to Bockstiegel,4 = ; I jW n”(D) dD 2
p(1)
Derivation
and rearrangement
Integration
of equation
(1)
gives
2 gives $‘3(4
=
1 _
2 ~_N1
nN3
f’(l)
1
(3)
Lineal Analysis Multiplication
Determination by 1 and derivation F(Z)
Putting
163 gives
+ 1 &F(Z)
=
1 -
J&
-F(Z) ;a
1 = 0 gives (4)
Thus
7ri
v=
(5)
2(d2/dZ2) F(0)
By definition,
g(Z) = Iv327+f which
“(I)
(6)
Z
(7)
can be written g(z)
-Adfl(l)
=
31 Definition Suppose
dZ
of x that a point
is randomly
chosen
p(Z) dl, that the point
probability,
on a straight
is on an intercept
line in a specimen.
with length
between
The 1 and
Z + dZ is p(Z) dZ = +(Z) We now chosen
define point.
x as the mean
length
dZ
of intercept
(8) drawn
through
a randomly
Thus
j-aZ f
ii =
f l(Z) dZ = f [” Z2f1(Z) dZ
0
YO
The definition of x is valid for any grain shape, and so is equation 9. The value of x can always be calculated from lineal analysis data. Thus x is a general structure parameter. The mean value,
& , of Z with respect
to g(Z), for spherical
grains
is
i, = _f* t-g(Z) dZ = $ j-m Z”f’(Z) dZ = ; ii 0
0
(10)
Lineal Analysis Some general function
relations
between
have been derived.
Before
different
quantities
we use these
and the lineal distribution
relations
for numerical
calcula-
K. G. Stjernberg
164 tions, a method of determining
the function F’(Z) will be described.
As some
corrections have to be made, it will be shown in detail which quantities can be determined by lineal analysis and how they must be corrected. In the following discussion we consider a material consisting of two phases, a and j3. The a-phase consists of separate grains which are embedded in continuous B-phase. The grain shape is such that only a slight error is introduced by regarding it as spherical. Furthermore,
we are interested only in the lineal distribution
of the a-phase. As to the p-phase, intercept length.
we are content with determining
We also assume the lineal analysis to be done on photographs. on photographs,
the method
function the mean
(By measuring
can also be used with an electron
microscope.)
All quantities connected with the quantity I (for example, L and A) are lengths in the specimen. The corresponding lengths on the photograph are denoted by analogous notation connected with the letter x (for example, L and 1 correspond to X and 5, respectively).
Thus if the magnification
is G,
x = Gl On the photographs length X.
Total
the intercepts
accumulated
are measured
lengths
along N lines of average
in the 01- and B-phases
are measured
separately. The total lengths also include the lengths of those fractions of intercepts that are at the beginning and the end of every measuring line. By lineal analysis it is possible to determine the following quantities: Xtot = total measured length on the photographs. X,
= total measured length in a-phase on the photographs.
Xs = total measured length in B-phase on the photographs. Ni”
= number
of measured
a-intercepts
along Xt,t
which are less than xi ,
where xi; i = 0, I, 2 ,..., p, is the grain class boundaries. N,”
= number of measured a-intercepts
along Xt,,t .
IV,” = number of measured B-intercepts
along Xt,,t .
(The upper indices, m, in the last three quantities mean that the quantities are those initially determined. It will be shown below that they need to be corrected. The corresponding corrected quantities.)
quantities
without
upper
indices
With this notation it is possible to calculate some important
represent
the
quantities:
c, = volume fraction of a-phase5 = X,/Xtot .
(114
cs = volume fraction of B-phase = 1 - c, .
Ulb)
j
a
_
1
x, _ mean __
G N,
intercept length in a-phase.
Lineal Analysis Determination 1
4
=LxB=
G NB
&NE--___
N,
165
mean intercept length in p-phase.
