Vous êtes sur la page 1sur 7

American Industrial Hygiene Association Journal

ISSN: 0002-8894 (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/aiha20

Uncertainty in particle counting and sizing



To cite this article: DAVID LEITH & MELVIN W. FIRST (1976) Uncertainty in particle counting
and sizing procedures, American Industrial Hygiene Association Journal, 37:2, 103-108, DOI:

To link to this article: https://doi.org/10.1080/0002889768507419

Published online: 04 Jun 2010.

Submit your article to this journal

Article views: 11

Citing articles: 8 View citing articles

Full Terms & Conditions of access and use can be found at

A method is described for calculating confidence intervals for
particle or fiber concentration, and for dust corllector
penetration. The span of the interval depends upon the value of
fiber concentration or collector penetration
reported and upon the number of particles or fibres counted.

Uncertainty in particle counting and

sizing procedures
H?r'ivard School of Public Health, Department of Environmelntal
~ e $ l t hSciences, Boston, Massachusetts 02115

Introduction a representative number of fields under the

mic~oscope.All particles seen are segregated
The concentration of particles in a gas can be
into convenient, continuous size categories.
determined by passing a known gas volume
Subsequent traverses note entries for only
through1 a filter and counting particles on
those size ranges in which particles are present
representative filter polrtions. Particle concen-
in relatively small numbers. The average num-
trations are valuable to determine compliance
ber of particles in each size range per traverse,
with legal standards, as for asbestos fibers in
workroom air, or to determine the particle size x, is then calculated as shown in Table I.
collection efficiency of a dust collection device Stratified counting is a way to emphasize those
by makilng counts of simultaneous upstream particles whose concentrations are most diffi-
and downstream samples. For both applica- cull. to assess with statistical reliability because
tions, it is impolrtant to estimate the. count of their relative rarity.
reliability. Although enough particles must be
counted to establish the validity of the result
within acceptable limits, it is wasteful to insist
upon excessive counting to obtain needlessly A -inverse of the fraction of total filter
high reliability. area examined per traverse
Particles for microscopic counting are G -inverse ofthe total volume of gas
~onvenientlycollected on membrane filter~l.~ passed through a filter
3r electron miscroscope g r i d s l ~ ~The
, ~ . "stratifi- m -number of an equal area, counting
:ation" particle counting illustrated outward from filter center
n Table I is often used to reduce counting M -total number of equal areas into
ime. An initial traverse is made by examining which a filter is divided

David Leith, Assistant Profes- Melvin W. First, Professor

sor at the Haward Uni- of Environmental Health
versity School of Public' Engineering a t the Harvard
Health, holds Bachelors and University school of Public
Masters degrees in chem- Health, is a diplomate of
ical engineering from the the American Academy of
University of Cincinnati, Environmental Engineers
and a Doctorate in Environ- and a Director of th,e Amer-
mental Health Sciences ican Board of Industrial
from Harvard. His interests Hygiene. Dr. First was a Di-
lie in industrial hygi,ene rector of Al HA from 2964-
and air pollution control. 1967,

American Industrial Hygiene Associatio~i Journal 103


Data Tabulation for Particle Sizing by Stratified Counting5
< 0.45- 0.63- 0.89- 1.30- 1.80- 2.50- 3.50- 5.00- 7.10- TOTAL PER
0.44 0.62 0.88 1.29 11.79 2.49 3.49 4.99 7.09 10.00 TRAVERSE-
1 57 87 54 36 2:L 24 6 12 0 3 300
2 1 2 6 3 0 21
3 3 0 3
4 2 1 3
5 2 1 3
6 1 2 3
Total, N 57 87 54 36 211 24 18 18 11 7 333
Mean Count
per Traverse, 57 87 54 36 2:L 24 9 9 1.8 1.2 300
Std. Deviation
of Mean, u; 7.55 9.33 7.35 6.00 4.58 4.90 2.12 2.12 0.55 0.45 I
95% Confidence 72 105 68 48 30 34 13 13 2.9 2.1
Interval to to to to ts0 to to to to to ?f
42 69 40 24
12 15 4.5 4.8 0.7 0.3
- 'i

