United States Office of Air Quality EPA-450/2-78-025a
Environmental Protection Planning and Standards August 1978
Agency Research Triangle Park NC 27711
Air
Primary Aluminum
Background
Information:
Proposed
Amendments
-------
EPA-450/2-78-025a
Primary Aluminum
Background Information:
Proposed Amendments
Emission Standards and Engineering Division
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air, Noise, and Radiation
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
August 1978
-------
This report is issued by the Environmental Protection Agency to report
technical data of interest to a limited number of readers. Copies are
available free of charge to Federal employees, current contractors and
grantees, and nonprofit organizations - in limited quantities - from the
Library Services Office (MD-35) , U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711; or, for a fee, from the
National Technical Information Service, 5285 Port Royal Road, Springfield,
Virginia 22161.
Publication No. EPA-450/2-78-025a
-------
Table of Contents
Chapter 1. INTRODUCTION AND SUMMARY
Chapter 2. BACKGROUND
Chapter 3. SUPPLEMENTAL INFORMATION AND
ENVIRONMENTAL IMPACT ....
Paqe
Chapter 4. OTHER ISSUES 10
APPENDIX A. CORRESPONDENCE 16
Patrick R. Atkins to Don R. Goodwin
December 14, 1977
APPENDIX B. CORRESPONDENCE 59
Don R. Goodwin to Patrick R. Atkins
April 27, 1978
-------
1. INTRODUCTION AND SUMMARY
This document supplements information contained in the preamble for
the amended Standard of Performance for Primary Aluminum Plants (40 CFR
Part 60, Subpart S). The amendments are being proposed in response to
arguments raised by four aluminum companies who filed petitions for
review of the standard of performance.
The principal amendments to Subpart S would be as follows:
1. Following the initial performance test, performance testing
would be required at least once each month during the life of a new
primary aluminum plant.
2. Performance test results above the level of the current potroom
standard [0.95 kg/Mg (1.9 Ib/ton) for prebake plants and 1.0 kg/Mg (2.0
Ib/ton) for Soderberg plants] would be allowed if an owner or operator
can establish that the emission control system was properly operated and
maintained at the time the excursion above the current standard occurred.
Emissions above 1.25 kg/Mg (2.5 Ib/ton) would not be allowed under any
condition.
3. The owner or operator of a new primary aluminum plant may apply
to the Administrator for an exemption from the monthly testing requirement
for primary and anode bake plant emissions.
Additional information, including the amendments to Reference
Method 14 - Determination of Fluoride Emissions from Potroom Roof
1
-------
Monitors of Primary Aluminum Plants, may be found in the preamble and
regulation for the proposed amendments in the Federal Register.
-------
2. BACKGROUND
A standard of performance for new primary aluminum plants was
promulgated on January 26, 1976 (41 FR 3826), and shortly thereafter
petitions for review were filed by four U. S. aluminum companies. The
principal argument raised by the petitioners was that the standard was
too stringent and could not be consistently complied with by modern,
well-controlled facilities. (Facilities which commenced construction
prior to October 23, 1974, are not affected by the standard.) Following
discussions with the petitioning aluminum companies, EPA conducted an
enission test program at the Anaconda Aluminum Company plant in Sebree,
Kentucky. The Sebree plant is the newest primary aluminum plant in the
U.S., and its emission control system conforms with what EPA has
defined as the best technological system of continuous emission reduction
for new facilities. The purpose of the test program was to aid EPA in
its reevaluation of the standard by expanding the emission data base.
The test results were available in August of 1977 and indicated that
there is some probability that the result of a performance test conducted
at a modern, well-controlled plant would be above the existing standard.
EPA has concluded that this justifies revising the standard.
During discussions and in correspondence with the petitioners,
several issues were raised. These may be summarized as follows:
3
-------
1. The level of the emission standard is too stringent and cannot
be consistently achieved by modern, well-controlled plants.
2. Operating practices, in particular the bath ratio, can adversely
affect fluoride emissions.
3. Anode bake plant control is not cost-effective.
4. EPA Method 13 consistently understates the measured level of
fluoride emissions.
5. The EPA emission test program at the Sebree plant of the
Anaconda Aluminum Company yielded questionable results.
6. Dry scrubbing causes increased secondary emission rates.
EPA has determined that only the first of these issues justifies
amending the regulation. This issue is discussed in the preamble for
the amended standard of performance, and supplemental information is
contained in Chapter 3. The remaining issues are discussed in Chapter 4.
For more information see:
U. S. Environmental Protection Agency. Proposed Standards of Performance
for New Stationary Sources (Primary Aluminum Industry). 40 CFR Part 60,
Subpart S. Washington, D. C. Federal Register (39 FR 37730). October 23,
1974.
U. S. Environmental Protection Agency. Promulgated Standards of Performance
for New Stationary Sources (Primary Aluminum Industry). 40 CFR Part 60,
Subpart S. Washington, D. C. Federal Register (41 FR 3826). January 26,
1976.
Background Information for Standards of Performance: Primary Aluminum
Industry (Volume 1: Proposed Standards; Volume 2: Test Data Summary; and
Volume 3: Supplemental Information). Office of Air Quality Planning and
Standards, U. S. Environmental Protection Agency. Research Triangle Park,
North Carolina. Report number EPA 450/2-74-020(a); (b); and (c). October 1974;
October 1974; and January 1976. 122, 48, and 47 pages.
-------
3. SUPPLEMENTAL INFORMATION
Sebree Test
The results of the Sebree test program are presented in Table 1.
EPA personnel observed potroom operations during the test program and
afterwards concluded that although operations could have been improved,
the variability of emissions was largely inherent in the production process
and beyond the control of plant personnel. The emission data base EPA
used to establish the original standard did not reflect this variability.
Analysis of Sebree Results
A performance test is defined in 40 CFR Part 60.8(f) as the arithmetic
mean of three separate runs, except in situations where a run must be
discounted or cancelled and the Administrator approves using the arithmetic
mean of two runs. To determine the probability that a performance test
would be above the level of the current standard, EPA calculated the arithmetic
mean of all 504 possible combinations of three of the nine Sebree runs;
8.3 percent of the means are above the level of the current standard of
0.95 kg/Mg (1.9 Ib/ton). Thus, EPA estimates that there is about an 8
percent chance that a performance test conducted at a well-controlled
plant would be above the current standard. The petitioners, using different
The Sebree test program is described in:
Air Pollution Emission Test: The Anaconda Company Sebree Reduction Plant,
Henderson, Kentucky. Office of Air Quality Planning and Standards. U. S.
Environmental Protection Agency. Research Triangle Park, North Carolina.
Report number 77-ALR-6. October, 1977. 700 pages.
-------
Table 1. EPA TEST RESULTS AT THE ANACONDA PRIMARY
ALUMINUM PLANT IN SEBREE, KENTUCKY
FLUORIDE EMISSIONS
Test
Series
1
1
1
1
2
2
2
3(4)
3
0\
nu.. }
la
b
2a
b
3a
b
4a
b
la
b
2a
b
3a
b
1
2
Date
(1977)
3/29
3/30
3/31
4/1
6/6-7
6/7-8
6/8-9
6/29-30
6/30-7/1
Primary
Emissions
(Ib/ton)
0.055
0.033
0.021
0.023
°'°33(3)
11
II
II
tf
Roof
Monitor
Emissions Total Potroom
(Ib/ton) - (Ib/ton)
147 avg=1.43 1.49
086 av9=0-85 °-88
]'5n avg=1.73 1.75
2-gg avg=2.72 2.74
J-8,3 avg=1.81 1.84
g'g3 avg=0.82 0.85
J-" avg=1.15 1.18
1.12 3
1.18 1.21
1 . 44 1 . 47
Emissions^2*
(kg/Mg)
0.75
0.44
0.88
1.37
0.92
0.43
0.59
0.61
0.74
(1)
(2)
During the EPA tests, two sampling trains were run simultaneously at the roof
monitor sampling site. One served as a backup to the other, in case of malfunction.
Total potroom emissions were calculated by adding the primary and roof monitor
emissions.
' ' Primary emissions for test series 2 and 3 are estimates based on the average of
the primary emissions from the first test series.
* Test series 3 was performed by Anaconda using EPA test methods.
-------
analytical methods, have estimated chances of failure ranging from 2.5
to 10 percent. (For example, see page 24 of Appendix A.)
The relatively small number of runs performed at the Sebree plant
introduces some uncertainty into the method used to calculate the probability
that a performance test would be above the current standard. For example,
if the highest of the nine Sebree runs, 1.37 kg/Mg (2.74 Ib/ton), is
increased just one percent, the probability of a performance test being
above the standard increases to 10.7 percent. Similarly, if 1.37 kg/Mg is
decreased by one percent, the probability decreases to 5.9 percent. Thus,
it: is difficult to determine exactly the probability of a performance test
being above the standard.
The average of the three highest Sebree runs, 1.06 kg/Mg (2.11 Ib/ton),
is well below the proposed never-to-be-exceeded limit of 1.25 kg/Mg (2.5 Ib/ton).
EPA believes it is reasonable to expect emissions from a modern, well-controlled
plant to be below 1.25 kg/Mg all of the time.
Costs
The major costs incurred by the amendments are associated with the
periodic emission testing requirement. Testing of the primary control
system, secondary emissions, and the anode bake plant (for prebake plants)
would be required at least once each month during the life of the primary
aluminum plant. However, once the emission characteristics of the primary
or bake plant control systems are known, the owner or operator may apply
to the Administrator for an exemption from the monthly testing requirement
for primary and bake plant emissions. The estimated costs of performance
tests conducted by local private contractors, by nonlocal private contrac-
tors, or by plant personnel are presented in Table 2. Due to the higher costs
-------
associated with contractor tests, the owner or operator of a new plant
would probably arrange to conduct testing with plant personnel if
required to test as often as once each month. Thus, the estimated costs
for testing conducted by plant personnel are more representative of what
actual costs would probably be.
Table 2. ESTIMATED COSTS FOR PERFORMANCE TESTING
AT NEW PRIMARY ALUMINUM PLANTS
Estimated costs per performance test
Optional
Tester
Plant
Personnel
Local Private
Contractor
Nonlocal
Private
Contractor
Secondary
Emissions
$ 4,800
7,700
13,200
Primary
Control
$ 4,000
6,400
11,400
Bake Plant
Control
$ 4,000
6,400
11,400
These estimated costs assume that: (1) each monthly performance
test would consist of the average of three 24-hour runs; (2) testing
would be performed by two crews working 13-hour shifts; (3) primary control
system sampling would be performed at a single point within the stack; and
(4) Sebree inhouse testing costs would be representative of average costs
for other new plants. Although these assumptions may not hold for all
situations, EPA believes they provide a representative estimate of what
testing costs would be for new plants.
There are also costs associated with recalibration of anemometers
following performance tests, as required by the revisions made to
Reference Method 14. These costs, however, would be minor.
-------
Environmental Impact
One of the proposed amendments would allow emissions to be above
the current limits, but not above a higher limit of 1.25 kg/Mg (2.5 Ib/ton)
if the owner or operator can establish that the control system was properly
operated and maintained during the excursion above the current limits.
This amendment is intended to account for the inherent variability of
emissions from aluminum potrooms. Since the amendment would simply make
the current regulation more realistic with regard to emission rates from
what EPA considers the best system of continuous emission reduction,
environmental quality would not suffer. There would be no other environmental
impacts associated with this amendment.
Another of the proposed amendments would require performance testing
at least once each month during the life of a plant. (Under the current
regulation, a performance test is only required to be performed within
60 days after achieving the maximum production rate, but not later than
180 days after initial startup of a new plant.) EPA believes that this
amendment would provide an incentive to properly operate and maintain
control equipment throughout the life of a plant rather than just during
the initial performance test. This, in turn, would result in improved
air quality in the vicinity of an aluminum reduction plant. There would be
no other environmental impacts associated with this amendment.
The other amendments are basically clarifying in nature and would have
no impact on the environment. Thus, the proposed amendments would have an
overall beneficial impact on environmental quality.
-------
4. OTHER ISSUES
During negotiations with the litigants, several major issues were
raised. EPA has evaluated these issues and has determined that only
one - the contention that the standard is unachievable - requires
amending Subpart S. These amendments are discussed in the preamble to
the proposed amendments and in Chapter 3. The other major issues are
summarized and responded to below.
Method 13
One litigant contended that EPA Method 13 consistently understates
tne measured level of fluoride emissions, and thus, the data base is
sjspect. An EPA study was used in support of this contention.
