Part. Charact.

I(1984) 89-95
89

Representation and Evaluation of Particle Size Analysis Data

Kurt Leschonski”
(Received: 29 March 1984)

Abstract
While editing the first issues of Particle Characterization it has and graphs representing particle size analysis data instantly rec-
been recognized by the editors that the papers submitted differ ognizable by using agreed upon standardized nomenclature the
widely in the representation of data and in the nomenclature following paper has been submitted by one of the editors. It was
chosen to characterize such common variables as particle size, first presented at the Harold Heywood Memorial Symposium,
different equivalent particle diameters, average particle sizes, September 1974, at the University of Technology, Lough-
cumulative and density values of a distribution etc. In order to borough, U. K.
make papers published in this Journal more readily assimilable

Table 3 : Representation of Particle Size Analysir Data.

1
1 Introduction __-.

a) type of quanlily b) measure of quantity

The individual elements of particulate matter always form a
distribution the elements of which can be classified according to Y = 0: number Q, : cumulative distribution
their size. The distribution is, therefore, characterized by the r = 1: length 4,: density distribution
quantity, for instance the number or the mass of particles present a,b Y = 2: area, surface
in individual size classes. The elements can be solid particles, Y = 3: volume, weight qr ( x ) = d Qr (x)/dx
droplets, bubbles or pores. Particle size analysis must provide the *
physical property chosen to characterize the size of particles
necessary data and a multiplicity of methods can be used to ___-___ ____-- ~

analyze particulate matter. Each group of measurement techni-

ques based on the same physical principle provides, in general,
different distribution curves. The reason for this must partly be The independent variable is plotted on the abscissa describing the
attributed to the fact that, in particle size analysis, every possible physical property chosen to characterize the size of the particles.
physical principle has becn used to determine these distributions. The dependent variable, which characterizes type and measure of
Due to many reasons [l] the results will, in general, not be the quantity, is plotted on the ordinate.
same. One of the main problems still encountered with most If one counts particles having the same physical property, the
methods in use is their unknown absolute accuracy. However, type of quantity used is a number. If one weights them, the type
apart from this the distributions differ due to two reasons. Firstly of quantity is a mass. Particles laid end to end form a line of a
different physical properties are used in characterizing the “size” certain length but they also represent a ccrtain projected area.
of individual particles and, secondly, different types and Thus length and area are other posyible types of quantity used in
different measures of quantity are chosen. particle size analysis.
The possible types of quantityare: number, length, area, volume
or with constant particle density: mass.
This is the decisive difference from ordinary statistics where the
type of quantity used always is a number. Since number and
2 Graphical Representation volume distributions of the samc powder may considerably
of Particle Size Analysis Data differ, the type of quantity measured or chosen when represent-
ing the data has to be taken into account. One further must, how-
2.1 Particle Characteristicsand Types and Measures of Quantity ever, distinguish between two measures of quantity. This is
necessary because one may for instance in a counting method
In a graphical representation of particle size analysis data (see either count all particles smaller than a certain size or those found
Table 1) the independent variable is plotted on the abscissa in certain size classes.
describing the physical property chosen to characteriLe the size of The different relative amounts of particles measured in certain
the particles. The dependent variable, which characterizes type size intervals form the so-called density distribution qr (x)which
and measure of quantity, is plotted on the ordinate. represents the first derivative of the cumulative distribution
Q,(.u). In this notation x characterizes a physical property
* Prof. Dr.-Ing. K. Leschonski, Institut fur Mechanische Verfahrcns- uniquely related to particle size. The subscript rindicates the type
technik, Technische Universitat Clausthal, Leibnizstr. 15, of quantity chosen. Table 2 shows the types of distribution that
D-3392 Clausthal-Zellerfeld (Federal Republic of Germany). can be distinguished.

0Verlag Chemie GmbH, D-6940 Weinheim 1984 0176-2265/84/03i1-0089$02.50/0

90 Part. Charact. I(1984) 89-95

Table 2: Different types of distributions. 'Table 4: Intercorrelation of equivalent diameters.

dcnsity cumulative
distribution distribution

number: r = 0 40 (4
length: r = l 41 (4
area : r = 2 92 (4
volume: r = 3 43 (x)
general
description: r == r 9 r (4 In principle the following inequality holds:

Physical properties that characterize particles are listed in

Table 3. These are one-, two- or three-dimensional particle prop- 2.2 Cumulative and Density Distribution, Histogram
erties, such as linear dimensions of chords of different definition,
the projected area of particles, their surface and their volume. The graphical representation of particle size analysis data is
Other characteristics are the mass of particles, their settling rate shown in Figures 1 - 3. The term x represents an arbitrary equi-
or terminal velocity in a viscous fluid or the response of an valent diameter, the dimension of which is a length such as pm.
electrical or an optical field to the presence of particles.