-number of traverses performed in a filter directly downstream of the gas inlet.
stratified counting procedure :Dennis and co-workers8 found that such radial
concentration gradients did not occur for
-total number of particles of a certain
size counted through all traverses particles smaller than 20 micrometers diameter
-N for filter downstream of a particle ?whenfilter holders were used which had a ten
collector degree included angle between gas inlet and
filter surface. However, filter holders of this
-N for filter upstream of a particle: design are not always practical because c~ftheir
llarge size.
-number of particles per volume of
To avoid bias when using a conventional
holder with small inlet, fields are usuallly
-P for gas downstream of a particle
selected at random from a pie shaped piece of
the circular filter paper. About twenty fields
-P for gas upstream of a particle imust be examined2to complete one unbi'ased
collector estimate of the particle size distributialn.
- dust collector penetration (1 - ef-- 'Totally random field selection gives an un-
ficiency), Pdown/Pup biased estimate of particle concentration when
-distance from center of filter to a sufficiently large number of fields is ex-
point where microscope is to be amined. However, the same result can bc:
focused for equal area m obtained with fewer fields by an ordered ap-
-radius of filter proach. Because the pressure drop across a
-number of particles in a certain size imembrane filter is sufficient to assure uniform
range present in one traverse gas velocity normal to the filter surface, the
volume of gas flowing through each equal,
-mean number of particles in a cer- concentric area on the filter will be the same,
tain size range found per traverse unless the central areas become plugged be-
- standard deviation of x cause of excessive particle deposition there.
- standard deviation of x The overall dust concentration for the gas
-standard deviation of Pt sampled will be the average of the concentra-
tions found in the gas passing through each
of the equal areas. To utilize an orde'red
approach, the microscope should be foc~rsed
;st the center of each equal area ring preslent
Microscopic field selection jln the sector, and a single field examined.
When a filter holder has a small diameter in- ,After one field in each equal area has been
let relative to the filter diameter, the largest inspected, all data can be combined to make
particles may concentrate on that part of the ;an entry for one traverse, as shown in Ta.ble I.

February, 1976
Additional traverses are made along different comprising in traverses. A Poisson distribution
radii of the filter sector. describes the variation in these x values I2-l4.
This equal area traverse method for The standard deviation of these values, cr,
therefore equals the square root of the mean
locating counting fields is analogous to the
method used for positioning a pilot-static tube
number of- particles in that size range per
when determining the average gas velocity in traverse, x, i.e.
a round duct. Average gas velocity could be
determined by measuring the velocity at many
random points within the duct cross section The standard deviation of the mean, ux , is the
and averaging the results found. However, the standard deviation, u,divided by the square
number of measurements needed to reach the root of n, the number of traverses.
same statistical reliability using this approach
is greater than for the equal area method, and
the random approach is not used. By analogy,
it is as logical to use an ordered approach for
locating counting fields on a membrane filter From Equations 2 and 3,
as it is to use it folr locating pitot-static meas-
uring points in a duct.
The distance, r,, from the center of 'a
filter of :radius rf to the midpoint of each of where N is the total sum of all particles in a
M equal areas can be found from certain size range counted through all traverses.
Equation 4 is an application to stratified
counting of the expression given by Chapman
and RUM for the relative error associated with
repeated counts of particles in liquid suspen-
Here, m is the number assigned to an equal
sion.15 Appropriate values for the standard de-
area, starting from the filter center and count-
ing outwards. Alternatively, values can be viation of the mean, cr;; , are given for the data
found in a reference giving the relative distance: shown in Table I. The number concentration
. from the, wall at which a pitot probe should be of particles in a certain size range, P, is
proportional to x, the mean number of
blaced in order to have an unbiased estimate particles of this size per traverse, to the inverse,
slf average gas velocity in a circular duct.gJO A, olf the fraction of total filter area examined
per traverse, and to the inverse, G, of the total
95% confidence intervals gas volume passed through the filter.
After stratified counting procedures have been
employed and me,an concentrations for particles
The standard deviation of a product can be
in each size range calculated, it is important to
determine colllfidence intervals for these values. found f r omlG
Systematic sources of error such as anisokinetic
sampling, inaccuracies of flow measuring ~de-
vices, iinproper microscope calibration and the
like can be minimized through careful experi-
mental technique.11 Assignment of particles to1
improtper size categories is not a significant
problem when trained observers use a standard
Porton graticule for determining particle With careful technique, the standard error
diamete~r.~~ assoiciated with A and G can be made small
com~paredto that for x. The substitution of
A source of non-systematic sampling error Equation 4 into Equation 6 gives the relative
that cannot be eliminated by control of experi- standard error associated with a measurement
mental procedures is associated with random of concentration, P.
variations in the number, x, of particles in a
certain size range which are present in each
traverse of a stratified counting procedure