EPA carefully examined the original data base in light of the cited
EPA study and found no basis for concluding that the data base is biased.
Fjrthermore, the litigant was misinterpreting the study since it showed a
large bias only when a Whatman paper filter supported by a glass frit was
used at the front of the sampling train. EPA data were not gathered
using this method; rather a glass fiber filter was used. There is no
evidence of bias when using a glass fiber filter. In addition, the data
collected at Sebree were obtained using a sampling train with the filter
holder mounted between the third and fourth impingers, a position for
wnich no bias has been observed.
10
-------
Another study submitted to EPA raised questions about the efficiency
of paper filters placed in front of the Method 13 impingers. Since a
significant level of inefficiency is indicated, it is EPA's intention to
revise the reference method, disallowing the use of the paper filter in
the front of the impingers (an option which had been originally included
in Method 13 in response to industry comments) and recommending use of
a more efficient glass fiber or similar filter. This revision will be
made at a later date.
Additional information may be found in Aopendices A and B.
Bake Plant Control
One of the litigants argued that the anode bake plant emission limit
is not cost-effective and should be revised from 0.05 to 0.25 kg/Mq (0.1
to 0.5 Ib/ton) aluminum equivalent.
Standards of performance are based on the degree of emission reduction
achievable through the application of the best system of continuous
emission reduction which (taking into consideration the cost of achieving
such emission reduction) has been adequately demonstrated. In considering
cost, the Agency determines if the industry can procure the necessary
equipment and personnel to meet the standard at a cost which will allow
it to remain a viable competitive force. The cost consideration is thus
one of affordabi1ity rather than cost effectiveness. EPA has determined
that the bake plant standard is affordable and represents application of
the best system of continuous emission reduction. This is borne out by
the fact that the litigant recently started up a new bake plant in
compliance with the standard.
11
-------
Operating Practices
One litigant suggested that the aluminum industry standard could
not be achieved without production losses when operating a plant in a low
bath ratio (high current efficiency) mode.
The proposed amendments are based on EPA's experiences at the Sebree
Reduction Plant of the Anaconda Aluminum Company. The Sebree plant is the
newest primary aluminum plant in the U. S. and operates with a low bath
ratio. Since the Sebree plant can comply with the proposed amended
standard, EPA believes that other new plants will be able to comply with
the standard without production losses.
Sebree Test
It is argued that deviations from EPA's prescribed test methods in
the Sebree test program caused the fluoride emission measurements to
understate the true level of fluoride emissions. One of the more signifi-
cant departures from Method 14, it is claimed, involved the number of
anemometers used and the location of the anemometers. EPA used three
anemometers rather than two as called for under Method 14.
The use of three anemometers exceeds the requirement of Method 14
which specifies that only two are needed for a roof vent of this size.
(This type of variation from the reference test method is allowed under the
provisions of section 60.8 of 40 CFR Part 60.) Therefore, the accuracy of
the reported flow rates should be equal to or greater than the accuracy
that would have been obtained by following the Method 14 requirement exactly.
The contention that significantly higher emission rates would have resulted
if two anemometers had been used is purely speculative, since no comparisons
were made between the use of two and three anemometers. The Sebree data
12
-------
show that a substantial velocity gradient existed in the roof monitor;
because of this (and in the absence of data to indicate otherwise), the use
of three anemometers instead of two must be considered more representative
of the actual conditions in the monitor.
It is also contended that the velocity and fluoride concentration in
each eight-hour test were measured for different lengths of time, and this
had the effect of understating secondary emissions.
During the Sebree tests, velocity was monitored for the entire 480
minutes of each 8-hour test period. Therefore, the calculated volumetric
flow rates are considered to be representative. Fluoride concentration,
on the other hand, was measured only for the first 400 minutes of each
8-hour period. Since the majority of the activity in the potrooms occurs
in the first half of each shift, fluoride emissions tend to be higher
during this period. Consequently, measuring the fluoride concentration
for only 400 of the 480 minutes would tend to bias the results toward the
high side. Therefore, an emission rate calculated from the product of an
average fluoride concentration measured over the first 400 minutes times an
average volumetric flow rate measured over the entire 480 minutes would
tend to be higher than the true rate, rather than "understating the
secondary fluoride emission values," as contended.
Another argument was that EPA did not calibrate the anemometers properly,
and that the back-calculating EPA did in an attempt to rectify the erroneous
anemometer readings made the results highly suspect.
It is true that the velocity data from the Sebree tests were adjusted
after-the-fact based upon information obtained from the anemometer
manufacturer and from a series of post-test anemometer calibrations. How-
ever, these adjustments were made in order to increase the accuracy of the
13
-------
reported emission rates. (It should be noted that, in general, the
adjustments made in the Sebree data tended to increase the total flow and
emission rates, and thus, were in the industry's favor.) The contention
that industry would have no right to make similar adjustments in NSPS
performance test data is unfounded. Section 60.8 provides authority for
the Administrator to approve minor deviations from the reference methods
if it can be shown that equivalent or or more representative data will result.
Note that as a result of EPA's experience with the anemometers, EPA has
proposed revising Method 14 to include the proper calibration procedure
for anemometers.
EPA's use of monthly averages for determining the rate of aluminum
production was questioned since Subpart S expressly provides that the
actual rate of aluminum production is to be determined for the test run.
EPA considered using actual metal tapped in the calculations but
decided against this in favor of using the monthly average per pot produc-
tion rate, since the amount of metal tapped on any given day is not
necessarily the amount of metal produced on that day. For example, it
would be possible to not tap any pots on a given day in which case metal
tapped would be zero, but the pots would still be producing aluminum and
evolving fluorides. Alternatively, it would be to the advantage of the
source to tap a larger amount of metal than the amount actually produced
on the day of a compliance test, thereby reducing the calculated emission
rate. Thus, EPA believes that the monthly average gives a better
indication of actual metal production than amount tapped during the test.
The proposed amendments would make this an explicit provision.
More information on the Sebree test issues appears Appendices A and B.
14
-------
Dry Scrubber Emissions
One litigant has experienced secondary emission increases following
the installation of dry scrubbers at several of its facilities and
worries that emissions from new plants equipped with the scrubbers could
be in violation with the standards. It is theorized that the problem of
increased secondary emissions is related to the increased fines content
and increased fluoride content of the recycled alumina.
The proposed amendments are based on EPA's experiences at the Sebree
plant. Since the Sebree plant successfully utilizes dry scrubbers, EPA
believes this is evidence that new plants which are subject to Subpart S
would be able to comply with the amended standard.
15
-------
APPENDIX A
16
-------
/M.UMIMUM COMPANY OF AM ERICA
ALCOA BUILDING PITTSBURGH, PENNSYLVANIA 15219
ALCOA
1977 December 14
Mr. Don R. Goodwin
Environmental Protection Agency
Emission Standards and Engineering Division
Research Triangle Park, N. C. 27711
Dear Mr. Goodwin:
Aluminum Company of America ("Alcoa") does not agree with EPA's
conclusion that the results of the fluoride emission tests con-
ducted at The Anaconda Company's Sebree Reduction Plant demon-
strate that a new aluminum prebake reduction plant "...
which incorporates the best system of emission reduction,
considering cost, and which is properly operated and maintained
..." can achieve compliance with the New Source Performance
Standards for the Primary Aluminum Industry, promulgated on
January 26, 1976, 41 Fed. Reg. 3826 et seq. (See Letter dated
September 2, 1977, from Don R. Goodwin (EPA) to Dr. Patrick
R. Atkins (Alcoa), hereinafter referred to as the "Goodwin
Letter"). To the contrary, if the Sebree Emission Test results
demonstrate anything, they support Alcoa's contention that the
fluoride emission limitation set forth in the New Source Per-
formance Standards cannot be achieved on a never-to-be-exceeded
basis. The following are Alcoa's comments with respect to the
report published by EPA in connection with the Sebree Emission
Test (EPA Emission Test Report No. 77-ALR-6).
The Sebree Emission Test Data
Clearly Show that a "New" Pre-
bake Aluminum Plant Would Exceed
the Fluoride Emission Limitation
in the New Source Performance
Standards a Significant Percen-
tage of the Time
One of the most striking aspects of the Sebree Emission Test data
is the variability in the secondary fluoride emissions. The 24-
hour secondary fluoride emission values calculated by EPA from
17
-------
Mr. Don R. Goodwin
1977-12-14
Page Two
sets of three consecutive 8-hour measurements range from a low
of .80 pounds of fluoride per ton of aluminum produced ("Ibs.
F/TAP") to a high of 2.85 Ibs. F/TAP. The spread in the indiv-
idual 8-hour secondary fluoride emission measurements is even
more pronounced, ranging from .29 Ibs. F/TAP to 3.75 Ibs. F/TAP
(Sebree Emission Test Report, p. 5).
The Goodwin Letter of September 2, 1977, implies that the varia-
bility of secondary emissions is a function of operation and
maintenance practices in the potroom. It suggests that the
higher emission values found during Phase I of the Sebree Emis-
sion Test are attributable to alleged poor operating and
maintenance procedures "not conducive to low emissions;" the
implication being that the emission values found during Phases
II and III of the Sebree Emission Test are attributable to
improvements made by Anaconda in operating and maintenance
conditions. However, it must be pointed out that test runs 2a
and 2b in Phase I of the Sebree Emission Test produced secondary
fluoride emission values of .84 Ibs. F/TAP and .86 Ibs. F/TAP,
while six of the eight test runs in Phases II and III, which
were supposedly performed under conditions conducive to low fluor-
ide emissions, yielded secondary fluoride emission values far in
excess of .86 Ibs. F/TAP. The point is that the level of secon-
dary emissions is not a function of operation and maintenance
conditions alone; rather, it is the function of a complex
interaction of numerous factors which, in addition to operation
and maintenance conditions, include pot evolution, meteorological
conditions, wind direction and velocity, raw material quality
fluctuations, etc. Consequently, the level of secondary fluoride
emissions is highly variable. Furthermore, the Sebree Emission
Test data clearly indicate that there are measurable sources of
variation intrinsic to the fluoride sampling methods themselves.
Because of the variability of secondary fluoride emission test
results, the Sebree Emission Test data must be analyzed through
the application of standard statistical methods to determine
whether or not it supports the never-to-be-exceeded limitation of
1.9 Ibs. F/TAP in the New Source Performance Standards. Alcoa
submitted the Sebree Emission Test data, along with the emission
test data for Plants "B," "C," and "D" in the Background Documents
published by EPA in connection with the initial promulgation of
the New Source Performance Standards for the Primary Aluminum
Industry (EPA 450/2-74-020a & b), to Dr. Morris DeGroot, Ph.D,
Professor of Statistics and Industrial Administration at Carnegie-
Mellon University in Pittsburgh, Pennsylvania, for the purpose
18
-------
Mr. Don R. Goodwin
1977-12-14
Page Three
of obtaining an independent statistical evaluation of the data.
Dr. DeGroot's report, which is attached to these Comments as
Appendix A, conclusively establishes that, based upon the emis-
sion test data cited by EPA to support the 1.9 Ibs. F/TAP emis-
sion limitation in the New Source Performance Standards, a modern
prebake aluminum reduction facility will exceed the emission
limitation up to 10 percent of the time.
In interpreting Dr. DeGroot's Report, it should be observed that
Methods I, II and V used by him to analyze Sebree Emission Test
data are based on the assumption that the 8-hour test segments
are independent of each other and can be mixed taking a 0:00 to
8:00 test run from one day, a 8:00 to 16:00 test run from another
day and a 16:00 to 24:00 test run from a third day. In fact, that
assumption is false since meteorologic, raw material quality and
operating condition impacts, as well as many other factors will
frequently have an effect on fluoride emissions which extend beyond
an 8-hour period. The consequence is that the estimates obtained
by Methods I, II and V are lower than those obtained by the other
methods employed by Dr. DeGroot, which were based on 24-hour test
runs.
Based upon Dr. DeGroot's analysis, Alcoa reiterates that if the
emission limitation in the New Source Performance Standards is
to be a never-to-be-exceeded limit, it must be increased signifi-
cantly to allow for the highly variable nature of secondary fluor-
ide emissions which occur even in the best controlled facilities.
