EPA-600/2-77-050
August 1977
Environmental Protection Technology Series
STUDY ON THE FEASIBILITY AND DESIGN OF
AUTOMATIC PARTICULATE SIZE DISTRIBUTION
ANALYZER FOR SOURCE EMISSIONS
Municipal Environmental Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, Ohio 45268
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5. Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7. Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the ENVIRONMENTAL PROTECTION TECH-
NOLOGY series. This series describes research performed to develop and dem-
onstrate instrumentation, equipment, and methodology to repair or prevent en-
vironmental degradation from point and non-point sources of pollution. This work
provides the new or improved technology required for the control and treatment
of pollution sources to meet environmental quality standards.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/2-77-050
August 1977
STUDY ON THE FEASIBILITY AND DESIGN OF
AUTOMATIC PARTICULATE SIZE DISTRIBUTION ANALYZER FOR
SOURCE EMISSIONS
by
Pedro Lilienfeld
Daniel P. Anderson
Douglas W. Cooper
GCA/Corporation
GCA/Technology Division
Bedford, Massachusetts 01730
Contract No. 68-03-2154
Project Officer
John 0. Burckle
Solid and Hazardous Waste Research Division
Municipal Environmental Research Laboratory
Cincinnati, Ohio 45268
This study was sponsored by
Industrial Environmental Research Laboratory
Research Triangle Park, North Carolina 27711
MUNICIPAL ENVIRONMENTAL RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
CINCINNATI, OHIO 45268
-------
DISCLAIMER
This report has been reviewed by the Municipal Environmental Research
Laboratory, U.S. Environmental Protection Agency, and approved for publication.
Approval does not signify that the contents necessarily reflect the views and
policies of the U.S. Environmental Protection Agency, nor does mention of
trade names or commercial products constitute endorsement or recommendation
for use.
11
-------
FOREWORD
The Environmental Protection Agency was created because of increasing
public and government concern about the dangers of pollution to the health
and welfare of the American people. Noxious air, foul water, and spoiled
land are tragic testimony to the deterioration of our natural environment.
The complexity of that environment and the interplay between its components
require a concentrated and integrated attack on the problem.
Research and development is that necessary first step in problem
solution and it involves defining the problem, measuring its impact, and
searching for solutions. The Municipal Environmental Research Laboratory
develops new and improved technology and systems for the prevention, treat-
ment, and management of wastewater and solid and hazardous waste pollutant
discharges from municipal and community sources, for the preservation and
treatment of public drinking water supplies, and to minimize the adverse
economic, social, health, and aesthetic effects of pollution. This publica-
tion is one of the products of that research; a most vital communications
link between the researcher and the user community.
This study deals with exploratory research aimed at establishing the
feasibility of a continuous monitor for the measurement of the particle size
distribution of source emission aerosols. The approach was based on dividing
the aerosol into several size ranges and detecting the amount of particles
in the size ranges. Successful development of such a monitor will provide
a valuable tool to aid in a better understanding of particulate source
emissions and their control, especially with respect to the control of fine
particulate emissions.
Francis T. Mayo, Director
Municipal Environmental
Research Laboratory
ill
-------
ABSTRACT
The objective of this program was to evolve a method for the automatic
determination of the size distribution of particulates within stack gas ef-
fluent streams. This device was designed to cover the typical mass concentra-
tion range encountered upstream as well as downstream of emission control
systems, and to segregate the particles by means of a cascaded virtual inertial
impaction configuration to be inserted into the effluent stream. Several al-
ternative particle detection techniques compatible with this size segregation
method were investigated in the course of this program and a stage filter pres-
sure drop sensing technique was selected. The prototype device was subjected
to laboratory and stack testing showing very good correlation with an Andersen-
type impactor. The salient advantages of this instrument are: capability for
extended operation (of the order of hours), real-time indication of size dis-
tribution of particulates in the stack environment, relatively low cost, and
simplicity of operation.
iv
-------
CONTENTS
Foreword iii
Abstract iv
Figures vi
Tables ix
Unit Conversion Table x
Acknowledgments xi
1. Introduction 1
2. Summary 3
3. Conclusions 6
4. Recommendations 7
5. Theoretical Aspects 8
6. Particle Sensing and Detection Methods .... 26
7. Virtual Impactor Development 43
8. Particle Sensing Methods 56
9. Virtual Impactor-Pressure Drop Lab and Field Tests 65
References 95
Appendix
Token Flow Corrections to Virtual Impactor Data Collected in
New Bedford 98
-------
FIGURES
Number Page
1 Schematic of a Virtual Impaction Stage 12
2 Penetration, or Response, Function for Ideal and Nonideal
Instrument 17
3 Error Factor Versus Number of Geometrically Increasing Sizing
Intervals for Impactor 19
4 "Figure of Negative Merit" as a Function of Size Interval Ratio
and Number of Stages 20
5 Data, F2*, and Input Distributions, FI, Regained by Correcting
That Data, for a Three-Stage Virtual Impactor With 10 Percent
Secondary Flow 25
6 Optical Sensing Configuration 29
7 A Contact Electric Sensor Technique 33
8 Filter Pressure Drop Technique 33
9 Efficiency of Glass-Fiber Filters as a Function of Particle
Size and Flow Rate 34
10 Measured Dependency of Ap on Time (Three Successive
Experiments) . . 35
11 Comparison of Collection Efficiency E and Pressure Drop AP in
Gravel and Fiber Filters at Darcy Air Flow Velocity vj) = 2.0
cm/sec and Gravel Layer Thickness L = 30 cm. 36
12 Resistance Increase as a Result of Solid Particle Accumulation
in Tests 7 Through 10 37
13 Filter Resistance During Operation 38
14 Pressure Drop Versus Collection Density at Various Filter Face
Velocities (GCA Data) 39
15 Pressure Drop as a Function of Time for Extended Operation at
Nearly Constant Concentration (GCA Data) 40
vi
-------
FIGURES (Continued)
Number Page
16 Expected Variation of Pressure Drop as a Function of Accumulation
and Particle Size 42
17 Cross-Section of Initial Version of Virtual Impactor Breadboard . . 45
18 Various Orifice and Nozzle Designs 46
e
19 Redesigned Collection Nozzle 50
20 Collection Efficiency Versus Particle Size 54
21 Electronic Schematic 57
22 Optical Sensing System 58
23 Aerosol Number Concentration by Gravimetric Versus Optical
Detection for Arizona Road Dust, Aerosol Particles Greater
Than 3.5 ym 60
24 Aerosol Number Concentration by Gravimetric Versus Optical
Detection for Arizona Road Dust, Aerosol Particles Less Than
3.5 urn But Greater Than 1.5 ym 60
25 Electrical Sensing Design 62
26 Electronic Sensing Data 63
27 Virtual Impactor Schematic 66
28 Final Version of Prototype Virtual Cascade Impactor, Desassembled . 67
29 Final Version of Prototype Virtual Cascade Impactor, Assembled
With Backup Filter 67
30 Data For Test 23 70
31 Test Number 27, Coal Dust 73
32 Test Number 29, Fly Ash 74
33 Test Number 32, Iron Oxide 75
34 Test Number 35, Arizona Road Dust 76
35 Cumulative Size Distribution for January 29, 1976 81
36 Virtual Impactor Pressure Drop Data for January 30, 1976 83
vii
-------
FIGURES (Continued)
Number Page
37 Cumulative Size Distribution Data for January 30, 1976 84
38 Pressure Drop Versus Time for Run No. 1, 24 March 1976 89
39 Pressure Drop Versus Time for Test in Which Concentration Was
Varied 90
40 Pressure Drop Versus Time for Stage 1, 26 March 1976 92
41 Pressure Drop Versus Time for Stage 2, 26 March 1976 93
viii
-------
TABLES
Number Page
1 Concentration Extremes to be Sensed 27
2 Approximate Signal Currents Resulting From an Electric Charge-
Wire Sensor, 30 Lit/Min Total Flow Rate 31
3 Effects of Velocity and Particle Size on Filtration Mechanisms. . . 32
4 Virtual Impactor Test Data (First Version) 47
5 Virtual Impactor Dimensional Data 47
6 Virtual Impactor Test Data (Second Version) With Monodisperse
Aerosol 51
7 Impactor Losses 53
8 Cut Size Data 55
9 Virtual Impactor Dimensional Data for Tests 15 Through 23 ..... 68
10 Virtual Impactor Test Data (With Arizona Road Dust) 69
11 Virtual Impactor Data (Polydisperse Dusts) 71
12 Andersen In-Stack Impactor Data Obtained During In-Stack Tests
at New Bedford 79
13 In-Stack Virtual Impactor Data Taken at New Bedford (Corrected
for Token Flow Effect) 80
14 Virtual Impactor Filter Pressure Gain Compared With Loading as
Reported by the Andersen Impactor 85
15 Virtual Impactor Data From Preliminary Field Test 85
16 Andersen In-Stack Data 87
17 Virtual Impactor Data 88
18 Virtual Impactor Pressure-Drop Drag Coefficients 94
ix
-------
UNIT CONVERSION TABLE
1 inch = 2.54 cm
1 inch H20 = 248.6 Newtons/m2
1 inch Hg = 3386 Newtons/m2
1 psi = 6895 Newtons/m2
1 cfm = 4.719 x ICT11 m3/sec
1 X,pm = 1.667 x 1(T5 m3/sec
-------
ACKNOWLEDGMENTS
The collaboration and support of Bill Kuykendal of the U.S. Environmental
Protection Agency and John Langley of GCA during laboratory and field tests
are gratefully acknowledged.
Our thankful appreciation is due to Phil Morrow, Walter Park, and Gordon
Park of the supervisor staff, as well as to the members of the electrical shop
of New Bedford Power and Light, for their invaluable help during the perfor-
mance of the field tests.
xi
-------
SECTION 1
INTRODUCTION
Monitoring of particulates flowing in stack gases is a challenging problem
for the instrumentation scientist. Stack sampling and sizing of particulates
has, in general, been performed by extracting a sample, under manually adjusted
isokinetic flow conditions, and collecting the particles by filtration or by im-
paction. The collected material is then evaluated gravimetrically; i.e., by
weighing before and after collection. The average mass concentration as a func-
tion of size is then calculated on the basis of the total volume of air sampled
during the collection period. This method is tedious, inefficient and incom-
patible with the continuous recording, automatic data transmission and process-
ing that are required to handle the increasing volume of information resulting
from intensive air pollution monitoring programs. Furthermore, these manual
methods do not provide real-time information, or the temporal resolution from
which to determine emission variability both from the point of view of total
emission as well as size distribution.
Presently available methods for recording or indicating type particulate
monitoring are based on such indirect sensing principles as measuring certain
parameters related to either particle area or particle number concentration. If
light scattering or extinction are used, variations in particle size, shape and
index of refraction, such as are found in the stack environment, can introduce
serious particle sizing errors. A recently developed technique, based on piezo-
electric mass detection, utilizes either electrostatic precipitation or inertial
impaction to deposit particulates on a quartz crystal whose oscillation fre-
quency is a function of the collected mass. Operational difficulties, and the
basic incompatibility of this technique with prolonged unattended operation,
have restricted its use to certain laboratory applications.
The generally preferred solution to the problem of automated mass measure-
ment of particulates from stack gases is based on beta radiation absorption
sensing of material collected, within the stack, on a suitable substrate. This
method when applied to particle sizing, however, results in instrumentation
with significant mechanical and electronic complexity as exemplified by the
seven-stage automatic beta-cascade impactor recently completed by GCA/Technology
Division under another EPA contract.
The objective of the program reported herein was to determine the feasi-
bility of the development of an instrument capable of rapid, real-time deter-
mination of the relative size distribution up and downstream of a particulate
control device in typical stationary source environments. Cost and simplicity
of operation were essential objectives, as well as reliability and compatibility
with typical stack gas conditions.
1
-------
Two general aspects were considered within this development program: the
particle sizing method, and the sensing technique of the segregated particle
fractions. Inertial sizing by virtual impaction was selected for the first of
these aspects. Only inertial sizing methods are intrinsically and reliably
related to particle dynamics and most of their interactions with the environ-
ment including health effects. Aerodynamic size provides the most useful par-
ticle characterization with the exception of visibility effects, although even
there, particle persistence in the atmosphere and thus its temporal effect on
visibility are related to aerodynamic properties.
The second and crucial aspect of this instrument development program was
the method of sensing of the pariiculate fractions segregated by the inertial
technique. This detection method had to be compatible with the measurement
requirements as well as with the environmental constraints imposed by source
conditions. Several candidate approaches to particle detection were investi-
gated within this program; one of them—stage filter pressure drop sensing—was
selected as the most promising technique.
-------
SECTION 2
SUMMARY
The objective of this program was to perform a feasibility investigation
consisting of the selection, development, fabrication and testing of a proto-
type instrument capable of providing a real-time or continual information on
the size distribution of particulates contained in effluent stack gases. The
basic measurement philosophy and criteria underlying this program were:
1. Particle sizing by inertial separation.
2. Particle collection and sensing within the stack itself (i.e.,
no sample extraction).
3. Particle sensing of the inertially separated fractions compatible
with the stack environment and capable of providing continuously
recordable signals to the outside of the stack.
4. Sensing technique had to be reliable and rugged.
At the outset of the program virtual cascade impaction was selected as the
method of inertial particle size segregation. This technique was considered
compatible with several candidate particle sensing approaches and presented
several significant advantages over other inertial methods; e.g., large sample
retention capability, relative simplicity and adequate size selectivity.
An initial theoretical effort was directed at the development of a math-
ematical approach for the optimization of the number of cascade sizing stages
and the sizing separation between consecutive stages. This model took into
consideration the nonideal size selectivity of typical virtual impaction stages
and resulted in a "figure of merit" characterization indicating that a size
interval ratio of about 2.5 provides an optimum compromise between selective
size information and stage overlap effects. Another mathematical effort was
directed at developing a data inversion approach to incorporate the effect of
the secondary or token flow at each virtual impaction stage in the calculation
of the actual particle size distribution. Both of these mathematical tools
should prove very useful in the development and characterization of related
devices and techniques.
During the initial phases of the program a three-stage virtual impactor
prototype was designed and fabricated. This initial version was then subjected
to a series of laboratory tests to determine and characterize particle selec-
tivity as well as interstage particle losses. As a result of this experimental
evaluation a second version was fabricated incorporating several modifications
-------
and improvements designed to minimize particle losses and facilitate disassem-
bly and replacement of various nozzles of different dimensions. This second
version was then in turn subjected to laboratory testing using fluorescent
particle aerosols, and the results, reported in detail within the body of this
report, were entirely satisfactory.
In parallel with the virtual impactor development, four alternative par-
ticle sensing methods were investigated experimentally in order to select one of
these techniques and incorporate it within the three-stage prototype virtual
cascade impactor. These four sensing principles were:
1. Light extinction within each virtual impactor stage stagnation
volume, using fiber optics as source and signal light carriers.
2. Impact momentum sensing of the particles entering the stagna-
tion volumes, using a pressure sensitive transistor.
3. Triboelectric particle sensing by placing a detection electrode
in a small constriction within each of the stage token flow
lines.
4. Pressure drop sensing across a filter at each stage, resulting
from the token flows through the accumulating dust layer.
The tests conducted with each of these methods indicated that of the first
three techniques listed above, light extinction and triboelectric or contact
charging appeared as feasible candidates although their implementation in a
practical field-worthy system would entail considerable effort without guarantee
of ultimate success. The latter technique; i.e., flow drag sensing across each
stage filter, however, provided the most reliable results with a minimum of
system complexity, and it was thus selected for integration into the virtual
cascade impactor for subsequent field testing. This sensing method presented
the further advantage of incorporating its own calibration reference since the
total weight increment on each stage filter can be used to normalize the average
slope of the pressure drop versus time signal obtained from each stage.
The three-stage virtual impactor (two inertial stages and a back-up filter,
with 50 percent cutoff points of 7 and 2 ym for the first two stages) was
modified incorporating pressure sensing taps, and hardened to withstand the
elevated stack temperatures by using heat-resistant gasketing and tubing, and a
series of tests were undertaken within a stack of an oil-fired power plant.
Concurrent samples were taken with an Andersen in-stack cascade impactor and
the results of these tests showed a reliable correlation between stage filter
pressure drop slopes and stage collection, as well as excellent correlation be-
tween the stage filter collection of the virtual impactor and the reference
Andersen jet-to-plate impactor.
Similar results were obtained during a subsequent test series performed at
two different EPA wind-tunnel facilities with concurrent comparisons with
Andersen and Brink cascade impactors. During these latter tests it was deter-
mined that collections of the order of 3000 mg on the first stage filter and
-------
150 rag on the second stage filter could be collected without loss of pressure
drop versus mass linearity.
The tests performed with the prototype cascade virtual impactor with in-
dividual stage pressure drop sensing demonstrated the following crucial aspects:
1. Mass concentrations and size distributions agreed remarkably
well with concurrent measurements performed with an Anderson
in-stack cascade impactor.
2. Sampling times of several hours in duration can be performed
without sample retrieval or operator intervention.
3. Stage loads of the order of hundreds of milligrams can be col-
lected without loss of pressure drop versus loading linearity
with a device whose cross sectional dimensions are compatible
with insertion through a 4-inch diameter port.
It is worth pointing out that the originally stipulated performance ob-
jectives of this program were in most cases either met or exceeded by this
collector-sensor combination.
-------
SECTION 3
CONCLUSIONS
The instrumental combination of a multiple-stage virtual impactor and
pressure drop sensing across stage filters provides a relatively straightforward,
simple and reliable method for the in-stack determination of the time dependent
size distribution of particulates. Linear increase of filter pressure drop
versus mass loading provides a reliable indication of both cumulative mass
(average size distribution), as well as short term or nearly real-time concen-
tration (instantaneous size distribution) with time resolutions in the range of
1 to 10 minutes.
Salient advantages and other characteristics of this instrumental approach
to stack effluent particulate monitoring are:
1. Compatibility with high temperature and corrosive environments.
2. Compatibility with large cumulative collections and thus
capability for extended measurement times without overloading.
3. Minimum maintenance requirements.
4. Self calibrating operation by gravimetry of stage filters.
5. Simplicity of read-out and signal transfer from stack.
6. Operation in a specific stack may require only occasional
self-calibration checks. .Transfer to other sources requires
self-recalibration. Calibration against an independent
reference method may prove to be unnecessary.
7. Adaptability to high and low source concentration levels (up
and downstream of controls).