American Industrial Hygiene Associatiorl Journal

until the standard is no longer contained
within the confidence interval, as one can then
state with 95 % certainty whether or not the
standard has been met. When the fiber con-
centration is close to the standard, it will be
necessary to count a larger number of particles
to establish with 95 % confidence whethm or
not the standard is met, than when the con-
centration is clearly well above or below the
5 10 100
I l l U
NUMBER OF PARTICLES COUNTED, N For example, after counting 50 asbestos
Figure 1-Number of particles counted versu's fib,ers on a membrane filter, one might find
relative error of particle c:oncentration. that the mean concentration in the air passed
through the filter was 1.5 fibers/cc. Equation
8 and Figure 1 show that one can state with
95 % confidence ,that the true fiber concentra-
tion was 1.5 t1.96 X 1.5 X (1/50)%, or
from 1.08 to 1.92 fibers/cc. The upper bound
of the confidence interval for this example is
below the 1976 OSHA standard of 2.0 fibers/
cc. Counting additional fibers would make the
95 % confidence interval smaller, but would
be unnecessary if 95 % confidence that the
standard is met is sufficient.
Equation 4 can also be used to determine
confidence intervals about an experimentally
determined value for d~zstcollector penetration
or efficiency. Penetration for particles of a
certain size is the ratio of particle conceritration
in the downstream gas to the analogous con- ,
NUMBER OF PURTICLES ON DOWNSTREAM centration in the upstream gas.
Figure 2-Number of particles of one size counted Pt = -
o n downstream filter versus number of particles of
that size counted o n upstream filter, with relative
error of penetration as parameter. The standard deviation of this quotient i;slG

A 95 % confidence interval about P will ex-

tend plus and minus 1.96 times the standard
deviation for P.
P + 1.96 P -\/I/N (8)
Figure 1 is a plot of the relative standard errlor The substitution of Equation 7 into Equ,ation
of particle concentration, m / P , against the 10 yields
number of particles counted, N.

Therefore, a 95 % confidence interval about
When asbestos fibers are counted to determine Pt will extend plus and minus 1.96 times the
compliance with an applicable standard, it is standard deviation of Pt, as shown in Equation
prudent to determine periodically the mean 12.
fiber concentration and associated confidence
interval. The count should be continued only

February, 1976
Equation 11 indicates that the relative error determined by counting. The techniques out-
in penetration for particles of a certain size lined above can be used for data from auto-
is only a function- of the total number of inatic counting devices such as optical instru-
particles clf that size counted upstream, N,,,, ments working on light scattering principles,
and downstream, Ndow,,,of the collector. This as well as for data from other automatic
relationship is plotted in Figure 2. counting devices.

Equation 11 and Figure 2 show, for ex-

ample, that to be 95 % confident that pentra-
tion of particles in a certain size range is
between 401 and 60 % (1.96 crpt = 10% with
mean penetration of 50%) it will be necessary ,4n equal area traverse method is described for
to count 200 particles in this size range on the selecting microscopic fields when counting
upstream filter and 200 particles in the same particles or fibers on a membrane filter. This
size range on the downstream filter. Alter- method is anadogous to that used to position a
natively, 500 particles counted upstream and pitot-static tube in a duct when determining
125 downstream would give the same result. irverage gas velocity. The equal area traverse
However, the fewest total particles that must approach is ;m aid in avoiding inadvertent
be counteld to achieve a given relative error counting bias due to nonrandom field selection.
in penetration will always be found when the Although confidence intervals are im-
number of particles counted on the upstream portant to establish the significance of particle
and downstream filters is equal. This can be concentration or collector efficiency data, they
proven by differentiating Equation 11 with inre seldom calculated or reported. When
respect to Nu,, setting the derivative equal to particle size data are generated by a stratified
zero, and proceeding in the usual manner. counting procedure, the method described can
When the particle deposit is less dense on the loe used to establish with 95% confidence
downstream filter, it becomes necessary to whether or not mean concentrations are below
make more traverses for that filter in order to or above a fixed value.
' count about the same number of particles as Confidence intervals about values of pene-
are counted on the upstream. Or, a larger field tration or efficiency can be calculated in a
size could be used for the less dense filter. similar manner. The stratified counting ap-
When the particles observed on a filter are proach allows the microscopist's efforts to be
separated into many size categories to de- concentrated onto those particle size ranges
termine collector particle size efficiency, it is .where small confidence intervals for penetra-
necessary to observe a large total number of tion are the most difficult to achieve. Charts
particles to generate adequate confidence in lhave been prepared that make it possible to
the penetration or efficiency results for each determine easily the number of particles or
size range considered. fibers which must be counted to assure desired
confidence intervals.
When it is desirable to maintain a 95 %
confidence interval of constant size, i.e. a
constant d u e of mt for all size r~uzges,
Equation 12 shows that fewer particles need
be-counted in each size range as penetration 1. SILVERMAN, L., C. E. BILLINGS and M. w. FIRST:
decreases. The stratified counting technique Particle Size Analysis in Industrial Hygiene.
can be used with good effect to concentrate Academic Press, New York (1971 ) .
the microscopist's efforts on the size ranges 2. EDWARDS, G . H. and J. R. LYNCH: The Method
where penetration is high. Therefore, it is Used by the U. S. Public Health Service for
worthwhilie to identify these size ranges as soon Enumeration of Asbestos Dust on Membrane
Filters. Ann. Occup. Hyg. 11:l (1968).
as possible by making a preliminary estimate
3. MORROW, P. E. and T. T. MERCER: A Point to
of penetration based on an initial traverse of Plane Electrostatic Precipitator for Particle Size
the upstream and downstream filters. Sampling. Anz. Ind. Hyg. Assoc. J . 25.3 (1964).
4. BILLINGS, C. E. and L. SILVERMAN: Aerosol Sam-
This method of calculating confidence pling for Electron Microscopy. J. Air Pollut.
intervals applies whenever concentrations are Control ASSOC.12:5186 (1962).