Table 3: Physical properties that characterize particles.

a) linear dimcnsion
b) surface, projected area
c) volume I
95 I
d) mass I
e) settling rate I
t) response of electrical, optical or acoustical field I
-___ I
I
I [length1
n
v
Each of these properties can be used to characterize the size of a Xmin x~ X 2 x 5 ~ X- %ax
particle. In principle, however, they canusefully be applied to the
Fig. 1 : Cumulative distribution Q,(x).
characterization of particles only if there exists a unique relation-
ship between the physical property used and a one-dimensional
property unequivocally defining "size". With irregular particles Figure 1 shows the normalized cumulative distribution Q, (x).
this can, in general, be achieved only approximately. Each individual point of Q,(x) specifies the relative amount of
One defines therefore the so-called equivalent diameters. These particles smaller than, or equal to, a certain equivalent diameter.
are diameters of spheres which yield the same value of a certain The kind of quantity chosen is indicated by r. The normalized
physical property when analyzed under the same conditions as curve Q, (x) runs between zero and one:
the irregularly shaped particle. The relationship between the
physical property chosen and the equivalent particle diameter
Qr(x = x m i d =0
must either be evaluated theoretically or measured experiment-
ally. Q, (x = x,,,) = 1 (condition of normalization)
An equivalent diameter represents the diameter of a sphere that
has the same physical property as the irregularly shaped particle.
One distinguishes, for instance, the following:
xp: equivalent projected area diameter,
xs: equivalent surface area diameter,
xv: equivalent volume diameter,
x,: equivalent settling rate diameter.
For a given sphere all equivalent diameters are the same. They
differ from each other if the particles are irregularly shaped.
Under certain assumptions and after having defined a shape
factor that describes the deviation in shape from a sphere the Fig. 2: Histogram of a density distribution q, ( x ) .
equivalent diameters given above can be intercorrelated using the
shape factor sphericity, iy, as defined by Wadell [2]:
Figure 2 shows the normalized histogram of a density distribution
,lu = xqx; . (1) q, (xl,xz). The difference A Q, of two values of the cumulative
distribution Q, (x)between two equivalent diameters x, and x2 =
Using Cauchy's theorem [3] and a force balance applied to a x1 + A x is represented by:
particle settling in a viscous fluid under laminar conditions, one
arrives at Table 4.
Part. Charact. I(1984) 89-95 91
_ _ _ ~ -

The histogram can then be obtained from: pore volume. Usually one figure, or at the most two figures,
must represent the size distribution adaequately. The problem
of defining a n average particle size (a position parameter) and
(3) another parameter describing the spread of the distribution (a
spread parameter) then arises. Some of the existing
q, (xl,x2) represents the constant height of the rectangular possibilities will be explained in the following sections using
shaded area of Figures 2 and 3 of width A x . The shaded area is the concept of moments, as proposed by Rumpf and Ebert in
therefore given by: 1964 [4].

3 The Concept of Moments

and represents the relative amount of particles within the size
interval Ax. 3.1 Complete and Incomplete Moments, Definitions

The concept of moments is old and well known in statistics and

mechanics. The following integral represents, for instance, the
so-called expected value of xk, E(xk), which equals the
k-th moment of x.

Xrnax
XkqO(x)dx= E ( x k ) . (7)
Xrnin

With k = 1, the expected valueE(x) = R,,,, i. e. the arithmetic

Ax mean of the number distribution [4 - 61.
c
[length] The integral
Xmin x- Xmax
Xrnax
Fig. 3 : Density distribution q r ( x ) . S (X - X)kqo(X)dx = E[(x - .Qk]
Xmin

equals the expected value of (x - X)k and represents the k-th

If Q,(x), as plotted in Figure 1, is a differentiable curve the
central moment of x.A well known example with k = 2 is the
density distribution qr(x)is obtained from
variance or the square of the standard deviation (Eq. 12).
The moments of Eqs. (7) and (8) are based on number distribu-
tions, indicated by the subscript nought of q. For other types of
quantity a diffcrent notation is therefore suggested. The
complete k-th moment of q, (x)is defined by:
which is plotted in Figure 3. It represents the first derivative or the
slope of Q, (x).It should be noted, that q, (x)has the dimension of Xmax

an inverse length. Irrespective of the type of quantity plotted the Mk,r= xkq,(x)dx. (9)
overall area underneath qr(x)equals unity or 100 percent.
I *min

iecond subscript: type of quantity: r = 0, 1,2,3

first subscript: power of x:k = . . +- 3 , - 2, - 1,0,1,2,3,. . .