-------
SECTION 4
RECOMMENDATIONS
The positive results of the laboratory and field tests performed with the
cascade virtual impactor-pressure drop system, as well as the attractive sim-
plicity and basic nature of the measurement principle applied therein strongly
favor further development work on this instrumental approach with the specific
objectives of more fully characterizing this type of in-stack particle sizing
device, and principally to develop a practical and fully field-worthy instru-
ment compatible with prevailing testing facilities and easily adaptable to the
entire range of particulate concentrations encountered within source environments.
Additional refinements and improvements to be considered for such a follow-
on development program are in the area of pressure drop signal recording and
processing. The use of dial-type gauges and liquid manometers was justified
within the reported program and may be acceptable for other research type mea-
surements but for more routine and extensive measurement programs electric
pressure transducers with continuous recording and slow-differentiation of the
pressure drop-versus-time signal would provide more useful and less labor-
intensive data retrieval.
Extensive field testing under a wide variety of source environments and
emission characteristics should be an important phase of such follow-on
activities.
-------
SECTION 5
THEORETICAL ASPECTS
PARTICLE SIZING METHOD
Aerodynamic Particle Diameter
The aerodynamic fate of the emitted particulates and their effect on human
health are the primary areas of concern on which control technologies must be
based, and thus the sizing approach for the measurement of these particulates
must be as representative as possible or as directly "coupled" as possible to
those criteria. Particle sizing methods based on optical, electrical, acous-
tical, etc., techniques are not directly related to the aerodynamic behavior of
particulates and to their mass. Inertial methods, on the other hand, provide a
size selection mechanism that is based on the aerodynamic properties of the
particles and is directly relared to the atmospheric transport and sedimenta-
tion characteristics, as well as to the respiratory tract deposition and clear-
ance behavior of airborne particulates.1"3 Thus the selected sizing method had
to be based on the inertial separation of particles providing a measure of their
aerodynamic diameter.
The aerodynamic diameter characterizes particle behavior under the force of
gravity and under inertial acceleration in fluid flow fields, such as accelera-
tion caused by curvature of the flow. The mechanisms of impaction and sedimenta-
tion and, indirectly, diffusion, depend on the aerodynamic diameter, and those
are the predominant mechanisms in filtration and in lung deposition. Further-
more, the behavior of particles in such control devices as cyclones, Venturi
scrubbers, electrostatic precipitators, filters, and settling chambers also
depends directly or indirectly upon aerodynamic diameter. The aerodynamic diam-
eter comes into play in the dispersion of particles in the atmosphere, con-
trolling the sedimentation of particles, their capture by obstacles in a wind,
and their scavenging by rain.
The optical equivalent diameter (or light scattering cross-section), on the
other hand, is significant in the effect of particles on visibility.
Other measures of particle size have special contexts that are less directly
related to particle behavior and are of interest for special and more restrictive
particle studies.
The aerodynamic diameter of a particle of arbitrary shape is that diameter
which would give the same terminal settling velocity for a sphere of unit den-
sity. Mathematically,
-------
v = fp D2 C/18u] g
s L o ae J
where vs = terminal velocity
po = unit density
Bae = aerodynamic diameter
C = Cunningham correction factor (calculated for Dae)
g = gravitational acceleration
y = gas viscosity
The term in brackets, the "relaxation time" of the particle, governs the
behavior of particles falling due to gravity (sedimentation) or crossing the
streamlines of a gas to strike an obstacle deflecting the gas (impaction). The
choices for separating particles on the basis of their aerodynamic diameters
narrow down to: gravitational sedimentation, centrifugation and jet impaction.
Of these alternatives, a variation of jet impaction (virtual impaction) was
selected because, among other reasons, it provides a more controllable collection
geometry and better size segregation selectivity as well as compatibility with
several sensing techniques.
Sampling Considerations
The need for isokinetic sampling does not have to be re-emphasized here.
The problems unique to inertial sizing devices, however, preclude flow rate
adjustment as one of the means to match stack and sampler inlet velocities. The
only method to obtain such sampling conditions without affecting the size seg-
regation properties of an inertial separator is by changing inlet cross-sectional
area; i.e., by selecting an inlet nozzle of proper size. This method, however,
is not compatible with any straightforward automation, but remains at present
the only feasible approach and was thus adopted for the instrumentation
problem at hand.
The problem of sample extraction versus nonextractive in-stack sampling
will now be discussed. The conclusions presented in a comprehensive GCA/
Technology Division literature review on source testing instrumentation4 will
simply be restated verbatim:
"Most stack sampling devices used for routine measurements are based on
the sample extraction approach; i.e., ah isokinetic inlet nozzle followed by a
duct, commonly called a 'probe' whose length is determined by the cross-sectional
dimensions of the stack at the point of sampling. The particle-laden gas flows
through this probe, and the particulate matter is collected on filters or other
devices on the outside of the stack wall, through which the probe passes. This
procedure is specified for stack sampling by the Environmental Protection Agency.
This method has several drawbacks that are, although acceptable for manual opera-
tion, not compatible with the meaningful -operation of an automated device, and
in particular for a particle sizing instrument. These problem areas are (1) par-
ticulate losses along the probe or, extraction duct, and (2) condensation within
-------
the exterior section of the probe and at the collector exposed to lower tem-
peratures than the stack environment (or conversely the need for heating of
these elements, not only above the water dewpoint, but also above the acid mist
condensation temperature).
"The first problem (i.e., particulate losses within the probe) is caused by
the following mechanisms:
1. Impaction in tube bends.
2. Sedimentation due to gravity.
3. Electrostatic scattering due to mutual repulsion by charged
particles.
4. Turbulent deposition.
5. Electrostatic precipitation due to image charges.
6. Diffusional loss to the tube wall.
7. Thermal precipitation due to temperature difference between gas
and tube wall.
All of these loss mechanisms are dependent on particle size, thus changing not
only the total concentration but also the shape of the size distribution. Such
losses in stack tests have been shown to occur.5 Further changes in the size
distribution can come from:
1. Coagulation, enhanced by shear in the gas velocity profile and
by vapor condensation.
2. Re-entrainment.
3. Condensation.
The approximate magnitude of such influences can be estimated from formulas
available in the literature.6*7 In any case these losses are unacceptable for
an automated-recording type instrument for the obvious reason that the fraction
deposited in the probe is not incorporated in the measurement. The EPA specified
procedure for manual sampling includes the requirement for probe washout and
addition of the mass of particulates retained in the probe to the collected mass
on the filter. Such a procedure must be ruled out for an automated measurement
system. No practical method has yet been found to eliminate the deposition of
particulates in probes for size distributions with mass median diameters of 10
to 20 ym and Og between 2 and 5, typical of stack environments, and for probe
lengths of the order of, or in excess of 4 meters required to sample repre-
sentatively a stack with a flow duct area of up to about 50 m2 (approximately
500 square feet). The only method by which probe deposition can be reduced,
albeit not eliminated, is by increasing the flow velocity to very high values
(of the order of 20 m/s or more) to produce re-entrainment of particles
10
-------
collected on the probe walls. This procedure may be marginally acceptable for
total mass measurements, but not for accurate sizing purposes since particle
agglomeration and clustering can be expected in any situation where re-entrain-
ment plays a significant role.
No line losses are acceptable unless they constitute a known fraction of
the inlet loading, a highly unlikely condition indeed. Obviously the conclusion
is that since the probe length cannot be minimized if the collection-sensing
device is exterior to the stack, the only rational solution is to place the
device within the stack duct to be sampled. This method also resolves in large
measure a second problem: the condensation of water and acid mist at the mea-
surement site. By operating the sampler wholly at the source temperature, con-
densation can be precluded. At typical stack gas temperature of 150°C to 260°C
(300°F to 500°F) all the water is in the vapor phase and if the collection-
detection head is exposed to the same environment, no condensation will occur.
The measurement of particulate mass and especially of the size distribu-
tion by mass of stack emissions can only be determined meaningfully at two loca-
tions: (1) within the stack itself, or (2) downstream from the stack exit
(stack plume). The first of these approaches can provide information on the
particulate emissions as they are generated, and within the high temperature
environment of the stack itself. The second approach would provide a measure of
the particulates as they will eventually disperse in the atmosphere. Both of
these approaches have their merit.
Sample extraction, by lengthy probes and requiring thermal control, is an
approach that we believe is an unacceptable compromise between the two al-
ternatives presented above, and it is especially incompatible with representa-
tive size measurements.
Cascade Virtual Impaction Advantages
Inertial particle separation by cascade impaction has been used extensively
in a number of different configurations, and the literature on the subject of
impaction and its application to sizing through the cascade arrangement is
copious.8"11 The use of cascade impaction in source testing has received in-
creasing attention in the last few years and several versions of the multi-
orifice (e.g., Andersen) configuration have been designed for direct insertion
into the stack environment.12 Such devices are merely size-selective collectors
of particulates, and the evaluation of the collected material is performed sub-
sequently by gravimetry, chemical analysis, etc.
All these devices, however, are based on the jet-to-plate configuration;
i.e., particle interception by a solid (usually coated) collection surface. A
variation on this generalized approach is the cascade centripeter, aerosol con-
centrator or virtual impactor. Figure 1 is a schematic of a virtual impactor
stage.
The method of virtual impaction was originally developed by McGinn and
MacWaters13 and then by Hounam and Sherwood11* who called the device a "cascade
centripeter." Similar devices have been described by Conner15 and Loo.15 All
11
-------
MAIN GAS FLOW
MAIN GAS
FLOW
FILTER
STAGNATION
AND
SENSING
VOLUME
TOKEN
FLOW
Figure 1. Schematic of a virtual impaction stage.
12
-------
these devices are based on "impaction" into a stagnation volume rather than onto
a solid surface normal to the flow direction as in the case of jet-to-plate im-
pactors. Particles larger than a characteristic cut-off dimension penetrate
into the void due to their inertia, while the smaller particles are entrained
by the air that circumvents the entrance to the void volume. Particles entering
the void can then be collected on a filter, or detected as they are being
extracted by means of a small secondary flow typically of the order of several
percent of the primary sampling flow. This small flow rate can be obtained
either by relying on the jet impact pressure or by a small diversion flow line
from the vacuum pump.
The equation governing the collection of particles by means of an impactor,
and thus equally applicable to a centripeter or virtual impactor, is
,1/2
where i|> is the impaction parameter (geometry dependent)
pp is the particle density
v is the jet velocity
Dp is the particle diameter
y is the coefficient of viscosity of the gas
Wj is the jet width
C is the Cunningham correction factor.
For the experimental set pp of Conner,15 ^ = 0.6, and for the experimental
set up Hounam and Sherwood11* ^ = 0.73, a value considerably higher than the
typical round jet impactor value of about 0.3 cited by Mitchell and Pilcher,9
but very close to the value reported by Lundgren^ for a rectangular jet (slit)
type impactor (^ = 0.6).
The advantage of the virtual impaction method with respect to the usual im-
paction approach is the vastly increased mass which can be collected on each
stage without overloading and concomitant re-entrainment . Particle bounce
problems due to uncoated or poorly coated impaction surfaces17 are eliminated.
In addition, filter collection of each of the particulate size fractions is made
possible because lower collection flow rates can be used; i.e., the small
"token" extraction flow mentioned above, permitting large collection densities
on such filters without excessive pressure drops, and thus extended sampling
times between filter replacements become possible. Filter collection offers
other advantages with respect to the usual impaction methods in that samples
may be collected for subsequent physical or chemical analysis; or small samples
may be collected, the filter material being rendered transparent by use of im-
mersion oil, and the sample inspected using light transmission microscopy.
Furthermore, for certain particle sensing techniques this method provides a
13
-------
significant and necessary enhancement of concentration within the stagnation
volume wherein the particles are retained. For those sensing methods not re-
quiring actual collection and retention of particulates (i.e., nonintegrating
techniques), the virtual impaction approach obviates the need for any frequent
and routine collection surface renewal or cleaning, as opposed to the usual
type of impaction.
Theoretical Cascade Virtual Impactor Optimization
The resolution or size selectivity of a cascade impactor depends on the
number of stages used to cover the desired total size interval to be assessed.
In practice, however, because the retention characteristics of impactor stages
are not ideal; i.e., they deviate significantly from true step functions, the
resulting overlap between stages becomes the limiting factor tending to degrade
the information as the size interval between successive stage is decreased.
A computer program was thus developed to characterize this cascade impactor
behavior and provide the criteria for its optimization. This program was based
on a cumulative log-normal stage retention curve with a geometric standard
deviation of 1.4 as a reasonably close approximation to the collection efficiency
curve found by Conner15 for a cascade centripeter or virtual cascade impactor.
This program provides a measure of the error amplification or degradation in the
measurement accuracy, of a cascade separator of nonideal characteristics; i.e.,
with gradual particle-size cuts rather than a step-function. In fact, as a
result of this deviation from ideal conditions, what is retained on any stage
of a series of such stages is really the result of the interaction of that stage
with those preceding it. This problem of cross sensitivities is obviously
crucial in determining how closely spaced the impactor size intervals can be
and yet contribute constructively to the size characterization of an aerosol.
There is an extensive literature concerning the solution of the integral
equations which are typical of some data analysis problems:
F*(a*) = I p(a*,a) f(a) da
where F*(a*) is the reading (or the penetration, transmission, etc.) which re-
sults when an instrument which has the reading or response p(a*,a) to a mono-
disperse distribution (a delta function) is used to measure a polydisperse dis-
r<*>
tribution f(a). By definition I f(a) da = 1 and p(a*,a) and F*(a*) are di-
Jo
mensionless and are in the range 0 to 1. This kind of formulation has been used
by several authors in the aerosol analysis context18"20 and occurs in many other
fields as well. The method of solution is to convert this equation into a set
of linear equations by approximating the integral by a summation.
Of the ways to convert the integral equation to a set of linear equations,
we discuss only one, the use of a quadrature formula. A specific upper limit
14
-------
is chosen, a^ and a lower limit, ag. The interval of integration is broken into
segments in terms of the known kernel, p(a*,a), and the unknown function f(a)
evaluated at specific points within the segments:
•aoo N
p(ai*,a) f(a) da = 2^ p(a±*, a ) f(5 ) Aa..
j = 1
o
or
N
F*(a.*)= £ Pij[^a]_
j = 1 J
We have made the decision to evaluate the segments at their midpoints, a^,
which means we are using a "midpoint quadrature," the error of which is known.
Usually, the number of points at which the data are obtained (ai*) will be the
same as the number of points at which the integral is evaluated (aj). The last
equation is conveniently written in vector-matrix notation as
F* = PAF
where the vector AF has components fj A aj. If the set of linear equations,
standard computer program the solutions can be ex-
F* = PAF, is solved with a
pressed as
AF = P~1F*
and the matrix P~ is referred to as the inverse of P.
Twomey21 used such a method to invert the data from a diffusion battery
method and found that the solutions were highly oscillatory functions, even
sometimes negative. Twomey22 and Phillips" and others have tried to circumvent
the oscillatory results by "smoothing" techniques, which solve the equations
with the constraint that the solutions be as smooth as possible and still come
within a prescribed error in giving results which match the data. Rust and
Burrus21f and Cooper,19 among others, have discussed such methods at length; the
choice of the degree of smoothing is formidable. The oscillations result from
magnification of small errors which come from the formulation (approximating
the integral) or from the data. The error in the solution, 6 (AF), due to a
small error in the data SF* can be shown to be
-1
6F*
?* II
15
-------
where the notation || P || indicates the magnitude or norm of P. One set of
vector and matrix norms is defined as follows (the maximum norm):
F* (I = max |F.*|
N
p H=o max £ IP-M
1 < i < N i = 1 J
The contribution of an error in the matrix, 6P, is akin to that of the con
tribution of an error in the data. The product || P|| • || P"1 || is referred to
in the literature of linear equation solving as the condition number of P,
cond (P). For the Lundgren impactor,^ this cond(P) is about 25; and is very
much larger for diffusion batteries, meaning that a few percent error in the
data will be greatly magnified by attempts to improve the results by solving
the linear equations. The condition number will get larger as the number of
data points (subintervals or segments) increases or as the sharpness of size
discrimination (e.g., dgi^/d^o) decreases, as discussed elsewhere.19
The role of the condition number, dominated by || P"1 || , is examined in
detail by Cooper19 with respect to its influence on data inversion. What does
not seem to have been realized previously is that || P"1 || is important even in
situations where the data are not to be inverted, but used directly.
Figure 2 shows an ideal cascade impactor response (unity within the assumed
sizing interval and 0 outside) with a nonideal response function superimposed
upon it. The nonideal instrument shows "cross-sensitivity," a response in a
given channel to a particle size which is actually outside the bounds of that
channel. A channel for a cascade impactor is simply the impactor stage, and
the size associated with that stage is generally taken to be the difference be-
tween dso for that stage and d$Q for the stage upstream from it. Because the
multi-channel interpretation of the impactor data is related to differences,
the r'esponse of the £th channel with respect to the jtn particle size is defined
here as
If the impactor were ideal, that is if the response functions were step
functions, then they would be characterized by the equations which charac-
terize the ideal instrument shown in Figure 2; that is,
AF* = I AF
where I is the identity matrix (I±i = 1, I-jj = 0 for j + i). The difference
between what is registered by the nonideal instrument and what is actually in
the assumed size intervals associated with the stages is given by:
16
-------
Q2» Q3*
SIZE PARAMETER,a
04*
Figure 2. Penetration, or response, function for ideal and nonideal
instrument.
17
-------
- AF = (I - AP"1) AF*
where AP"1 indicates the inverse. When impactor data are used directly, it is
assumed that they represent the aerosol to be found within the limits associated
with the itn stage, but this assumption is incorrect by the factor of (I - AP"1),
and thus, its magnitude is a measure of the error associated with the instrument.
The linear equations were solved to obtain an error factor, || AP"1]) , for sev-
eral cases to show the severity of the problem or cross-sensitivity for a cas-
case impactor. Two points are important here: this is a factor related to in-
strument error which applies whet'heT OP not any data inversion techniques are
tried; || AP"1]] » 10 indicates that data inversion will be difficult.19
The integral equations were set up using the response function for a cas-
cade virtual impactor ("cascade centripeter") given by Conner,15 approximated
by cumulative log-normal distributions with the medians (dso) chosen to be the
size interval demarkation (ai*) and their geometric standard deviation chosen
to be 1.4. The range for the impactor tests was ao = 0 and aTO = 1 , with the
device assumed to have one stage a« and in absolute filter at a0.
One set of numerical experiments was done by subdividing the range into
equal sizing intervals. For N measurements, there were N sizing intervals,
each having a width of 1/N. Values of N from 2 to 10 were used, representing
two-state measurements, three-stage measurements, etc. As N increases, the
interval between the stages decreases and the degree of overlap of the response
functions, the cross-sensitivities, increases. The values for cond (AP) and
|| AP"1!! are nearly the same (because || AP|| = 1).
