Impactors and Particle Size Distribution (2)

National Institute for Occupational Safety and Health

Division of Respiratory Disease Studies
Field Studies Branch

Ju-Hyeong Park, Sc.D., M.P.H., C.I.H.

Particle size statistics

1
Aerosol measurement and size distribution

Concentration
Increasing

Complexity

Diameter

1. Concentration 2. Size distribution

Particle mass, surface area Concentation versus particle size

or number per unit
volume

Aerosol particle sizing

Assign to size bin

Sample 0.4 µm < d < 0.7 µm

Size particles Bins

0.1 0.2 0.4 0.7 0.9 1.0 1.2

Diameter / µm

Particles are assigned to bins according to particle diameter

2
 Conversion of a discrete particle size distribution
to a continuous distribution
0.2
Discrete Distribution

Number
concentation is
0.15 proportional to
the height of
Number concentration

each bar

0.1
Number

Particle size bins

0.05

0
0 1 2 3 4 5 6

Particle Diameter /

0.4

Continuous Distribution
0.35
Number concentation
0.3 is proportional to the
Number Concentation / Bin

area of each bar

0.25

0.2

0.15

0.1

0.05

0
0 1 2 3 4 5 6

Particle Diameter /

3
 0.

Continuous Distribution
0.3
The smooth continuous
distribution is obtained
0.
by joining the bin
Number Concentation / Bin

mid-points
0.2
d)
dn/d(

0.

0.1

0.

0.0

0
0 1 2 3 4 5 6

Particle Diameter /

Continuous graph
™ Y-axis
‰ Differential particle concentration (or number)
¾ Particle number normalized by range of particle diameter
of the interval (or bin)
¾ dn/dd
™ X-axis
‰ Particle diameter range of the interval
¾ dd
™ Integrated area under the curve=total # of particle

∫ dn
dd
dd = n

4
Properties of particle size distribution
™Asymmetrical (or skewed) distribution
‰ Long tail to the right
¾ Large number (fraction) of small particles
¾ Small number (fraction) of large particles
‰ Large range of particle size
¾ Several orders of magnitude in particle diameter
‰ No negative particle size
™ Mode < Median < Mean
™ Geometric mean (dg or GM)

Log dg=(Σnilogdi)/N Æ dg=exp{(Σnilogdi)/N}

Arithmetic mean and GM

™Arithmetic mean
‰ da=(Σnidi)/N = (summation of all areas of the
bars) / (total number of particles)

™Geometric mean
‰Log dg=(Σnilogdi)/N Æ dg=exp{(Σnilogdi)/N}

5
 0.

Continuous Distribution
0.3
The smooth continuous
distribution is obtained
0.
by joining the bin
Number Concentation / Bin

mid-points
0.2
d)
dn/d(

0.

0.1

0.

0.0

0
0 1 2 3 4 5 6

Particle Diameter /

Asymmetric (skewed) distribution

1
Modal Lognormalormal Size Distribution
ModeDiameter
Normalized distribution (n=1)
0.8 Area inder curve = 1
Arithmetic mean

Note that when plotted as dn/d(d),

0.6
the modal and count median diameters
dn/d( d)

Geometric
of a lognormalmean (Median)
distribution are different

0.4
Distribution skewed to
smaller particle diameters

0.2

0
0 1 2 3 4 5
CMD
Diameter / µm

6
 Normal distribution
1
Normalized distribution (n=1) Normal Size Distribution
Area inder curve = 1
0.8 dp
Mean = Mode = Median
68% of all particles
0.6
are between ± σ
dn/d( d)

0.4

Distribution characterized
by n, d and
p σ
0.2

0
0 1 -σ 2 +σ 3 4 5

Diameter / µm

Normal distribution

(d p −d p )
2

n −
df = 2σ 2
e dd p
σ 2π
™⎯dp: arithmetic mean particle diameter
™ σ: standard deviation
™ ddp: particle diameter interval
™ df: frequency of occurrence of particles of
diameter dp