The abbreviation Mk,r is used to characterize the type of moment,

the power of x and the type of quantity. It is important to agree
Each of the three curves shown in Figures 1- 3 represents the
upon and to bear in mind, that the first subscript indicates the
information needed for further evaluation of the data. If the
power, k, of x while the second subscript, r, indicates the type of
cumulative distribution, Q, (x),has been obtained by multiple
quantity chosen. M is used for moments related to the origin
analyses, statistical methods allow the most probable distri-
(x = 0) of the abscissa, and m for central moments related to the
bution curve to be found. If certain assumptions with regard to
abscissa of the center of gravity of the density distribution q, (x)
an analytical function which fits Q, (x)are possible, regression
(see section 3.3.2). Thecompletecentralmoment, mk,r,thereforeis
analysis using least square methods will yield the most probable
given by:
parameters of the analytical function. The main problems in
evaluating more data from these curves originate mainly from
two sources:
a) The distribution as plotted has been analysed with several size
analysis methods using different types of quantity and
different physical properties defining “size”. Therefore the If the lower and upper limit of the integral coincides with the
values plotted on the ordinate as well as on the abscissa have to smallest (xrnIn)
and largest particle size (xrnax)
of the distribution, the
be recalculated and to be reduced to the same form. It may be moment is called a complete one. For other limits x, and x, the
necessary, for instance, to transform a number distribution of moment is called incomplete.
volumes into a volume distribution of equivalent sizes.
b) The distribution curves obtained have t o be correlated to
other material properties, as for instance strength, solubility,
92 Part. Charact. l(1984) 89-95

The subscripts of the above integrals do not necessarily have to be a Average particle sizes to be calculated from a given size
single figure or letter. In formulae of general applicability, sums of distribution, qr(x), can be determined from Eq. (18):
letters and figures are usually found. The letter r , indicating the
type of quantity chosen, also varies, mainly if two types of quantity
appear in the same equation.
Well known examples describing a complete central moment are
the variance, .$, and the coefficient of variation, C,, of a distri- It is most important to indicate a n average particle size by two
bution q, (x): subscripts, k and r, as shown in Eq. (18).

3.3.1 Arithmetic Average Particle Sizes

Arithmetic average particle sizes are always calculated from a

number distribution q0(x). Their general notation is Xkk.O.

3.2 Calculation of Any Moment from a Given Distribution

In the general application of the concept of moments, it is often

Three arithmetic mean sizes are given in Eqs. (20) - (22), showing
necessary to evaluate the k-th moment of the q, (x) distribution
the notations used by Heywood [7] on the left hand side and the
from the basis of a number distribution. By introducing Eq. (27)
relevant moments on the right hand side:
into Eq. (9) it can easily be shown, that M,,, can be calculated from
the ratio of two moments:

Eq. (14) can be used to interpret the physical significance of any

given moment, Mk,rr since the first, second and third moment of a
number distribution reprensents its average length, area or
volume. If for instance k = - 1 and r = 3 one arrives at:

3.3.2 Weighted Average Particle Sizes

According to Eq. (15) the first negative moment of a volume
distribution equals the ratio of the second to the third moment of The general notation of all weighted mean sizes is 21,r.
the number distribution. This ratio is directly proportional to the
ratio of surface area to volume (see section 5 ) .
xs,i = Ms.r = ~ I + ~ , O ~ ~ L O ~ (23)
Eq. (16) must be used if the k-th moment of the qr(x)distri-
bution, Mk,i,has to be calculated from a given arbitrary distri-
Three weighted mean sizes are again given in Eqs. (24) - (26):
bution qe(x), with e = 0, 1 , 2 or 3:

If one calculates, for example, M - 1 , 3 from a surface or projected

area distribution with e = 2 one arrives at:

M-l,3is inversely proportional to the first moment of the surface

or projected area distribution.