A second set of numerical experiments was made by dividing the range
a0 = 1) into segments which increased geometrically in size. The sizing inter-
vals were chosen so that their boundaries were a± = Ai-N for i from 1 to N and
ao = 0,* A was chosen to be 2^, 2, 3, 10. As can be seen in Figure 3, this leads
to error factors that tend toward an equilibrium value at large N, are smaller
than the corresponding case for intervals sized at 1/N, and are not much larger
than 1 except for the A = 2^ case. To select an optimum value of A, the ratio
between impactor stage cut-offs (d^Q values), the error factor || AP"1 || should
be combined with some measure of advantage to form a figure of merit. Gen-
erally, closely spaced data are desired (small A) and so are small values of
|| AP~*|| . Figure 4 is a plot of the resulting "figure of negative merit"
defined by the product of the sizing interval ratio A times the above mentioned
error multiplication parameter || AP"1]] resulting from stage size separation
overlap. This product A x || AP"1]! is to be minimized for maximum size resolu-
tion and minimum overlap error. The variable N is the number of stages of the
impactor such that the total range covered is equal to A^:!.
It is obvious from Figure 4 that an optimum condition exists for a size
interval ratio of about 2.5 for most values of N. For a typical total range of
Where A is defined as the sizing interval factor; i.e., the ratio between adja-
cent particle size cuts.
18
-------
20
10
i
O.
g 3
I
g 3
(T
CL
UJ
o 2
•o 3
-o 10
456789
NUMBERS OF CHANNELS,N
10
Figure 3. Error factor versus number of geometrically increasing sizing
intervals for impactor.
19
-------
8
K>
Figure 4. "Figure of negative merit" as a function of size interval ratio and number of stages. For a
log-normal stage collection efficiency curve whose geometric standard deviation is equal to
1.4 (typical of virtual impactor stages).
-------
20:1 (e.g., 0.5 ym to 10 ym) the corresponding number of stages would be 3.3.
A somewhat more practicel value results by setting N = 4 which corresponds to
A = 2.11, a value sufficiently close to that of the minimum product
A x || AP~l|| . For N = 5, and a range of 20:1, A = 1.82 which corresponds to a
higher "figure of negative merit;" i.e., a less advantageous measurement con-
dition. A compromise between measurement, figure of merit and practical design
considerations appears to be a four-stage configuration with a sizing interval
ratio A = 2 and a corresponding total range of 16:1 (e.g., 0.5 ym to 8 ym) which
could be extended by the addition of a fifth stage if so desired.
Data Inversion Incorporating Secondary Flow
As shown in Figure 1 (page 12), the aerosol enters the first impaction stage
and virtually all particles with aerodynamic diameters larger than the cut-off
particle diameter of that stage are captured by "impaction" into the nearly-
stagnant volume of that stage. A small flow, the secondary flow, is removed to
draw the particulate material to the filter, where its mass can be determined by
several different techniques. The same process occurs at the subsequent stages,
after which the remaining aerosol flows to and is captured by the final filter,
the last sizing stage. During this program it was found that the losses of
particulate matter in the impactor can be minimized if a fairly substantial
secondary or token flow (as much as 10 percent of the main flow) is employed;
this flow could cause a data analysis problem because it contains a sample of
particles of all sizes not already captured upstream. It thus becomes necessary
to develop a mathematical procedure in order to correct for this secondary flow
contribution. This method should also be applicable to other particle sizing
situations in which there is a constant fraction of the aerosol material either
lost or added to a sizing stage.
If the centripeter worked like an ideal multichannel sizing device, each
stage would have 100 percent collection efficiency for particles within its
assumed sizing interval and 0 percent efficiency for particles outside this
interval. Unfortunately, there are three nonideal aspects of the centripeter:
(1) the secondary flow collects a fraction of all the particulate matter present
in the gas stream approaching the nearly-stagnant region of each stage; (2) the
series (cascade) nature of the centripeter means that succeeding stages have
their effective capture efficiencies influenced by the preceeding stages; i.e.,
they cannot capture what does not reach them; (3) even without these two inter-
ferences, each centripeter stage does not change abruptly from 0 percent to
100 percent efficiency at the stage cut size. Correcting for the latter two
factors has been the subject of the preceding section of this report and will
not be discussed further here. We shall now assume that the secondary flow
factor is a predominant error factor, which should be true for centripeters
with secondary flows of more than a few percent and with sizing intervals large
compared with the particle size differences needed to go from nearly 0 percent
to nearly 100 percent impaction efficiency for the stages.
Let the fraction (by count, mass, etc.) of the aerosol which is truly in
size range j be labeled Fj. Let the efficiency of the itn sizing stage for
particles which are in the jth size range be EIJ. Then the fraction, Fi*, of
the total collected aerosol material which is collected on the ith stage will
be given by (for a device of k stages)
21
-------
j - 1
In the case of a three-stage centripeter this becomes:
Fl + E12 F2 + E13 F3
E21 Fl + E22 F2 + E23 F3
If the secondary flow is a constant fraction, e, of the primary flow (sec-
ondary flow assumed to have same size distribution as primary flow), and if the
capture efficiency of the stage is essentially 100 percent for particles larger
than the cut size and essentially 0 percent for smaller particles, one has:
(1) Fx + (e) F2 + (e)
(0) F, + l(l-e) F0 + e(l-e)
= (0)
(0)
The set of linear equations can also be written as the transformation of a
vector by a matrix:
•+* ->.
F - E F
-v*
where the vector F has as its components
-»•*
F -
The matrix is just the array of the coefficients of the linear equations:
E =
"l
0
_0
e
Kl-e)
0
e
e(l-e)
l(l-e(l-e)-e)_
22
-------
One can solve the linear equations by the usual methods21* to obtain (for
E2i = E31 = E32 = 0):
F2 - (F2 - E23F3)/E22
These equations show that the true distribution can be obtained by sub-
tracting the material caught due to secondary flow and correcting for the loss
of the material before it reaches the sizing stage (EH < 1). Another way of
solving the system of equations is to obtain the inverse matrix, E"1 , and thus
F from:
-1 -*
= E F
In this case the inverse matrix can be shown to be
-1
1
11
0
0
(-E12/EUE22)
_
(E12E23E22 ~ E13)/E11E33
-1
^33
which becomes:
1 0 T
1 -(e + e + eJ + ...)
0 (1 + e1 + e2 + e3 + ...)
1 2 3
-(e + e + eJ + ...)
-(e1 + 2e2 + 3e3
3e2 + 4e3
The equations presented above are sufficient to correct for the secondary
flow effect for a three-stage device. For four or more stages, it is probably
more convenient to use computer subroutines to obtain solutions to the linear
equations (or to obtain the inverse matrix).
23
-------
A number of simulations were performed to determine the magnitude of the correc-
tions.^ Hypothetical distributions were "sampled" by the centripeter. (Actu-
ally, F* was obtained from E F.) The "data," F.^* were then used to obtain,
once again, the input distribution, W±. (This is the same as obtaining F£ from
g-1 F^.) xhe results are shown in Figure 5; the size cuts were shown for 2 ym
and 7 ym (same as those obtained for the actual prototype to be described later),
but the actual cut sizes are immaterial to this analysis. Figure 5 shows five
comparisons between corrected and uncorrected data obtained by mathematical
simulation for a three-stage centripeter (e = 0.1). The first case has an
aerosol having all the particles actually within the first size range (Fj = 1);
the uncorrected and corrected data produced the same result. In the second case
the aerosol is all within the second size range (F2 = 1). The uncorrected data
show aerosol in the first two sizing intervals, but the corrected data give the
right result. The third case shows the uncorrected and corrected data for an
aerosol simulated to be wholly within the smallest size range (F3 = 1). (These
three cases correspond to sampling aerosols with sufficiently narrow size dis-
tributions that each was wholly within one sizing interval.) The fourth case
is for an aerosol equally divided between all three size intervals and the fifth
has an aerosol which is split between the two smallest size intervals. In all
cases but the first, the corrections are appreciable. In all cases, the true
aerosol size distributions are recovered by using E""1.
24
-------
THREE STAGE VIRTUAL IMPACTOR
TOKEN FLOW* 10%
V
5 °5
o
£72
0.5
M
5
SIZE
INTERVALS
(a)
SIZE
INTERVALS
(b)
(C)
SIZE
INTERVALS
(d)
SIZE
INTERVALS
(e)
Figure 5. Data, F2*, and input distributions, Fi, regained by correcting that data, for a three-
stage virtual impactor with 10 percent secondary flow, flow = 10 percent; (a) Fj = 1,
(b) F2 = 1, (c) F3 = 1, (d) F! = F2 = F3 = 1/3, (e) F2 = F3 = 1/2.
-------
SECTION 6
PARTICLE SENSING AND DETECTION METHODS
Once inertial or aerodynamic size segregation of the particles has been
effected, the objective of this instrument development program was to determine
by some method of particle sensing or detection, the relative amounts in each
size fraction and this information had to be made available as an output signal
which can be displayed and/or recorded.
Several reviews of applicable techniques have been published, among them
Dorsey and Burckle presented a comprehensive tabulation. 5
The method of sensing the particulates in each size fraction had to fulfill
several conditions which are dictated by the specifications of the pertinent RFP
as well as by the overall utilization objectives of the instrument.
In reviewing some of these specifications, it became obvious that very few
sensing techniques are available fulfilling these conditions. From Table 1,
based on the original specification of the request for proposal, it becomes ap-
parent that the required measurement range extends over about 4-1/2. decades;
i.e., from about 0.07 mg/m3 to as high as about 2000 mg/m3.
If, in addition, one considers that the low limit must be detected within
typically ±10 percent the overall sensing range becomes a staggering 5-1/2
decades, by no means a trivial objective. For a cumulative method; i.e., where
the particulates are collected and the measurement is based on an integrating
phenomenon, such as filter collection, the required range is further extended
because of the stipulated time requirements of 10 minutes minimum resolution and
2 hours minimum operational time without servicing. Superimposed on this rather
demanding range requirements were other very rigorous operational constraints,
some of them stated specifically in the RFP and others, although not mentioned
there, of an even more fundamental nature which have been discussed in a pre-
ceding section; i.e., the need for direct insertion of any sampling-sizing-
sensing configuration into the stack or flue environment, with the attendant
problems of temperature compatibility and corrosion resistance.
Last, but not least, the overriding requirement for low cost and simplicity
introduce further severe limitations in the selection of a sensing method.
The combination of all the preceding conditions, specifications and qual-
itative objectives made the selection of a particle detection method exceedingly
difficult. Thus, after careful review of all applicable techniques, it was
decided to explore three of the most promising approaches within this program
with the objective of making a final selection to be incorporated in the proto-
type device.
26
-------
TABLE 1. CONCENTRATION EXTREMES TO BE SENSED
Outlet loading Inlet loading
Size interval, urn Loading, mg/m3 Size interval, pm Loading, mg/m3
>6.0
3.75-6.0
2.53-3.75
1.71-2.53
1.09-1.71
0.52-1.09
0.30-0.52
0.18-0.30
<0.18
0.23
1.15
0.092
0.115
0.069
0.184
0.391
0.391
1.265
>3.4
2.0-3.4
1.3-2.0
0.7-1.3
0.43-0.7
0.23-0.43
<0.23
1021
156
529
1992
1309
472
577
The detection methods that were studied during this program were:
1. Optical sensing.
2. Electrical sensing.
3. Pressure sensing.
In addition, some experiments were conducted using an impact detection
scheme. The theoretical aspects of the first three methods will now be treated
in some detail.
PARTICLE SENSING METHODS
Optical Sensing
The volume concentration of an aerosol may be measured by optical means
provided a certain degree of optical characterization has been performed upon
the specific aerosol. Two methods commonly have been used. In the first, the
extinction of a beam of light is measured as it passes through a known aerosol
path length from source to detector. In the second method, light scattered by
aerosols is collected and measured. Optical methods of aerosol measurement
suffer from a number of disadvantages. The scattering cross-section is not a
simple function of size but varies in a complex way which yields ambiguities in
size versus intensity of scattered light. Departures from sphericity and dif-
fering refractive indices further complicate the method. However, where the
size separation is performed inertially as in the current situation, and large
ensembles of particles are involved whose diameters span a factor of two or so,
a substantial amount of smoothing occurs and a single calibration constant is
27
-------
required for each stage for each type of dust mainly due to differing re-
fractive index.
A method using scattering must eliminate the primary beam from the measure-
ment and, since particles whose diameter exceeds the wavelength of light scatter
mainly in or near the direction of incident light, either a rather sophisticated
optical system is required or rather inefficient light collection is achieved.
The nature of the "in-stack" environment under consideration strongly suggested
the use of extinction as a measurement method. A schematic of the sensing con-
figuration explored within this program is shown in Figure 6. All particles
with diameters larger than d enter the optical scattering volume, attenuating
the light entering I0 to the level I according to:
T ,, -k n A L
I/I = e
o
where L is the optical path through the aerosol
A is the mean particle cross section area
n is the number concentration
k is the average extinction efficiency.
The values of I and Io are obtained from detectors which respectively mea-
sure the light exiting the optical chamber and a sample of the incoming beam at
the light source itself. A and k may be determined from the aerosol size cut-
offs employed in the particular stage and the nature of the aerosol, and L is, of
course, characteristic of the instrument. Adjustment of the amplifier gain
associated with the detectors is set to I = Io when no aerosol is present. The
ratio is obtained directly with an electronic divider circuit. Thus, a value
of n may be obtained and by operating for a known time at a specific flow volume
comparisons may be made with a gravimetric sample collected on the filter.
In practice, within the stagnation volumes of a virtual impactor, the con-
centration of particles undergoes a substantial enhancement that must be in-
corporated in the calculations of n. To calculate the response time of such a
system; i.e., the concentration of particulates in the particulate size interval
of a given virtual impaction stagnation volume, the following equation applies:
c(t) =ce+ E+1 c
s
where c(t) is stagnation volume concentration
co is initial concentration in the stagnation volume
cs is unenhanced or inlet concentration for the size fraction
retained by the stage under consideration
28
-------
AEROSOL
(MPACTOR NOZZLE
FIBER OPTIC LIGHT
PIPE FROM LASER
VO
TICAL SENSING
VOLUME
FIBER OPTIC LIGHT
PIPE TO DETECTOR
FLOW 0.51 /min
FLOW 151/min
Figure 6. Optical sensing configuration.
-------
v is stagnation volume
t is time
E is collection efficiency of the stage
Ac is centripeter opening area
Aj is jet area
Q is primary flow rate
q is secondary or extraction flow rate.
The factor E-rr ^ + 1 is the enrichment factor or concentration enhancement
Aj q
factor, which in many cases where E = 1 and Ac = Aj, can be approximated by
Q/q. Thus the free stream number concentration no can be calculated as follows:
nQ = n q/Q
Electrical Detection
Two electrical sensing approaches of the particulate fraction captured by
each centripeter stage will now be considered. The first of these consists of
an electric charging section preceding the cascade centripeter. This charging
section can be either a corona discharge stage or a radioactive source with
unipolar charge removal (by means of an electric field). All entering partic-
ulates (larger than 0.1 ym approximately) would receive a charge (by a combina-
tion of field and diffusion charging for the corona case, and diffusion for the
radioactive source) commensurate with particle size (the actual size-charge
relationship would be unimportant as long as it remains constant), these charged
particles would then be sensed by impaction on a receiving electrode36 such as
a thin wire placed at, or immediately after each centripeter inlet. Upon im-
paction on the insulated wire the particles transfer their charge to the wire
and are then re-entrained and enter the stagnation volume from which they are
continuously removed by the previously mentioned secondary extraction flow. The
signal from each of these sensing wires consists of a current which can be trans-
mitted to the outside of the stack. An empirical calibration constant for each
stage would relate total charge per unit time (signal current) to the number of
particles in each size interval. The accuracy of this method obviously depends
on how narrow the segregated size fractions are.
Table 2 shows typical parameters relevant to this detection scheme. Al-
though only three size fractions are shown (for simplicity), as mentioned above
a larger number of size subdivisions would be desirable. This table is based on
a 20 percent detection efficiency; i.e., approximately 20 percent of the avail-
able charges are sensed.
As Table 2 shows the range of current to be detected exceeds 5 decades and
while it is simple to detect a current of 2 x 1Q~8 amperes at the high end, it
is by no means trivial to do so for 10~13 amperes, especially within the
30
-------
TABLE 2. APPROXIMATE SIGNAL CURRENTS RESULTING FROM AN ELECTRIC
CHARGE-WIRE SENSOR, 30 LIT/MIN TOTAL FLOW RATE
Signal current
Size range Control inlet Control outlet
1
0.
>3
-
3
urn
3
-
pm
1 ym
8 x
- 5 x
2 x
ID'11
lo-1
10-8
0
ampere
ampere
ampere
io-13
10
3 x 10
-13
-12
ampere
ampere
ampere
far-from- ideal environment under which this device must operate. An alterna-
tive to current sensing is the use of a charge integration approach wherein a
small capacitance is charged up by the sensing current. This accumulated charge
can then be processed electronically by a high input impedance charge detector
and displayed with typical integration times of the order of 100 seconds cor-
responding to an accumulated charge of 10"11 coulombs (e.g., 0.1 volt across a
100 pf capacitor) . This capacitance could be obtained directly from a coaxial
conductor feeding the signal to the outside of the stack. Range switching
would be accomplished by external paralleling of additional capacitance.
The disadvantage of this method resides in the relative complexity of the
electric charging stage which would require a high voltage feed or possibly a
radio isotope.
An alternative electrical detection technique is based on contact electri-
fication or turbulent charge transfer, a method which does not require particle
charging. This phenomenon is the basis for two commercial particulate monitors
(Konitest and Ikor) . Although their strict correlation with particulate mass
concentration; i.e., mass flow, is questionable for a large variety of particle
compositions, this technique nevertheless, can be applied to the performance of
such relative measurements as those intended for the present application. The
signal from the sensing electrode is a current Is which has been reported to be
proportional to the mass flow m and the flow velocity v squared26
I « m v
The frictional charge removal is performed by a high velocity flow, im-
pinging either rectilinear ly or centrifugally on a sensing electrode, which for
the Konitest consists of a semiconductor surface (magnesium hydrosilicate) . 27
Velocities of the order of 100 m/s or more are required at the sensing surface,
both for collection and for the continuous removal of particles by turbulent re-
entrainment. Concentration ranges of more than IO5 have been reported.26 This
sensing technique can be combined with the cascade centripeter sizing system by
using the secondary flow and scaling down the usual Konitest geometry to a flow
31
-------
rate of the order of a few liters/min (the secondary or token flow rate), thus
measuring the particles as they are extracted from the stagnation volume (see
Figure 7). The resulting electrical signal is then carried to the exterior of
the stack. If the flow velocity at the sensor is maintained sufficiently high,
no significant particle deposition is to be expected and a minimum of maintenance
would be required.