The EPA recently recognized the need for relief from a never-to-be
exceeded emission limitation in the case of highly variable emis-
sions when it amended the New Source Performance Standards for
the copper smelter industry published on November 1, 1977, in
the Federal Register (42 Fed. Reg, p. 57125). In the preamble to
that amendment, EPA noted that analysis showed that affected facil-
ities under the New Source Performance Standards for the copper
industry could be expected to exceed the emission limitation set
forth in the Standards a certain percentage of the time. To
reconcile this probability of excursions with the emission limi-
tation, EPA amended the Standards to allow these predicted excursions.
Deviations From EPA's Prescribed Test
Methods in the Sebree Emission Test
Caused the Fluoride Emission Measure-
ments Calculated by EPA to Understate the
Level of Fluoride Emissions
Following the promulgation of the New Source Performance Standards,
both Alcoa and Anaconda submitted to EPA test data which support
the conclusion that the 1.9 Ibs. F/TAP emission limitation set
19
-------
Mr. Don R. Goodwin
1977-12-14
Page Four
forth in the Standards cannot be achieved on a never-to-be-
exceeded basis by "affected facilities" in the primary aluminum
industry. EPA refused to consider any of this data because the
emission tests performed by Alcoa and Anaconda were not in strict
compliance with EPA testing procedures, e.g., Method 13 and
Method 14 (Goodwin Letter, pp. 1-2). The express purpose of
the Sebree Emission Test was to measure emissions from a modern
prebake aluminum reduction facility, utilizing EPA test methods,
for the purpose of re-evaluating the achievability of the fluoride
emission limitation in the New Source Performance Standards.
However, in conducting the Sebree Emission Test, EPA chose not to
adhere to the specifications of the prescribed test methods.
One of the more significant departures from EPA Method 14 involved
the number of anemometers used and the location of the anemometers.
For purposes of its test program, EPA utilized three anemometers
in the area of the Sebree potroom being tested. As EPA acknow-
ledges in the Sebree Emission Test Report (p. 27), this is one
more anemometer in that area than called for under Method 14.
The data show, and EPA acknowledges in a "Trip Report" dated June
29, 1977 (Sebree Emission Test Report, Appendix 16) that the
center anemometer consistently recorded higher exit velocities
than the two end anemometers. In the "Trip Report", EPA con-
cluded that the difference was not the result of the calibration
error found. Furthermore, a subsequent "Trip Report" included
in the Sebree Emission Test Report as Appendix 17 concludes that
magnetic fields in the potroom had little or no effect on the
signal characteristics of the anemometers. Therefore, it appears
clear that the difference in anemometer measurements can be at-
tributable to an actual velocity gradient in the roof monitor.
If the Sebree Emission Tests were conducted using only two anemom-
eters located in compliance with EPA Method 14, the air volume cal-
culation would have been significantly higher than the air volume
reported in the Sebree Emission Test Report. This higher air volume
determination would have resulted in a higher secondary emission
level value since fluoride emissions are calculated by the direct
multiplication of fluoride concentration times velocity times
monitor area.
In addition to the mislocation of the anemometers, the sampling
procedure employed in the Sebree Emission Test also had the
effect of understating the secondary fluoride emission values
measured by EPA. As explained in the Sebree Emission Test
Report, a separate sampling run was conducted for each 8-hour
shift at the Sebree Plant. However, in the case of the secon-
dary emission samples, the actual sampling time for each shift was
only 400 minutes out of a possible 480 minutes (Sebree Emission
Test Report, p. 4). In every instance, the sampling run was
20
-------
Mr. Don R. Goodwin
1977-12-14
Page Five
started at the beginning of the shift and terminated approximately
80 minutes prior to the end of each shift. EPA reasoned that
since the majority of the activity in the potrooms occur in the
first half of each shift, fluoride emissions are potentially
greater during that time (Sebree Emission Test Report, p. 4).
On the other hand, anemometer measurements of air volume were
taken continuously and, therefore, covered all 480 minutes in
the 8-hour shift. For purposes of calculating the fluoride
emission levels, EPA determined a 24-hour .average roof monitor
flow rate.
The effect of this discrepancy in sampling time is to understate
the level of fluoride emissions measured. This conclusion is
required by the following reasoning: As EPA assumes, the
majority of activity in the potrooms occurs in the first half
of each shift. By the same token, higher air volume values
are obtained when the pots are being worked during the first
half of the shift. The average air volume per minute over the
entire 480 minutes is substantially less than the average air
volume per minute over the 400 minutes when fluoride samples were
being collected. By including the air volume measurements
taken during the last 80 minutes of the shift when virtually no
work is being done, EPA has effectively reduced the air volume
factor to be used in computing emission levels. Since the fluor-
ide emission level is calculated from the direct multiplication
of concentration times velocity times monitor area, the effect
of lowering the velocity factor is to lower the fluoride emis-
s-on level measured.
The Validity of the Sebree
Emission Test is Highly
Questionable
In reporting the results of his statistical analysis of EPA's
emission test data, Dr. DeGroot felt constrained to conclude
that there were too few data points available to enable one
confidently to assess the fluoride emission behavior of
modern aluminum reduction facilities (DeGroot Report, p. 2).
Alcoa concurs in this conclusion. It is equally important to
note that the limited data collected by EPA from the Sebree
Test is so affected by deficiences in the apparatus and test
procedures employed in the emission testing that its validity
must be seriously questioned.
For example, in the Sebree Emission Test Report (p. 28 and
Appendices 2 and 3) EPA concedes that the anemometers used
21
-------
Mr. Don R. Goodwin
1977-12-14
Page Six
for the testing were not calibrated properly. Anemometer values
have direct impact upon the determination of the fluoride emis-
sion value; therefore, accurate calibration of the anemometers
is of critical importance to the accuracy of the emission test
results. It is obvious that EPA did a great deal of back-calcu-
lating in an attempt to rectify the clearly erroneous anemometer
readings. Such manipulations necessarily make the results of the
Sebree Emission Test highly suspect.
A second source of inaccuracy in the Sebree Emission Test is the
method by which EPA determined the level of production at the
Sebree Plant. Section 60.194(d) of EPA's New Source Performance
Standards Regulations expressly provides that the actual rate of
aluminum production is to be determined for the test run period.
For purposes of the Sebree Emission Test, EPA merely used a
monthly average production figure (Sebree Emission Test Report,
p. 33). Obviously, since the emission limitation set forth in
the New Source Performance Standards is stated in pounds of fluor-
ide per ton of aluminum produced, an accurate determination of
actual tonnage is essential. EPA's failure to adhere to its
own regulations in this regard brings the accuracy of the Sebree
Emission Test results into question.
The foregoing are merely two examples of instances where EPA
has failed to adhere to its own methodology for determining
fluoride emissions from an aluminum reduction facility. These
deviations from EPA methodology have a substantial impact upon
the results of the emission test. Furthermore, it is clear that
industry would have no right to employ such deviations in per-
formance tests conducted under the New Source Performance Stan-
dards.
Another way in which the conduct of these tests differed from
what the owner of a source would be permitted in a performance
test under the New Source Performance Standards is the alleged
correction of a number of practices and conditions which were
not "conducive to low emissions" (Sebree Emission Test Report,
p. 7). Replacement and repair of hoods, cleaning raw materials
from pot decks and special hood cover placement were performed
specifically for Phases II and III of the Sebree Emission Test
program; all quite beyond the possibility of any normal long-
term practice in an aluminum smelter. In addition, mechanical
floor sweeping was suspended and the floors were swept by hand
brooms, specifically to reduce fluoride collection during the
tests.
In essence, EPA required the Sebree Plant to make special changes
in equipment and operating procedures prior to Phase II in order
22
-------
Mr. Don R. Goodwin
1977-12-14
Page Seven
to obtain lower fluoride emission measurements than would have
been obtained under normal conditions.
Finally, although this comment does not pertain solely to the
Sebree Emission Test, Alcoa wishes to reiterate its contention
that EPA Method 13 consistently understates the measured level of
fluoride emissions. Attached to these Comments as Appendix B are
(1) an article entitled "Retention of Fluoride by Sintered-Glass
Filter Supports" dated May, 1977 by Mitchell and Midgett and (2)
a report entitled "Investigation of EPA Method 13 for Total Fluor-
ide Determination," dated September 29, 1977, by Jerry W. Jones
of Revere Copper and Brass, Inc. Both of these reports support
Alcoa's position in this regard.
After you have reviewed these Comments, if you have any questions
please contact me.
Sincerely,
Patrick R. Atkins, Manager
Environmental Control
PRA:smk
cc: Ronald Hausmann, Esq. - EPA
23
-------
STATISTICAL ANALYSIS OF
EPA EMISSION TEST DATA
Morris H. DeGroot, Ph.D.
December 12, 1977
24
-------
1. Introduction
The purpose of this report is to describe our analysis of the emission
test data presented in Tables I, II, III, and IV attached to this report.
Each of the four tables contains data pertaining to emissions of gaseous
and particulate fluorides from different aluminum reduction plants. For
each plant the analysis was carried out in order to estimate the proportion
of measurements that would exceed the New Source Performance Standards for
the Primary Aluminum Industry. We estimated this orooortion for each nlant
by the probability that the arithmetic mean of three separate 24-hour
emission measurements exceeded the applicable EPA standard of 1.9 Ibs/ton
for prebake reduction plants and 2 Ibs/ton for horizontal stud Soderberg
reduction plants. Several methods were used to analyze each set of data.
2. General Considerations
Before we proceed to our specific problem, we believe it is important
to note that the specification of emission limitations on a "never-to-be-
exceeded" basis necessarily contradicts the inherent random nature of emission
measurements. The stochastic nature of these measurements implies that some
proportion of measurements will exceed any given value. In particular, the
lognormal distribution, which has often been used to model the random
behavior of emission measurements, allows arbitrarily large observations.
Furthermore, the imposition of a penalty when the standard is exceeded,
no matter how small the excess, seems inappropriate when the emissions
measurements follow a continuous distribution. A nenslty which is a con-
tinuous increasing function of the excess over the standard seems more
appropriate.
25
-------
A major obstacle to our analysis of the d?ts in Tables I-IV w?s the
limited number of observations. It is difficult to estimate the behavior
of the average of three 24-hour measurements, as reauired by the EPA
standard, when only a few measurements of short duration are available.
In fact, for plant C there are only two observations on the primary
emissions, each of duration less than 7 hours. Thus, it must be kept in
mind that all the estimates derived from these data could change drastically
if more measurements were available.
Our analysis takes into account the fact that there is a 24-hour
production cycle in these plants. Virtually all of the measurements for
plants B, C, and D were for durations of less than 24 hours. We do not
know from which part of the production cycle each of these measurements
was taken, and each measurement is itself a time average. Therefore, in
some of our methods of analysis we have assumed that the variance of a
measurement is proportional to the reciorocal of its time duration. For
the Sebree plant, we utilized measurements from each of the three 8-hour
shifts, thus covering the entire production cycle.
It seems to us that the EPA standard should specifically reouire that
each test run be for the duration of the production cvcle. The standard
as" presently written of at least 8 hours for primary control system
emissions and at least 8 hours but up to 24 hours for secondary emissions
is unsatisfactory because it could potentially lead to different results
for different durations.
26
-------
We did not seriously consider the possibility that some observations
were aberrant and should be excluded from the analysis. With so few obser-
vations to begin with, we felt that we could not afford to exclude any.
3. Plant B
Plant B is a horizontal stud Soderberg-type plant for which the data
can be summarized as follows:
Total F Ibs/T ;.L
Test Time Duration Primary Secondary
18:25 1.10
12:15 2.01
4:05 0.410
4:10 0.382
10:55 2.28
3:27 0.422
7:35 2.85
These data were analyzed by four different methods.
Method I: Our first method of analysis is to assume that the three
primary measurements form a random sample from a lognormal distribution,
and that the four secondary measurements form an independent random sample
from some other lognormal distribution. The lognormal distribution is
commonly accepted in pollution data analysis.
2
For any group of numbers X.. , ..., X , we shall let X and s
denote the sample mean and sample variance, defined as follows:
-ln 2ln -?
X = - £ X. and s = - £ (X. - X)"
n . 7 i n . - i
1=1 1=1
27
-------
Some statisticians might use the divisor n-1 rather than n in the
definition of s in order to obtain an unbiased estimator of the variance
2
CT when the observations are normally distributed. Since this property is
not directly relevant to our analysis, we have chosen to use n in our
2
definition of s which will yield maximum likelihood estimates for both
normal and lognormal distributions.