3.3 Averages, Average Particle Size

Particle size analysis is performed t o characterize particulate 3.4 Calculation of Any Density Distribution, q, (x),
matter and to link the size information with some other technical from a Given Distribution, qe(x)
property as for instance strength, flowability, pore volume,
solubility. In general, a correlation can be obtained much more It is well known to particle technologists that a wide size
easily if a single number, describing a mean or average particle distribution cannot always be analyzed by one principle or by one
size, characterizing the distribution can be used. One should, method of size analysis. Sometimes the same principle, for
however, bear in mind, that at least two figures are necessary to instance sedimentation, can be applied to all particles, the finest
approximately describe a distribution: ones being analyzed in a centrifuge the bigger ones in a
a) a position parameter that is an average particle size, and gravitational field. O n the other hand quite often, with the
b) a parameter for the spread of the distribution, for instance its different methods applied, the type of quantity varies. One then
standard deviation. has to reduce the results to the same type of quantity. Eq. (27)
Part. Charact. l(1984) 89-95 93

describes the transfer of a q,(x) distribution into a q r ( x ) The second function is the Rosin-Rarnmler-Sperling-Bennett
distribution, with r and e being either 0, 1, 2 or 3: (RRSB) function [lo-12, 141:

1 - Q3(x)= exp [ - (x/x')n] (33)

With r = 3 and e = 0 one obtains Eq. (28) which describes the x ' represents the position parameter when the function
calculation of a volume distribution from a number density 1 - Q3(x')= e ' = 0.368. The spread parameter, n, rep-
distribution. resents the slope of the straight line obtained on special graph
paper when plotting lglg (1 /R)on the ordinate and lg x o n the
abscissa. The straight line is represented by:

Graphically one alters the height of the ordinates at certain x lglg(l/R) = n lgx - n lgx' + lglge . (34)
values, that is one alters the overall shape of the distribution
curve. The density distribution curves are shifted to larger sizes, The third analytical function is the normal distribution [13]
when moving from number to volume density distributions. The given here in its general form:
shift is the bigger, the wider the spread of the distribution is.

3.5 Substitution of the Abscissa

with z being a dimensionless variable describingparticle size.
In many cases, for instance in a graphical representation of the The cumulative distribution is obtained from:
data, not only the ordinate but also the abscissa must be changed.
7
In order not to overrun the condition of normalization the
ordinates have to be altered again. The basic equation is easily Q:(z) = j qr*(5) d < . (34)
-m

found if one takes into account that an area beneath the density
distribution curve in a certain size range represents the relative Normal distribution differ according to the definition of z:
amount of particles present and this area should remain constant The Gaussian normal distribution with linear abscissa:
irrespective of changes of the abscissa. If one substitutes
<
x = f(<)or = p(x) one arrives at:
x50,r
equals the median of the Q , ( x ) distribution, s represents
the standard deviation.
or Due to the fact that Q: (1)is tabulated for certain zvalues the
standard deviation s can be calculated from:

If for instance < = lnx, then one obtaines due to d</dx = x - I :

The log-normal distribution:
The cumulative distributions on the other hand remain un-
altered.
or according to Herdan 1131:

z = lg (x/xso,r)/lg sg = In (x/xso,,)/ln sg . (41)

By comparison of Eqs. (40) and (41) one obtains:

4 Distribution Functions

Several analytical particle size distributions have been found to s = Ins, or: sg = es . (42)
approximately describe empirically determined particle size
The standard deviation can be calculated from:
distributions. Special graph papers have been developed where
cumulative distributions following these functions form a
straight line. Three of the best known analytical distribution
functions are:
The complete k-th moment of a q,(x) distribution for a log-
a) The power function normal distribution can be calculated by introducing Eqs. (35)
and (40) or (41) into Eq. (9). One arrives at:

= x&, exp(k2 ln2sg/2) .

h'k,r (44)
also known as the Gaudin-Schumann distribution [8, 91 with
the position parameterxmaxand the spread parameter m. The Average particle sizes, as defined by Eq. (lS), can be calculated
latter parameter indicates the slope of the straight line which from:
one obtains when the above function is plotted on a log Q3 -
logx paper.
94 Part. Charact. l(1984) 89-95

5 The Calculation of Specific Surface Area Because of Eq. (44) the total volume related surface area of a log-
normal distribution can be calculated from
5.1 Definition
S,X~,=
, ~ 6fexp(ln2sg/2) (56)
Specific surface area may be defined in two ways:
or
The ratio of surface area, S , to volume, V, equals the volume
related surface area S , : S,x,,,, = 6 f exp(-2.51n2sg) . (57)

S" = S / V . (46) 5.2 Calculation of Specific Surface Area

Using Special Graph Papers
The ratio of surface area, S, to mass, rn, equals the mass
related surface area: If it is possible to represent a size distribution on a special graph
paper in size intervals by v successive straight lines, then the total
S, = S/m . (47) surface area can be calculated from Eq. (58):
S, and S, differ by the solids density p,:

S" = s, ' ps . (48)

The volume related surface area S , of particles of a size interval

A x = (x2 - xl) can be calculated from a number distribution
40 (4. Here x,, is the smallest size of the first straight line and x,, the
x2 x2 biggest size of the v-th interval. The relevant moments can be
sv (xi 9 Xz) = S ks (X)2qo(x)dx/S kv (x)x340 (x)d x . (49) calculated if assumptions with regard to the analytical function
X1 Xl representing the size distributions in the different size classes are
ks (x) = surface area shape factor, made.
k, (x) = volume shape factor.