Pressure Drop Method
A prime candidate method for the determination of mass accmulation for each
stage of the virtual impactor is to measure the pressure drop across each stage
filter as a function of time (see Figure 8). To do this, the dependence of
pressure drop on time, filter loading, particle size, particle characteristics,
filter face velocity, and filter efficiency, must be understood.
Mechanisms for aerosol collection by fibrous filters are described in Fuchs**
as impaction, diffusion, interception, and sedimentation. Table 3 may be con-
structed which shows whether the efficiency of each mechanisms increases or de-
creases as filter velocity, Vo, and particle size, Dp, are increased:
TABLE 3. EFFECTS OF VELOCITY AND PARTICLE SIZE ON
FILTRATION MECHANISMS
Mechanism VQt Dp+
Impaction t t
Diffusion 4- 4-
Interception No effect f
Sedimentation 4- t
Note: t designates increase;
I designates decrease.
Many investigators have predicted an aerosol size which has maximum pene-
tration through fibrous filters, based on theoretical single fiber filtration
mechanisms as listed above, but usually ignoring electrostatic charge effects
which generally act to increase filter efficiency. This maximum penetration has
been verified experimentally in some studies,28 whereas other studies report a
continuously increasing penetration as particle size decreases.29 The work of
Thomas and Yoder3^ as reported in Fuchs is shown in Figure 9. This work is in-
cluded here rather than some others available,29'31 because the filter face
velocities are on the order of those used within the virtual impactor stages.
Calculation shows that for a 5 percent token flow of a 1/2 cfm (14.15 £pm)
virtual impactor with 50 mm diameter filters the resulting face velocity is ap-
proximately 0.62 cm/sec.
Inspection of Figure 9 indicates that penetration increases with filter
face velocity and that the size at minimum collection efficiency (maximum
penetration) decreases with filter face velocity.
32
-------
u>
STAGNATION CHAMBER
(ONE FOR EACH STAGE)
SIGNAL OUTPUT
EXTRACTION FLOW
CONTACT ELECTRIC
SENSOR
V
FILTER FOR INLET
CONCENTRATION
MEASUREMENTS
PRESSURE
DROP TAPS
FILTER
FOR
OUTLET
CONCENTRATION
MEASUREMENTS
EXTRACTION
FLOW
Figure 7. A contact electric sensor technique.
Figure 8. Filter pressure drop technique,
-------
UJ
z
UJ
Q.
IO
5
I
0,5
0.2
UJ
g O.I
UJ
a.
0.05
0.02
0-01
0.94 cm/MC
0.42 cm/stc
0.21 cm/ttc
0.094 cm/tec
i I
RESIN IMPREGNATED FIBERGLASS FILTERS;
MONODISPERSE OOP
.2
.4 .6 .8
PARTICLE DIAMETER,
1.0
1.2
Figure 9. Efficiency of glass-fiber filters as a function of particle size
and flow rate.
Very little work has been found in the literature describing filtration
during cake buildup, and none at the low face velocities (<2 cm/sec) of interest
to this project. One of the investigators, Anderson,32 studied filtration ef-
ficiencies of various filter media collecting cotton picker machine effluent
(mostly lint free) at face velocities of approximately 50 cm/sec and 100 cm/sec.
Penetration decreased as the filter cake and pressure drop grew,.and the re-
sultant dependence of pressure drop on time was exponential. In fact, after a
cake had formed, pressure drop was an excellent indication of collection
efficiency.
Figure 10 shows pressure drop versus time, and therefore loading, for mono-
disperse latex particles (Dp = 0.088 ym), collected on nuclepore filters, at a
face velocity of 5 cm/sec.36 Figure 11 shows pressure drop and collection ef-
ficiency versus cake thickness for polydisperse ammonium chloride fume (Dp =
0.1 to 3.0 ym) collected by gravel bed filters and fiber filters at a face
velocity of 2.0 cm/sec.33 Figure 12 shows pressure drop versus cake thickness
for various face velocities and various fiber filters.3^ The work of LaMer as
reported by Billings34 is presented in Figure 13. Experiments performed at
GCA/Technology Division with polydisperse General Motors AC fine air cleaner
test dust (Arizona Road Dust) are presented in Figures 14 and 15.
-------
o
N
X
a.
<
Dp =0.088 m
V0 = 5 cm/sec
6 8 10
2O 4O 60 80100
t, minutes
200
500
IOOO
Figure 10. Measured dependency of Ap on time (three successive
experiments). 36
35
-------
UJ
o -6
1 .4
0 .2<
*
o GLASS FIBER HI E 60g
A GLASS FIBER HI E 40g
D GRAVEL d«l.5 mm
» GRAVEL d = 3.0 mm
0 COTTON FIBER INITIAL VALUE
.05
.15
o
.05 .1 .15
PARTICIPATE LOAD W,kfl/m2
Figure 11. Comparison of collection efficiency E and pressure
drop AP in gravel and fiber filters at Darcy air
flow velocity VD = 2.0 cm/sec and gravel layer
thickness L = 30 cm.33
36
-------
E
en
<0
CM
oo
6
0.20
0.15
0.10
O.OS
V0= 29 CM/SEC
200
400
600
800
1000
ACCUMULATION, I06 p/CM2
0.025
«o
111
M
<
111
cc
u
OT
OT
Ul
cc
ui
r 0.020
O.OIS
0.010
0.005
SYMBOL
.V0 = 14 CM/SEC
100
200
300
400
500
ACCUMULATION, 10* p/CM2
Figure 12. Resistance increase as a result of solid particle
accumulation in tests 7 through 10. Test mat
numbers refer to various glass fiber filter mats
(see reference 34).
37
-------
5.6
4.8
I
OJ
N 4.0
CD
Ul
(si
5
at
o
u
(k
LJ
U
z
55
ee
3.2
1.6
0.8
V0 =28 CM/SEC
N0=6xl06
LAMER ET AL.,
CC-5 FILTER
a -s 6.5 ^m
a = 0.126
L = 0.048 CM
WAX AEROSOL
o •= 0.3 ft m
13 CM/SEC
7 p/CMS
V0 « 4.7 CM/SEC
N0 = I07 p /CM3
TIME, minutM
Figure 13. Filter resistance during operation.31*
38
-------
90-
ARIZONA ROAD DUST
FIBERGLASS TYPE A
47 M.M. DIA.
x 0.53 cm/sec
O 1,07 cm/sec
A 1.64 cm/sec
D 10.7 cm/sec
10 15 20 25
COLLECTION-,*«/cm2
30
Figure 14. Pressure drop versus collection density at various filter face
velocities (GCA data).
39
-------
u
c
a."
<3
24
23
22
21
20
19
18
17
16
15
13
12
II
10
9
8
7
6
5
4
3
2
I h
ARIZONA ROAD DUST
FIBERGLASS TYPE A, 2
47,iun DIA, COLLECTION AREA=l5.5cm
Q - I I pm
V0= 1.07 cm/sec.
FULL LOADING = 3l.8mg/cm2
*
*
X
X
23456789 10
12 13 14 15 16 17 18
Figure 15. Pressure drop as a function of time for extended operation at
nearly constant concentration (GCA data).
40
-------
The foregoing curves shows that fiber filters behave quite differently from
nuclepore filters or gravel bed filters, in that fiber filters display a rather
steady increase in pressure drop as accumulation proceeds. Although it appears
at first glance that pressure drop versus loading is linear for a time for
nuclepore filters as shown in Figure 10, closer examination reveals that nucle-
pore filters clog at a very low pressure drop, in this case at about 2 in. H20.
Comparing Figures 14 and 15 with Figure 10 it is obvious that nearly linear
pressure drop increase proceeds to much higher values in the case of fiber
filters. The shape of a pressure drop versus time curve for a fibrous filter
under constant loading can be expected to be somewhat concave upward because the
penetration will decrease as the filter cake builds up, if it is assumed that
pressure drop is proportional to accumulation. From Billings,3k
^= V N /l-P(A))
dt o o \ /
where A = accumulation
V0 = filter face velocity
No = aerosol concentration
P(A) = aerosol penetration.
and
6 Ap = S p V D 6A
pop
where Ap = pressure drop
y = gas viscosity
Dp = particle diameter
SD = particle resistance coefficient
and assuming that S_ 1 f(A).
From this it can be understood why the pressure drop versus time curves of
Anderson's high face velocity tests showed an exponential form. Initial penetra-
tion was quite high (as much as 30 percent), and decreased as the cake formed.
The curve in Figure 15 from data gathered at GCA/Technology Division as well as
the curves in Figure 12 only hint at upward concavity. From the earlier dis-
cussion of filter penetration as a function of particle size and face velocity,6
it is anticipated that the filters behind the final virtual impactor stages
(those collecting the smallest particles) are the most likely to display upward
concavity-
41
-------
This effect may be minimized by reducing filter face velocity (increasing
filter area), or by choosing a filter of higher initial efficiency. From the
foregoing data it can be expected that pressure drop versus time curves are
nearly linear for the low filter face velocities to be used in the virtual im-
pactor. Another aspect of importance to this method of measuring accumulation,
is the slope of the pressure versus time (or cake thickness, A) for varying
particle size.
The Kozeny equation is presented in Happel and Brenner35
Ap
VoyL
36 (1 - e)'
E3D2
P
describes pressure drop across a porous bed
where e = porosity, fraction of void volume; and,
K = Kozeny factor (5.0 approx.)
L = length of bed.
The other parameters being the same as defined for the previous equation.
For close packing of spheres, e = constant, and it is expected that pressure
drop increases as particle size decreases for the same mass loading as shown in
Figure 16. Experimental data in the literature has not been found to support
the conclusions from Kozeny1s equation. For real aerosols e is probably a com-
plicated function of particle size, shape, compressibility, hygroscopicity, etc.
DP,
-------
SECTION 7
VIRTUAL IMPACTOR DEVELOPMENT
METHODOLOGY- DESIGN TOOLS
The main three laboratory tools used for the experimental development of
the virtual impactor were a Research Engineers, Ltd. air driven spinning disc
generator for monodisperse aerosol production, a G. K. Turner fluorometer for
quantitative analysis of uranine tagged aerosol, and an American Optical Series
10 microscope for particle sizing.
With this equipment a monodisperse, fluorometrically active aerosol was
introduced into the virtual impactor prototype. Aerosol deposition on the
virtual impactor filters as well as undesired aerosol deposition within the
virtual impactor was quantitatively measured with the fluorometer. The measure-
ment range of the fluorometer from 1 yg/liter of solution (deionized water for
these experiments) to 250 yg/liter, allowed quantitative analysis of extremely
small amounts of material.
The experimental procedure was as follows: (1) All impactor parts were
thoroughly washed with deionized water and the wash from each part was fluoro-
metrically analyzed to insure that the fluorescence level was not much different
from that of the deionized water, (2) the impactor parts were assembled with the
aid of a Jones and Lamson Machine Company optical comparator to accurately mea-
sure jet orifice diameter, collection nozzle diameter, and orifice to nozzle
distance, (3) 51 mm diameter, RAWP, Millipore filter media were weighed and
mounted in the stage filter holder, backup filter holder, and second (indepen-
dent) probe filter holder, (4) the completely assembled impactor stage with the
backup filter holder and the second probe filter holder were connected to iden-
tical probes in the wind tunnel, (5) critical orifices were u'sed to control flow
through the impactor, the token flow through the impactor stage filter, and the
second probe, (6) a solution of deionized water, ethyl alcohol, and methylene-
blue and fluorescein was prepared and placed in a syringe to supply the spinning
disc, (7) appropriate spinning disc speed and liquid feed rate were chosen and
the test was run taking care to exchange wind tunnel probes between the impactor
and second probe filter assembly to minimize any possible bias between probe
location in the wind tunnel, (8) after a suitable collection period the impactor
was disassembled and the filters were sectioned to provide wedges for microscopy
and for fluorometry, (9) the second or reference probe filter was photomicro-
graphed in several fields to record enough aerosol particles (usually around
50) to determine aerosol size, (10) wedges from the stage filter, backup filter,
and second probe filter were weighed and then washed in a known amount of
deionized water and the wash analyzed with the fluorometer to determine aerosol
concentration, and (11) the impactor parts were washed with a known amount of
43
-------
deionized water and the wash analyzed with the fluorometer to determine aerosol
losses.
Aerosol size was determined by particle diameter measurements from the
photomicrographs. The average diameter determined by measurement was converted
to aerodynamic diameter by
D
meas.
aero magnification power
where p is assumed to be 1.3 for uranine.
TESTS WITH FIRST VIRTUAL IMPACTOR BREADBOARD
Figure 17 is a cross-sectional view of the first stage of the initial ver-
sion of the virtual impactor breadboard, which was designed so that orifices
and collection nozzles could be changed and the spacing between the orifice and
nozzle could be varied.
The design also allowed for stages to be cascaded. Figure 17 is labeled
to indicate the manner in which internal surfaces were washed to measure aerosol
deposition.
Seven tests were performed with this breadboard using monodisperse uranine
(fluorescein) aerosol. Six tests were done using only one impactor stage, but
one test used two stages in a cascade arrangement. These tests resulted in the
redesign of the virtual impactor to further minimize aerosol collection by parts
other than the filters.
The size of the aerosol was kept in the 4 to 6 ym aerodynamic diameter
range, although great effort was not expended toward accurate size analysis for
each test because attention was primarily focused on minimizing losses within
the impactor. Further, there was a nonneglible satellite population produced
by the spinning disc generator whose size appeared to be near the limit of
resolution of the optical microscope used for size analysis, probably in the sub
to 1 micrometer size range. The mass contribution of this satellite population
was estimated to be about 10 percent of the total aerosol concentration.
Table 4 summarizes data collected during these tests. For each test, con-
centrations are reported in yg/m^ (of uranine), and as a percentage of the con-
centration measured by the second probe for the purpose of comparison between
tests that may contain incomplete data. For the column labeled "Can" the con-
centrations are from the surfaces labeled "Can, outer surface" and "Can, inner
surface" as labeled in Figure 17. For the column labeled "All other," concen-
trations are summed for those surfaces such as "Outer can," "Outlet plate,"
etc., that do not contribute significantly to the total impactor losses.
Figure 18 shows the various orifice and nozzle configurations that were tested.
Table 5 indicates which nozzle and orifice designs were used for each test, as
well as orifice diameter, nozzle diameter, and spacing between orifice and
44
-------
INLET TUBE
AND PLATE
STANDOFFS
NOZZLE
OUTLET
PLATE
CAN,OUTER SURFACE
CAN, INNER SURFACE
OUTER CAN
0 RING ASSEMBLY
0 RING
STAGE FILTER
FILTER SUPPORT
SCREEN
FILTER HOLDER
Figure 17. Cross-section of initial version of virtual impactor breadboard.
45
-------
Figure 18. Various orifice and nozzle designs.
46
-------
TABLE 4. VIRTUAL IMPACTOR TEST DATA (FIRST VERSION)
Date
8/4/75
No. 1
8/6/75
No. 2
8/20/75
No. 3
8/21/75
No. 4
8/26/75
No. 5
8/29/75
No. 6
Stage 1
Stage 2
9/4/75
No. 7
Stage
31.5
13.
7.8
3.
71.9
30.
62.1
29.
24.3
13.
23.6
22.
11.
11.
33.6
20.
filter
yg/m3
5Z
Wg/m3
3%
yg/m3
0%
yg/m3
2%
yg/m3
8%
Mg/m3
8%
7%
1%
yg/m3
9%
Backup
filter
_
-
20.4
8.5
19.3
8.0
45.3
21.3
24.0
13.6
14.5
14.0
55.8
34.7
Can
4.0
1.7
_
-
20.0
8.3
35.4
16.6
2.2
1.2
18.5a
17. 9a
17.3
0.6
3.9
2.4
Nozzle
60.0
25.8
132
55.2
9.3
3.9
18.4
8.7
35.3
20.0
15. 7a
15. 2a
12.2
3.0
41.3
25.7
Orifice
8.4
3.6
25.9
10.8
17.8
7.4
7.0
3,3
—
-
4.6a
4.4a
4.1
0.4
13.2
8.2
0-ring All
assembly other
_.
-
_
—
_
-
19.1
9.0
15.8
8.9
2.4a
2.3a
0.8
1.5
9.4
5.8
2.6
1.1
1.1
0.5
4.3
1.8
15.4
7.2
_
-
1.5
1.5
-
—
—
—
I
Inpactor
106.
45.
187.
78.
142.
59.
202.
95.
101.
57.
80.
78.
-
—
157.
97.
5
7
2
3
6
4
7
3
6
5
8
1
2
6
Second
probe
233
100
239
100
240
100
212.7
100
176.6
100
103.4
100
-
—
161
100
"stages 1 and 2.
TABLE 5. VIRTUAL IMPACTOR DIMENSIONAL DATA
Test
No.
No.
No.
No.
No.
No.
No.
1
2
3
4
5
6
7
Orifice
0.
0.
0.
0.
0.
0.
0.
0.
diam, In.
1195
1195
1195
1207
1235
1861
1193
1848
Nozzle
0.
0.
0.
0.
0.
0.
0.
0.
diam, in.
1187
1862
0660
0660
0660
0902
0661
1838
Orifice-nozzle
distance, in.
0.
0.
0.
0.
0.
0.
0.
0.
1204
1196
0615
0657
0635
0902
0652
0915
Design
Orifice
a
a
c
d
s.
c
c
c
Nozzle
b
b
f
f
f
f
f
f
Designed
aerodynamic
D50, urn
3
3
3
3
3
7
3
7
.5
.5
.5
.5
.5
.0
.5
.0
47
-------
nozzle. The aerodynamic cut diameter is also included, as predicted from the
impaction equation, when a value of $$ = 0.66 is chosen as a compromise between
the data from Conner,15 ifrS = 0.6, and the data from Hounam and Sherwood,14
^ = 0.73, when the impactor was operated at 1/2 cfm (14.15 £pm) as was the case
of these tests.