We took the logarithms of the measurements in the table in order to
obtain normally distributed values. For these logarithms, we computed
2
the following values of X and s
2
X s
Primary
Secondary
-.9056 .00175 !
.6662 .12429 |
For any given values of X and s , the estimates m and v of
the mean and variance of the corresponding lognormal distribution are
1 2
m = exp (X + -T s )
2 2
v = m [exp(s ) -1] .
The appropriate values are
m v
Primary
.4047 .000286
Secondary 2.0717 .5680
The total emission for a single test run will be the sum of an
observation P from the primary source and an observation S from
the secondary source. The EPA standard specifies that the average of
three test runs shall not exceed 2 Ibs/ton. This average A can be
represented as 1
A ' 5 (?1 + Sl + P2 + S2 * P3 * V '
28
-------
The distribution of the sum of these six independent lognormal
variables is not known exactly, and so we use the Central Limit Theorem
to approximate this distribution. The distribution of A is approximately
normal with mean 2.4764 and variance .1894.
The probability that the average of three test runs will exceed the
EPA standard is Pr(A > 2). Using our approximation, we find that this
is equal to the probability that a standard normal variable exceeds -1.095.
Hence, we estimate the probability that the average of three test runs will
exceed the EPA standard to be .863.
Method II: In the application of our second method, we noted that
the average duration of the three primary measurements is 3 hours and 54
minutes. As discussed in Section 2, the variance of an observation will
be approximately proportional to the reciprocal of the test time duration.
Hence, the variance of a 24-hour primary measurement will be approximately
.0000466. Similarly, the average duration of the four secondary measurements
is 12 hours and 17 minutes. Hence, the variance of a 24-hour secondary
measurement will be approximately .2909. From these values, we find that
the distribution of A will be approximately normal with mean 2.4764 and
variance .0970. Hence, we estimate the probability that the average of
three test runs will exceed the EPA standard to be .937.
Method III: Our third method is to assume that the three primary
measurements form a random sample from a normal distribution, rather than
a lognormal distribution, and similarly that the secondary measurements
form an independent random sample from some other normal distribution.
29
-------
For these measurements, we find the following values of X and s
2
Primary
Secondary
X
.4047
2.060
s
.000281
.3992
It now follows that the distribution of A is normal with mean
2.465 and variance .1331. Hence, we estimate the probability that the
average of three test runs will exceed the EPA standard to be .899.
Method IV: Our fourth method is to assume again the primary and
secondary measurements have normal distributions, and to adjust the
variances as in Method II. In this way, we find that the distribution
of A is normal with mean 2.465 and variance .0683. Hence, we estimate
the probability that the average of three test runs will exceed the EPA
standard to be .962.
4. Plant C
Plant C is a prebake plant for which the data can be summarized
as follows:
Tot al F Ibs/T Al
Test Time Duration Primary Secondary
9T25 1-12
6^35 0.267
20C40 1-37
6:52 0.168
23:03 1.23
22:54 1.10
We applied the same four methods to these data as were used for plant B.
Method I: After taking logarithms, we compute the following values
- 2
of X and s :
30
-------
Primary
Secondary
X s
-1.552 .05366
.1826 .00762
The corresponding values of m and v are:
m
Primary I .2176 .00261
Secondary i 1.2049 .01111
The distribution of A is approximately normal with mean 1.4225
and variance .00457. For a prebake plant, the EPA standard is 1.9 Ibs/ton.
Hetice, we estimate the probability that the average of three test runs will
exceed the EPA standard to be the probability that a standard normal variable
exceeds 7.06. This probability is essentially zero.
Methods II, III, and IV yield the same result.
5. Plant D
Plant D is a prebake plant for which the data are as follows:
Total F Ibs/T AL
Test Time Duration Primary Secondary
4:15 0.612
24:00 0.8615
3:45 0.519
3:40 0.504
20:00 1.475
3:50 0.321
3:30 0.495
3:30 0.381
3:30 0.438
We applied the same four methods to these data as were used for plant 3
Method I: After taking logarithms, we compute the following values
- 2
of X and s :
31
-------
Primary
Secondary
X
-.7803
.1198
2
s
.03958
.07229
The corresponding values of m and v are:
m
Primary .4674 .00882
Secondary 1.1688 .10241
The distribution of A is approximately normal with mean 1.6362 and
variance .03708. Hence, we estimate the probability that the average of
three test runs will exceed the EPA standard of 1.9 Ibs'ton to be .085.
Method II; The average duration of the seven primary measurements
is 3 hours and 43 minutes. Hence, the variance of a 24-hour primary
measurement will be approximately .00137. Similarly., the average duration
of the two secondary measurements is 22 hours. Hence, the variance of a
24-hour secondary measurement will be approximately .09388. From these
values, we find that the distribution of A will be approximately normal
with mean 1.6362 and variance .03175. Hence, we estimate the probability
that the average of three test runs will exceed the EPA standard to be .069.
2
Method III; We find the following values of X and s directly
from
-
the observations:
_.
X
Primary
Secondary
.4671
1.1682
2
s
.00792
.09410
It now follows that the distribution of A is nornal with mean
1.6353 and variance .03401. Hence, we estimate the probability that the
average of three test runs will exceed the EPA standard to be .076.
32
-------
Method IV: By adjusting the variances as in Method II, we find that
the distribution of A is normal with mean 1.6353 and variance .02916.
Hence, we estimate the probability that the average of three test runs will
exceed the EPA standard to be .060.
6. The Sebree Reduction Plant
We note that the measurements of the primary emissions for both
8-hour and 24-hour periods have very small variation relative to the varia-
tion of the secondary emissions. For these reasons, we assumed throughout
our analysis that the primary emissions had a constant value of .033 Ibs/ton,
which is the average of the twelve 8-hour measurements and also is the average
of the four 24-hour measurements.
Some of the measurements of secondary emissions were made on two test
trains a and b . We noted that the measurements on the two trains were
not independent and that the variability between the trains was small relative
to the variability within the trains. Furthermore, there were two days.for
which measurements were reported on only one train. For these reasons we
have chosen to use the average of the measurements for the two test trains in
our analysis whenever both trains are presented. It should be remarked that
whet, measurements are made according to the performance standard only a
single value will be available for each time period. The average of two
measurements necessarily has a smaller variance than a single measurement of
the same quantity. For this reason our estimates, based on the average of
33
-------
the measurements for the two test trains, will be slightly smaller than esti-
mates based on single measurements.
By averaging the measurements made on the a and b test trains, we
obtain the following table of secondary measurements:
Total F Ibs/T Al for Sebree Plant
0:00-8:00 8:00-16:00 16:00-24:00 24-hr.
.795
.720
1.050
2.065
.970
.595
1.385
r.ooo
.700
.770
.640
1.660
2.285
1.525
.290
.915
.840
1.080
2.980
1.190
2.645
3.655
3,555
1.585
1.180
1.670
2.380
1.430
.850
1.730
2.715
1.810
.815
1.145
1.180
1.440
These data were analyzed by six different methods. In each of the methods
of analysis we have assumed that measurements made during different time peri-
ods were independent of each other. We are aware that this assumption is vi-
tiated somewhat by meteorologic and other conditions that may persist for pro-
longed periods. Any positive dependencies between these measurements would
increase their variance. For this reason our estimates, which ignore such
dependencies, will be no larger than estimates which account for positive de-
pendence of the measurements.
34
-------
Method I; Our first method of analysis is to assume that the nine
8-hour measurements for the time period 0:00-8:00 form a random sample
from & lognormal distribution, that the nine 8-hour measurements for the
period 8:00-16:00 form an independent random samole from some other log-
normal distribution, and that the nine measurements for the neriod 16:00-
24:00 form a third independent random samnle from a third lognormal dis-
tribution. After taking the logarithms of the measurements in the table,
- 2
we computed the following values of X and s ;
0:00- 8:00
8:00-16:00
16:00-24:00
X
-.04051
-.04184
.75656
2
s
.13086 1
.32734 !
.17336
The estimates m and v for the lognormal distribution corresponding
to each time period are:
m v
0:00- 8:00
8:00-16:00
16:00-24:00
1.0252
1.1296
2.3239
.14695
.49414
1.02225
The secondary emission for a single 24-hour test run will be
approximately the average of the secondary emissions from each of
the three 8-hour shifts. We are aware that the actual 24-hour measure-
ments reported in the table are not merely the averages of three
successive 8-hour measurements but are based on the formula given in
Table 1-1 of Appendix 1 of EPA Report No. 77-ALR-6. However, because
of the cyclical nature of the production process, we believe that the
average of the 8-hour measurements from the three different shifts will
provide a less variable estimate of a 24-hour emission than will a single
24-hour measurement.
35
-------
Therefore, the arithmetic mean B of three 24-hour test runs for
the secondary emissions can be reoresented as the average of nine
independent measurements, three from each of the three shifts. Thus,
B " 9 (J1 + Kl + Ll + J2 + K2 + L2 + J3 + K3 + V
where J., J^, J, are independent measurements from the lognormal dis-
tribution for the 0:00-8:00 period; K , K«, K. are independent measure-
ments from the lognonnal distribution for the 3:00-15:00 period; and
L-, L_, L- are independent measurements from the lognormal distribution
for the 16:00-24:00 period. The distribution of the sum of nine independent
lognormal variables is not known exactly, so we use the Central Limit Theorem
to approximate this distribution. Thus, the distribution of 3 is
approximately normal with mean 1.4929 and variance .06161.
The total of Che pricary and secondary emissions will be B -t- .033.
Therefore, the probability that the average of three 24-hour test runs will
exceed the EPA standard is ?r(B + ,033 > 1.9). We estimate that this proba-
bility is .066.
Method II: Our second method is to assume that each of the 3-hour
secondary measurements has a normal distribution, rather than a lognormal
- 2
distribution. We compute the following values of X and s :
0:00- 8:00
8:00-16:00
16:00-24:00
X
1.0311
1.1117
2.3156
2
s
.18351 j
.33067 |
.82126 !
36
-------
With this assumption, the distribution of 3 is normal with mean
1.4861 and variance .04946. Thus, we estimate the probability that the
average of three 24-hour test runs will exceed the EPA standard to be
Pr(3 4- .033 > 1.9^ = .043.
Method III: In this method, we use only the ninp 74-hour secondary
measurements. We assume that these measurements form a random samnle from
a log-normal distribution. After taking the logarithms of these measurements,
_ 2
we computed the values X = .3107 and s = .12845. Therefore, for the
Icrgtvormal distribution, we obtain m = 1.4549 and v = .29014.
The arithmetic mean B 'of three 24-hour test runs for the secondary
emissions can now be represented as the average of three independent measure-
ments from the lognormal distribution. By the Central Limit Theorem, this
distribution of B will be approximately normal with mean 1.4549 and variance
.09671. Thus, we estimate the probability that the average of three 24-hour
test runs will exceed the EPA standard to be Pr(B + .033 > 1.9) » .093.
Method IV; Our fourth method is to assume that each of the 24-hour
secondary measurements has a normal distribution, rather than a lognormal
JTT f\
distribution. We comoute the values X = 1.459 and s = .3045.
The distribution of the arithmetic mean 3 of three 24-hour secondary
emission measurements can now be represented as a normal distribution with
mean 1.459 and variance .1015. Thus, we estimate the probability that the
average of three 24-hour test runs will exceed the E?A standard to the
?r(3 + .033 > 1.9) = .100.
37
-------
Method V; The method is different from all the previous methods
described in this report in the sense that no specific distribution, such
as lognormal or normal, is assumed for the observations. It: is a distribution-
free procedure. As in Method I we assume that the arithmetic mean 5 of
three 24-hour test runs for the secondary emissions can be represented as
the average of nine independent measurements, three from each of the three
8-hour shifts.
From the nine values for each of the three shifts given in the first
table of this section, we computed all 592,~04 possible arithmetic means
that could be formed by choosing three values from each of the shifts
To each of these means, we then added .033 to reoresent the nrimary
emissions. With this method, we estimate the probability that the mean
of three 24-hour tests will exceed the standard to be the proportion of
these values greater than 1.9. This proportion is found to be .025.
Thus, we estioate the probability that the average of three 24-hour test
runs will exceed the EPA standard of 1.9 Ibs/ton to be .025..
Method VI: This method is a distribution-free method similar to
Method V based only on the nine 24-hour secondary emission measurements.