= continuity.6
To get a correct value of the total number of intercepts
found on the length
Xtot , it is necessary to add half the number of end points of the measuring lines which are in the or-phase. The probability that an end point is in the a-phase is c,.~ Thus the following equations are valid: N, = N,” + 42Nc, NB = N,”
= N,” + NC,
(124
+ NcB
(12b)
Determination of Fl(x) In practice, Fl(x) is determined
by measuring the different intercepts
along
lines on photographs. This is most easily done by using a device which has certain reference lengths xi; i = 0, 1, 2 ,..., p, built in. The number of intercepts less than xi is determined
for each value xi,& The
probability,
F”(xJ,
that a
measured intercept is less than x is given by
It is now convenient to discuss the way of choosing the xI’s. In the literature the relation xi+i = kxi is often seen. This choice is suitable when n3(Z) is to be determined because n”(Z) often seems in reality to be logarithmiconormaldistributed.ssg
Experience
shows that it is better to use a lineal dividing when
determining g(Z), because g(Z) is almost
symmetric
even if n3(Z) seems to be
logarithmiconormal (see Fig. 1). Thus xi is chosen here as xi
=
ih;
i = 0, 1) 2 )..., p
where h is the difference between consecutive values of xi . Suppose that the total measuring line, Xtot , is divided into N parts (they need not be of the same length). Those intercepts that have one end point outside the
& The author has used an instrument where the reference lengths are contact bars. After measuring on the required number of photographs, N,“, No”‘, and N,“; i = 1, 2, 3,..., p, are read on electrical counters, and the lengths Xtot and X,g on mechanical counters.
K. G. Stjernberg
166 photograph
cannot be measured.
The probability
C, ;&) An expression between
that an end point is on an
of length between x and x + dx is
a-intercept
can be written
x and x + dx which
(compare with equation 8)
dx
for the number
of intercepts
would have been
counted
with lengths
on a continuous
measuring line of length Xtot :
N&f’(x) dx = Nmmfm(x) dx + 2Nc, ; ;jl(z)
dx
This can be written as
f’(x) =
-%I&)fm(x)
N,
If X = Xt,,t/N and N,/Nem = K, the equation becomes
Multiplying
by K(X
-
X
~-
f’cx) =
K(X
_
x)
f
“(‘)
(13)
x) and integrating gives
KX ,; f ‘(x) dx -
K s; xf’(x)
dx = X c
f “(x) dx
Thus (14) The value of K can be calculated from equation 13:
sco
K=
Multiplying
o
equation 13 by xK(X
-
-+&f
m(4dx
(15)
x) and integrating gives
f_X-& where l = GA and xm = s,” xf”(x) dx. Equations numerical calculations of K and A, respectively.
(16) 15 and
16 are used
for
Numerical Calculation of g(Z) We now assume that we have an array of experimentally the function F”(xi)
determined values of
= Nim/Nam, where xi = ih; i = 0, 1, 2,..., p. The values of
Lineal Analysis P(x,)
Determination
167
are considered to be so exactly determined
derivation
and the dividing so finethat
can be done as follows:
;Fqxi+J= fyXi+$)
A new array of values offm( x ) is obtained. Equation
f’@i+J = jr(x)
is approximated
fqx
i’
1
13 gives
f”(Xi+J
xi+;)
by a polynomialfl(x)
0, 1,2 )..., p -
i =
= ; [F”(x,+,) - FV’(Xi)l
= ax2 +
bc +
c,
where a, 6, and c
are determined in such a way that fr(x,+J when j = i -
= ax:++ + bxj+* + c
1, i, and i + 1. Using equation 7 gives
_--
By solving the system of equations above, (ax2 the quantitiesfr(xj+;);
g(xi+J
=
j = i -
1 x. (axf+r 3% z+t 1
c)
c) can be expressed in terms of
1, i, and i + 1, and it is possible to write
& I”i+ifl(x$++) ++
[fl(Xi-J
-fY%+*lli
The following notation is now introduced:
X
Yi+t =
X
-
xi+'
f
*(xi+*)
I
G and
Q =
6X(K
-
1)
It is easily shown that g(Z,++) is given by
g(Z,++) = Q
Estimation
[2xi++yi++ + kXf++(J’i-+- yi+s)]
(17)
of the Errors
The number of intercepts with length between two given values is a stochastic variable that according to statistical laws is binomically the probability 3
distributed.