American Industrial Hygiene Association Journal 107

5. SICHEL,I.1. S.: On the Size Distribution of Airborne 11. HAWKSLEY, P. G. W., S. BADZ10Cl-Iand J. H.
Mine Dust. J. S. Afr. Inst. Min. Met. 58:171 BLACKETT: Measurement o f Solids in Flue Gases.
(1957). British Coal Utilization Research Assn., Leather-
6. HOEL,P. G . : Introduction to Matlzenzatical head, Surrey, England ( 1961).
Statistics. Wiley, New York (1949). 12. FAIRS,G. L.: XII-Developments in the 'Technique
7. WHITBY, K. T.: Determination of Particle Size of Particle-size Analysis by Microscopical Ex-
Distribution-Apparatus and Teclzniqzm for amination. J. Roy. Microscop. Soc. 71:2O9
Flour Mill Dust. Univ. of Minn. Eng. Expt. (1951).
Station Bull. No. 32 (January, 1950).
13. CORN, M. : Statistical Reliability of Particle Size
8. DENNIS, R., I,. SILVERMAN, C. E. BILLINGS, Distributions Determined by Microscopic Tech-
E. KRISTAL, D. M. ANDERSON AND P. DRINKER: Air niques. Anz. Itzd. Hyg. Assoc. J. 26:s (1'965).
Cleaning Studies Progress Report for July 1,
1955 to June 30, 1956. A. E. C. Contract No. 14. HERDAN,G . : Small Particle statistic.^. 2nd ed.
AT(30-1) 841 (March 16, 1959). Academic Press, New York (1960).
9. American Conference of Governmental Industrial 15. CHAPMAN, H. M. and R. c. RUHF: Dust Counting
Hygenists : Industrial Ventilation. 13th ed. P. 0 . Reliability. Anz. Ind. Hyg. Assoc. Quart. 16:201
Box 453, Lansing, Michigan ( 1974). (1955).
10. HEMEON, W. c.: Plant and Process Ventilation. 16. IN, H. and R. R. COLTON: Statistical Methods.
2nd ed. Industrial Press, New York (1963). 5th ed. Barnes and Noble, New York ( 1970).
Accepted October 15, 9975

submission The American Jndustrial Hygiene Detailed guidelines covering

of new Association JOURNAL, as the manuscript preparation to suit the
manuscriptS official publication of the Associa- new format requirements of the
tion. provides a medium f~ortimelv JOURNAL will be published
of scientific articles - frequently. Reprints of guidelines
and technical reports covering the also are available to authors. These
broad fields of industrial hygiene current guidelines should be re-
and occupational health. These viewed carefully and observed,
relate to the detection, evaluation before the final draft of a manu-
and control of problems con- script is prepared. Requests for
cerned with occupatio~ial, en- reprints of guidelines for authors
vironmental a n d radidlogical and submission of manuscripts
health, as well as air pollution, should be made to the editor,
human and animal toxicology, American Industrial Hygiene Asso-
product safety and related sub- ciation JOURNAL, 66 S. Miller,
jects. Rd., Akron, OH 443 13.
Worthy contributions to the Galley proofs will be provided
literature are welcomed. 14 anu- for final reading, but not for re-
scripts will be acknowledged and writing or revision, just prior to
reviewed for acceptance promptly. publication. Reprint orders are
When approved, they will be made available to authors with
scheduled for publication at the galley proofs.
earliest possible date.
Inquiry regarding current
Only manuscripts subimitted status of a manuscript previously
exclusively to the AIHA J OUR- submitted for consideration should
NAL, not published elsewhere and be made in writing. Be sure to
not being considered by another mention the manuscript's JOUR-
publisher will be reviewed. Manu- NAL number, if known, plus the
scripts not meeting these stipula- first author's name and the manu-
tions should not be offered. script's full title.

February, 1976