For the first three tests, the "0-ring assembly" which holds the stage
filter in place, was not washed for its contribution to impactor concentration
measurements because the 0-ring was glued to the metal assembly with rubber
cement and probably for this reason the fluorescent background levels were never
able to be brought down to an acceptable level. At that time it was reasoned
that the 0-ring assembly was small and its aerosol collection potential could be
ignored. After Test No. 3, however, the 0-ring was separated from the assembly,
both parts were carefully scrubbed with steel wool and solvent removing the
rubber cement, and these parts were analyzed for aerosol collection in subse-
quent tests. Another procedure instituted after Test No. 3 was to repeatedly
wash each impactor part until the background fluorescence level was reached,
record the wash volume and concentration for each repetition, and sum all these
contributions to determine the concentration for that part. With these pro-
cedural modifications, good agreement was achieved in Test No. 4 between concen-
trations determined by the second probe and concentrations determined by sum-
ming all impactor part concentrations. Test No. 5 was an abbreviated test
because the orifice design used drastically increased aerosol deposition on
the nozzle.
Evaluating impactor design in regard to losses within the impactor, tests
No. 1 and No. 2 demonstrated that orifice type a and nozzle type b were unequal
to the assigned task. In both tests, aerosol deposits were observed inside and
on the outside tip of the collection nozzle. Nozzle deposition measurements
confirmed what was observed visually. Test No. 3 resulted in much reduced nozzle
deposition, although orifice losses seemed unaffected. Test No. 4 attempted to
reduce deposition within the orifice by making the gas flow channeling less
abrupt, but this resulted in increased deposition on the can and nozzle. Micro-
scopic examination of the backup filter showed quite a few primary particles,
unlike other tests, indicating that this geometry reeuced the sharpness of size
cut. Orifice type e. was tried in Test No. 5, again toward the purpose of re-
ducing deposition within the orifice, but this resulted in heavy aerosol deposits
forming on the outside tip of the nozzle. The type e orifice made for this test
had galled in the inlet plate and had to be machined out, thus the missing data
point for Test No. 5.
From these tests, orifice type c and nozzle type f gave the best results.
It seems that an abrupt change in flow contour is needed within the orifice to
properly focus aerosol particles into the collection nozzle. Test No. 6 was
designed to give information about impactor losses and size cut characteristics
when the aerosol was somewhat smaller than the designed aerodynamic cut diameter
of the stage. An orifice of design c and a nozzle of design f were made with
larger diameters to provide a calculated cut diameter of 7.0 ym. For this test
two stages were cascaded, with the calculated cut diameters of 7.0 and 3.5 ym
48
-------
bracketing the test aerosol particle size. The result indicated that the
aerosol was unable to negotiate the sharp deflection of the airstream around
the nozzle and can, evidenced by deposition on the outside of the nozzle and by
annular deposition on the can at the nozzle base. Test No. 7 generally rein-
forced this conclusion, although deposition was less on the can and more on the
nozzle.
Further information regarding impactor loss sites was obtained from a test
conducted with high concentrations of polydisperse Arizona road dust (AC Fine
Air Cleaner Test Dust). During this test 550 mg of dust was collected on the
stage plus backup filter, whereas 714 mg of dust was washed from the various im-
pactor parts. Deposition was heaviest on the inside and outside of the nozzle,
on the can top (outer surface), within the can near the 0-ring assembly, and
on the inlet side of the orifice. Deposition to a lesser degree was also
observed on the standoffs and on the outer can (not to be confused with the
"can, outer surface") where the airstream turns to follow the passage between
the outer can, and the "can, outer surface" on its way toward the next stage.
To resolve some of these difficulties, the breadboard was redesigned as
presented in Figure 19. A taper of 20 degrees within the can was expected to
reduce losses within the nozzle. The flange that connects the can to the inlet
plate by means of the standoffs was lowered by almost 1/2 inch and reduced in
diameter compared to the former can to reduce obstruction to the gas flow.
The 0-ring assembly which holds the stage filter in place within the can
had also been a site of aerosol deposition, collecting from 2.3 to 9.0 percent
of the total aerosol concentration. The redesigned breadboard eliminated the
0-ring assembly from the internal volume of the can which allowed this pre-
viously lost fraction of aerosol to be collected on the stage filter.
As mentioned above, the satellite population of small aerosol produced along
with the primary particles by the spinning disc generator is estimated to be
10 percent of the total population. This is based on microscope examination of
the backup filters, where primary particles were observed to a significant degree
only in Test No. 4 and Test No. 7. In Test No. 4, the presence of primary par-
ticles on the backup filter was ascribed to poor separation characteristics of
orifice type d, and in Test No. 7, by experimental design, as the calculated
aerodynamic cut of the stage was greater than the test aerosol size. It was
further assumed that the secondary aerosol population should not be inertially
deposited on impactor surfaces because of their much smaller size. In fact,
satellite aerosols were rarely observed on stage filters.
Microscope examination was encouraging in the sense that primary particles
are found on the stage filter, and satellite particles on the backup filter,
indicating good size segregation.
TESTS WITH SECOND VIRTUAL IMPACTOR BREADBOARD
Seven tests were performed with the redesigned virtual impactor, whose re-
sults are presented in Table 6.
49
-------
HOLE SIZE
TAKE FROM SAMPLE
FILTER
0>cn
Figure 19. Redesigned collection nozzle.
50
-------
TABLE 6. VIRTUAL IMPACTOR TEST DATA (SECOND VERSION) WITH MONODISPERSE AEROSOL
1
Date
9/22
9/25
9/29
9/30
10/6
10/7
10/10
2
Test
no.
8
9
10
11
12
13
14
3
Stage,
25.8
34.4
27.4
33.8
13.3
66.4
5.9
4
Backup ,
34.5
39.4
35.2
38.0
93.3
0.8
84.9
5
Inner
nozzle,
07.7
18.0
17.8
12.5
-0
14.7
-0
6
Outer
nozzle,
__
3.9
6.1
-0
—
—
—
7
Submerged
nozzle ,
28. 9a
1.3
3.2
0.2
5.6a
2.9a
2.0a
8
Orifice,
5.1
6.2
1.9
6.9
2.8
1.1
2.6
9
£
impact or,
105.4
103.2
91.6
91.4
115.0
85.9
95.4
10
Second
probe ,
ug/m3
102.6
120.2
118.6
169.1
2.85
45.3
3.05
11
D
aero,
jam
6.54
7.69
5.18
6.33
-------
A pure uranine aerosol was used for tests 8, 9, 10, and 11, the aerosol
particles being less than spherical, although regular enough in shape to permit
size analysis. Geometric standard deviation was typically 1.2. For tests 12,
13, and 14, the aerosol was composed of 90 percent uranine with 10 percent
methylene blue added to enhance sphericity. The aerosol produced for tests 12
and 14 was largely submicron. The 3.66 pm aerosol generated for test 13 was
very good, the particles being spherical with a geometric standard deviation of
1.14, although a satellite population about 1/3 as numerous as the primary
population was sized at 1.38 yra, aerodynamic diameter. The satellite population
was calculated to contribute less than 2 percent of the aerosol mass and is
therefore considered insignificant.
Tests 8, 9, 10, and 11 were conducted for the purpose of determining col-
lection nozzle diameter and orifice to collection nozzle distance that minimized
losses. Column 12 of Table 6 indicates orifice diameter, Wj (width of jet), in
terms of the calculated aerodynamic cut diameter at the test flow rate of
1/2 cfm (14.15 £pm). Column 13 indicates both the collection nozzle diameter
and the orifice-nozzle spacing in terms of the orifice diameter. It can be ob-
served that test 11, using the largest nozzle diameter and spacing resulted in
the lowest loss within the impactor. Table 7 is presented to elucidate data
presented in Table 4. In Table 7 percentages are calculated in terms of the
total concentration measured within the impactor, whereas percentages in
Table 6 are calculated in terms of concentration reported by the second, inde-
pendent probe.
Test 8, in addition to providing information concerning nozzle diameter and
orifice-nozzle spacing, was performed with great care to evaluate the impactor
redesign. All parts were carefully washed to eliminate spurious fluorescence
prior to testing, and all parts were analyzed for aerosol collection. Results
were encouraging; only 3.3 percent of the aerosol was collected on parts other
than the filters, nozzle, and orifice. In subsequent tests, only the filters,
nozzle, and orifice were analyzed for aerosol collection, therefore the sum of
aerosol concentrations for the impactor (column 9) is generally less than con-
centration reported by the independent probe (column 10).
Work by Langmead and O'Connor37 in the U.K. in the late sixties on the de-
velopment of a personal centripeter indicated that losses were maximum for
aerosols slightly smaller than the cut diameter of the impactor stage. With
this observation in mind, tests 12, 13, and 14 were conducted for the purpose
of evaluating losses when aerosol size was much different from the cut diameter
of the impactor stage. For the case in which the aerosol has much smaller
particle size than the size which the stage was designed to collect, losses are
much reduced (tests 12 and 14). This is fortunate, because as an aerosol passes
through several impactor stages, that fraction destined for the last stages will
not be significantly depleted by losses in the preceding stages. On the other
hand, test 13 indicated that substantial losses could be anticipated when the
aerosol was composed of particles greater in size than the cut diameter.
52
-------
TABLE 7. IMPACTOR LOSSES
Test no. Stage and backup % Loss in impactor %
8
9
10
11
12
13
14
57.2
71.5
68.4
78.6
92.7
78.2
95.2
42.8
28.5
31.6
21.4
7.3
21.8
4.8
It may be noticed that token flow has not been mentioned in the preceding
description of the impactor design. Token flow for the design phase was held
constant at 0.5 fcpm because it was felt that introduction of this variable
would serve to obscure effects of design changes. Token flow was varied in
^subsequent tests with polydisperse aerosols (while keeping impactor geometry the
same) with the important result that increased token flow decreased losses, but
probably at the cost of reduced sharpness of cut.
The development phase reported above resulted in the decision that the
optimum impactor geometry would be as it was in tests 11, 12, and 14; that is
with the orifice to collection nozzle distance equal to 5/4 of the orifice diam-
eter and the collection nozzle diameter also equal to 5/4 the orifice diameter.
COLLECTION EFFICIENCY VERSUS PARTICLE SIZE
An important determination for an impactor is the curve of collection effi-
ciency versus particle aerodynamic diameter for a given stage, from which the
particle size which has a 50 percent collection efficiency may be found. This
is called the stage cut size or D5Q. The slope of this curve at the DSQ point,
or the Dgi,. minus the Die P°i-nts of this curve divided by the D5g point (geo-
metric standard derivation) are measures of the stage sharpness of cut. Fig-
ure 20 illustrates this concept.
It was intended, at the onset of this program, to completely define this
curve for at least one stage of the final recommended design. The GCA aerosol
laboratory has the necessary equipment to perform this calibration given that a
fairly accurate measure of particle density may be assumed.
However, as the program continued, other laboratory and field tests were
deemed more important toward developing an instrument capable of providing on-
site size distribution information and a calibration curve was not obtained.
As noted in an earlier section, values of ^ for the impaction equation from
the work by Conner15 and from Hounam and Sherwood14 were considered in reasonable
53
-------
100
D50 D84
AERODYNAMIC DIAMETER
Figure 20. Collection efficiency versus particle size.
54
-------
agreement with each other. The calculated cut size of the virtual impactor,
obtained by taking an average of these values to obtain ty% = 0.66, was con-
firmed for those tests performed with monodisperse aerosol. Data in Table 6
supports the proposition that the calculated cut sizes are quite correct when
one compares percentages caught on the stage and backup filters with the mea-
sured size of the aerosol and the calculated cut size.
Table 8 presents data taken near the end of the program toward establish-
ing the cut size for the virtual impactor. The aerosol was 90 percent uranine
and 10 percent methylene blue, displayed good sphericity, and was assumed to
have a density of 1.3 g/cm3 for the purpose of determining aerodynamic diameter.
The aerosol was tainted, however, by the presence of the usual secondary or
satellite population. Concentrations obtained by fluorometry therefore misrepre-
sent the true segregating ability of the virtual impactor. In an effort to more
accurately measure the capture efficiency, equal areas of the stage, backup, and
second probe filters were examined under the microscope for test no. 37. Par-
ticle counts of the primary aerosol, also presented in Table 8, indicate better
segregation than was indicated by the fluorometric analysis. Again, data in
Table 8 supports the calculated cut size.
TABLE 8. CUT SIZE DATA
Test no.
37
38
37
By particle count
Stage 1,
7.0 urn
136 ug/m3
69.3%
156 vm/m3
35.1%
275
85.7%
Stage 2,
2.0 jim
56
28.5%
283
63.8%
46
14.3%
Backup
<2.0 urn
4.3
2.2%
4.9
1.1%
0
0
Second probe,
total
235
120%
Not obtained
483
150
Daero
aerosol
7.4 urn
5.3 pro
Coaaent
1.5 pm secondary
aerosol present
Secondary present
By count of primary
7*4 pm aerosol in
55
-------
SECTION 8
PARTICLE SENSING METHODS
PRESSURE SENSITIVE TRANSISTOR
In addition to the three particle sensing techniques described in the
original proposal; i.e., optical, electrical and pressure drop, a fourth method
was investigated experimentally: impact momentum (or energy) detection. This
technique is based on the detection of particles impacting on a diaphragm which
in turn is part of a capacitor or other component whose electrical properties
are altered by the deflection of the diaphragm. The device used in this par-
ticular set of tests was a pressure sensitive transistor (trade name PITRAN)
whose collector current is a function of a force (or pressure) transmitted by
the integral diaphragm onto the base-emitter junction. A full scale output of
2 volts is obtained with a pressure of 1/4 psid applied to the transistor top
which forms the diaphragm.
The PITRAN was installed in an impactor configuration such that the dia-
phragm was used as the impaction surface. By recessing the transistor within a
stagnation volume behind an orifice facing the impactor nozzle, the system was
made into a centripeter-sensor combination. The output of the pressure-sensitive
transistor was a.c.-amplified and displayed on an oscilloscope. The main
problem encountered was the noise level expectedly associated with the turbulence
of the air jet. This noise level was found to be a strong function of flow-
rate; i.e., of air velocity of the jet. At about 30 m/s the noise component of
the signal was about 2 mV, whereas at 240 m/s it reached 50 mV. When large par-
ticles of Arizona road dust (estimated size >20 ym) where injected into the im-
pactor inlet distinct signal pulses were discerned on the oscilloscope display,
with a signal/noise ratio of the order of 2 to 4. Smaller particles could thus
not be detected as their signal was buried in the noise. An attempt was made to
perform a frequency analysis of the noise-signal combination in order to dis-
tinguish the signal by a possible frequency discrimination but this attempt was
not successful.
The turbulence-generated noise appeared to be an intrinsic limitation of
this technique, which was therefore not pursued.
OPTICAL SENSING OF PARTICLE CONCENTRATIONS WITHIN THE VIRTUAL
IMPACTOR STAGE
The optical extinction method of measurement was pursued to perform aerosol
concentration sensing within the centipeter stage. The electronic circuitry
designed for this purpose is depicted in Figure 21. The optical path is shown
in Figure 22. A reference source is derived from the laser and transferred to
56
-------
10
TELEDYNE
PHILBRICK 4455
REFERENCE f ~t ~
BOTTOM VIEW
BELL a HOWELL 529
Figure 21. Electronic schematic.
-------
00
VIRTUAL
IM FACTOR
STAGE
FIBER OPTICS
REFERENCE DETECTOR
SIGNAL DETECTOR
Figure 22. Optical sensing system.
-------
the reference diode by fiber optics directly. It may be noted that where
several optical stages were employed, only one reference channel is required.
The signal is taken from the output of the centripeter stage (Figure 6). Suit-
able gain was provided by the feedback shown and the two outputs connected to
the divider circuit (Teledyne Philbrick 4455) which yields 10 I/IO volts output.
Two successful experimental runs were made. In the first, a Wright dust
feeder was connected to the inlet of the virtual impactor stage with a calcu-
lated cutoff of 3.5 vim. All larger particles therefore entered into the optical
detector volume. From the size fraction data provided with the AC fine air
cleaner test (Arizona road dust), a mass median diameter of 14.5 ym was derived.
The procedure adopted after some experimentation was as follows. With the en-
tire apparatus clean a fairly low dust feed rate was provided to the impactor
inlet for a measured 5 minutes. A reading on a digital voltmeter of the output
of the divider was taken prior to the initiation of dust feed and again after
the dust had been drawn completely out of the optical path. During the run,
measurements of I/IO were taken every 15 seconds and subsequently averaged. The
filter was weighed to obtain a gravimetric measurement. Knowing the flow rate,
and assuming a value for particle density p = 2.5, optical extinction coefficient
k = 2 and A = 1.65 x 10~^ cm2, two values of number concentration may be derived,
the gravimetric and the optical value. The results are plotted for several in-
creasing dust feed rates in Figure 23. Some scatter is observed in the data and
it was noted that I0 had declined to 70 percent of its initial value after five
runs at fairly high overall concentration, due to dust deposition on the optics.
A second run was made with another impactor stage ahead of the optical mea-
surement stage. This first stage had a nominal cutoff of 3.5 urn and the im-
pactor at the optical stage had a nominal cutoff of 1.5 ym so that the optical
volume received particles between 1.5 and 3.5 ym. The measurements were made as
before except that the readings of I/Io were only used after an equilibrium had
been reached, usually after about 1 minute. The results of this run are shown
in Figure 24. As can be seen there is less scatter in this data, but the slope
is lower.
In both cases (Figures 23 and 24), the optical readings gave higher number
concentrations than the gravimetric measurements which is most probably due to
simplifications in the optical model. However, losses to the gravimetric method
were not trivial and may reasonably have contributed to this slope.
Improvements may be made to the optical system by inhibiting the contamina-
tion of the optics. In the system tested the surfaces were recessed about 0.5 cm
into the chamber wall and this undoubtedly reduced the problem. Further
recessing as well as simple baffles would result in additional improvement.
Purge air around the lenses would improve the situation considerably at the cost
of much greater complexity. In any redesign of the instrument for optical
sensing, a reduction in the volume without a reduction in optical path length
would decrease the time required to obtain a steady reading of I/I0.
ELECTRICAL SENSING
Two methods have been considered for measuring the output of a virtual im-
pactor stage using electrical sensing. One of these is to use corona charging
59
-------
OPTICAL, NUMBER/ml
Figure 23. Aerosol number concentration by gravimetric versus optical
detection for Arizona road dust, aerosol particles greater
than 3.5 ym.
10
3O 40 50 60 70
OPTICAL .NUMBER/ml xlO4
80
90
100
Figure 24. Aerosol number concentration by gravimetric versus optical
detection for Arizona road dust, aerosol particles less
than 3.5 ym but greater than 1.5 ym.