As in Method III, the arithmetic mean B of three 24-hour test runs can
be represented as the average of three of these nine measurements. We
computed all 84 possible means that could be formed in this way, ^dded
.033 representing the primary emissions to each of these me
-------
7. Summary
For each of Che four plants that we considered, we obtained several
different estimates of the probability that the EPA standard will be
exceeded. These estimates are summarized in the following table:
Method
Plant I II III IV
B .863 .937 .899 .962
C 0000
D .085 .069 .076 .060
Method
I II III IV V VI
Sebree .066 .043 .093 .100 .025 .095
We note that for each of the plants -3, C and D the estimates obtained
from Methods I - IV are all more-or-less of the same general magnitude.
For the Sebree plant, the estimates obtained from Methods I, II and V are
of the same general magnitude and smaller than the estimates obtained from
Methods III, IV, and VI. The first set, Methods I, II, and V, are based
on the 27 eight-hour measurements; and the second set, Methods III, IV,
and VI, are based on the 9 twenty-four hour measurements. Since the esti-
mates in the first sat are based on the averaging of a. larger number of eight
hour measurements (assumed to be independent) than the estimates in the second
set, the estimated variances for the distribution in the first set are smaller.
Therefore, it is natural that the probability estimates in the first set will
also be smaller. We have not attempted to choose among these estimates or to
combine them in any way.
Finally, as previously mentioned, these estimates are based on very
few observations and could change drastically in the light of future data.
39
-------
i_l
in CN
O
E"1
X;
CO
x:
< &^r
r3 CN
\ G
w o
£i U
rH G)
CO 5-1
EH x:
rH CO
o
E-i E-i
CO ^
W XI
E-I
rr
»3* S- rsl
*£< i*^ ^N
O r-l
H ca
co g
CO -H
M ^ Vj
2T vo (Xi M
H 1 X!
Ci
H E-i iJ co
2 <
K < 1
Q &, P"
E-i H "c
WO O
c£ 2 >r"i
(3 -P
W f3
CO -y ^
CU ?
- rH
co r o">
CM
m
*
rH
VO P** P "
r*^ r^* ^y
CN
in
m
o
^
rH CN
ro i>
in co
rH
*3" GN Cn P^ P^ P**
CTl P^ rH P~ VC rH
1"*") rH
"5- 00
" *
r~f
ro p** CJN f*^ en in
co p~ o r-» vo rH
m rH
m
fT)
o
a
^" T CN
O O O
__^ ,->, ^^ ^^
x~> O O --^ O O
O 0 0 000
o o
» CO CO CO CO
CO ' ' CO *~* '
-~ O O -^ 0 O
coo o o o
C^ f-^ « «*
VO T t< V0 *3*
CO r-i CN CO rH CN
III III
O O O O O O
o o o o o o
O CO VO O CO VO
o
CN rn
\ X.
cr\ o
CD r»
rH
in cr> in CN ^ IH
r- vo CN o o vo
rH rH rH CN
VD CO
CO VO
* *
rH
rH in CO O CN
-------
co
w
o
H
W
CO
M *-»
!T" ^D
a i
H E-I I-J
, ^ <
K ^ 1
E-i W
H O
c: s
Q
H
a
* i)
^ CH
Q -"
2
O
U
jz
<
rH
f-"^
r.
\
C/l
<-;
rH
t*
rH
4->
O
5-1
W fN
o
i-l
JC
CO
^
^J
,5
c
o
O
0
tO 5-1
CO
^
x:
T
5-1
f^
e
H
5-1
x;
CO
'c
O
H
4-)
2
3
D
^"^
O
6
H
E-i
to
G
^
C
H
a
H
OT
0
r-
co
fN
VD rH CTi
rH Lfl r^-
CM fN n
in PO cri ro o r~- fN
co co r^ co co i i i i
»
fS fH iH iH iH
*
^ O LD r-H CO ^3* O VO fH ^5* CT\ 1 f^O O O^ C^ VD O P"* r 1 rn
CN(NrOrOiHrOHiH rH »HiH iH iH
^-»* O O O < » O O *-^* O O X-N O O * » O O »***, O O - ^ O
OOO OOO OOO OOO OOO OOO OOO
CD * CD CD CD CD CD * CD
> CO CO CO « CO CO < CO CO CO CO CO CO t. CO CO CO
* * O O O *' O O °*^ O O ^"^ O O ^"^ O O *"' O O **-* O
ooo ooo ooo ooo ooo ooo ooo
o o » o o o * o o *
ii ^o 'g* ^f \C rj- VO ^" VO ^J VO TT «t V£) ^r VO
COrHfN fNCOrH CNCDrH fNCOrH CNCOrH CNCOrH fNCOrH
III III III III 111 III III
OOO OOO OOO OOO OOO OOO OOO
ooo ooo ooo ooo ooo ooo ooo
OCOO V£>OCO VOOCO VDOCO VOOCO VDOCO VOOCO
i 1 i 1 rH rH rH rH rH
.a fl a fl J2 rC A
O
r-
i
co
I
CO
I
r-
cn
I
CO
I
CO
o
-------
iH
E-<
CO
X!
rH
tu
rH
rd
-t-J
o
EH
en
H
EH
2
O
H
CO
W
* 10
«=- V»
H 1
e:
H E-" t-3
2 <
K < 1
r3 IH" t-»
CO CU r-
<
E-» w
H 0
OS 2
a
H
CO 4J
cu
* 0
-l
J=.
*f
cn CN
_j
r~i
(3
i J
4-1
O
El
)M
JS
CO
J-i
x:
>i
M^p
^^
(3 CN
T5
C
0
U
OJ
CO V4
X
CO
co ^r
rH -*
»
rH iH
o n r- o
t *
rH rH CN rH
1 .
M
JC
^r
>iCN
^
<3
E
H
S-l
cx, ^
X
CO
"c
O
ri
4J
(d
J-;
3
Cl
0
e
H
E-<
-p
en
CJ
EH
0 O -~ 0 ^ O
O O O O O O
! O C
CO CO CO » CO
o o *-* o > o
o o o o c o
t« »« o 9* O
VC ^3" ^* VO
rH CN CO CN CO rH
III III
O O O O O O
o o o o o o
» »
OO Vfl O VO O CO
rH rH
C
H
(3
H
E-i
4J
0)
O
CD
iJ
fy
D
W
o
C-i
O
rn
I
I
c
42
-------
w
^
<
a
w
u
1
H EH
H Z H
< K
K J ir.
t-3 PJ CJ
Q <
< < EH
5~i p^ ^
H K)
2
<"
O
CJ
iH
<3
4J
rH
<
EH >_
\ M
ro , *
to
3 -a
-1 c
o
fa U
(U
H y}
ra
ll
^-^
o
EH
>1
i^
f3
C
H
H
Ol
^-.
G
0
H
4J
(3
a
o
*
0)
g
H
EH
-JJ
tn
E^
GJ
JJ
rd ^
Q CN
r-
!-l CN
W rH
EH
er-
rs
\
u
o
Q
CN [*> COO
,_! rn CN iH
» t
H ^H i-H rH
r~ co
VD VO
CN rH
O O
^-* ^ . > .
x-» * > o % n ^r
in in rr CN o in
CN ro in ** *
CO * C^O CN
C*> WO CN ^O CN CN
in o o in n o
^ o o rn m rH
rH VD OO in CT> ^
rH rH rH rH
1 1 1 1 1 1
o in o m o wo
CN CN CN ^f LO rH
CN CTi rH CO O in
rH rH rH
r*i ^* m
i i i
rH CN CN rn m q-
\x \^ X^ X^ ^\ N^
LO in in in in m
CN C\ CN CJ\ CN
-------
tn
t,
o
EH
a
4J
o
a
q
o
o
CD
m
ID
oo
CS
vo
tH o
in in
o o
iH in H CD
CN c> oo ro
O O O O
M
H
H
H
^
e
<*
EH
a
Q K
PH
£H
2 1
<^
^ 2
CLi H
O
^
-------
CO
CO
CD
fc4
03
U
O
EH
(0
4J
O rH
iH O
r-i fS
CO
(N
H
K
i_H
O
<
EH
CQ
E1
2
<;
i_5
eu
e^
^
W
W
M
>
u
z
o
«.
CO
Q
j
O
2
K
K
^^
c
o
H
4-1
(0
S-l
3
Q
**
OJ
E
H
4J
W
«^ tH in o m r> in
o r-i eg m
CO fsj f^t
"I rH "^ «3« iI CO f-
cs
o o o o in o
N ^r CM CN r-t
o m cr\ co rs in
CS r-i rH iH rH r-t
o in in o in co in
o H rH ro CN in n
m co en in r- co t*-
Q OJ
r-
CN m in in vo vo r»-
r-l rH iH t I il iH tH
m
..
m in in m m
U
s
I I I I I I
o\ oo CD a\ co cr\
CN CN (N
(N CM CM
'4!
-------
Morris H. DeGroot
General Information
Date of Birth: June 8, 1931
Place of Birth: Scranton, Pennsylvania
Citizenship: United States
Marital Status: Widower, two children
Degrees
B.S., 1952 Roosevelt University
M.S., 195^ University of Chicago
Ph.D., 1958 University of Chicago
Positions Held
(l) Faculty member, Carnegie-Mellon University, 1957-present:
Professor of Mathematical Statistics, 1966-present;
Head, Department of Statistics, 1966-1972;
Associate Professor, Mathematics and Industrial Ad-
ministration, 1963-1966;
Assistant Professor of Mathematics, 1957-1963.
(2) Consultant, Western Management Science Institute, University
of California at Los Angeles, 1960-1961.
(3) Principal Investigator on research contracts and grants from
Air Force Research Division, 1961-1963; Office of Ordnance
Research, 1961-1964; National Science Foundation, 1962-present,
(4) Professor, European Institute for Advanced Studies in Manage-
ment, Brussels, 1971-
Honors and Professional Activities
Fellow, American Statistical Association
Fellow, Institute of Mathematical Statistics
Fellow, Royal Statistical Society
Visiting Lecturer Program in Statistics, 1967-1970, 1975-
Associate Editor, Journal of the American Statistical Association,
1970-1974
Book Review Editor, Journal of the American Statistical Association,
1971-1975
Committee on Health Effects of Nuclear Reactors, appointed by
Governor of Pennsylvania, 1973
Assembly of Behavioral and Social Sciences, National Research
Council, 1973-
46
-------
Associate Editor, Annals of Statistics, 1974-1975
Pittsburgh Statistician of the Year, American Statistical Association,
Pittsburgh,Chapter, 1975
Committee on National Statistics, National Research Council, 1975-
Editor, Theory and Methods, Journal of the American Statistical
Association, 1976-
Council, Institute of Mathematical Statistics, 1975-
Member of Steering Committe, Study Group on Environmental Monitoring,
National Research Council, 1975-1977.
47
-------
BIBLIOGRAPHY
Morris H. DeGroot
(l) "Efficiency of gene frequency estimates for the ABO System,"
American Journal of Human Genetics, Vol. 8 (±956), pp. 39-43.
(2) "The covariance structure of maximum likelihood gene frequency
estimates for the MNS system, " American Journal of Human
Genetics, Vol. 8 (±956), pp. 229-235-
(3) "Some aspects of the use of the sequential probability ratio
test, " Journal of the American Statistical Association,
Vol. 53 (1958), PP- 187-199 (with Jack Nadler).
(4) "Unbiased sequential estimation for binomial populations, "
Annals of Mathematical Statistics, Vol. 30 (1959), pp. 80-101.
(5) "Simplified method of estimating the MNS gene frequencies,"
Annals of Human genetics, London, Vol. 24 (1960), pp. 109-115
(with C. C. Li).
(6) "Minimax sequential tests of some composite hypotheses, "
Annals of Mathematical Statistics, Vol. 31 (19.60), pp. 1193-1200.
(7) "The essential completeness of the class of generalized se-
quential probability ratio tests," Annals of Mathematical
Statistics, Vol. 32 (196l), pp. 602-605.
(8) "Uncertainty, information and sequential experiments," Annals
of Mathematical Statistics, Vol. 33 (1962), pp. 404-419":
(9) "Stochastic models of choice behavior," Behavioral Science,
Vol. 8 (1963), pp. 47-55 (with G. M. Becker and J. Marshak).
Reprinted in Decision Making, ed. by Ward Edwards and Amos
Tversky, Penguin Books, Ltd., Baltimore, Maryland, pp. 353-378
(1967).
(10) "Some comments, on the experimental measurement of utility, "
Behavioral Science, Vol. 8 (1963), PP» 146-149.