Let SF, be
that the intercept length is between the two given values. The
168
K. G. Stjernberg
standard deviation, ogF, of SF, when one attempts to determine SF by measuring N intercepts, is 06F
=
I
6F,(l - SF,)
(18)
2-N
If only the error due to the statistical uncertainty is considered (this is supposed to be the largest part of the total error), yi+* can be written:
u
the standard deviation of the value
X
y*+:=
_y - xi+’ I
Differentiation of equation 15 gives
Assuming that the numerical values of all dy’s are equal and that the combination of signs is unfavorable, the standard deviation of g(1) can be roughly estimated to be
ug - TX
(1+ 5)-&
J
f”
(:,-f-) (19)
N
Diagram for g(1) The function g(Z) is represented
diagrammatically
in the usual manner (see
Fig. 1). In the diagram is also included the function g = cl3 for different values of c. According to equation 6, c = (n/6) n”(l). Thus along a curve g = c13, n”(l) = constant = 6c/~r. This fact can be used when plotting the function n”(1). The values of 1 are read for the intersections between g(1) and the curves g = c13, and the corresponding 1 and n3(l) values are plotted in a diagram. There is also another case where the curves g = cl3 are of great importance. This is in studying grain growth when it is expected that all grains of the same size grow at the same rate. Then all grains of a certain size will grow or shrink by equal amounts. This means that n3 is a constant. In the diagram this means that the points on g(l) move along curves g = ~13. As an example, the functions g(l) and n3(1) for a specimen tungsten-titanium-carbide
consisting
of
sintered with cobalt are shown in Fig. 1. Some 1000
grains are measured. The value of g(l) is calculated by means of equation 17. The errors in g(l) are estimated by means of equation 19. The errors in n”(l) are easily obtained because the relative errors of g(1) and n3(l) are equal.
Lineal Analysis
Determination
1
2
3
169
L
5
Diameter
6
7-
i-9
10
[micron]
FIG. 1. The volume frequency distribution function and the spatial grain size frequency distribution function for the same material. (The curve parameters are the values of the spatial grain size frequency distribution function.)
Discussion From equations 4 and 5 it is clear that it is not possible to determine accurately the total number of grains per unit volume or the mean volume of the grains, because the determination of these quantities involves the determination of the second derivative at the origin of the lineal distribution function. Smith and Guttman3 showed that it is possible to obtain a value for the specific grain surface, s, from lineal analysis data from the formula s = 4/i independent of structure.
Thus the mean diameter according to Heyn (that is, I) is in reality
not a length but a ratio between a volume and an area. The mean diameter as defined by Jeffries has even less meaning. In this paper another structureindependent parameter, A, is introduced which in fact is a certain length in the structure.
Thus for any structure it is possible by lineal analysis to calculate two
parameters of importance, namely i and A. Not much will be said about the corrections.
Experimentally
it is found that
the correction in i is about 10 o/owhen NIN, is between 0.1 and 0.3. As a consequence the corrections cannot be neglected even though there are ten intercepts on the average per line. In Fig. 1 it is easy to see the difference between g(Z) and n”(l). The part of n3(Z) to the right which is usually called a tail occupies in fact more than half the volume. Furthermore
in this case half the number of the smallest grains occupies
only a few percent of the volume.
170 The
K. G. Stjernberg position
of the maximum
in g(l) is of importance
when
studying
grain
growth. At least in this case it is found that grains with this diameter neither grow nor shrink. Another important fact is that the mean value of I with respect to the functiong(l)
is related
in a simple
manner
to x (equation
10).
Summary Some
structure
parameters
and
distribution
functions
have
been
defined.
Assuming spherical grains, some valuable relations have been derived. Lineal analysis has been treated in detail. Specifically it has been shown corrections
have to be made when
A method described.
for calculating A type of diagram
one is not measuring
the volume
frequency
has been introduced
along a continuous
distribution which
help when studying grain growth. It has also been shown distribution function can be obtained from the diagram.
function
is found
which line.
has been
to be of great
how the size frequency
References 1. 2. 3. 4. 5. 6. 7. 8.
Z. Jeffries, A. H. Kline, and E. B. Zimmer, Trans. &ME, 54 (1917) 594. E. Heyn, Metallogruphist, 6 (1903) 37. C. S. Smith and L. Guttman, Trans. AIME, 197 (1953) 81. G. Bockstiegel, Z. MetaUk., 57 (1966) 647. A. Rosiwal, Bull. GeoZ. Sot., 14 (1903) 466. J. Gurland, Trans. Met. Sot. AZME, 212 (1958) 452. A. A. Glagolev, Eng. Mineral. J., 135 (1934) 399. E. J. Meyers, First International Conference on Stereology, Congressprint, Austria, 1963. 9. H. E. Exner and H. Fischmeister, Arch. Eisenhiittenw., 37 (1966) 418.
Accepted May 6, 1969
Vienna,