60
-------
of the dust as in an electrostatic precipitator, and to collect the dust on an
appropriate electrode. While this method would undoubtedly yield usable sig-
nals, the disadvantages are several. Highly charged dust precipitates every-
where, thus much of it may not be collected on the desired surface. The elec-
trical noise of the corona discharge will tend to interfere with the small
electrical signal currents and implementation is inherently complicated.
The alternative was to use contact charging, a relatively little understood
phenomenon whereby dust contacting an electrode at high velocity (100 m s-1)
yields a charge to the electrode. The response of such a system is not quite
linear and exhibits some temperature dependence.
Figure 25 shows the design of the contact charging device. The dust laden
gas is forced through a constriction which (1) forms a critical orifice to en-
sure constant gas flow and (2) impacts the dust particles onto an electrode.
This electrode is coupled to a sensitive electrometer for current measurement.
Initial tests were performed with clean filtered air to determine back-
ground current signal levels. These were determined to be about 10~12 A using
the first test fixture which was fabricated entirely out of teflon with the ex-
ception of the sensing needle (steel). Subsequently a series of tests was per-
formed with Arizona road dust at concentrations ranging from about 1 mg/m3 to
300 mg/m3, and a monotonic increase of signal current with dust concentration
was observed. During these tests an unusual phenomenon was observed: after one
or two hours of operation the polarity of the signal suddenly reversed from an
initial positive current to a negative one. This phenomenon was consistently
observed each time the test was performed. It was suspected that a gradual
charging up of the teflon body of the test fixture resulted in a sudden break-
down transferring this charge to the sensing needle when the potential reached
a critical value. In order to preclude this problem a similar test fixture was
then fabricated out of aluminum with a small insulating ring separating the
body of the fixture from the sensing needle. The metallic body was then elec-
trically grounded to provide a leakage path for any charge buildup. This new
device was tested in a similar manner as the teflon fixture and no evidence of
polarity reversal was observed. Background signal current dropped to about
10~13 A at the same time as signal currents 10~10 to 10~9 A were detected for
dust concentrations between about 30 and 300 mg/m3, respectively. These read-
ings were obtained with a flowrate of 0.55 liters/min typical of the token flows
required for a virtual impactor stage.
This test fixture was further tested to determine repeatability and reli-
ability of the method. Arizona road dust concentrations were varied between
about 3 mg/m3 to 300 mg/m3 as measured by a GCA model RDM 101 dust monitor which
had previously been calibrated against gravimetric samples over the subject con-
centration range. For concentration levels up to about 120 mg/m3 the current
from the sensing needle remained steady for test periods of about 20 to 30 min-
utes, but for concentration levels greater than about 120 mg/m3 to about
300 mg/m3 the current was observed to decay from a higher value down to values
associated with the 120 mg/m3 concentration level. Data for the tests up to
about 120 mg/m3 (X's) and the data collected earlier (circles) are presented in
Figure 26 which indicates a measure of repeatability for this method.
61
-------
-FROM VIRTUAL IMPACTOR
-TO PUMP
-CRITICAL ORIFICE
-CONTACT ELECTRODE
ELECTRICAL SIGNAL TO ELECTROMETER
Figure 25. Electrical sensing design.
The test fixture was run at temperatures elevated above ambient. After some
experimentation to find a heating method which did not disturb the normal opera-
tion of the device, the following was adopted. A few turns of resistance wire
was bifilar-wound around the device and connected to the secondary winding of a
filament transformer. The center tap of this winding was grounded to the body
of the test device. The input power to the primary of the filament transformer
was connected to a Variac transformer. The whole assembly was insulated ther-
mally with asbestos cloth, and a thermocouple probe attached to the body of the
device, thermally insulated from direct contact with the heater element.
Trial runs were made with the heater energized at very low power to check
that no disturbance to the normal operation was occurring on account of the
62
-------
IX10"
IX10'
,-10
z
U
CC
-------
device frequently responded more as a voltage source than as a current source.
This could be determined by the indicated electrometer current being strongly
dependent on input impedance.
In some instances this voltage source characteristic (whether positive or
negative), and the polarity reversal disappeared if the device was permitted to
cool slightly, and in other cases it required a return to room temperature. The
rate of heating and cooling was a significant factor, giving more erratic
responses for higher rates. This latter effect is presumably due to thermo-
electric effects which have their most dramatic impact with greatest temperature
differences. Efforts were made to make all the device components of aluminum.
The collector electrode was originally stainless steel and a new aluminum one
was fabricated, but with no improvement in stability. Since the electrode lead
attachment could not be made at the same point as the ground return an intrinsic
difficulty is seen, owing to inevitable temperature differences.
It is believed, however, that most of these problems could be substantially
mitigated. The change in sensitivity after a period with high concentrations is
most probably due to the aggregation of dust in the vicinity of the impaction
site and would surely yield to improved aerodynamic design that would enhance
scouring. Careful attention to intermetallic connections would also help to
make the unit operable at sufficiently elevated temperatures. However, it is
judged that in the scope of the current program that other approaches are more
viable, especially when one considers the small magnitude of the current (down
to 10~12 A) that has to be measured accurately at a location several meters from
the device itself. Thus, work was discontinued on the contact electriciation
method of particulate detection and measurement within this program.
PRESSURE DROP SENSING METHOD
The pressure drop sensing method was the selected method within the present
program because of its simplicity, reliability, repeatability, and self-
calibrating ability. Data, which are more conveniently presented in the next
section on lab and field tests, support this selection.
64
-------
SECTION 9
VIRTUAL IMPACTOR-PRESSURE DROP LAB AND FIELD TESTS
LAB TESTS
After the impactor design had been reasonably well optimized using mono-
disperse aerosol, the impactor was modified by fitting it with pressure taps
to permit the sensing of the pressure drop across the stage filters. Figure 27
is a schematic of the modified three-stage virtual impactor prototype. Fig-
ures 28 and 29 are photographs of this final version of the device as used during
the subsequently reported laboratory and field tests.
Tables 9 and 10 contain data collected during the first trials of the pres-
sure drop sensing method. During Tests 15, 16 and 17, some further modification
was performed by varying the second stage collection nozzle diameter. Arizona
road dust was used for all tests in Tables 9 and 10 in concentrations high
enough that deposition on the various internal impactor surfaces could be
readily observed, thus facilitating those changes in the second stage collection
nozzle required to minimize losses. It was encouraging that the ratio of col-
lection nozzle diameter to orifice diameter determined to result in minimum
losses using Arizona road dust was 5/4 as had been determined from the mono-
disperse aerosol tests. However, the orifice to collection nozzle distance was
closed down to one orifice diameter, rather than 5/4 orifice diameter, without
any apparent detrimental effect. This, it was felt, would result in a sharper
cut size.
In regard to the pressure drop sensing method, difficulties were encountered
in Tests 15 through 22 in that a linear pressure drop versus time curve was not
obtained for Stage 1. Careful inspection of the first stage filter holder,
revealed a leak between the downstream side of the filter and the surrounding
volume which carries the main impactor flow. Sealing of this leak resulted in
linear pressure drop versus time curves in Test 23, which are presented in Fig-
ure 30. Curves for Stage 2 were all linear for these tests.
For the following tests, data for which are presented in Table 11, the im-
pactor was set up and operated in the same manner as it had been for Test
No. 23 (Table 9). Exceptions included two tests during which the token flow was
increased, and one test in which a novel geometry was tried in the second stage.
Aerosols sampled during these tests included Arizona road dust (AC Fine Air
Cleaner Test Dust), coal dust, fly ash, and iron oxide. The calculated aero-
dynamic diameter cut size was 7.0 ym for the first stage and 2.0 ym for the
second stage for those tests conducted with a nominal token flow of 0.5 £pm;
i.e., for those tests conducted as in Test No. 23. For the two tests in which
token flow was increased, total inlet flow also increased due to the fact that
65
-------
GAS INLET
NOZZLE
ORIFICE
BACKUP FILTER
AP,
xxxxxxxxxxxxxxx
FILTER
TOKEN FLOW
TOKEN FLOW
V MAIN GAS FLOW
Figure 27. Virtual impactor schematic.
66
-------
Figure 28. Final version of prototype virtual cascade impactor, disassembled.
Figure 29. Final version of prototype virtual cascade impactor, assembled
with backup filter.
67
-------
TABLE 9. VIRTUAL IMPACTOR DIMENSIONAL DATA FOR TESTS 15 THROUGH 23
CO
Date
10/19
10/27
a.m.
10/27
p.m.
10/28
10/29
10/31
11/3
11/4
11/5
Test
no.
15
16
17
18
19
20
21
22
23
Stage 1
orifice,
inches*
0.185
0.185
0.185
0.185
0.185
0.185
0.185
0.185
0.185
Stage 1
nozzle,
inches
0.229
0.229
0.229
0.229
0.229
0.229
0.229
0.229
0.229
Stage 1
orifice-
nozzle
distance,
inches
0.184
0.187
0.187
0.187
0.187
0.187
0.187
0.187
0.187
Stage 2
orifice,
inches
0.0835
0.0835
0.0835
0.0835
0.0835
0.0835
0.0835
0.0835
0.0835
Stage 2
nozzle,
inches
0.0802
0.0802
0.1417
0.1026
0.1026
0.1026
0.1026
0.1026
0.1026
Stage 2
orifice-
nozzle
distance,
inches
0.0817
0.0803
0.0887
0.0787
0.0787
0.0787
0.0787
0.0787
0.0787
Filter
media used
Fiberglass A
VM-1
VM-1
VM-1
VM-1
Fiberglass A
Fiberglass A
Fiberglass A
Fiberglass A
The use of inches was dictated by the available length measuring
instrumentation.
-------
TABLE 10. VIRTUAL IMPACTOR TEST DATA (WITH ARIZONA ROAD DUST)
ON
VO
Test
no.
15
16
17
18
19
20
21
22
23
Stage 1
cone. ,
mg/m3
0.98
torn
66.1
65.5
72.4
30.8
48.1
44.4
34.1
Stage 2
cone. ,
mg/m'
1.17
131.4
27.1
116.2
113.3
76.8
130.7
124.4
123.0
Backup
cone. ,
mg/m3
0.56
46.9
134.8
43.8
56.6
44.4
61.1
67.3
96.7
Total3
impact or
cone. ,
mg/m3
2.71
—
228.0
225.5
242.3
152.0
239.9
236.1
253.8
Second
probe
cone . ,
mg/m3
3.97
250.9
263.1
234.6
261.3
162. 8C
244.2
243.1
269.4
Stage 1,
%
36.2
—
29. Od
29. Od
29. 9d
20.3
20.1
18.8
13.4
Stage 2,
%
43.1
—
11.9
51.6
46.7
50.6
54.4
52.7
48.5
Backup,
%
20.7
—
59.1
19.4
23.4
29.2
25.5
28.5
38.1
Impact or
as % of
2nd probe,
%
68.3
—
86.7
96.1
92.7
93. 4e
98. 2e
97. le
94.2
Impactor
loss, %
31.7
—
13.3
3.9
7.3
6.6
1.8
2.9
5.8
Test
duration,
minutes
173
35
36
36
36
43
60
43
60
Impactor flowrate = 14.25 (main flow) + 0.6 + 0.5 (token flows) « 15.35 Kpm.
Second probe flowrate = 15.0 fcpm.
CThe Wright Dust Feeder failed; therefore concentration is reduced in this test.
Replica tests for first stage; note repeatability for 7.0 \an size cut.
eReplica tests (all conditions similar).
-------
0.9
I I I I
0.8
-0°
0 AP,
0.7
oo
oo
O.S
.00
5 0.5 -
o
5
tn
o>
o 0.4 -
c
oT
<3 *xx
0.3
oo
.00
opeiTxx'
xxo*
oo
.000
ooo
XXX
XX
XX X
xxxxx
xxxxxxx AP,
r
READOUT
RESOLUTION
(0.01 inches
0.2
O.I -
I I I
J I I I
04 8 12 16 20 24 28 32 36 40 44 48 '52 56 60
TIME, minutes
Figure 30. Data for Test 23.
70
-------
TABLE 11. VIRTUAL IMPACTOR DATA (POLYDISPERSE DUSTS)
Test
no.
23
24
25
34
35
36
26
27
28
29
30
31
32
33
Stage 1
cone. ,
34.1
31.1
34.3
34.5
35.4
64.4
9.4
30.6
_ a
126.9
42.6
34.8
41.3
29.2
Stage 2
cone. ,
123.0
113.4
108.7
132.2
130.3
142.5
13.2
62.4
124.1
98.9
38.8
105.7
94.0
86.6
Backup
cone. ,
96.7
79.8
84.6
105.0
80.1
57.1
2.7
9.8
12.2
8.2
34.8
128.7
112.0
64.3
total probe ^"^ fi,£* Dust TMt
Bg/m3 na/m3 m n- P01
253.8
224.3
277.6
271.7
245.8
264.0
25.3
102.8
-
234.0
116.2
269.2
247.3
180.1
269.4
274.1
266.4
321.6
267.0
276.9
24.5
143.2
349.1
288.0
138.6
357.1
316.4
294.7
60
36
36
36
26
30
60
72
60
84
40
60
60
60
0.5
0.5
0.5
0.5
1.0
1.7
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
Arizona road dust
Arizona road dust
Arizona road dust
Arizona road dust
Arizona road dust
Arizona road dust
Coal dust
Coal dust
Fly ash
Fly ash
Iron oxide
Iron oxide
Iron oxide
Iron oxide
23
24
25
34
35
36
26
27
28
29
30
31
32
33
Inpactor distribution
Stage 1,
13.4
13.9
15.1
12.7
14.4
24.4
37.1
29.8
-
54.3
36.7
12.9
16.7
16.2
, Stage 2,
t
48.5
50.5
47.7
48.7
53.0
54.0
52.2
60.7
-
42.2
33.4
39.3
38.0
48.0
, Backup,
1
38.1
35.6
37.2
38.6
32.6
21.6
10.7
9.5
-
3.5
29.9
47.8
45.3
35.8
•rrs 'Tr
2nd probe lo"'
* *
94.2
81.8
85.4
84.5
92.1
95.4
103.4
71.8
-
81.2
83.8
75.4
78.2
61.1
5.8
18.2
14.6
15.5
7.9
4.6
-3.4
28.2
-
18.8
16.2
24.6
21.8
38.9"
Stag* 1
10' sec-1
3.67
3.96
3.01
2.88
2.93
3.55
0.69
0.90
0.42b
0.32
0.47C
1.92
1.44
1.81
Stage 2
3.48
3.09
4.43
2.90
3.71
5.20
1.17
1.57
1.10
1.02
1.85
2.20
2.11
2.20
Total
**£*' 1st half
CD,
105 ,ec-l
-
-
-
14.6 8.37
13.8 8.87
15.3 8.88
-
7.44
8.99
6.94
72.0>>
7.81
6.70
6.88 5.19
filter
2nd half
CD. ,
105 sec"1
— Fiberglass A
Gelnan VM-1 aedla
Mllllpore RAHF Bella
9.98 Fiberglass A
9.30 Fiberglass A
9.02 Fiberglass A
- Fiberglass A
Fiberglass A
— Fiberglass A
~ Fiberglass A
— Khataan backup
Fiberglass A
Fibers ls»e A
5-1' Fiberglass A
"Filter fumbled. d*t« lost.
*Bi)se*i on estimate of Stage 1 fait, by subtracting Stage 2 t backup Ant froa 82 percent of 2nd probe Awe.
Unreliable because pressure change here vas only 0.02 in. HjO.
dA Whatman type filter paper us* used here, which apparently IB unsuitable for sir Banpllng.
*Hlgh loss due to trial of novel Stag* 2 geometry.
-------
the same critical orifice was used downstream of the backup filter for flow con-
trol. Accordingly, the calculated aerodynamic cut diameter decreases. Since
the dependence is inversely proportional to the square root of the total flow,
the actual particle cut size was barely affected.
The total flow and token flows as well as the flow through the independent
filter were measured before and after each test with calibrated rotometers. The
total flow and independent filter flow decreased during the tests, always by a
small amount, because of the method of flow control employed; that is, the use
of critical orifices between the filter and the air pumps. As pressure builds
across the filter, the gas density between the filter and the critical orifice
decreases and the volume flow upstream of the filter, where pressure is near
ambient, decreases. The flow change for the token flows was always undetectable
because the pressure drop was very slight, usually less than 1 in. H20 =
0.003 atm, but the flow change for the independent filter was often significant,
with pressure drops across that filter sometimes as great as 2 psi = 0.14 atm.
After a certain number of tests, the area of the impactor backup filter was in-
creased, by a factor of about 4, to decrease the pressure drop across it, thus im-
proving impactor flow control. The test data as presented in Table 11 have been
corrected for flow variation. Concentrations, as reported, have been calculated
using average flow, as determined by flowmeter measurements. The nominal impactor
flow for these tests was 1/2 cfm = 14.15 £pm.
Discussion of the test data in Table 11 will begin with the assessment of
the pressure drop particle sensing method in mind. Pressure drop across the
filters was recorded as a function of time. For the tests performed with
Arizona road dust and coal dust a Wright dust feeder was employed, which can be
relied upon to generate dust at a fairly constant rate. For the tests performed
with fly ash and iron oxide an improvised auger type feeder with an aspirator
for dust pickup and dispersion was employed, the feed rate for which was probably
not as constant. Figures 31, 32, 33 and 34 show the pressure drop versus time
relationship for coal dust, fly ash, iron oxide, and Arizona road dust, respec-
tively. It may be observed that the plots are rather straight, after some
initial transient. The reasons for the initial transient, or nonlinear response,
are unclear, although initial cake formation and initial buildup of aerosol con-
centration in the dust chamber are easily cited. In these figures, API denotes
the pressure drop across the first impactor stage, and Ap2 denotes the pressure
drop across the second stage. It will be observed that the initial pressure drop
across the first and second stage filters are different in Figures 31, 32, and
33. This is because the critical orifices for the token flow are exceedingly
small and slightly different. The Stage 1 token flow was measured as 0.6 &pm,
and the Stage 2 token flow as 0.5 £pm. The token flows in Figure 34 were more
nearly equal, at slightly less than 1.0 £pm, resulting in equal initial pressure
drop across both stage filters. It will be observed that the initial pressure
drop across the backup filter is higher for Figures 30 and 31, compared with
Figures 32 and 33 because a larger area backup filter was used in the latter
two tests. It may also be observed that some of the Ap versus t curves for the
JU
The use of English units of pressure; e.g., inches H20 and inches Hg was in-
dicated by the scale calibrations of available pressure drop sensors.
72
-------
9
BACKUP
XXX
X X
X X
XXX
I
o
CM
X
«. /v
~ 0.4
0.