(11) "Bayes estimation with convex loss," Annals of Mathematical
Statistics, Vol. 3^ (1963), pp. 839-846 (with M. M. RaoJ.
(12) "An experimental study of some stochastic models for wagers,"
Behavioral Science, Vol. 8 (1963), pp. 199-202 (with G. M.
Becker and J. Marschak).
(13) "Probabilities of choices among very similar objects: An
experiment to decide between two models," Behavioral Science,
Vol. 8, (1963), pp. 306-311 (with G. M. Becker and J. Marschak).
(14) "Stochastic give-and-take," Journal of Mathematical Analysis
and Applications, Vol. 7 (±9b5), PP- 489-498 I with M. M. Rao).
48
-------
Bibliography - Morris H. DeGroot
(15) "Measuring utility by a single-response sequential method,"
Behavioral Science, Vol. 9 (1964), pp. 226-232 (with G. M.
Becker and J. Marschak).
(16) "Optimal allocation of observations," Annals of the Institute
of Statistical Mathematics, Vol. 18 (1966), pp. 1.2-26.
(17) "Multidimensional information inequalities and prediction,"
Proceedings of the International Symposium on Multivariate
Analysis, ed.by P.RlKrishnaiah, Academic Press,1967,
pp. 2«7-313 (with M. M. Rao).
(18) "Correlations between similar sets of measurements,"
Biometrics, Vol. 22 (1966), pp. 781-790 (with C. C. Li).
(19) "Some problems of optimal stopping," Journal of the Royal
Statistical Society (B), Vol. 30 (196«), pp. 10S-122.
(20) "Optimal two-stage stratified sampling," Annals of Mathematical
Statistics, Vol. 40 (1969), pp. 575-582 (with N. Starr).
(21) "Bayesian analysis and duopoly theory," Journal
of Political Economy, Vol. 78, No. 5 (1970), pp. 1168-1184
(with R.MTCyert).
Reprinted wi c.h revisions in Studies in Bayesian Econometrics
and Statistics (S. E. Fienberg and A. Zellner, edsj, North
Holland, 1974, pp. 79-97.
(22) "Multiperiod decision models with alternating choice as a
solution to the duopoly problem," Quarterly Journal of
Economics, Vol.84 (1970), pp. 410-429 ^with R. M Cyert).
(23) "Relations between Pitman efficiency and Fisher information,"
Sankya, Series A, Vol. 32, Part 3 (1970) pp. 319-324 (with M.
Kagnavachari.
(2L) "Matchmaking," Annals of Mathematical Statistics, Vol. 42
No. 2 (1971) pp. 578-593 (with P. I. Feder and P. K. Goel).
(25) "Social preference orderings and majority rule," Econo-
rnetrica, Vol 40 (1972), pp. 147-157 (with 0. A. Davis and
M. J. Hinlch.)
(26} "Bayesian statistics and modern decision theory," Indus-
trial Engineering Handbook, 3rd ed. (H. B. Maynard, ed.),
McGraw-Hill (1971J, Sec. 10, pp. 98-114.
(27} "Interfirm learning and the kinked demand curve," Journal
V of Economic Theory, Vol. 3 (1971), pp. 272-287 (with R. M.
Cyertj.
(2£) "Sequential statistical decisions and optimal stopping."
Behavioral Approaches to Modern Management, Vol. I: Informa-
tion Systems and Decision Making (ed. by Walter GoldbergJ.
Gothenburg Studies in Business Administration, Goteoorg,
Sweden (1970), pp. 323-338.
49
-------
Bibliography - Morris H. DeGroot
(29) "Statistical studies of the effect of low-level radiation from
nuclear reactors on human health,""Proceedings of the Sixth
Berkeley Symposium on Mathematical Statistics and Probability,
University of California Press, (1972), Vol. VI, pp. 223-234.
(30) "An Analysis of cooperation and learning in a duopoly context,"
American Economic Review, Vol. 63 (1973), pp. 24-37 (with R. M.
Cyert).
(3D "Doing what comes naturally: Interpreting a tail area as a
posterior probability or as a likelihood ratio," Journal of the
American Statistical Association, Vol. 68 (1973), pp. 966-969.
(32) "Reaching a consensus," Journal of the American Statistical
Association, Vol 69 (1974), pp. 118-121.
(33) "Rational expectations and Bayesian analysis," Journal of Polit-
ical Economy, .Vol. 82 (1974), pp. 521-536 (with R. M. Cyert).
(34) "Adaptive Utility," Adaptive Economic Models (R. H. Day
and T. Groves, eds.), Academic Press, 1975, pp. 223-246
(with R. M. Cyert) .
(35) "A mean squared error approach to optimal design theory,"
Proceedings of the 1976 Conference on Information: Sciences
and Systems, The Johns^Hopkins University, 1976, pp. 217-
221 (with G. T. Duncan).
(36) "Sequential strategies in dual control problems," Theory
and Decision, Vol. 8 (1977), pp. 173-192 (with R. M. Cyert).
(37) "Statistics" to appear in Encyclopedia of Physics (R. G.
Lerner and G. L. Trigg, eds . ) , Dowden, Hutchinson, and
Ross, Inc. (\rJ\ik "PO.U/
(38) "The matching problem for multivariate normal data" to
appear in Sankhya (with P. K. Goel) .
(39) "Optimal peremptory challenges in trials by juries: a
bilateral sequential process," to appear in Operations
Research (with A. Roth and J. B. Kadane) .
(40) "Sequential investment decisions with Bayesian learning,"
to appear in Management Science (with R. M. Cyert and C. A.
Holt).
Books:
Optimal Ste£istical Decisions, McGraxtf-Hill Book Co., New York,
1970.Russian translation,Mir Publishers, Moscow, 1974.
Probability and Statistics, Addison-Wesley Publishing Co.,
Reading, MA, 1975.
50
-------
; RETENTION OF FLUORIDE BY S1NTERED-GLASS FILTER SUPPORTS-
By William J. Mitchell and M. Rodney Midgett
Quality Assurance Branch
Environmental Monitoring and Support Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park. North Carolina 27711
May 1977
SECTION 1
INTRODUCTION
In May 1975, the Quality Assurance Branch (QAB) of the Environmental Monitoring and Support Laboratory of EPA
at Research Triangle Park. North Carolina, sampled for fluoride at a primary aluminum reduction plantfThe purpose of
the QAB testing was to establish the collection efficiency for fluoride of the EPA Method 13 sampling train when the
Whatman No. 1 filter is placed between the probe and the impingers [1] -
The results (2] of these tests indicated that the sintered-glass filter support that is commonly used in source sampling
trains retained some of the fluoride that had passed through the filter. However, the experimental test design did not
allow us to elucidate the magnitude of this retention phenomenon for a single sampling run. So, we conducted further
studies at a primary aluminum plant to determine the magnitude of the fluoride retention phenomenon. In this study,
the four train sampling arrangement previously developed and field tested by QAB was used. This sampling arrangement
allowed four independent sampling trains to sample simultaneously and isokinetically at essentially the same point in
the stack. Thus, during a sampling run all the trains sampled essentially the same pollutant concentration [3].
Six sampling runs were accomplished using the four train sampling arrangement. In each run two of the trains were
assembled with the Whatman No. 1 filter located between the probe and the first impingar and the other two trains
were assembled with the filter located between the third and fourth impingers. Single-point, isokinetic sampling was
dona in all runs, that is, no attempt was made to traverse the stack.
Table 1 summarizes the experimental design of the test, including the location of the filters in each sampling train. For
all six runs, sample train integrity was maintained by taking care that the probes, impingers and control console v/ere not
interchanged among trains. Prior to the test, the calibration factors for each control console were determined using a
spirometer.
Table 1. Sampling Schema.
Sampling
Run No.
1
2
3
4
5
6
Filter
Support
fritted glass
fritted glass
fritted glass
fritted gloss
wire screen
wire screen
Train 1
B
B
f
f
F
B
Sampling Train
Train 2
8
B
F
F
F
B
Filter Location3
Train 3
F
F
B
B
B
F
Train 4
F
F
B
B
B
F
a"F" indicates the fit cer was locaced between tha probe and the first tmpinger.
"8" indicates the filter was located between the third and the fourth impingers.
The Whatman filters were supported on sintered-glass filter supports, except in the last two runs, when they were sup-
ported on a 20 mesh stainless steel screen designed by QAB and constructed by Scott Environmental Technology.
Because the purpose was to determine the magnitude of the retention phenomenon that could occur in a sampling run,
a sintered-olass filter support was replaced after it had been used in a sampling run. The filters were not heated in any of
these six runs, because condensation was not a problem.
51
-------
SECTION 2
SAMPLING RESULTS
The fluoride concentrations in milligrams fluoride per dry standard cubic meter (mg F/scm) determined in each run are
presented in Table 2. The means for the two filter positions in each run were subjected to a comparative statistical analy-
sis to determine if they were statistically different at the 90 percent confidence level. This statistical analysis showed
that the means were not statistically different for those runs in which the Whatman filter was supported by the screen.
However, the two means ware statistically different for those runs in which the Whatman fitters were supported on the
Wintered glass frit. Further, when the percent difference between the two means in a run is calculated, one observes that
the amount of fluoride retained by the frit varies from run-to-run. That is, for runs 1 through 4, the mean values obtained
for the filter in front location were lower than those obtained^for the filter in back location as follows: run 1,22%;
run 2. 32%; run 3, 4%; and run 4,15%. It should be noted that attempts to quantitatively recover the fluoride retained
by the sintered-glass frit using a sodium hydroxide extraction were not successful.
Table 2. Sampling Results in mg F/Dry sent.
Sampling
Run No.
1
2
3
4
5
6
Trains with Filter
in Front of Impingars
0.233
0.148
0.164
0.154
0.178
0.191
0.185
0.120
:- 0.2003
-'>"" 0.183
- - 0.154
. . -11 '- 0.194
Trains with Filter
Behind Impingers
0.283
0.155
0.218
0.184
0.166
0.196
/, 'I'L n-2S2
' ' ' ', 0.237
.'/» 0.161
. ;' 0.212
,'-' 0.174
. ; *Z 0.188
aFiltar was dropped on floor during sampla recovery
SECTION 3
CONCLUSIONS
The actual mechanism by which the fluoride is retained by the sintered-glass filter support is unknown. Two possible
mechanisms that would explain this retention are: (1) small particles containing fluoride passed through theWhatmart
No. 1 filter (50% retention for 0.3 micron DOP particles at a 5 cm per second face velocity) and were physically
filtered by the sintered-glass frit; and (2) the fluoride reacted with the glass in the filter support. However, regardless
of the mechanism that caused the retention, it is apparent that a sintered-glass frit should not be placed in front of the
impingers when sampling for fluoride. In addition, since the retention might be caused by a physical filtaring mechanism,
sintered-metal frits are also of questionable value when sampling for fluoride.
Some preliminary results from an on-going study in this laboratory indicate that a similar retention phenomenon may
occur when sampling for arsenic using glass fiber filters [4]. This observation, in conjunction with the fluoride reten-
tion phenomenon, indicates that retention of sampled materials by filter supports might be a significant cause of
sampling error in many sampling methods. Although stainless steel screen filter supports were found to work well in
the fluoride study, the durability and suitability of these screens under other sampling conditions has not been
established.
Further, a sintered-glass filter support is frequently used to sample many different sources over a short period of time.
It could happen therefore, that under some sampling conditions (e.g. condensibles by EPA Method 5) materials previ-
ously retained by the frit at one source might be released at another source. For example, material collected at a source
where the filter was not heated, might be thermally released at the next source if the filter was heated.
52
-------
\
SECTION 4
REFERENCES
1. Federal Register. 41, 3825-3830, January 26, 1976. ,
2. W. J. Mitchell and M. R. Midgett, "Adequacy of Sampling Trains and Analytical Procedures Used for Fluoride.'
Atmospheric Environment. 10. 865-872 (1976).
3. W. J. Mitchell and M. R. Midgett, Environmental Science and Technology, 10, 85-88 (1976).
4. J. E. Knoll, unpublished.
53
-------
nvestigation of EPA Method 13
Total Fluoride Determination
by
Jerry W. Jones
-September 29, 1977
^INTRODUCTION
has been rouch controversy as to the accuracy and precision of Method 13
^'Determination of Total Fluoride Emissions From Stationary Sources" since its
introduction by EPA in the October 23, 1974 Federal Register. At the request
-of the Analytical Task Force of the Aluminum Association, the Environmental
--Control Department of Revere Copper & Brass, Inc., Reduction Division,
sScottsboro, Alabama agreed to participate in an evaluation of Method 13.
evaluation is concerned with accuracy of the method or, more specifically,
collection efficiency of the method. The sources tested were (1) cross
---flow-packed bed wet scrubber outlets (potroons), and (2) electrostatic preci-
pitator outlet (carbon bake plant). The tests were performed from mid-July 1977
-1977 to mid -Sept ember 1977.