7
Log transformation of continuous graph
™ Y-axis
‰ Differential particle concentration (or number)
¾ Particle number normalized by range of log-transformed
particle diameter of the interval (or bin)
¾ dn/dlog(d)
™ X-axis
‰ Range of log-transformed particle diameter of the
interval
¾ dlog(d)
™ Integrated area under the curve=total # of particle
‰ ∫{dn/dlog(d)}dlog(d)

Lognormal distribution with arithmetic scale

1
Modal Diameter Lognormalormal Size Distribution
Normalized distribution (n=1)
0.8 Area inder curve = 1

Count median diameter (CMD)=GM

Note that when plotted as dn/d(d),
0.6
the modal and count median diameters
dn/d( d)

of a lognormal distribution are different

0.4
Distribution skewed to
smaller particle diameters

0.2

0
0 1 2 3 4 5
CMD
Diameter / µm

8
 Mathematical function of lognormal
distribution
−
(Log (d p )− Log (CMD ))2
dLog(d p )
n 2Log (σ g )
2

df = e
2π Log(σ g )

(d p −d p )
2

n −
df = 2σ 2
e dd p
σ 2π

Two ways of log-transformed graph

™Transform the original particle size data using

logarithm, and then plot them on normal
arithmetic scale of the graph
‰To calculate all statistics mathematically
‰Exponentiate log-transformed statistics
™Transform x-axis scale of the graph, and then
plot the original particle size data on it
‰Do not transform the data
‰Only change the scale of the graph

9
 Log normal distribution (first approach)
1
Normalized distribution (n=1) Normal Size Distribution
Area inder curve = 1
0.8 dp
Mean = Mode = Median
68% of all particles
0.6
are between ± σ
dn/d( d)

0.4

Distribution characterized
by n, d and
p σ
0.2

0
0 1 -σ 2 +σ 3 4 5

Diameter / µm

Lognormal distribution with log scale

1.5
Normalized distribution (n=1) Lognormal Size Distribution
Area inder curve = 1

Count Median Diameter 50% of all particles

(CMD) are less than CMD
1
dn/dLog(d)

68% of all particles

are between CMD / σ
g

and CMD × σ
g

0.5 84% of all particles

are less than CMD*σg
16% of all particles
are less than CMD/σg Distribution characterized
by n, CMD and σ
g

0
0.01 0.1 CMD/σ 1 CMD × σ 10 100
g g

Diameter / µm

10
 Cumulative size distribution
1
Normalized distribution (n=1) Lognormal
Area inder curve = 1 Size Distribution

0.8 84%
Fraction of particles smaller than d

0.6

CMD
50%
0.4

0.2

16%

0
0.1 CMD/ σ 1 CMD × σ 10
g g

Diameter / µm

Probability scale of lognormal distribution

™Percent of particles less than a given particle
diameter
‰ <CMD/(2σg)- 5%
‰ <CMD/σg- 16 %
‰ <CMD (median)- 50%
‰ <CMD*σg- 84%
‰ <CMD*(2σg)- 95%
™Any pattern??
‰ Symmetry of probability

11
 Lognormal distribution with log scale
1.5
Normalized distribution (n=1) Lognormal Size Distribution
Area inder curve = 1

Count Median Diameter 50% of all particles

(CMD) are less than CMD
1
dn/dLog(d)

68% of all particles

are between CMD / σ
g

and CMD × σ
g

0.5 84% of all particles

are less than CMD*σg
16% of all particles
are less than CMD/σg Distribution characterized
by n, CMD and σ
g

0
0.01 0.1 CMD/σ 1 CMD × σ 10 100
g g

Diameter / µm

Cumulative plot to log-probability plot

1
Normalized distribution (n=1) Lognormal
Area inder curve = 1 Size Distribution

0.8 84%
Fraction of particles smaller than d

0.6

CMD
50%
0.4
Switch and change to probability scale

0.2

16%

0
0.1 CMD/ σ 1 CMD × σ 10
g g

Diameter / µm

12
 Log-probability paper
100
90
80
70
60
50
40
Log scale (particle size: µm)

30

20

10
9
8
7
6
5
4

1
1 2 5 10 20 30 40 50 60 70 80 90 95 98
Probability scale (probit)