0
0.3
AP,
x x
x x x $ » •
§ 4
e e
e o
o o o o o o
e o
0.2
O.I
8 16 24
32 40
TIME, minutes
48 56 64 72
Figure 31. Test number 27, coal dust.
73
-------
8
M
«
•*=
O
0.5
0.4
x x
•u
0.2
k.
O-'O"
i i
.
x x * APBACKUP
* X
" «
" «
K X
AP
xxxx
X X
• • •
...
...
••
••••
..
«•
*•
i i i
.i - 1 - 1 - 1 - 1 - 1 - 1
1 - 1 - 1
24 32 40 48
TIME, minutts
56 64 72 80 8t
Figure 32. Test number 29, fly ash.
-------
.
7.0
6.0
5.0
4.0
/I
1 T
x
x * APBACKUP
X X
X X
t- x«
X X
•g 0.5
•
o.
0.4
0.3
0.2
O.I
AP,
X X X X X X
X X X X
' x x x x
XXX _ ©
X • •
e o
e e
e e
AR,
X X
16 24 52
TIME, minutes
40 48 56
Figure 33. Test number 32, iron oxide.
75
-------
OVJ
25
20
1 5
10
x
* X
x x
X X
X X
x x
x x ^p
* x TOTAL
x x x FILTER
- X X X
x xx «
6.. xi
I.I
1.0
0.9
0.8
0.7
0.6
O5i
^
S
-
o
e
o ^
o • °
o
0 » *
O
©
o
o ° AR
e * x x x
,-ft.g-x f M x f " 1 " i * . * i i i i i i ,
12 16 20 24
TIME,minutes
Figure 34. Test number 35, Arizona road dust.
76
-------
backup filters display some slight upward concavity, probably due to the com-
bined effects of higher face velocity and small particle size, as discussed in
Section 5.
From the pressure drop information and the mass collected on the filters a
drag coefficient, CD, representative of the slope of a pressure drop versus cake
weight curve (different from pressure drop versus time as presented in Fig-
ures 31 through 34), may be calculated as follows:
Ap_
*2 2.
c v A = Ap (A2)
D ~ Awt Awt Q (Awt)
where CD = drag coefficient
Ap = final pressure drop across filter minus initial pressure
drop across filter
v = filter face velocity
Awt = weight of material collected by filter
A = filter area
Q = flowrate through filter = vA.
Examination of the drag coefficient data in Table 11 impactor loss data in-
dicates that increased token flow reduces impactor losses judging from Test
No. 35 and Test No. 36. Test No. 23 appears to be spurious and Test No. 26 was
done at a much lower dust concentration than usual.
NEW BEDFORD FIELD TESTS
Virtual Impactor Stack Tests
Two trips were made to New England Power and Light which graciously allowed
GCA to test their No. 3 oil-fired boiler downstream of an electrostatic pre-
cipitator, within a horizontal duct with an approximate square cross section of
2m x 2m. The stream temperature was 155°C (310°F).
Tests conducted on January 21, 1976 were of preliminary nature. Problems
were encountered in control of gas flows due to vapor condensation and icing in
the flow lines. Initial filter pressure drop data were worthless because of
poor flow regulation. These shortcomings were corrected for the secoul visit by
installing critical orifices for token flow control within the stack environ-
ment and by controlling main impactor flow using a calibrated orifice and dry
gas meter downstream of a condenser and dessicant column. This same method was
used for flow control of the Andersen In-Stack impactor, which was run in par-
allel with the virtual impactor in these tests.
77
-------
Flowrate through the Andersen reference impactor was maintained near
14.15 £pm (1/2 cfm) at stack gas conditions. Manipulation of the impaction
equation yields:
D
1/2
where the subscript 1 refers to stack conditions and the subscript o to ambient
conditions. Application of this relationship to Andersen 2000, Inc. calibration
data results in the 50 percent aerodynamic cut diameters of Table 12.
Flowrate entering the virtual impactor was maintained at 18.3 £pm, that flow
being determined by the sum of the 14.15 £pm (1/2 cfm) through the backup filter
and the 2.1 &pm token flow (see Appendix for token flow calculation) through
each of the two stage filters.
However, for the conditions under which these tests were conducted the
predicted aerodynamic size cut is altered less than 5 percent from the cut sizes
of 7.0 um for the first stage and 2.0 vim for the second stage, which had been
tentatively assigned to the virtual impactor at a flowrate of 14.15 &pm (1/2 cfm)
in the lab.
Virtual impactor data obtained during the New Bedford in-stack tests were
corrected for token flow effect as discussed in a preceding section of this
report. The Appendix, at the end of the report, is included to show how this
correction procedure is implemented.
The test run on 29 January was a parallel test between the virutal impactor
and the Andersen impactor. Results for these tests are presented in Tables 12
and 13. Salient features of the 29 January test are that total average con-
centration agreed well, 13.3 mg/m3 for the Andersen impactor and 15.4 mg/m3 for
the virtual impactor, and that size distribution is almost identical for the
two devices. Figure 35 shows the cumulative size distribution as measured by
the two impactors.
Unfortunately, this test was marred by a failure in the pressure sensing
line of the first stage of the virtual impactor, which necessitated its removal
from the stack for repairs. The 4-inch port available for these tests just
barely allowed the insertion of the virtual impactor prototype into the stack,
and the attendant shaking and jarring of the impactor most likely shook loose
some of the cake formed on the stage filters; thus pressure drop across the
stages was not wholly regained after restarting. It is interesting to note that
the pressure drop across the backup filter was regained, probably due to the
much tighter cake formed by the fine particles at the much higher face velocity
and the much higher pressure drop resulting therefrom.
78
-------
TABLE 12. ANDERSEN IN-STACK IMPACTOR DATA OBTAINED DURING IN-STACK TESTS AT NEW BEDFORD
vo
January
Stage
Probe
1
2
3
4
5
6
7
8
Backup
Total
Aero
dia,
>15.5
> 9.8
> 6.3
> 4.5
> 2.8
> 1.4
> 0.9
> 0.6
< 0.6
Awt,
mg
5.2
3.8
2.4
0.9
0.8
0.7
1.0
0.4
0
3.6
18.8
29
Cone. ,
mg/m3
3.69
2.70
1.70
0.64
0.57
0.50
0.71
0.28
0
2.55
13.3
January 30
•
28
20
13
5
4
4
5
2
0
19
Cum.
48
61
66
70
74
79
81
81
100
Awt,
mg
2.5
2.0
1.9
0.9
0.5
0.7
0.5
0.3
0.1
(7.6)a
(17.0)
Number
Cone . ,
mg/m3
2.94
2.36
2.24
1.06
0.59
0.82
0.59
0.35
0.12
(8.95)
(20.0)
1
*
15
12
11
5
3
4
3
2
1
(44)
Number 2
Cum.
27
38
43
46
50
53
55
56
100
Awt,
mg
1.1
5.4
2.3
1.5
0.8
0.9
0.8
0.6
0.1
6.1
19.6
Cone. ,
mg/m3
1.30
6.36
2.71
1.77
0.94
1.06
0.94
0.71
0.12
7.18
23.1
*
6
27
12
8
4
4
4
3
1
31
Cum.
33
45
53
57
61
65
68
69
100
Awt,
mg
5.7
5.3
3.2
1.1
1.8
2.1
1.8
0.9
0.1
2.7
22.7
Number
Cone . ,
mg/m3
6.71
3.89
3.77
1.30
2.12
2.47
2.12
1.06
0.12
3.18
26.7
3
«
25
14
14
5
8
9
8
4
1
12
Cum.
39
53
58
66
75
83
87
88
100
Filter weight gain considered unreliable; filter appeared stiff as though some condensation and subsequent
evaporation took place, leaving some condensate.
-------
TABLE 13. IN-STACK VIRTUAL IMPACTOR DATA TAKEN AT NEW BEDFORD (CORRECTED
FOR TOKEN FLOW EFFECT) (SEE APPENDIX)
Probe
Stage 1
(7.0 pm)
Stage 2
(2.0 pm)
Backup
Total
Probe
Stage 1
(7.0 pm)
Stage 2
(2.0 pm)
Backup
Total
Awt,
mg
11.1
12.4
3.5
8.1
35.1
20.2
37.6
11.3
(21.0)a
(90.1)a
Corrected
Awt, mg
January
11.1
10.9
2.6
10.5
35.1
January
20.2
33.4
9.3
27.2
90.1
C°^C3' Percent Cumulative %
mg/mj
29
4.85 32
4.76 31 63
1.14 7 70
4.59 30 |
i
15.3
30
6.0 22
9.91 37 59
2.76 10 69
8.07 30
26.7
a
Calculated
from pressure
drop data as
APF
ATTf — f,
AwtF A
follows:
(Awt,)
P™
where Awtp = weight gain, Friday, January 30
pressure gain, Friday, January 30 = 5.7 in. Hg
weight gain, Thursday, January 29 = 8.1 mg
Ap«p = pressure gain, Thursday, January 29 = 2.2 in. Hg.
80
-------
PERCENT LESS THAN STATED SIZE
99.99 99.9 99.8 99 98 96 9O SO 70 60 50 40 3O 20 10 0
100
90 -
80 -
70 -
60 -
50 —
40
30
2 I O.5 0.2 0.1 O.O9 0.01
<
Q
O
O
10
•
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
I
1
x -ANDERSEN IMPACTOR
o-VIRTUAL IMPACTOR
®-VIRTUAL IMPACTOR
CORRECTED FOR TOKEN
FLOW
® o
X
& o
0.05 0.1 0.2 0.3 I
2 5 10 20 30 40 50 60 TO 80 90 95 98 99
PERCENT GREATER THAN STATED SIZE
99.8 99.9 99.99
Figure 35. Cumulative size distribution for January 29, 1976.
81
-------
The test run on 30 January was designed to demonstrate that the virtual
impactor flow could be shut off, leaving the impactor in the stack (with the
probe turned 180° to the duct velocity) and restarted some time later to regain
the pressure drops recorded before shut-down. Figure 36 shows virtual impactor
pressure drop data for the second 2 hours of the 30 January test. The 30 Jan-
uary test was divided into three 1-hour sections, during which a freshly loaded
Andersen impactor was run each hour. Once again, pressure sensing lines had
failed in the 4200 ft/min stack gas velocity, requiring removal and repair after
the first hour. Between the second and third hours of this test, the virtual
impactor prototype was left off for 45 minutes, and identical pressure drops
were regained when flow was again started. Total concentration agreement be-
tween the virtual impactor (26.7 mg/m3 for the 3-hour test), and the Andersen
impactors (20.0, 23.1, and 26.7 mg/m3 for each hour) was good. Size distribution
information from the virtual impactor indicated 69 percent above 2.0 ym, and
59 percent above 7.0 ym, whereas the average of the three Andersen impactor
samples indicated 66 percent was above 2.0 ym, and 49 percent above 7.0 ym, as
shown in Figure 37. These results are based on an assumed value for the
material collected by the virtual impactor backup filter, which was in poor con-
dition from rust stains when the virtual impactor was disassembled in the lab
after returning. Presumably, this filter was wetted by vapor that had con-
densed in the 3/4-inch iron pipe that carried the main virtual impactor flow and
supported the whole assembly inside the stack. This could have happened if the
impactor was held at a lower level than the connecting pipe upon extraction from
the stack; a not unlikely event as movements of that day are mentally recon-
structed. A similar type of event seems to have occurred to the backup filter
of the first hour Andersen sample, but to a much lesser degree. That value, it
will be noted, is enclosed in parenthesis in Table 12. It may also be noted
that had the weight gain of that first hour Andersen backup filter been less,
the cumulative size distribution curve for that run, in Figure 37, would have
agreed better with the other Andersen samples and the virtual impactor data as
well.
Table 14 compares virtual impactor pressure drop data with the weight gained
in the same size fraction by the Andersen impactor. These data were obtained
from the cumulative size distribution curves of Figure 37 and the total weight
gain of the Andersen impactors reported in Table 12. For example, for the No. 2
Andersen sample of 30 January, the 7 ym line and the cumulative size distribution
curve of that run intersect at 52 percent greater than stated size. It will be
noted that 52 percent of 19.6 mg, the total weight gain of the Andersen im-
pactor, is 10.2 mg as reported in Table 14. During that same time period the
first stage filter pressure gain of the virtual impactor was 0.06 in. l^O. Ex-
amination of Table 14 shows that pressure drop roughly followed weight gain as
reported by the Andersen impactor samples. It will be noted that the only dis-
crepant data point is reported for the case where the Andersen backup filter
weight was questionable. It will also be seen from the data that if drag
coefficients, representative of the slopes of the pressure drop versus cake
weight gain curves, are computed, that the finer size fractions resulted in a
greater pressure gain; a conclusion reported for tests with various laboratory
generated aerosols.
82
-------
a.
0
g 1.20
M
Q.
7
6
5
4
3
2
1
0
^«
S*
0
0
10
o
})
....... |[ ,
UJ
j^
s
, x 2 *
« o
X ^
X
X K
. x 2
1 z
o
Q
3
X
CO
»•»•••
^*»
0
"•
a
o
1-
< *
o °- o
o ° s °
o —
0 °
o J
o 5
a: ,
x * * > » '
X «
X X
1 1 1 1 1 III
*x
x x
, x x BACKUP
, * " FILTER
-
-
«
.
•
.
00
0° STAGE 2
^ 0 FILTER
0 °
5
X*
x * STAGE 1
_x x * FILTER
K *
I 1 I I I i
60 80 100 120 (I 120 140 160 180
OPERATING TIME.minuUc
Figure 36. Virtual impactor pressure drop data for January 30, 1976.
83
-------
,oc9
90
60
70
60
50
40
90
20
E .0
*• 9
of 9
UJ 7
t^t 6
< 5
O
4
O
I ,
< 3
Z
O
O 2
HI
1
0.9
0.8
0.7
0.6
0.8
0.4
0.3
0.2
O.I
PERCENT LESS THAN STATED SIZE
1.99 99.9 99.8 99 98 95 90 80 70 60 90 40 90 20 10 5 2
* i i i — i 1 1 r — i — i — i — i — i — i 1 1 r
_
X '/30 I
,
A '/
* '30 3
o VIRTUAL
® VIRTUAL
1 0.5 0.2 O.I 0.05 0
— | 1 1— T i
^
^
ANDERSEN
SAMPLES
•w
^
IMP ACTOR
IMPACTOR
CORRECTED FOR TOKEN
FLOW
-
X QA
~ X D A
•
-
no
x QA
_
X D A
H.
X D A
A O
X D A
_
x D A
-
-
x O A
-
-
k*
,
*
—
"
-
m
-
^
_
-
.
-
-
-
-
-
-
-
i i i
0.01 0.1 0.2 0.5 I 2 5 10 20 30 40 50 60 70 80 90 95 98 99
PERCENT GREATER THAN STATED SIZE
99.6
99.99
Figure 37. Cumulative size distribution data for January 30, 1976.
84
-------
TABLE 14. VIRTUAL IMPACTOR FILTER PRESSURE GAIN COMPARED WITH LOADING AS
REPORTED BY THE ANDERSEN IMPACTOR
Andersen
test
1/29
1/30, No. 1
1/30, No. 2
1/30, No. 3
> 7.
Andersen,
64
40
52
58
0 pm
Awt,
mg
12.0
6.8
10.2
13.2
V.I.
A p
"H20
_
—
0.06
0.10
< 7.
> 2.
Andersen,
12
11
12
22
0 um,
0 pm
Awt,
mg
2.3
1.9
2.4
5.0
V.I.
A p
"H20
__
—
0.10
0.12
< 2.
Andersen,
24
49a
36
20
0 pm
Awt,
mg
4.5
8.3a
7.1
4.5
V.I.
A P
"H20
2.2
1.0
2.6
2.0
This Andersen Impactor test contains a backup filter weight gain considered
unreliable.
TABLE 15. VIRTUAL IMPACTOR DATA
FROM PRELIMINARY FIELD TEST
Col-
lection
stage
Stage 1
Stage 2
Backup
Total
January 21,
concentration,
mg/m3
No. 1
7.9
1.6
5.6
15.1
No. 2
5.6
2.5
5.4
13.5
85
-------
Table 15, which reports concentrations determined during the preliminary
field trip, is included to show that the virtual impactor, even though operated
with poor flow regulation, gave results consistent with the later measurement;
i.e., that particles within this stack are mostly large and small with few in
the 1 micrometer range. Concentrations determined during these earlier tests
are seen to agree well with later data.
The main conclusions of the stack tests are here summarized:
1. The virtual impactor reports concentrations and size dis-
tribution in good agreement with Andersen impactor data.
2. The virtual impactor may be left with zero flow in the
stack for some time after a collection period, and upon
restarting the flow the pressure drops at each stage will
return to the values recorded just before the flow was
interrupted.
3. Stage filter pressure drops indicate well the size
fractionated concentrations.
RESEARCH TRIANGLE PARK TESTS
Tests were conducted with the virtual impactor at two different EPA wind
tunnels to demonstrate the viability of this instrument.
On 24 March 1976 a test was run at the Environmental Sciences Research
Laboratory (ESRL) wind tunnel in the Beaunit building with the objective of
seeing how large a load the virtual impactor could handle before the pressure
drop output data became useless. An Andersen sampler was run for 20 minutes
during the 2-hour run, data for which may be found in Table 16, which is
labeled Andersen Run No. 1. Virtual impactor data is presented in Table 17.
Pressure drop versus time data for the virtual impactor are shown in Figure 38.
The curve for the first stage displays some upward concavity, the cause for
which is debatable, although it must be admitted that upon disassembly of the
virtual impactor a large amount of loose fly ash was contained on the first
stage, some of which was spilled and some of which was weighed as probe catch.
Change of weight for Stage 1 in Table 16 is unreliable. It will be noted that
pressure drop versus time for Stage 2, as shown in Figure 38, is well-behaved.
Virtual impactor run No. 2, also performed in the ESRL wind tunnel, was a
test to check the responsiveness of the instrument to a change of concentration.
After 30 minutes of this 60-minute test had elapsed, the setting of the Acrison
dust feeder rate control was reduced from 70 to 18. Andersen run No. 2 was a
5-minute sample during the first 30 minutes of this test and Andersen run No. 3
was a 20-minute sample taken during the last 30 minutes of this test. Pressure
drop versus time are plotted in Figure 39 for this test, from which it is
evident that the lower concentration for the second half of the test resulted
in reduced slopes. Further, the magnitude of the pressure gain for the first
stage, which was 0.33 in. H£0 for the high concentration segment and 0.09 in.
for the low concentration segment indicates the high concentration was
86
-------
TABLE 16. ANDERSEN IN-STACK DATA
00
Stage
Probe
1
2
3
4
5
6
7
8
Final
Total
Q = 14
Aero.
size
cut,
Vim
13.6
8.6
5.6
4.0
2.5
1.3
0.8
0.54
£pm
Andersen run
Awt
48.1
113.3
67.8
43.2
16.4
8.4
5.0
0.8
0.4
0.1
303.5
At =
Cone . =
%
15.9
37.4
22.3
14.2
5.4
2.8
1.6
0.3
0.1
0
100
no. 1
Cum.