.-^DESCRIPTION OF TEST PROGRAM
-cSanrple Train Configuration - The EPA Method 13 sampling train (Figure 1)
-with filter holders in both the regular and optional locations was used,
.5The first filter holder (A) contained a Whatnan #1 filter paper (Filter
-sas required by Method 13. The second filter holder (B) contained a Geliaan
-0.8 y membrane filter (Filter #2) treated with Citric Acid plus a back-up
-^alkali impregnated Ivhatman *4 paper .(Filter #3). The second filter holder
-^theoretically traps any fluoride escaping the Method 13 sample with the
vescaping particulates captured on the membrane filter (^2) and the gaseous
fluoride captured by the alkali impregnated paper (*3).
^-Sampling - The sampling was performed according to EPA Methods 1, 2, 3 and
13 except the minir.un sample volume was not achieved on five of the tests
-^sdue to the pressure drop from the second filter holder. The sampling
-"duration was eight hours each on four potroom scrubber samples and four
each on three electrostatic precipitator stack samples.
'^-Analysis - The Method 13 portion of -the samples (Filter #1 and impinger
-"solution) was analyzed according to the procedure except the actual analysis
-'twas made on the Technicon Auto Analyzer. The fluorides on filters **2 and
- -$3 were determined separately by the Auto Analyzer.
r-^Af ter evaluating the results, of the first two tests the logical conclusion
>**ras field contamination of filters #2 and #3. However, a field blank used
-vith tests 3, 4, 5 and 6 disproved that theory. The average lab blank for
filter "2 was 0.44 PIR F with a range of 0.2 to 0.8 np. F. The averapc field
blank was 0.54 mg F with no value exceeding the maximum Jab blank. The filter
*3 average for the lab blank was 0.06 mj; F with a range from 0.04 to 0.12 mp, F.
The field blanks averaged 0.07 mg F and did not exceed the maximum lab blank.
54
-------
.September 29, 1977 Page 2
-SU EJECT; Investigation of EPA Method 13 For Total Fluoride Determination
NUMMARY OF RESULTS AND CONCLUSION
The average wet scrubber total fluoride escaping the Method'13 train was
8.4% while the average gaseous fluoride lost (Filter #3) was only 0.6%.
This indicates that most of the escaping fluoride is particulate. Ander-
-son impactor particle size tests have shown better than 80% of the particles
-are less than ly in diameter. The Whatman 81 Filter would be a poor col-
lector of these small particles and the collection efficiency of the im-
rpingers would not be good at the low sampling flow rate.
The uverage electrostatic precipitator total fluoride escaping Method 13
was 0.3%. Results from prior tesl3 indicate most of the fluoride is in
"the gaseous form and is captured efficiently in the impingers.
From the test data (Attachment A), it is concluded that Method 13 is an
-adequate method for gaseous fluorides and a biased method for particulates.
-JBowever, it should be noted that the bias is on the low side and in favor
of the regulated industries.
55 -
-------
SUGGESTED EXPERIMENTAL DESIGN
FOR THE INVESTIGATION OF EPA METHOD 13
Sample Train Configuration: Use EPA Method 13 except an extra filter holder
Is" placed between the third and fourth impingers. The filter holder which
contains the IvTiatman #1 filter is placed in front of the first impinger and
is designated filter #1 in the table below. The filter holder which is
placed between the third and fourth impingers is designated filter #2 in
the table. below.
Sample Set No.
I
Filter 91
Filter
III
Whatman #1 with 20-mesh
stainless steel filter
support . . ,
Whatman #1 with fritted
glass filter support
0.8 um membrane filter
with 20-mesh stainless
steel filter support
0.8 urn membrane filter
treated with citric acid
plus alkali-impregnated
back-up filter (Whatman
84 or similar)
Same as I above
Same as I above
2. Sources To Be Sampled: Rcof vents, dry scrubber outlets, wet electrostatic
precipitator outletSi t ..-
3. Analysis: Each filter is to be analyzed separately by fusion, distillation
. and selective ion electrode determination. Impinger coni.cnts to be analyzed
by SIE. The analysis of the membrane filter which is placed in the *2 posi-
tion will indicate whether particulate fluoride is passing through the train
and the analysis of the alkali-impregnated filter will give an indication of
the amount of gaseous fluoride which might .be escaping the train.
4. Number of Samples: It is suggested that each, sample set consist of at least .
3 samples for each type of source sampled. A single Method 13 sampling
train can be used since there is no need for simultaneous sampling. The
objective is to determine how much, if any, particulate and gaseous fluorides
are passing through these three variations of.the Method 13 train.
NOTE: It might also be desirable to-conduct simultaneous sampling with sampling
train configurations II and III. This would give an indication of how much fluo-
ride is lost in the fritted glass filter support..
56
-------
r-l
(!
O
r:
i,
'«<
f-1
P
p.
lu rJ to
O r-n
rt uj TJ
o . o
o o .c:
'O -H 1)
ro
*H ^~"\
ri M
O R
UH tO
at
M
O
U +J
H r-l
+-» -H
M U.
rt
0,
CO
0
"X3 M
r-l 0)
O r 1
3 rj ^ >
r-< III ?)0
i1-* EJ
f-4 v~-'
r-( 0)
rt 4->
4-> r-<
0 -H
( U.
i C
e o
t3 bo >t .1-1
HE +->
JH '-^ uy 3
O r-l
3 tO r-<' O
r-H r-l :ic tO
P-
*T3 r-* £-*
rj ,c: 4J c/
M -M r-4 C
O O -r-4 -H
t- .:. U. d
b to
<-4 Q
(i. v '
n o
U^ o
O fn
O r X
S) M U
ij .M ^
H ' ' p
O 1^ ^^
> rt
Q "
^r
O
O
Di
O
(X
^
c^
03
" 1
Ci
CJ
to
f_
UJ
^c
,
"-"
b
,-j
a.
en
-/ J
o
a:
0
^f
o*
OO
o
o
0
t^
o
1 (
*
CO
f*v
0
in
.
P*,
G\
in
ci
f^
o
.
(N|
CO
cs
CO ..
.
"tf*
OO
o
o
o
00
o
OO
CO
*
o
in
»
f""**
r-4
O^
?o
o
in
rl
.
vO
CO
... .«<* . .
»
to
rH
in
o
*
o
o
in
r^
in
in
-t
0
o
r-l
en
to
o
o
*
vO
00
-
.*-
*-""''--
- -
£
*
f^
r-4
t-4
'eo
*«f
O
o
tN
in
rH
»o
"*\ - ^
O
co
r-4
O
o
*">
o
O.J
f)
<
C3
r-
O^
o
^H
-
vO
to
CD
0)
o
t^-»
CO
/
in
0
_
rH
«
o
0
r*
O
*
i 4
in
.
O
t^
-
C7i
O
o
t-o
o
DO
'-/"J
» <
^
to
to
c
E
o
O
to
X)
in
3
rX
C
o
i-H -H
-C 4->
O
*e o
r-i f-f
H O
D3
03-
r-l Jc
O tO C
r-l C 3
CU 3 Oi
e
c o
O r-l
to
^-< OO !
OO O O*O <
tu
c:0 oo oc
W
O
DO W
£ ^_3
O
C* ci
O 4-»
10
tn j3
rt D
3 w
^A in
10
rt
O
H
O
O)
<-< o
-Q rX O
c
r-l 05 ^4
-it r-l f4
XJ rt
r4 r-l
O J3 VO 43
+J cj
H "v x:
d, <; LO t-,
57
-------
_
\\ p/ \ / : / -, .
M V a __ / ft_Lr-n^ V -f*^ *~-»-m * ^ fc * wn
COPPER AND 3SASG
I'.O. ..,.: ,i. .
SCOTTSIMtiU*. .'.IA. ,'j'e-.
Filter Holder A equipped with Filter #1 (Whatman #!)
Filter Holder B equipped with Filter #2 (0.8w membrane)
and Filter #3 (Whatman #4)
HEATED AREA
PROQE
FILTER HOLDER A
\ / Filter Holder
THERMOMETER
ler/
PlTOTTU
c-TYPH V^
TU3£ Jx V£UOC|
x>* PRESSU
IMPINGEHS ict OATH
FINS COMTROL VALVE
PART1CULATH SAiV.PUMG TRAIN
,CHCCK VALVE
VACUUM LINb
Yigure 1
58
-------
APPENDIX B
59
-------
APR27
Or. Patrick R. Atkins, Manager
Environmental Control
Aluminum Company of America
Alcoa Building
Pittsburgh, Pennsylvania 15219
Dear Dr. Atkins:
In the past few months most of the communication between EPA and
Alcoa has been through attorneys. However, we have not responded to
several of the points raised 1n your December 14, 1977, letter, for
which we feel responses for the record must be made. The purpose of
this letter 1s to respond to your comments regarding the Sebree
emission tast (EPA Emission Test Report No. 77-ALR-6).
We agree that the Sebree test demonstrated the variability of
secondary emissions from one 24-hour period to another 24-hour period.
Variability in individual test runs is precisely why compliance with
all standards of performance is based on the average of three runs.
We do not find it remarkable that the spread of the individual eight-
hour secondary emissions measurements Is more pronounced than the
spread of the twenty-four-hour runs, since, as you well know, different
operations occur durinq each shift. Thus, it is incorrect to compare
the 0.29 Ib F/TAP which occurred during a tapping shift with the 3.75
Ib F/TAP which occurred during a carbon setting shift. Hone of the
eight-hour measurements could be used alone in a compliance test, since
they are not representative operating conditions. The nlnlmum accept-
able run time must be representative of all plant operations.
We believe that operation and maintenance can contribute to the
variability of secondary emissions; for this reason, we insisted that
this parameter be minimized for Phases II and III. Your statement
that the Sebree Emission Test data clearly indicate that there are
measurable sources of variation intrinsic to the fluoride sampling
methods themselves must be placed in perspective, since the same may
be said of any test method. The test results of Phases I and II show
clearly that the precision of the fluoride method is within the
acceptable limits for stack sampling methods 1n general. The emissions
obtained from 2 trains sampling simultaneously varied from their
average less than +3 percent for 6 of the 7 runs, and less than +5
percent for the 7th run.
60
-------
The statistical analysis of the data performed by Dr. DeGroot 1s
1n substantial agreement with our analysis; I.e., the Sebree data
would Indicate that there Is at least a 90 percent chance for a
properly controlled and operated aluminum reduction plant to pass a
performance test. He also agree with Dr. DeSroot that the necessarily
limited number of data points makes statistical Inference risky.
Your Interpretation of our actions regarding copper smelters Is
Incorrect. We did not amend the standard of performance, compliance
with which 1s determined by the average of three six-hour runs, to
provide "relief froa a never-to-be exceeded standard." The standard
of performance for copper siwlters requires the Installation and
operation of a continuous monitor to measure SO- emissions. The
standard also requires the owner or operator to report periods of
excess e»1ss1ons. There 1s no comparable requirement for primary
aluminum plants because there 1s no continuous monitoring system
available for seasurlng total fluoride emissions. The amendment we
made to the standard for copper smelters merely changed the reporting
requirement for excess emissions as measured by the continuous
monitor. We did not amend the level of the standard and the require-
ments that apply to a copper smelter during a performance test.
You contend that "... deviations from EPA's prescribed test
methods in the Sebree emission test caused the fluoride emission
measurements calculated by EPA to understate the level of fluoride
emissions." Two specific examples were given: (1) three propeller
anemometers were used instead of two; and (2) the velocity and
fluoride concentration 1n each eight-hour test were measured for
different lengths of time. Before answering the above objections,
Alcoa's statement should be clarified by adding the word "true"
before "level," i.e., ". . . to understate the "true" level of
fluoride emissions."
The use of three anemometers exceeds the requirement of Method 14,
which specified that only two are needed for a roof vent of this size.