In general, one assumes k, and k, to be independent of size and 6 Nomenclature

introduces a new shape factor, f , according to Heywood [7]
which takes into account deviations in shape from spherical C coefficient of variation
particles. If one introduces E expected value
f Heywood shape factor
k power of x in moments, or subscript indicating the same
kS surface area shape factor
one arrives at: kv volume shape factor
m mass
M moment
4 density distribution
S, (xl,x2) can therefore be calculated from the ratio of two Q cumulative distribution
incomplete moments of the number distribution. r subscript indicating the type of quantity chosen
Multiplying Eq. (51) withM3,,/M3,, and taking into account the S standard deviation
determination of a cumulative distribution using the concept of 54 geometrical standard deviation
moments one finally arrives at: S surface area
V volume
Q3@) = M3,0(Xrnin,x)/M3,0 X particle size, equivalent particle diameter
XP
equivalent projected area diameter
and one obtains: XS equivalent surface area diameter
XV equivalent volume diameter
xw equivalent settling rate diameter
2 dimensionless variable
r variable
A similar equation is obtained for a volume distribution: W sphericity
PS solids density
V coordinate

With x, = x,, and x1 = x,,,, Q3(xrn,,) = 1 and Q3(xrn,") = 0 7 References

the total surface area of all particles can be calculated from the
following expression : [ I ] K . Leschonski: Present and Future State of Particle Size Analysis,
Proceedings of the Conference on Particle Size Analysis 1970,
(55) Society for Analytical Chemistry, London 1972, pp. 409 - 419.
Part. Charact. Z(1984) 89-95 95

[2] H . Wadell: Volume, Shape and Roundness of Rock Particles. J . 191 R. Schumann: Principles of Comminution I - Size Distribution
Geology 40 (1932) 443 - 451. and Surface Calculations. American Institute of Mining and
[3] A. Cuuchy: Note sur divers theoremcs relatifs a la rectification des Metallurgical Engineers, Techn. Publ. No. 1189 (1940), pp. 1 - 11.
courbes. C. R. Hehd. Seances Acad. Sci. I3 (1841) 1060-1065. [I01 P. Rosin, E. Rammler, U. Sperliny: Korngrol3enprohlcmc des
[4]H . Rumpf, K . F. Ebert: Darstellung von KorngroRcnvcrteilungen Kohlenstaubs. Bericht C 52 des Reichskohlenrates, Berlin (1933)
und Berechnung der spczifischen Oberflache. Chcm. Ing. Tech. 36 1-33.
(1964) 523 - 537. [131 E. Rummier: Gesetzmaaigkeiten in der Kornvertcilung zerkleincr-
[5] K . Leschonski, W. Alex, B. Koglin: Darstellung und Auswertung ter Stoffc. Z. Vcr. Dtsch. Ing., Beih. Verfahrenstcch. 5 (1937)
von TeilchengroRenanalyscn. Chem. Ing. Tech. 46 (1974) 23 -26. 101 - 168.
[6] K. I.eschonski, W. Alex, B. Koglin: Darstellung und Auswcrtung [121 I . G . Bennett: Broken Coal. J. Inst. Fuel 10 (1936) 23 - 39.
von TciIchengroBenanalysen. Chem. Ing. Tech. 46 (1974) 1131 G. Herdan: Small Particle Statistics. Butterworth, London 1960.
101 - 106. [I 41 M. Langemann: Zur Berechnung der spezifischcn Oberflachen von
[7] H . Heywood: Numerical Definitions of Particle Size and Shape, Dispcrsioncn. Chem. Ing. Tech. 27 (1955) 27 - 32.
SOC.Chem. Ind. Symp. Aggregates 56(1937) 149- 154.
[8] A . M. Gaudin: An Investigation of Crushing Phenomena. Trans.
Amcr. Inst. Mining, Met. Petrol. Engr. 73 (1926) 253 - 316.