%
84.1
46.7
24.4
10.2
4.8
2.0
0.4
0.1
0
20 min
1084 mg/m3
0.47 gr/ft3
Andersen run no. 2
Awt
11.4
12.5
10.5
7.5
4.4
2.1
1.2
0.3
0.1
0.4
50.4
At
Cone.
„ Cum.
/0 %
22.6
24.8 87.4
20.8 S2..6
14.9 31.8
8.7 16.9
4.2 8.2
2.4 4.0
0.6 1.6
0.2 1.0
0.8 0.8
100
= 5 min
= 720 mg/m3
= 0:31 gr/ft3
Andersen run no. 3
Awt
14.7
8.4
6.5
4.3
2.9
1.7
0.9
0.8
0
0.3
40.5
At
Cone.
%
36.3
20.7
16.1
10.6
7.2
4.2
2.2
2.0
0
0.7
100
Cum.
%
63.7
43.0
26.9
16.3
9.1
4.9
2.7
0.7
0.7
= 20 min
= 145 mg/m3
= 0.063 gr/ft3
Andersen run no. 4
Awt
36.5
4.2
3.8
4.3
2.2
1.6
0.3
0.2
0.1
0.1
53.3
At
Cone.
„ Cum.
10 at
to
68.4
7.9 31.6
7.1 23.7
8.1 16.6
4.1 8.5
3.0 4.4
0.6 1.4
0.4 0.8
0.2 0.4
0.2 0.2
100
= 2 min
= 1904 mg/m3
=0.83 mg/m3
Nozzle = 1/4 in. for isokinetic sampling at 1800 fpm.
-------
TABLE 17. VIRTUAL IMPACTOR DATA
Probe,
Stage 1
Stage 2
Backup ,
Total
At
Awt
Awt
Ap
Awt
Ap
Awt
Awt
Concentration
V
run
819
313
1
60
0
4
1,198
118
584
.1.
no. 1
.9 mg
.8 mga
.82 "H20
• 8 mg
.68 "H20
.0 mg
.5 mg
min
mg/m3
V.I.
run no
52.
406.
0.
15.
0.
3.
478.
60
458
. 2
9
2
42
6
15
4
1
V.I.
run no
345.
81.
4.
0.
3.
433.
5
4,983
. 3
2
1
0.10
0
05
2
5
V.I.
run no
2,068.
3,215.
1.
153.
1.
8.
5,444.
186
1,900.
. 4
0
5
28b
0
63
2
7
8
Unreliable.
Ap for first 55 minutes of run, after which pressure drop became erratic due
to overloading; see Figure 37.
88
-------
2.8
2.6
2.4
2.2
0
CM 2.0
X
A
O
.£ 1.6
oo
«> 1.4
1.2
1.0
0.8
n s
°' <
- -
x -
X
X
-
* STAGE 1
X
x
X
x
X
X o 0 -
o o
* o o ° STAGE 2
x °
x * ° °
-. & i ° °
> *
i i i i i i i i i i i i i i t i t i i i i t i i
) 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 IIO 115 120
t, minutes
Figure 38. Pressure drop versus time for run no. 1, 24 March 1976.
-------
I I I I I I I I
0 5
10 15 80 25 30 35 40 45 50 55 60
t, minutes
Figure 39. Pressure drop versus time for test in which concentration
was varied.
90
-------
0.33/0.09 = 3.67 times greater than the low concentration. Data from the
Andersen samples indicate a ratio of 720/145 = 5. Average concentration from
the two Andersen samples is 433 mg/m3 whereas the virtual impactor reports
458 mg/m3. The ratio of the Stage 1 gain to the Stage 2 gain is lower for the
low concentration portion of the test indicating a shift in particle size dis-
tribution toward smaller particles for the lower concentration; a result con-
sistent with that one expects from agglomeration at higher concentrations, and
a result borne out to some degree by the Andersen data.
A switch was made to the Industrial Environmental Research Laboratory (IERL)
wind tunnel and virtual impactor run No. 3 was taken simultaneously with a Brink
sampler. The probe wash dominates this sample and results in total concentra-
tion twice as great as expected. The virtual impactor was in the stack for ap-
proximately 1-1/2 hours while the Brink was readied and the probe may have been
contaminated somehow during this time.
The fourth Andersen run was taken at the IERL wind tunnel. It may be seen
that total concentration agrees with virtual impactor concentration for virtual
impactor run No. 4, which was a test to better define the ultimate overload
point. Pressure drop versus time data are plotted in Figures 40 and 41. The
pressure drop versus collected mass overload point was apparently never reached
on Stage 1 even after 3 hours of operation at the end of which more than
3000 mg had been collected on that stage. Stage 2 pressure drop remained linear
for the same period of time with a possible overload condition beyond 180 min-
utes at a collection of about 150 mg. The loadings on Stages 1 and 2 are in-
dicative of the notable collection capabilities of this device, which exceed
typical collection limits of commonly used in-stack jet-to-plate impactors by
factors between 10 and 100. The pressure drop of both stages tracked concentra-
tion as measured by an Ikor monitor. Three different slopes may be seen in
Figures 40 and 41, corresponding to three different concentration levels as
measured by the Ikor.
The above discussed performance also demonstrates the relatively long un-
attended measurement duration capability of this collection system, which would
be extended even more if a larger number of sizing stages is used.
Table 18 presents dust load drag coefficients derived from the virtual im-
pactor data, for the two virtual impaction stages. It is apparent that the
characteristic drag coefficient for these tests all fall within about ±20 per-
cent for each of the two stages.
In addition to the significant performance characteristics demonstrated by
the device investigated within this program, the cascade virtual impactor-
pressure drop sensor was found to be intrinsically self calibrating because the
collection on the individual stage filters can be assessed gravimetrically at
the completion of a run. This method can be applied either routinely or as a
means for sporadic spot checking the pressure drop-loading relationship at a
particular source under surveillance.
91
-------
8.2
7.8
7.4
7.0
6.6
6.2
5.8
5.4
5.0
O
-------
A WEIGHT OF FILTER =153.0 mg
' i i t ' ' i i i i i
100 120 140 160 180
Figure 41. Pressure drop versus time for Stage 2, 26 March 1976.
93
-------
TABLE 18. VIRTUAL IMPACTOR PRESSURE-DROP DRAG COEFFICIENTS3
V.I. V.I. V.I. V.I.
run no. 1 run no. 2 run no. 3 run no. 4
CD, Stage 1 — 0.21 x 105 sec"1 0.25 x 105 0.27 x 1Q5 b
CD, Stage 2 2.27 x 105 sec"1 1.95 x 105 2.54 x 105 2.16 x 105
3. o
C = 1.49 x 105 Q/^wt) sec"1, when A is in cm2,
Q is in £pm,
Awt is in mg,
Ap is in in.
Tr(4.4)2 2
A = — —, - — cm .
Q = 1.7 £pm (token flow).
Calculated by assuming Awt for first 55 min = 3215.5 mg x -—
-------
REFERENCES
1. Lippmann, M. and R. E. Albert. The Effect of Particle Size on the Regional
Deposition of Inhaled Aeros-ls in the Human Respiratory Tract. Am Ind
Hyg Assoc J. Vol. 30, p. 257, 1969.
2. Guide for Respirable Mass Sampling. Aerosol Technology Committee, American
Industrial Hygiene Association. Am Ind Hyg Assoc J. Vol. 31, p. 133,
1970.
3. Lippmann, M. Respirable Dust Sampling. Am Ind Hyg Assoc J. Vol. 31,
p. 138, 1970.
4. Lilienfeld, P. and D. W. Cooper. Literature Review and Selection of Design
Principle for the Development of a Fine Particulate Source Testing In-
strument. Prepared for U.S. Environmental Protection Agency, Contract
No. 68-02-1314. GCA Technical Report Number 73-22-G. October 1973.
5. Bird, A. N., Jr., J. D. McCain, and D. B. Harris. Particulate Sizing
Techniques for Control Device Evaluation. APCA Meeting. Paper Number
73-282. Chicago, Illinois. 1973.
6. Fuchs, N. A. The Mechanics of Aerosols. Pergamon Press, 1964.
7. Green, H. L. and W. R. Lane. Particulate Clouds: Dusts, Smokes, and
Mists. Princeton, New Jersey, D. Van Nostrand Co., 1964.
8. Lundgren, D. A. An Aerosol Sampler for Determination of Particle Concen-
tration as a Function of Size and Time. J Air Pollut Control Assoc.
Vol. 17, p. 225, 1967.
9. Mitchell, R. I. and J. M. Pilcher. Measuring Aerosol Particle Sizes.
Ind Eng Chem. Vol. 51, p. 1039, 1959.
10. Ranz, W. E. and J. B. Wong. Jet Impactors for Determining the Particle
Size Distribution of Aerosols. Arch Ind Hyg Occup Med. Vol. 5, p. 464,
1952.
11. Marple, V. A. A Fundmaental Study of Inertial Impactors. Ph.D. Thesis,
University of Minnesota, December 1970.
12. Pilat, M. J., et al. Source Test Cascade Impactor. Atmos Environ.
Vol. 4, p. 617, 1970.
95
-------
13. McGinn, J. H. and J. T. MacWaters. Ballistic Particle Separator. Rev Sci
Instrum. Vol. 31, p. 513, 1960.
14. Hounam, R. F. and R. J. Sherwood. The Cascade Centripeter: A Device for
Determining the Concentration and Size Distribution of Aerosols. Am Ind
Hyg Assoc J. Vol. 26, p. 122, 1965.
15. Conner, W. D. An Inertial-Type Particle Separator for Collecting Large
Samples. J Air Pollut Control Assoc. Vol. 16, p. 35, 1966.
16. Loo, W. B. Letter to the U.S. Environmental Protection Agency. January 10,
1974.
17. Dzubay, T. G., L. E. Hines, and R. R. Stevens. Particle Bounce Errors in
Cascade Impactors. Atmos Environ. Vol. 10, p. 229-234, 1976.
18. Priening, 0. The Cross Sensitivities of the Royco Aerosol Photometer
PC-200. Staub. Vol. 28, p. 29, 1968.
19. Cooper, D. W. Variable Jet Impactor and Aerosol Size Distribution
Analysis. Ph.D. Dissertation, Division of Engineering and Applied Physics,
Harvard University, 1974.
20. Picknett, R. G. A New Method of Determining Aerosol Size Distributions
From Multistage Sampler Data. J Aerosol Sci. Vol. 3, p. 185, 1972.
21. Twomey, S. The Determination of Aerosol Size Distributions From Dif-
fusional Decay Measurements. J Franklin Inst. Vol. 275, p. 121, 1963.
22. Twomey, S. The Application of Numerical Filtering of the Solution of
Integral Equation Encountered in Indirect Sensing Methods. J Franklin Inst.
Vol. 279, p. 95, 1965.
23. Phillips, D. L. A Technique for the Solution of Certain Integral Equa-
tions of the First Kind. J Assoc Comput Mach. Vol. 9, p. 84, 1962.
24. Rust, B. W. and W. R. Burrus. Mathematical Programming and Numerical
Solution of Linear Equations. New York, American Edison Company, 1972.
25. Dorsey, J. A. and J. 0. Burckle. Particulate Emissions and Process Monitors,
Monitors. Chem Eng Prog. Vo-. 67, p. 92-96, 1971.
26. Hyland, R. G. The Ikor Continuous Particle Monitor. APCA Conference.
Paper Number 72-38. 1972.
27. Prochazka, R. Recording Dust Measurement With the Konitest. Staub.
Vol. 26, No. 5, p. 22, 1966.
28. Jones, L., C. Lochboehler, and W. Magee, Jr. Aerosol Filtration by Fibrous
Filter Mats. Environ Sci Technol. Vol. 6, p. 821-826, 1972.
96
-------
29. Stafford, R. G. and H. J. Ettinger. Filter Efficiency as a Function of
Particle Size and Velocity. Atmos Environ. Vol. 6, p. 353-362, 1972.
30. Thomas, J. and R. Yoder. Arch Ind Health. 13, 545, 1956.
31. Clarenburg, L. A. A Probabalistic Theory of Aerosol Penetration Through
Glass Fiber Filters. Aerosol Science. Vol. 3, p. 461-490, 1972.
32. Anderson, D. P. Unpublished work at NIOSH, Cincinnati. 1971.
33. Miyamoto, S. and H. Bohn. Filtration of Airborne Particulates by Gravel
Filters: II. J Air Pollut Control Assoc. Vol. 25, No. 1, 1975.
34. Billings, C. Effects of Particle Accumulation on Aerosol Filter Life.
Proc., 9th AEG Air Cleaning Conference. 1966.
35. Happel, J. and H. Brenner. Low Reynolds Number Hydrodynamics. Prentice-
Hall, 1965. Chapter 8.
36. Spurny, K. R., et al. Aerosol Filtration by Means of Nucleopore Filters,
Filter Pore Clogging. Environ Sci Technol. Vol. 8, No. 8, 1974.
37. Langmead, W. A. and D. T. O'Connor. The Personal Centripeter—A Particle
Size-Selective Personal Air Sampler. Ann Occup Hyg. Vol. 12, p. 185-
195, 1969.
97
-------
APPENDIX
TOKEN FLOW CORRECTIONS TO VIRTUAL IMPACTOR DATA
COLLECTED IN NEW BEDFORD
Restating the matrix equation as presented in Section 5 of this report
which gives Ey values for the coefficients of the linear equations describing
aerosol collection for a three stage virtual impactor when token flow is assumed
a constant fraction of main flow:
E
1 e e
0 l(l-e) e(l-e)
0 0 Kl-e(l-e)-e)
(1)
which is a close approximation to the actual operating conditions which are
described by
1 e e
0 l(l-e) f(l-e)
0 0 l(l-f(l-e)-e)
(2)
where token flow is not a constant fraction of main flow.
For the tests run at New Bedford, total flow entering the virtual impactor
is equal to 14.15 £pm (1/2 cfm) through the backup filter plus 2.08 fcpm for the
second stage token flow plus 2.08 £pm for the first stage token flow equals
18.31 Jlpm. Therefore, e
for E-H gives
2.08
18.31
and f =
2.08
2.08
18.31-2.08 ~ 16.23'
values
" 1
0
. 0
0.1137
0.8863
0
0.1137 '
0.1138
0.7725_
(3)
98
-------
Choosing the exact approach and applying the method outlined in Section 5
of the report:
F2 ' (F2* - E23 F3)/E22
F '
- E12 F2 - F13 F3)/E11
where F]_* = 12.4 mg
F£* = 3.5 mg
F3* = 8.1 mg
from the data of January 29 in Table 13,
(4)
and
E12
E13
E22
E23
E33
1
0.1137
0.1137
0.8863
0.1138
0.7725
from matrix (3) above, one obtains the corrected change in weights shown in
Table 13.
Alternatively, the inverse matrix, in the report,
F = E'1 F*
(5)
may be applied as follows by expanding to
'Elf1 <-E12/EllE22> (*12E23E22~1 ~ E13)/E11E33
0
LO
-1
J22
"E23/E22E33
„ -1
(6)
99
-------
and evaluating by vector multiplication one obtains:
E12E23
T? ^fe "E* ifeTT
1 912
F, - ==- - J- „ + F *
E11E22 3 E11E33
= 0
F*
F - 0 F* + 0 F * +
l ^
which gives the same result as above.
100
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
. REPORT NO.
EPA-600/2-77-Q5Q
3. RECIPIENT'S ACCESSION-NO.
4. TITLE AND SUBTITLE
STUDY ON THE FEASIBILITY AND DESIGN OF AUTOMATIC
PARTICULATE SIZE DISTRIBUTION ANALYZER FOR SOURCE
EMISSIONS
5. REPORT DATE
August 1977 (Issuing Date)
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Pedro Lilienfeld; Daniel P. Anderson;
Douglas W. Cooper
8. PERFORMING ORGANIZATION REPORT NO.
6CA-TR-76-22-G
9. PERFORMING ORGANIZATION NAME AND ADDRESS
GCA Corporation
GCA/Technology Division
Burlington Road
Bedford, Massachusetts 01730
1O. PROGRAM ELEMENT NO.
INB458
11. CONTRACT/GRANT NO.
68-03-2154
12. SPONSORING AGENCY NAME AND ADDRESS
Municipal Environmental Research Laboratory--Cin., OH
Office of Research and Development
U.S. Environmental Protection Agency
Cincinnati, Ohio 45268
13. TYPE OF REPORT AND PERIOD COVERED
Final
14. SPONSORING AGENCY CODE
EPA/600/14
is.SUPPLEMENTARY NOTES Project Off icer: John 0. Burckle, SHWRD/MERL/ORD - This work
was performed under a research agreement for the Industrial Environmental Research
Laboratory/Research Triangle Park. ____
16. ABSTRACT
The objective of this program was to evolve a method for the automatic determination
of the size distribution of particulates within stack gas effluent streams. This
device was designed to cover the typical mass concentration range encountered up-
stream as well as downstream of emission control systems, and to segregate the
particles by means of a cascaded virtual inertial impaction configuration to be
inserted into the effluent stream. Several alternative particle detection techniques
compatible with this size segregation method were investigated in the course of this
program and a stage filter pressure drop sensing technique was selected. The proto-
type device was subjected to laboratory and stack testing showing very good correla-
tion with an Andersen-type impactor. The salient advantages of this instrument are:
capability for extended operation (of the order of hours), real-time indication of
size distribution of particulates in the stack environment, relatively low cost,
and simplicity of operation.
7.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
>.IDENTIFIERS/OPEN ENDED TERMS
COSATI Field/Group
Air Pollution
Flue Dust
Aerosols
Particle Size Distribution
Size Determination
Detectors
Monitoring
Source Emission Measure
ments
Virtual Impactor
Contact Change Sensing
Drag Flow Sensing
Light Extinction Sensin
13B
8. DISTRIBUTION STATEMENT
Release to Public
19. SECURITY CLASS (ThisReport)
Unclassified
21. NO. OF PAGES
113
2O. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
101
. S. GOVERNMENT PRINTING OFFICE l977-/5/-05b/bt79 Region No. 5-11
-------
|