(This type of variation from the reference test method 1s allowed
under the provisions of section 60.8 of 40 CFR Part 60.) Therefore,
the accuracy of the reported total flow rates should be equal to or
greater than the accuracy that would have been obtained by following
the Method 14 requirement exactly. Alcoa's contention that signifi-
cantly higher emission rates would have resulted if two anemometers
had been used is purely speculative, since no comparisons were trade
between the use of two and three anemometers. The Sebree data show
that a substantial velocity gradient existed 1n the roof monitor;
because of this (and in the absence of data to Indicate otherwise),
the use of three anemometers Instead of two must be considered more
representative of the actual conditions 1n the monitor.
61
-------
During the Sebree tests, velocity was rnonitored for the entire
480 minutes of each 8-hour test period; therefore, the calculated
volumetric flow rates are considered to be representative. Fluoride
concentration, on the other hand, was measured only for the first 400
minutes of each 8-hour period. Since the majority of the activity 1n
the potrooms occurs 1n the first half of each shift, fluoride emissions
tend to be higher during this period. Consequently, measuring the
fluoride concentration for only 400 of the 480 minutes would tend to
bias the results toward the high side. Therefore, an emission rate
calculated fro* the product of an average fluoride concentration
measured over the first 400 minutes t1n*s an average volumetric flow
rate measured over the entire 480 minutes would tend to be higher
than the true rate, rather than "understating the secondary fluoride
emission values," as Alcoa contends.
Alcoa contends that the validity of the Sebree emission test
results Is highly questionable because of Improper Initial calibration
of the anejmwteters and backcalculatlon 1n order to rectify the
erroneous anemometer readings. It 1s true that the velocity data from.
the Sebree tests were adjusted after-the-fact based upon Information
obtained from the anetnoraeter manufacturer and from a series of post-test
anemometer calibrations. However, these adjustments were made 1n order
to Increase the accuracy of the reported emission rates. (It should
be noted that 1n general, the adjustcwnts made 1n the Sebree data
tended to Increase the total flew rates and emission rates.) Alcoa's
contention that Industry would have no right to n»ke similar adjust-
ments 1n NSPS performance test data Is unfounded. Section 60.3 provides
authority for the Administrator to approve minor deviations from the
reference methods; If 1t can be shown that more representative data
will result, such deviations would be approved, "ote that as a result
of EPA's experience with the anemometers, Method 14 is currently being
revised to Include the proper calibration procedure for anemometers.
We disagree that the raethod of determining the level of production
Introduces a source of error. We considered using actual metal tapped
1n the calculations but decided against this 1n favor of using the
tnonthly average per pot production rate, because the airount of metal
tapped on any given day 1s not necessarily the amount of ratal produced
on that day. For example, it would be possible to not tap any pots 1n
which case netal tapped would be zero, but the pots would stm be
producing aluminum and evolving fluorides. Of course 1t would be to
the advantage of the source to tap a laroer amount of rcetal during a
compliance test; we understand there 1s seme discretion 1n the amount
of metal tapped per day and that 1t would be possible to "undertap"
on the days preceding a test and "overtap" during the test. We thus
felt that the monthly average gave a better Indication of metal produc-
tion than amount tapped during the test. !?e Interpret 40 CFR
6Q.194(d)(l) to allow our approach, and we would allow an owner or
operator to use the monthly per pot average In calculating emissions
62
-------
during a compliance test. Ife are considering a change to 40 CFR
60.194(d)(l) to make this an explicit provision.
Good operation and maintenance of the air pollution control system
1s required under 40 CFR 60.11 at all times. Thus not only would we
permit an owner or operator to have hoods in good repair and properly
placed and to have raw materials cleaned from the pot deck, but we
would require such actions during a performance test, as well as at
all other tiroes. We do not believe that such practices are beyond the
possibility of any normal long-term smelter operation, and since there
are no continuous monitoring requirements, observation of poor hood
placement, bent hoods, sloppy material handling, etc. would be grounds
for an inspector to require a performance test for an affected facility.
The only reason for requiring hand sweeping was that we observed
a large an»unt of dust generated from the mechanical sweeper, and
hand sweeping was the most expedient solution to temporarily minimize
this dust. We would expect that a properly designed and functioning
mechanical sweeper could be used that would not generate the dust that
was observed at Sabree.
You cited two reports to support your position that EPA Method 13
consistently understates the measured level of fluoride emissions.
The first report cited 1s entitled "Retention of Fluoride by Sintered-
Glass Filter Supports," dated Hay, 1977, by Mitchell'and Midgett. This
report has no bearing on the results of the Sebree tests because at
Sebree, the filter was placed behind the ircpingers (as allowed by
Method 13) rather than in front of them. It should be noted that even
if the filter had been inserted between the probe and first ircplnger,
a stainless steel screen filter support would have been used instead
of a glass frit (see the amendments to Reference "'sthods 13A and 13B
in the November 29, 1976, Federal Reg1star).
The second report cited is entitled "Investigation of EPA Method
13 for Total Fluoride Determination," dated September 29, 1977, by
Jarry VJ. Jones of Revere Copper and Brass, Inc. This report is not
applicable to the Sebree tests for the sane reason that the first
report does not apply. In addition, the findings of this study have
no effect on the data base used to set the fluoride emission standard,
because more efficient fiber glass mat filters (not paper filters, as
used by Mr. Jones) were employed in the gathering data for the
standard.
It should be noted, however, that the results obtained by
Mr. Jones raise questions about the efficiency of the fluoride train
in streams havinn significant particulate loadings. From the data
provided, it is not clear what the total irroact of the 13.4 percent
maximum inefficiency is on the total emissions, since the majority of
63
-------
the emissions are from the roof monitors and the level of emissions
from the wet scrubber is about 10 times lower than from the ESP.
Since a significant level of inefficiency is indicated, it is EPA's
intention to revise the reference method disallowing the use of the
paper filter in the front and recoranending use of a more efficient
glass fiber or similar filter.
As you have been informed through counsel, we are evaluating
several alternatives for amending the standard which address the
issue of variability. We look forward to your connents on our
proposal.
Sincerely yours,
Don R. Goodwin
Director
Emission Standards and
Engineering Division
Enclosure
cc: Ron Hausraann
Donald C. Winson
64
-------
Enclosure
Title 40Protection of Environment
CHAPTER IENVIRONMENTAL
PROTECTION AGENCY
(FRL 639-2]
PART 60 STANDARDS OF PERFORM-
ANCE FOR NEW STATIONARY SOURCES
Amendments to Reference Methods 13A
and,13U
On August 8. 1975 (40 FR 33151), the
Environmental Protection Agency (EPA)
Promulgated Reference Methods 13A and
13B In Appendix A to 40 CPR Part 60.
Methods 13A and 13B prescribe testing
and analysis procedures for fluoride
emissions from stationary sources. After
promulKation of the methods. EPA con-
tinued In eviilmilo Ilicni and us u rr:,ult
has determined the need for certain
amendments to improve the accuracy
and precision of the methods.
Methods 13A and 13 B require assembly
of the fluoride sampling train so that
the filter is located either between the
third and fourth impingers or in an
optional location between the probe and
first impinger. They also specify that a
fritted glass disc be used to support the
filter. Since promulgation of the meth-
ods, EPA has found that when a glass
frit filter support is used in the optional
filter location, some of the fluoride
sample is retained on the glass. Although
no tests have been performed, it is be-
lieved that fluoride retention may also
occur if a sintered metal frit filter sup-
port is used. However, in tests performed
using a 20 mesh stainless steel screen
as a filter support no fluoride retention
\MIS noted. Then-fore, to eliminate the
possibility of fluoride retention, sections
5 1.5 and 7.1 3 of Methods 13A and 13B
are being revised to require the use of
a 20 mesh stainless steel screen filter
support if the filter is located between
the probe and first implnger. If the filter
Is located In the normal position between
the third and fourth impingers, the glass
frit filter support may still be used.
In addition to the changes to sections
5.1.5 and 7.1.3, a few corrections are also
being made. The amendments promul-
gated herein are effective on November
29. 1976. EPA finds that good cause exists
for not publishing this action as a notice
of proposed rulemaking and for making
it effective Immediately upon publication
because:
1. The action is intended to improve
the accuracy and precision of Methods
13A and 13B and does not alter the
overall substantive content of the meth-
ods or the stringency of standards of
performance for fluoride emissions.
2. The amended methods may be used
immediately in source testing for fluoride
emissions.
Dated: November 17,1976.
JOHN QUARLES,
Acting Administrator.
In Part 60 of Chapter I, Title -*0 of the
Code of Federal Regulations, Appendix
A is amended as follows:
1. Reference Method 13A is amended
as follows:
(a) In section 3., the phrase "300
/ig/liter" is corrected to read "300 mg/
liter" and the parenthetical phrane " (see
section 7.3.6)" is corrected to read "(see
section 7.3.4)".
(b) Section 5.1.5 is revised to read as
follows:
5 l 6 Filter holderIf located between the
probe and first Impinger. borosllicate glass
with a 20 mesh stainless steel screen filter
support and a slllcone rubber gasket; neither
a glass frit filter support nor a sintered metal
filter support may be used If the filter Is In
front of the Impingers. If located between
the third and fourth impingers, bo'.-wlllcate
glass with a glass frit filter supporr and a
.'lilcone rubber gasket. Other materials of
ruiiHlriiftlnii may be UHed with approval from
the Administrator, e g.. If probr liner !> '.tivln-
leas steel, then filter holder may be .Inlets
steel. The hoMer design t>h«ll provide a pu.si-
tlve seal against leakage from the: outside or
around the filter.
(c) Section 7.1.3 is amended by re-
vising the first two sentences of the sixth
paragraph to read as follows:
7.1.3 Preparation of collection train.
Assemble the train as shown In Figure
13A-1 with the filter between the third and
fourth Impingers. Alternatively, t>m filter
may be placed between the probe and first
Implnger If a 20 mesh stainless steel screen
Is used for the filter support. *
* *
(d) In section 7.3.4, the reference in
the first paragraph to "section 736" is
corrected to read "section 7.3.5".
2. Reference Method 13B is amended
as follows:
(a) In the third line of section 3. the
phrase "300..g liter" k> corrected to read
"300 mg, liter".
(b) Section 5.1.5 is revised to read as
follows:
5.1.5 Kilter holderaf located net*,, n the
probe und fln.t, implnger, boronlllciue n\&b-i
with a 20 me li .stalnlcsb !>teel arrof" inter
support and a wllootie rubber gasne' i Ither
a glass frit filter support nor a sinteieu metal
filler support may be used If the filter Is In
front of the Impingers. It located -ween
the third and fourth Impingers, bui ^.ilcate
glass with n glass frit (liter support and a
silicone rubber gasket. Other mav>--',il.s ol
construction may be used with appr-> . irom
the Administrator, e.g, if probe liner : ^tain-
less steel, then filter hold, r may be stainless
- steel. The holder design shall provide a posi-
tive seal against leakage from the out;,1 le or
around the filter.
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO. 2.
EPA-450/2-78-025a
4. TITLE AND SUBTITLE
Background Information for Proposed Amendments to the
New Source Performance Standard For the Primary
Aluminum Industry
7. AUTHOH(S)
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Standards Development Branch
Emission Standards and Engineering Division
Research Triangle Park, N. C. 27711
12. SPONSORING AGENCY NAME AND ADORES5
DAA far Air Quality Planning and Standards
Office of .Air, Noise, and Radiation
U.S. Environmental Protection Agency
Rpcparrh Tyisnnlp Psrlc M T ?7711
3. RECIPIENT'S ACCESSION-NO.
5. REPORT DATE
August 1978
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
13. TYPE OP REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
EPA/200/04
15. SUPPLEMENTARY NORTHS
16. ABSTRACT
This document supplements information contained in the preamble to proposed amendments
for the new source performance standard for the primary- aluminum industry. The-
document contains additional information on the emission test results, the statistical
treatment of the test results, the costs, and the environmental impact of the proposed
amendments. Also, the document addresses other issues raised by litigants.:
17. KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
Air pollution
Pollution control
Standards of performance
Primary aluminum plants
13. DISTRIBUTION STATEMENT
Unlimited
b.lOENTIFIERS/OPEN ENDED TERMS
Air pollution control
19. SECURITY CLASS (Tin's Report)
llnr 1a<;<;if ipri
20. SECURITY CLASS (This page)
Unclassified
c. COSATI Field/Group
21. NO. OF PAGES
^
22. PRICE
EPA Form 2220-1 (9-73)
-------
50) O
$*-
a < .
3 Tffi
00&
------- |