Log-probability plot of lognormal distribution

10 16% 50% 84%

1
CMD × σ
g

2
1/2 = σg = 2/3
1 CMD
Diameter / µm

CMD/σ
g
3
0.1

Lognormal
Size Distribution
0.01
.001 .01 .1 1 5 10 20 30 50 70 80 90 95 99 99.9 99.99 99.999

% Distribution less than diameter d (probit)

13
 Count median diameter and σg
™CMD (count median diameter)
‰50% of all particles are less than CMD
™Geometric standard deviation
‰CMD*σg/CMD or
‰CMD/(CMD/σg)

CMD, SMD, and MMD

1.5
Normalized distributions
Number weighted ( X=n )
Surface-area weighted (X=s)
Mass weighted (X=m)

1
dX/dLog(d)

0.5

0
0.1 1 10 100
CMD SMD MMD
Diameter / µm

14
 Log-probability plots for count, mass, and surface
area lognormal distribution
100 16% 50% 84%

Number weighted
Surface-area weighted
Mass weighted
10
Diameter / µm

Plots are parallel - σ

g
(given by gradient) is the
same for each weighting
0.1

0.01
.001 .01 .1 1 5 10 20 30 50 70 80 90 95 99 99.9 99.99 99.999

% Distribution less than diameter d (probit)

Cascade impactor data reduction

Stage # Size Initial Final Net Mass Cumm.
Range
d50 Mass Mass Mass Fraction Mass
(µm) (mg) (mg) (mg) (%) Fraction
(µm)
(%)

1 >9.0 9.0 850.5 850.6 0.1 0.6 100.0

2 4.0-9.0 4.0 842.3 844.1 1.8 11.0 99.4

3 2.2-4.0 2.2 855.8 861.0 5.2 31.7 88.4

4 1.2-2.2 1.2 847.4 853.6 6.2 37.8 56.7

5 0.7-1.2 0.70 852.6 855.1 2.5 15.2 18.9

Back 0-0.7 0 78.7 79.3 0.6 3.7 3.7
filter
Total 16.4 100.0

15
 An example using count data

Data

William Hinds, Aerosol Technology, 2nd (1999)

16
 Plotting upper size range vs. fraction/µm
0.10

0.08
Fraction/µm

0.06

0.04

0.02

0.00
0 10 20 30 40 50 60
Particle size (µm)

Log scale of particle diameter (lognormal distribution)

0.10

0.08
Fraction/µm

0.06

0.04

0.02

0.00
1 2 4 6 8 10 20 40 60 80 100

Particle size (µm)

17
Cumulative lognormal plot (cum vs. size)
100

80
Cumulative fraction (%)

60

40

20

0
1 10 100
Particle size (µm)

Log-probability plot of count data

100
90
80
70
60
50
40
Particle size in log scale (µm)

30

20

10
9
8
7
6
5
4
3

16% 50% 84%

1
1 2 5 10 20 30 40 50 60 70 80 90 95 98
Cumulative fraction in probability scale/µm

18
 Take-home practice
Stage cut-point Mass concentration Cumulative mass % Cumulative mass % Cumulative mass
(diameter d) concentration greater than concentration greater than concentration less than
diameter d diameter diameter
(µm) (mg/m3) (mg/m3)
10 0.4209
5.62 1.3330
3.16 3.3770
1.78 6.4771
1 9.5248
0.5 12.7579
0.285 8.1861
0.158 5.2389
0.05 2.9853

Sum (mg/m3)

1. Fill out the blank columns.

2. Find out MMD (Mass Median Diameter) and Geometric standard deviation (GSD).

19